TerraMosaic Daily Digest: Feb 10, 2026
Daily Summary
Across 179 selected papers, the clearest shift is from static hazard mapping to process-resolved, testable prediction. Studies combine dense observations (TS/MT-InSAR, microseismic monitoring, UAV and optical products) with interpretable models to track evolving hazard states, from rapid urban subsidence in Guangzhou (up to 283.7 mm/yr) to activity-conditioned landslide susceptibility in the Three Gorges Reservoir.
At the same time, mechanistic work strengthens the physical basis of risk estimates: multiphase models for submarine landslide-pipeline impact and boulder-rich debris flow, coupled hydro-mechanical simulations for CO2-injection fault reactivation, and laboratory-to-field constraints on liquefaction, rockburst, and water-sand gushing. The practical gain is not model complexity alone, but clearer failure thresholds, uncertainty bounds, and infrastructure-relevant decision metrics.
Key Trends
- Dynamic hazard states replace static susceptibility maps: MT-InSAR/TS-InSAR products are increasingly fused with interpretable models (e.g., SHAP-RF) to update hazard levels in time, rather than issuing one-off susceptibility classes.
- Coupled flow-structure physics is moving into operational design: Flume-constrained and multiphase simulations now quantify impact loads and runout behavior for submarine landslides, debris flows, and pipeline exposure.
- Underground infrastructure monitoring is becoming quantitative and predictive: Digital twins, microseismic warning indices, and coupled seismic-fault tunnel solutions are translating monitoring data into actionable stability thresholds.
- Laboratory micromechanics is tightening constitutive assumptions: New evidence on liquefaction resistance, re-liquefaction, rockburst triggering, and confined aquifer gushing is improving parameterization of failure models.
- Trigger rates matter as much as trigger magnitude: Studies on rapid forcing (e.g., AMOC sensitivity) and human-induced loading (construction, fluid injection) emphasize rate-dependent hazard escalation and management.
Selected Papers
This digest features 179 selected papers from 1045 RSS items analyzed (out of 2707 raw RSS items scanned) across multiple journals. Each paper has been evaluated for its relevance to landslide and broader geohazard research and includes links to the original publications.
1. Evaluating the impact force of submarine landslide on pipelines: insights from flume tests and multiphase flow model analysis
Core Problem: Submarine landslides pose significant threats to marine pipelines, but the impact loads exerted by these landslides are not systematically assessed or quantified.
Key Innovation: Systematically assesses impact loads of submarine landslides on pipelines through flume tests and multiphase flow numerical analysis, quantifying drag and lift force components, their relationship with landslide velocity and other parameters, and establishing a consistent impact load evaluation model.
2. Integrating TS-InSAR and machine learning for spatiotemporal analysis and influencing factors identification of land subsidence in Nansha, Guangzhou, China
Core Problem: Quantifying significant land subsidence induced by rapid urban expansion and construction on compressible coastal soils in Nansha District and identifying its main driving factors.
Key Innovation: Integration of Time-Series Interferometric Synthetic Aperture Radar (TS-InSAR) with Sentinel-1A data and a Random Forest regression model, utilizing a multi-source database including detailed building-type classification and aquaculture areas, to precisely monitor subsidence rates (up to 283.7 mm/year) and quantitatively identify construction activity, soft soil thickness, aquaculture, and custom rural houses as leading contributing factors.
3. Correction: Spatial distribution patterns and landslide susceptibility analysis from a global–local perspective along the Zhuzhou-Guangzhou section of the Beijing–Guangzhou railway
Core Problem: Analyzing spatial distribution patterns and susceptibility of landslides along the Zhuzhou-Guangzhou section of the Beijing–Guangzhou railway.
Key Innovation: A correction to previous findings regarding the spatial distribution patterns and susceptibility analysis of landslides, refining the understanding and assessment of landslide risk along the specified railway section.
4. Deep learning-based landslides detection improved by the gray level co-occurrence matrix of RGB optical images
Core Problem: Accurate and automatic detection of landslides, especially rainfall-induced ones, using remote sensing imagery is challenging, and existing deep learning models can be improved by better feature integration.
Key Innovation: Developed a hybrid deep learning framework that integrates GLCM-based texture descriptors into U-Net, DeepLabv3+, and Mask R-CNN models, significantly enhancing landslide detection performance (e.g., 15.2% IoU improvement).
5. Integration of interpretable machine learning and MT-InSAR for dynamic enhancement of landslide susceptibility in the Three Gorges Reservoir Area
Core Problem: Landslide susceptibility mapping (LSM) faces challenges from heterogeneous triggers, reactivation due to engineering activities, and lack of interpretability in data-driven models, hindering practical application.
Key Innovation: Proposed a novel framework integrating interpretable machine learning (SHAP with Random Forest) and MT-InSAR-derived landslide activity levels to dynamically enhance landslide susceptibility mapping, accounting for different landslide types and significantly improving predictive capability and interpretability.
6. A two-point three-phase resolved coupling framework for modeling boulder-laden debris flows
Core Problem: Conventional single-phase models struggle to capture the complexities of boulder-laden debris flows, which have amplified destructive potential due to coarse boulders and multi-phase interactions, while strong-coupling methods are computationally prohibitive for practical hazard assessments.
Key Innovation: Proposed a semi-hybrid, fully resolved coupling numerical framework for modeling boulder-laden debris flows, treating the continuous phase with SPH and coarse particles with DCDEM, coupled via an efficient two-way resolved scheme, validated against physical experiments and a real-world debris-flow event.
7. Variability in Performance of a Machine-Learning Seismicity Catalog: Central Italy, 2016-2017
Core Problem: Machine learning (ML) seismicity catalogs contain many more earthquakes than routine catalogs, but their performance in phase picking and earthquake detection, including spatial variability, has not been fully evaluated.
Key Innovation: Developed station-level detection probabilities and spatial magnitude-of-completeness fields using logistic regression, demonstrating that an ML-based catalog for Central Italy (2016-2017) substantially increases detection sensitivity (median magnitude-of-completeness shifts from 1.6 to 0.5) but also shows greater spatial non-uniformity in performance.
8. Admissibility of Solitary Wave Modes in Long-Runout Debris Flows
Core Problem: Understanding the coherent wave structures (shock-like roll waves vs. weaker dispersive pulses) in long-runout debris flows, particularly their admissibility and contribution to mobility on gentle slopes where the base-flow Froude number is order unity.
Key Innovation: Obtains a Korteweg-de Vries (KdV) reduction from depth-averaged balances for gentle-slope, long-wave, low-amplitude debris flows, introduces a practical nonlinearity diagnostic, and organizes published cases into a Froude-slope diagram, showing that dispersive pulses are a regime-specific complement to roll-wave dynamics.
9. Characteristics of rockburst triggered by weak dynamic disturbance: insights from structural model tests
Core Problem: Understanding how weak dynamic disturbances influence rockburst behavior in deeply buried tunnels and distinguishing these characteristics from self-initiated rockbursts.
Key Innovation: Conducted structural model experiments under true-triaxial compression to compare self-initiated and weak dynamic disturbance-triggered (WDD-triggered) rockbursts. Findings indicate WDD-triggered rockbursts exhibit distinct characteristics (rapidity, violence, fractured band size, energy release) significantly influenced by the static stress level at the moment of disturbance.
10. Effects of secondary fractures on fault seismic rupture and aseismic slip during CO2 sequestration
Core Problem: Fluid injection-induced fault activation and seismicity pose significant risks to CO2 geological sequestration projects, and the influence of secondary fractures on these processes is not fully understood.
Key Innovation: Developed a computational model integrating two-phase fluid flow and geomechanics with a slip weakening model to investigate the influence of secondary fractures on fault activation and induced seismicity, revealing that secondary fractures initially delay slip but later increase seismicity, and documenting a novel mechanism of seismic slip cascades.
11. Multi attribute refined identification of flood-affected bodies based on multi-source data fusion
Core Problem: Current methods for identifying attributes of flood-affected bodies (urban land function, dynamic population distribution) suffer from accuracy constraints, data limitations, and insufficient spatial resolution, hindering effective flood prevention and mitigation strategies.
Key Innovation: Develops an ensemble learning model for optimal urban land function identification and a spatiotemporal human-land relationship matching method for high-resolution dynamic population distribution. Integrates these with hydraulic simulation and GIS for a multi-attribute diagnostic framework for flood-affected bodies.
12. Excavation-induced open-pit slope failures behaviors from microscopic insights using DEM analysis
Core Problem: The underlying mechanisms governing various failure morphologies in excavation-induced open-pit slope failures remain insufficiently understood, posing critical challenges in mining engineering.
Key Innovation: Employs a multi-scale approach integrating Discrete Element Method (DEM) simulations and analytical solutions to investigate slope arching failures, revealing the dominant role of initial material packing density in governing failure morphology and proposing a novel classification criterion based on incremental displacement ratios for distinguishing failure phases.
13. Comprehensive seismic hazard mapping for North Bengkulu, Southwest Sumatera, Indonesia: Planning for urban resilience and Safer cities
Core Problem: North Bengkulu, Indonesia, is highly prone to earthquakes from tectonic plate convergence and the Sumatran Fault System, posing a serious threat of earthquake-induced ground shaking and land deformation, requiring comprehensive hazard mapping for urban planning.
Key Innovation: Development of a comprehensive seismic hazard map for North Bengkulu by integrating secondary data (earthquake catalogs, PGA bedrock) and primary microtremor measurements (HVSR method) to map areas with high potential for soil deformation based on various elastic parameters (A0, SVI, PGA, GSS, MMI, Vs, Vp, Poisson's ratio, G, E). This provides insights for targeted development strategies and improved building quality.
14. Investigating the liquefaction resistance and initial shear modulus of silty sands from equivalent intergranular void ratio concept
Core Problem: There is a need to better understand the liquefaction resistance and initial shear modulus of silty sands, particularly how fines content affects these properties, and to refine the equivalent intergranular void ratio concept.
Key Innovation: This research conducted undrained cyclic triaxial and bender element tests on silty sands, deducing b-values for liquefaction resistance and shear modulus by fitting data to clean sand relations and introducing a fictitious active fines content (Fc). It found that Fc values differ for large-strain (liquefaction) and small-strain (shear modulus), leading to multiple cyclic resistance ratio (CRR) and shear wave velocity (Vs) relations, which advances micromechanical modeling of silty sand behavior.
15. Effects of non-plastic silt and soil aging on re-liquefaction resistance of sandy soils
Core Problem: Understanding the influences of non-plastic silt content and soil aging on the re-liquefaction resistance of sandy soils is crucial for assessing the stability of earth structures, especially after earthquake events.
Key Innovation: Conducted undrained triaxial tests on sand-silt mixtures and undisturbed/reconstituted sandy soils, revealing that liquefaction resistance initially decreases with fines content up to ~45%, and that aging effects can persist even after subsequent liquefaction histories, providing insights into re-liquefaction susceptibility.
16. Energy relief effect of real-time drilling to prevent rockburst in high-stress rock
Core Problem: Mitigating rockbursts in high-stress rock environments, which pose significant safety risks and operational challenges in underground engineering.
Key Innovation: Experimental demonstration that real-time drilling effectively weakens rock mechanical behavior and reduces rockburst intensity by dissipating energy, quantified by a proposed 'real-time drilling energy dissipation index (E RD)'. The study establishes a link between energy relief, pressure relief, and reduced rockburst proneness, validating drilling as a mitigation strategy.
17. Analytical solution for longitudinal responses of tunnels under combined effects of seismic waves and strike-slip faulting
Core Problem: Determining the longitudinal mechanical responses of tunnels subjected to the combined effects of seismic waves and strike-slip faulting, especially in high-intensity seismic zones, where such combined effects are likely to cause severe damage.
Key Innovation: Developed a novel analytical solution using an elastic spring-beam model, considering SH waves and S-shaped fault dislocation, superposition principle, and contact conditions. Validated the model and conducted a parametric study, classifying tunnel responses into seismic-dominated, faulting-dominated, and seismic-faulting coupled responses based on relative contributions.
18. Deformation warning of surrounding rock mass of underground powerhouse based on octree theory and microseismic monitoring
Core Problem: Effective early warning of surrounding rock mass deformation is crucial for the safety and stability of underground constructions, requiring quantitative assessment and spatial-temporal analysis.
Key Innovation: Introduced a novel risk early warning model based on multi-parameter fuzzy comprehensive evaluation, integrating microseismic monitoring and octree theory-based indices, to achieve 3D visualization and quantitative early warning of surrounding rock mass deformation in underground powerhouses.
19. Effects of confined aquifer on water-sand gushing disasters in soft soil
Core Problem: Water-sand gushing (WSG) disasters in confined aquifers pose significant challenges, and the impact of confined aquifer thickness and pressure on these disasters is not well understood.
Key Innovation: Developed a novel visual model test system to investigate the influencing characteristics and mechanisms of confined aquifer thickness and confined water pressure on water-sand gushing disasters in soft soil, revealing a two-stage gushing process and quantifying increased severity with greater aquifer thickness and confined water pressure.
20. Dynamic responses of Dagangshan high-arch dam under Luding earthquake: Insights from microseismic monitoring and digital twin model
Core Problem: Evaluating the dynamic response and potential damage of high-arch dams during earthquakes, and integrating MS monitoring with digital twin technology for self-updating mechanical parameters, remains under-explored.
Key Innovation: Proposed a framework for a damage-driven digital twin (DT) model of a high-arch dam, capable of self-updating mechanical parameters based on MS data, and used it to assess the dam's dynamic response and identify high-risk areas during the Luding earthquake, achieving higher accuracy than traditional methods.
21. Reactivation of rate-and-state faults induced by CO2 injection: Effects of pore pressure diffusion and fluid pressurization
Core Problem: Understanding the mechanisms of fault reactivation during early-stage CO2 injection, particularly the competing effects of pore pressure diffusion and fluid pressurization, is crucial for managing seismic risk in CO2 storage projects.
Key Innovation: Developed a coupled hydromechanical model incorporating rate-and-state friction laws to systematically investigate fault reactivation during CO2 injection, revealing that fluid pressurization enhances fault strength but simultaneously elevates seismic risk through amplified stress drops, and that positive Coulomb failure stress changes do not always correlate with actual slip zones.
22. Transmission patterns of progressive damage and reliability analysis of reservoir-induced landslides considering local tensile failure
Core Problem: Understanding the impact of local tensile failure and progressive damage on the stability and failure modes of reservoir-induced landslides, especially considering the role of tensile cracks and water level fluctuations.
Key Innovation: Incorporates potential tensile cracks into limit equilibrium and reliability analysis, quantifies landslide reliability under different tensile failure scenarios, and analyzes failure transmission paths (regular and skip patterns) validated by physical models. It highlights that cracks at the rear reduce stability, while cracks at the forefront reduce individual slide mass descent.
23. Development of a Dual‐Domain Karst Flow Model Under Consideration of Preferential Film‐Flow Dynamics and Analysis of Compartment‐Specific Parameter Sensitivities
Core Problem: Characterizing and managing karst systems is challenging due to heterogeneity and vulnerability to contamination, with highly conductive features complicating simulation of rapid recharge dynamics for flood and contamination risk assessment.
Key Innovation: Presenting a novel modeling strategy by extending MODFLOW-CFPv2 to simultaneously compute diffuse fluxes and film-flow in the vadose zone, simulating infiltration via preferential pathways. Global sensitivity analysis shows that considering film-flow and its controlling parameters becomes important during strong infiltration events, benefiting the characterization of such systems for risk assessment.
24. Failure to track a stable AMOC state under rapid climate change
Core Problem: The global warming threshold for Atlantic Meridional Overturning Circulation (AMOC) collapse may be misleading, as AMOC stability depends on the rate of radiative forcing and background climate state.
Key Innovation: Identifies an AMOC stabilizing mechanism operating on longer timescales, demonstrating that slow CO2 increase permits AMOC stability up to higher warming levels (+5.5C), while rapid forcing leads to collapse at substantially lower levels (+2.2C and +2.8C), highlighting the critical role of forcing rate.
25. TAMMs: Change Understanding and Forecasting in Satellite Image Time Series with Temporal-Aware Multimodal Models
Core Problem: Temporal Change Description (TCD) and Future Satellite Image Forecasting (FSIF) in Satellite Image Time Series (SITS) are disjointed tasks fundamentally limited by the challenge of modeling long-range temporal dynamics.
Key Innovation: Introduces TAMMs, the first unified MLLM-diffusion framework to jointly perform TCD and FSIF, incorporating Temporal Adaptation Modules (TAM) for enhanced long-range dynamics comprehension and Semantic-Fused Control Injection (SFCI) for fine-grained generative control, outperforming state-of-the-art specialist baselines.
26. Noisy-Pair Robust Representation Alignment for Positive-Unlabeled Learning
Core Problem: State-of-the-art Positive-Unlabeled (PU) learning methods substantially underperform supervised counterparts on complex datasets, especially without auxiliary information, due to the challenge of learning discriminative representations under unreliable supervision.
Key Innovation: Proposing NcPU, a non-contrastive PU learning framework that combines a noisy-pair robust supervised non-contrastive loss (NoiSNCL) for aligning intra-class representations and a phantom label disambiguation (PLD) scheme for conservative negative supervision, achieving substantial improvements in PU learning, including for post-disaster building damage mapping.
27. Hydro-mechanical behaviour of Nanyang expansive soil during the wetting–drying cycles
Core Problem: Expansive soils pose significant challenges to geotechnical engineering due to their distinctive hydraulic-mechanical behavior, particularly under varying environmental conditions like wetting-drying cycles, leading to long-term deterioration.
Key Innovation: Established a comprehensive testing framework to examine the hydraulic-mechanical properties of Nanyang expansive soil during moisture variations. Results showed that repeated wetting-drying cycles lead to crack development, significant reduction in shear strength, and increased permeability, providing valuable insights for reliable design and maintenance strategies in high-risk regions.
28. Accurate tidal level forecasting over an extended horizon using a deep learning method
Core Problem: Accurate and efficient tidal level forecasting over extended horizons is crucial for coastal activities and mitigating hazards like storm surges, but existing methods may struggle with capturing both periodic and aperiodic signals effectively.
Key Innovation: Proposes an innovative hybrid CNN-Attention model that integrates convolutional neural networks with a self-attention mechanism to capture both periodic and aperiodic tidal signals, enabling accurate 5-day forecasts and maintaining robust performance over extended horizons.
29. CloudBreaker: Breaking the cloud covers of Sentinel-2 images using multi-stage trained conditional flow matching on Sentinel-1
Core Problem: Cloud cover and nighttime conditions severely limit the temporal continuity and utility of multi-spectral optical satellite imagery (e.g., Sentinel-2), hindering effective Earth observation and disaster response, especially when existing SAR-optical fusion methods are inefficient or constrained by temporal misalignment.
Key Innovation: CloudBreaker, a novel generative framework, synthesizes high-fidelity, comprehensive multi-spectral Sentinel-2 imagery directly from single-modal Sentinel-1 SAR inputs, bypassing concurrent optical data requirements. It uses a multi-stage training strategy with conditional latent flow matching and cosine scheduling, demonstrating exceptional utility in disaster response by providing unobstructed optical visualizations and biophysical indices under dense clouds/ash.
30. Resilience evaluation of key land use/cover types in the middle reaches of the Yellow River under flood stress: Implications for slope regulation and storage measures
Core Problem: Unclear resilience (recovery and resistance) of critical land use/cover (forest, grassland, terrace) under flood stress in the Middle Reaches of the Yellow River, hindering effective soil and water conservation and slope regulation.
Key Innovation: Proposed evaluation methods for flood events and calculation methods for resistance and recovery to analyze the resilience of different LULC and sub-watersheds, finding that grass shows superior recovery, forest great resistance, and resilience decreases with increasing slope, providing a scientific basis for water resource management and ecological protection under flood stress.
31. Retrieval of snow depth using synthetic aperture radar: past, current, and future
Core Problem: Accurate and continuous retrieval of snow depth, a key parameter for water storage estimation and disaster prediction, is challenging due to limited spatial representativeness of ground observations and insufficient resolution of passive microwave remote sensing.
Key Innovation: Provides a comprehensive review of synthetic aperture radar (SAR) techniques (PolSAR, InSAR, PolInSAR, TomoSAR) for superior precision and fine-scale retrieval of spatiotemporally continuous snow depth, outlining past, current, and future directions for this critical measurement.
32. An offshore non-ergodic ground motion model for horizontal pseudo acceleration spectrum in the Japan trench area
Core Problem: Ergodic ground motion models overlook the significant spatial non-uniformity of ground motion, leading to large prediction errors, especially in offshore earthquake-prone areas.
Key Innovation: Development of a non-ergodic ground motion model for the Japan trench area by treating seismic source, path, and site effects as Gaussian random processes and introducing non-ergodic terms. This significantly reduces model uncertainty (23-38% overall, 22-61% between-event) compared to ergodic models, improving prediction accuracy for marine engineering and earthquake risk assessment.
33. Data-driven iterative calibration method for prior knowledge of earth-rockfill dam wetting model parameters
Core Problem: Wetting deformation is a critical factor influencing earth-rockfill dam safety, but existing mathematical models often lack prior knowledge of parameters, which is essential for accurate Bayesian parameter inversion and reducing uncertainty.
Key Innovation: Introduced a data-driven iterative calibration method to establish prior knowledge of wetting model parameters for earth-rockfill dams, significantly improving settlement prediction accuracy (RMSE of 5.18 mm vs. 11.97 mm for non-informative priors) and computational efficiency.
34. Volume change and creep behaviors related to stress-phase transition path in methane hydrate-bearing silty sand
Core Problem: Understanding the complex volume change and creep behaviors of methane hydrate-bearing sediments under various stress-phase transition paths, which is critical for assessing the stability of hydrate reservoirs and predicting potential ground instability during hydrate extraction or environmental changes.
Key Innovation: Experimental investigation using a customized apparatus revealing unusual shrinkage during hydrate formation under low stress, and demonstrating that hydrate formation increases yield stress but minimally affects compression/swelling indexes. It also shows significantly greater creep in hydrate-bearing sediment under high stress, providing insights into the mechanical response influencing reservoir stability.
35. Influence of native pores on the size distribution and predictability of rock failure
Core Problem: The influence of native pore structures and loading conditions on the fracture size distribution and, critically, the predictability of catastrophic rock failure remains unclear.
Key Innovation: Systematic evaluation using NMR and AE monitoring across four lithologies, demonstrating that native pore size primarily controls the AE b-value and failure predictability. Rocks with larger pores show higher b-values and better predictability, with a proposed inverse power law model showing higher accuracy for high-porosity rocks, highlighting the governing role of pore heterogeneity.
36. Effect of dominant fractures on triaxial behavior of 3D-printed rock analogs with internal fracture networks
Core Problem: Lack of understanding of how dominant fractures within complex internal fracture networks affect the macromechanical properties and failure modes of rock masses under triaxial stress.
Key Innovation: Used 3D printing to create rock-like specimens with internal fracture networks and conducted triaxial compression tests. Demonstrated that dominant fracture angle significantly affects compressive strength and elastic modulus, and that confining pressure improves strength and reduces the impact of fracture angle on deformation.
37. Experimental and theoretical investigation of face failure and ground collapse during slurry pressure-balanced shield tunneling in saturated sand
Core Problem: Understanding the mechanisms of face failure and subsequent ground collapse in saturated ground during slurry pressure-balanced shield (SPBS) tunneling due to seepage force and inadequate support pressure.
Key Innovation: Employed laboratory model tests with a slurry circulation system and theoretical validation to elucidate collapse mechanisms. Developed a theoretical solution for critical collapse pressure based on rotational failure, validated by experiments. Identified crucial tunneling parameters for filter cake formation and pressure maintenance, and characterized the continuous failure mode of ground collapse.
38. A new algorithm for high-speed identification of discontinuities on large-scale rock outcrop: A case study in Jinsha River suture zone
Core Problem: Conventional methods for identifying complex discontinuities in large-scale rock slopes are inefficient and limited, hindering accurate stability analysis.
Key Innovation: Developed an edge-first connection algorithm and RD ID software using UAV photogrammetry to rapidly and accurately identify planar discontinuities on large-scale rock outcrops, significantly improving computational efficiency and accuracy for rock slope stability assessment.
39. RNPC-net: Automatic recognition and mapping of weathering degree and groundwater condition of tunnel faces
Core Problem: Conventional methods for recognizing weathering degree and groundwater conditions in tunnels are subjective, complex, and time-consuming, hindering rapid and accurate rock mass quality evaluation.
Key Innovation: Proposed RNPC-net, a deep learning model with a hybrid feature extraction module and adaptive weighting auxiliary classifier, for rapid and accurate automatic recognition and mapping of weathering degree and groundwater conditions of tunnel faces, enabling quantitative rock mass quality evaluation (RMR system).
40. Microseismic characteristics and settlement analysis of concrete face rockfill dams on deep overburden layers during the filling process
Core Problem: Deep overburden layers beneath concrete face rockfill dams cause uneven settlement, posing a serious threat to dam safety, and real-time monitoring of microfracture in bedrock during construction is needed.
Key Innovation: Employed microseismic (MS) monitoring technology for the first time in dam filling engineering to real-time monitor bedrock microfracture, revealing fracture mechanisms, and analyzing relationships among slope deformation, dam settlement, and MS activity, demonstrating its value in predicting dam settlement risk.
41. A Continuum of Slow Slip Events in the Cascadia Subduction Zone Illuminated by High‐Resolution Deep‐Learning Denoising
Core Problem: Whether slow and fast fault slip arise from similar physical processes remains unresolved due to detection biases affecting noisy surface measurements and ambiguities in analyzing the source properties of slow slip events.
Key Innovation: Using daily geodetic time series denoised with a deep learning model, the study inverted for 15 years of slow slip evolution on the Cascadia subduction zone with unprecedented temporal resolution, revealing a continuum of slow slip events of various sizes controlled by subduction interface geometrical constraints.
42. Multi-Expert Learning Framework with the State Space Model for Optical and SAR Image Registration
Core Problem: Existing deep learning methods for optical and SAR image registration struggle with significant nonlinear radiometric variations, limited textures, and the trade-off between local (CNN) and global (Transformer) feature learning efficiency.
Key Innovation: Proposes ME-SSM, a multi-expert learning framework that constructs a learnable soft router to fuse features from various image transformations, integrates a Mamba state space model for efficient global contextual feature extraction with linear complexity, and uses multi-level feature aggregation, significantly improving optical and SAR image registration accuracy.
43. VIMD: Monocular Visual-Inertial Motion and Depth Estimation
Core Problem: Accurate and efficient dense metric depth estimation is crucial for 3D visual perception but challenging, especially in resource-constrained settings or with extremely sparse input data.
Key Innovation: Develops VIMD, a monocular visual-inertial motion and depth learning framework that leverages MSCKF-based motion tracking and iteratively refines per-pixel scale using multi-view information, achieving exceptional accuracy and robustness even with extremely sparse depth points.
44. Front Matter
Core Problem: This entry is a journal's front matter and does not present a specific research problem or gap.
Key Innovation: This entry is a journal's front matter and does not present a specific research innovation or contribution.
45. Transformation of the middle–upper ordovician sedimentary environment of the north China plate under the caledonian tectonic background: a case study of the ordovician majiagou formation in the qishan area
Core Problem: Understanding the abrupt sedimentary and petrographic transition and the causes of intensified syn-depositional disturbance and margin instability during the late stage of Majiagou deposition in the Qishan area.
Key Innovation: Detailed lithofacies analysis revealing a rapid shift from shallow-water platform to deeper-water slope/base-of-slope deposits characterized by slump breccias and gravity-flow facies, interpreted as a major increase in accommodation and margin instability potentially linked to intensified Qinling Ocean subduction, refining the depositional model and contributing to basin-orogen coupling discussions.
46. Emergency Response Reliability: An SPN-based framework for cross-departmental collaboration efficiency and dynamic optimization
Core Problem: Deficiencies in cross-departmental coordination frequently lead to resource misallocation and critical delays, undermining earthquake emergency response reliability, and existing static models fail to capture stochastic interagency dynamics.
Key Innovation: A formal analytical framework based on stochastic Petri nets (SPN) is developed to conceptualize and quantify 'collaboration reliability' in earthquake emergency responses, introducing a novel collaborative efficiency index and providing data-driven strategies for dynamic optimization.
47. CSTFSeg: A High-Resolution Chinese Tidal Flat Dataset and Multi-Scale Attention Semantic Segmentation Network
Core Problem: Fine-grained semantic segmentation of tidal flats is hindered by a lack of public, high-resolution pixel-level benchmark datasets and difficulties in precise segmentation due to spectral ambiguity at boundaries caused by hydrodynamic tides and sediment mixing.
Key Innovation: Introduced CSTF, the first high-resolution semantic segmentation benchmark dataset for Chinese tidal flats, and proposed CSTFSeg, a multi-scale attention network integrating fuzzy logic and edge guidance, which explicitly models boundary uncertainty and achieves superior accuracy and generalization for tidal flat mapping, aiding coastal disaster mitigation.
48. Evaluation of the relationship between freezing point and suction of sodium sulfate loess near 0°C
Core Problem: Understanding the mechanism of salt weathering of check dams in the Loess Plateau requires correlating the freezing point and suction of sodium sulfate loess, a task complicated by controversies over frozen soil suction measurement.
Key Innovation: This study developed a temperature-dependent calibration model for soil suction near 0°C, combining vapor equilibrium technique (VET) and sensor data. It established a close link between the freezing point and suction at 0.5°C and developed a modified Mizoguchi model for saline loess, shedding light on the nature of salt weathering and informing targeted treatments for check dams.
49. Improvement of dispersive soil properties by enzyme-induced carbonate precipitation technology: Mechanical, microstructural, and statistical analysis
Core Problem: Dispersive soils are highly susceptible to erosion, posing significant environmental challenges, and there is a need for environmentally friendly methods to improve their mechanical properties and hydrostability.
Key Innovation: Proposed a novel reinforced enzyme-induced carbonate precipitation (REICP) method using EICP-MgCl2-xanthan gum to significantly enhance the mechanical properties and hydrostability of dispersive soil, increasing cohesion by a factor of 2 and reducing permeability by 1.7 × 10^7 times.
50. Suffusion of sand–clay mixtures under stepwise increase in hydraulic gradient
Core Problem: Understanding the factors influencing suffusion (loss of fine particles) in sand-clay mixtures, specifically the critical hydraulic gradient, clay type, and ionic concentration, which is crucial for assessing the stability of soil structures.
Key Innovation: Experimental investigation revealing that critical hydraulic gradient is low (<0.1) for all sand-clay mixtures, and illite particles are more susceptible to suffusion than kaolinite. The study highlights the importance of clay type, sand-to-clay size ratio, and ionic concentration in suffusion behavior.
51. Shear fracturing behavior and mechanism of intact granite under thermal-mechanical coupling loading
Core Problem: Lack of understanding of how intact granite responds to real-time high temperature upon shear loading, particularly in subsurface resource recovery and deep underground space utilization.
Key Innovation: Conducted direct shear tests with a newly built apparatus under varied normal stresses and high temperatures, comparing real-time heating vs. thermal treatment. Found that real-time heating leads to earlier and greater cracking events and lower peak shear strength, and identified temperature-dependent strengthening (below 200°C) and weakening (200-400°C) mechanisms.
52. Damage and pore structure characteristics of sandstone subjected to the disturbance creep process
Core Problem: Limited understanding of the effects of periodic high-frequency stress disturbances on sandstone creep behavior and associated microstructural changes.
Key Innovation: Conducted high-frequency disturbance creep experiments on sandstone, analyzing damage value, porosity increments, and pore structure fractal dimension using NMR. Found that increased loading stress and disturbance cycles increase damage and porosity, while higher disturbance frequency reduces creep strain and rate.
53. Thermomechanical coupling analysis of granite fracture shear behavior: True triaxial test and numerical approaches
Core Problem: Immature understanding of the influence of thermomechanical coupling on the shear behavior and damage evolution of prefractured granite, particularly under true triaxial conditions.
Key Innovation: Conducted true triaxial laboratory tests and discrete element method simulations on rough granite fractures under varying confining pressures and temperatures. Found that high temperature and confining pressure increase peak strength, leading to more microcracks and gouges, and identified thermal strengthening at low temperatures under greater confining pressure and the role of fracture angle in strength and slip.
54. Vibrations induced by time-delayed double blastholes in underground rocks: Experimental study and theoretical analysis
Core Problem: The underlying mechanisms for vibration reduction in time-delayed blasting, especially for double-hole configurations, are unclear, limiting optimization for mitigating induced vibration hazards.
Key Innovation: Conducted experimental blasts and developed an exact full-field theoretical model to systematically analyze the effects of delay time, charge length, and borehole inclination on vibrations induced by time-delayed double blastholes, providing insights for optimizing blasting parameters to control vibrations.
55. Investigation of strata fractures during longwall mining: Original introscopic probe and image analysis methods
Core Problem: Accurately analyzing the evolution and extent of strata fractures caused by longwall mining is challenging, impacting methane emissions and overall ground stability.
Key Innovation: Developed an introscopic probe with a high-resolution camera and automatic image analysis methods to continuously record and classify fractures in deep boreholes during longwall mining, correlating fracture evolution with methane capture rates and determining the extent of the fracture zone.
56. Automated recognition of rock discontinuity in underground engineering using geometric feature analysis
Core Problem: Mainstream methods for identifying rock discontinuities in underground engineering suffer from accuracy degradation, omission of critical discontinuities, and require manual intervention, especially with uneven orientation density.
Key Innovation: Introduced a novel automated discontinuity identification method based on geometric feature analysis of point clouds, integrating an adaptive region growing algorithm and an adaptive hierarchical clustering algorithm to accurately detect independent discontinuities and determine optimal structural sets without manual intervention.
57. Effect of dynamic disturbance frequency on brittle failure of granite in deep excavation
Core Problem: Dynamic disturbances from various sources (blasting, TBM, rockbursts, earthquakes) with different frequencies can trigger diverse failure modes in deep excavations, and understanding this frequency dependency is crucial.
Key Innovation: Conducted true-triaxial compression tests on granite under dynamic disturbances of varying frequencies, revealing a U-shaped relationship between peak strength and disturbance frequency, and showing how specific frequencies and low-amplitude disturbances affect crack damage and expansion, providing insights into brittle failure mechanisms.
58. Energy budget in geomaterials fracture: analysis using non-local ductile damage model
Core Problem: A lack of a comprehensive approach to calculate and understand the energy budget components during progressive failure in cohesive-frictional geomaterials, which is crucial for understanding fracture propagation.
Key Innovation: Presents a novel non-local ductile damage model derived from thermodynamic formulation to calculate energy budget components during geomaterial fracture. It captures fracture in various scenarios, including slope stability, and provides a physics-based understanding of energy budget mechanisms.
59. Electric charge induction monitoring of deformation and failure behavior of igneous rock: Laboratory test and field application
Core Problem: Advancing the theoretical understanding, technological development, and field application of electric charge induction for monitoring rock deformation and failure, particularly for early warning of geohazards like coal bursts.
Key Innovation: Investigates induced electric charge signals during igneous rock deformation, develops an online downhole electric charge induction monitoring system, identifies optimal monitoring parameters and early warning indicators (max absolute value, arithmetic mean, dispersion coefficient, cumulative sum), and validates its field application for coal burst prediction.
60. Scalable and Reliable State-Aware Inference of High-Impact N-k Contingencies
Core Problem: Assessing higher-order N-k contingencies in power systems is computationally prohibitive, and existing heuristic screening methods lack formal guarantees for consistently retaining all critical contingencies.
Key Innovation: Proposes a scalable, state-aware contingency inference framework that directly generates high-impact N-k outage scenarios without enumerating the combinatorial space, employing a conditional diffusion model and a topology-aware graph neural network, and providing controllable coverage guarantees for severe contingencies.
61. Adaptive recurrent flow map operator learning for reaction diffusion dynamics
Core Problem: Learning stable operators to forecast long-term dynamics of reaction-diffusion equations from data is challenging due to error accumulation in autoregressive rollouts and degradation with out-of-distribution initial conditions, while physics-based regularization has drawbacks.
Key Innovation: DDOL-ART, a purely data-driven operator learner with adaptive recurrent training, uses a robust recurrent strategy with lightweight validation milestones to achieve stable long-term predictions and zero-shot generalization to strong morphology shifts across various reaction-diffusion systems, offering a strong accuracy and cost trade-off.
62. Energy-Efficient Fast Object Detection on Edge Devices for IoT Systems
Core Problem: IoT systems require energy-efficient and fast object detection, which end-to-end methods struggle with, especially for fast-moving objects, leading to lower accuracy and efficiency.
Key Innovation: Presents an IoT application using an AI classifier for fast-object detection via the frame difference method, implemented on edge devices. This lightweight algorithm significantly improves average accuracy, efficiency, and latency compared to end-to-end methods, making it suitable for fast-moving object detection in energy-constrained IoT systems.
63. SCA-Net: Spatial-Contextual Aggregation Network for Enhanced Small Building and Road Change Detection
Core Problem: Automated change detection in remote sensing imagery, critical for urban management and disaster assessment, often struggles with low sensitivity to small objects and high computational costs.
Key Innovation: SCA-Net, an enhanced architecture for precise building and road change detection, incorporates a Difference Pyramid Block, an Adaptive Multi-scale Processing module, multi-level attention mechanisms, a dynamic composite loss function, and a four-phase training strategy to improve accuracy and efficiency.
64. Contextual and Seasonal LSTMs for Time Series Anomaly Detection
Core Problem: Existing time series anomaly detection methods struggle to capture subtle anomalies (small point anomalies, slowly rising anomalies) in univariate time series data.
Key Innovation: Proposes Contextual and Seasonal LSTMs (CS-LSTMs), a prediction-based framework that uses a noise decomposition strategy and integrates contextual dependencies, seasonal patterns, and both time-domain and frequency-domain representations to robustly detect subtle anomalies in univariate time series.
65. PlugSI: Plug-and-Play Test-Time Graph Adaptation for Spatial Interpolation
Core Problem: Current graph-based Spatial Interpolation (SI) methods for sensor networks rely on pre-trained models, lack adaptation to larger and unseen graphs at test-time, and overlook the utilization of test data.
Key Innovation: PlugSI, a plug-and-play framework that refines test-time graphs through an Unknown Topology Adapter (UTA) for adapting to new graph structures and a Temporal Balance Adapter (TBA) for maintaining stable historical consensus, significantly improving existing graph-based SI methods.
66. GeoFormer: A Swin Transformer-Based Framework for Scene-Level Building Height and Footprint Estimation from Sentinel Imagery
Core Problem: Accurate, generalizable, and open-source three-dimensional urban data (building height and footprint) are scarce, hindering applications like climate modeling, disaster risk assessment, and urban planning due to reliance on proprietary sensors or poor cross-city generalization.
Key Innovation: GeoFormer, an open-source Swin Transformer framework, jointly estimates building height and footprint on a 100m grid using only Sentinel-1/2 imagery and open DEM data, achieving improved accuracy and cross-continent transferability compared to CNN baselines, and is publicly released.
67. Perception with Guarantees: Certified Pose Estimation via Reachability Analysis
Core Problem: Ensuring safety in cyber-physical systems by providing formally guaranteed, worst-case accurate pose estimates for agents, solely from a camera image and known target geometry, without relying on potentially untrustworthy external services or insufficient rough estimates.
Key Innovation: The paper presents a certified pose estimation method in 3D by formally bounding the pose, leveraging recent results from reachability analysis and formal neural network verification. This approach efficiently and accurately localizes agents in both synthetic and real-world experiments, providing guarantees on the pose estimate.
68. Spatio-Temporal Attention for Consistent Video Semantic Segmentation in Automated Driving
Core Problem: Existing semantic segmentation models for video process frames independently, failing to leverage temporal consistency, which limits both accuracy and stability in dynamic scenes.
Key Innovation: Proposes a Spatio-Temporal Attention (STA) mechanism that extends transformer attention blocks to incorporate multi-frame context, enabling robust temporal feature representations for consistent video semantic segmentation. This significantly improves temporal consistency and mIoU on datasets like Cityscapes and BDD100k.
69. Self-Supervised Learning Based on Transformed Image Reconstruction for Equivariance-Coherent Feature Representation
Core Problem: Existing self-supervised learning (SSL) methods often discard transformation information in images, which is crucial for computer vision tasks requiring equivariant features (e.g., segmentation, detection, depth estimation), while current equivariant approaches have restrictive assumptions.
Key Innovation: A novel SSL auxiliary task that learns equivariance-coherent representations through intermediate transformation reconstruction, allowing decomposition of feature vectors into invariant and equivariant parts, leading to substantial improvements on equivariance benchmarks and competitive performance on downstream tasks like segmentation, detection, and depth estimation.
70. Bridging Past and Future: Distribution-Aware Alignment for Time Series Forecasting
Core Problem: Adoption of representation-learning methods like contrastive learning in modern time series forecasters is limited, and a distributional gap often exists between historical inputs and future targets, hindering forecasting performance.
Key Innovation: Introduces TimeAlign, a lightweight, plug-and-play framework that aligns auxiliary features via a simple reconstruction task and feeds them back into any base forecaster, explicitly bridging the distributional gap between past and future representations for improved time series forecasting.
71. SNAP: Towards Segmenting Anything in Any Point Cloud
Core Problem: Current interactive 3D point cloud segmentation approaches are domain-restricted, limited in user interaction types, and suffer from negative transfer when trained on multiple datasets.
Key Innovation: Presents SNAP, a unified model for interactive 3D segmentation supporting both point-based and text-based prompts across diverse domains (indoor, outdoor, aerial) by training on 7 datasets with domain-adaptive normalization, achieving state-of-the-art performance.
72. Knowledge-Guided Masked Autoencoder with Linear Spectral Mixing and Spectral-Angle-Aware Reconstruction
Core Problem: Deep learning models often lack interpretability, generalization, and data efficiency, particularly in remote sensing applications, due to a reliance solely on data-driven optimization without embedding scientific domain knowledge into the self-supervised reconstruction process.
Key Innovation: A novel knowledge-guided ViT-based Masked Autoencoder that embeds scientific domain knowledge, specifically the Linear Spectral Mixing Model (LSMM) and Spectral Angle Mapper (SAM), as physical constraints within the self-supervised reconstruction process. This jointly optimizes LSMM, SAM, and Huber loss, enhancing reconstruction fidelity, stabilizing training, and yielding interpretable latent representations grounded in physical principles for spectral data.
73. Solving PDEs With Deep Neural Nets under General Boundary Conditions
Core Problem: Traditional numerical methods for Partial Differential Equations (PDEs) often struggle with high-dimensional or complex problems, and Physics-Informed Neural Networks (PINNs) face challenges in achieving high accuracy and handling complex boundary conditions.
Key Innovation: An extension of the Time-Evolving Natural Gradient (TENG) framework to address Dirichlet boundary conditions in PINNs, integrating natural gradient optimization with numerical time-stepping schemes (Euler and Heun methods) and incorporating boundary condition penalty terms into the loss function to ensure both stability and accuracy, establishing a foundation for broader PDE applicability.
74. Predictive Modeling of Power Outages during Extreme Events: Integrating Weather and Socio-Economic Factors
Core Problem: Predicting low-probability, high-consequence power outages caused by extreme events, and understanding how socio-economic factors contribute to community vulnerability during such conditions.
Key Innovation: A novel learning-based framework integrating EAGLE-I outage records with weather, socioeconomic, infrastructure, and seasonal event data to predict power outages, demonstrating that LSTM networks achieve higher accuracy and reveal patterns of community vulnerability.
75. Harbor resonance under coupled shear flow-wave forcing: Non-hydrostatic modelling and mechanism analysis
Core Problem: Understanding and accurately modeling harbor resonance under complex coupled shear flow-wave forcing, including its excitation pathways and modal modulation, which is an often underestimated driver of long-period harbor oscillations and potential damage.
Key Innovation: Examined harbor resonance using a three-dimensional non-hydrostatic model, validated against benchmark experiments. The study identified entrance shear flow as an important driver and modulator, showing it excites low-order modes, shifts resonant frequencies, reshapes amplification curves, and enhances nonlinear transfer to low-frequency long waves.
76. Ship Target Rapid Detection and Signal Extraction in Wide-Area Oceanic Scenes Based on ResNet-ID42 Network for Spaceborne SAR Range-Compressed Domain Data
Core Problem: Conventional full-scene imaging followed by target detection for spaceborne SAR data is highly inefficient for rapid ship detection in wide-area oceanic scenes characterized by strong target sparsity.
Key Innovation: Proposed a fast ship detection method in the SAR range-compressed domain using a ResNet-ID42 network with a physics-informed feature enhancement module. This enables rapid detection and complete target signal extraction without full-scene imaging, significantly improving processing efficiency (14.1x) and reducing data volume while maintaining high recall.
77. High-resolution atmospheric data cubes from the WegenerNet 3D Open-Air Laboratory for Climate Change Research
Core Problem: A lack of continuous, high-resolution atmospheric observational data at sub-kilometer and sub-hourly scales hinders the study of weather extremes, water vapor-cloud-precipitation interactions, and surface-atmosphere interactions in a changing climate.
Key Innovation: Released the first high-resolution atmospheric data cubes from the WegenerNet 3D Open-Air Laboratory, integrating data from an X-band radar, radiometers, and a GNSS network. This provides a multi-year, sub-kilometer, sub-hourly dataset of precipitation, cloud, and tropospheric profiles, serving as an independent resource for evaluating climate models and studying weather extremes relevant to climate change.
78. A novel integrated dependence assessment method for human reliability analysis under uncertainty and expert disagreement
Core Problem: Dependence assessment in human reliability analysis (HRA) is challenging, especially when dealing with uncertain and conflicting expert evaluations, leading to issues in calculating conditional human error probability (CHEP).
Key Innovation: A novel integrated method combining a consensus-reaching process (CRP), the full consistency method (FUCOM), and Dempster-Shafer evidence theory (DSET) to support dependence assessment and CHEP calculation in HRA, providing a transparent, traceable, and numerically plausible approach for managing expert disagreement and weighting consistency in safety-critical applications.
79. Multi-stage robust optimization of reliability improvement for power transmission cyber-physical systems under sequential coordinated attacks: an extreme attack scenario search-assisted framework
Core Problem: Power transmission cyber-physical systems (PTCPS) are increasingly vulnerable to sequential and coordinated cyber-physical attacks (CPAs), necessitating robust optimization frameworks to improve reliability and resilience, especially given the complexity of fault propagation and the need for efficient scenario identification.
Key Innovation: A multi-stage robust optimization framework for PTCPS under sequential coordinated CPAs, incorporating an extreme attack scenario search mechanism that identifies critical scenarios via objective-free optimization based on fault propagation, significantly reducing computational time and improving convergence efficiency for reliability improvement.
80. Electric bus system design reliable against stochastic power grid load shedding and uncertain charging efficiency
Core Problem: The rapid electrification of public transit introduces challenges related to energy supply uncertainties, specifically power grid load shedding and uncertain charging efficiency, which can lead to delays and disruptions in electric bus (eBus) systems.
Key Innovation: A hybrid robust-stochastic mixed integer linear program framework for simultaneously designing a reliable eBus system and its charging schedule, accounting for stochastic load shedding and robustly handling charging efficiency uncertainty, demonstrating cost-effective resilience and adaptability to increasing uncertainty probabilities.
81. Prediction of maximum ceiling temperature rise in inclined tunnel fire based on improved non-orthogonal ventilation plume model
Core Problem: Predicting fire-induced thermal environments and maximum ceiling temperature in inclined tunnels is challenging because existing models overlook additional physical mechanisms introduced by steeper slopes, leading to systematic deviations.
Key Innovation: Used Fire Dynamics Simulator (FDS) to investigate fire plume behavior in inclined tunnels, identified asymmetric entrainment and non-orthogonal ventilation effects, and developed an improved predictive model by introducing a modified dimensionless velocity and a slope-dependent entrainment correction factor, achieving substantially enhanced accuracy.
82. Inertial-buoyant coupling and bi-directional flow effects on plume deflection in inclined tunnel fires under natural ventilation
Core Problem: Plume deflection in inclined tunnel fires is inadequately understood, yet it critically influences smoke movement and evacuation safety.
Key Innovation: Conducted high-fidelity numerical simulations to classify flow-plume interactions into three regimes based on tunnel slope, proposed a modified Richardson number (Ri’) to characterize the critical transition for plume deflection (identified as Ri’=16.0 ± 1.5), and developed a predictive model for plume deflection angle.
83. A study on the impact of tunnel cross-sectional shape on the accidental dispersion and explosion characteristics of hydrogen
Core Problem: The impact of different tunnel cross-sectional geometries on the accidental dispersion and explosion characteristics of hydrogen is not well understood, which is crucial for preventing accidents and improving emergency response.
Key Innovation: Utilized CFD simulations (GASFLOW-MPI with DES turbulence models) to compare hydrogen leakage, dispersion, and combustion in arched and trapezoidal roof tunnels, analyzing dispersion distribution, concentration stratification, flame acceleration zones, and peak overpressure, providing theoretical support for tunnel design and management.
84. Urbanization-induced changes in rainfall and drought patterns: a study across six Indian states with mega-cities
Core Problem: Limited understanding of how urbanization impacts rainfall and drought patterns in Indian cities, and the implications for urban flood management.
Key Innovation: Utilized a dynamic classification system for rain gauges and both statistical and mesoscale model approaches to show a pronounced upward trend in heavy rainfall and annual monsoon precipitation over urban areas compared to rural counterparts, and exacerbated drought occurrences, underscoring the urgent need for enhanced urban flood management.
85. RRV (reliability, resilience and vulnerability)-based assessment of groundwater extraction sustainability in an over-exploited piedmont plain
Core Problem: Assessing the sustainability of groundwater extraction in over-exploited plains is complex, with existing frameworks often failing to adequately account for evaluation period length, precipitation contributions, and potentially overestimating system reliability based on groundwater storage changes.
Key Innovation: Develops an RRV-based framework to assess groundwater extraction sustainability, demonstrating that GWS-derived sustainability indices can significantly overestimate sustainability compared to those based on precipitation deficit. Highlights the dominant influence of drought periods and the necessity of dynamically adjusted pumping to mitigate overexploitation.
86. Explainable AI for predicting the strength of bio-cemented sands
Core Problem: Accurately predicting the unconfined compressive strength (UCS) of bio-cemented sands (MICP-treated) to optimize soil stabilization processes, which is complex due to numerous influencing variables.
Key Innovation: Development and optimization of explainable AI (XAI) models (e.g., LightGBM-CDO) to predict UCS of MICP-treated sands with high accuracy, and using XAI techniques (FIA, SHAP, PDPs) to interpret the complex non-linear relationships between input variables and soil strength, enhancing process optimization.
87. Effects of grain interfacial morphologies on microbially induced calcium carbonate precipitation process: Experimental evidence and numerical analysis
Core Problem: Lack of clear understanding regarding how grain interfacial morphologies influence the Microbially Induced Calcium Carbonate Precipitation (MICP) process, which is crucial for optimizing this eco-friendly soil improvement technology.
Key Innovation: Experimental and numerical analysis using 3D-printed flow cells demonstrating that rough interfaces promote greater bacterial adsorption and CaCO3 precipitation due to reduced flow velocities and increased surface area, leading to larger and more densely packed crystals, thus enhancing MICP effectiveness.
88. Effect of particle size on migration and retention of bacteria in sand and its biomineralization
Core Problem: Optimizing MICP-based in-situ biotreatment for soil stabilization requires understanding how soil particle size affects the migration and retention of bacteria and subsequent biomineralization, which is crucial for uniform and effective treatment.
Key Innovation: Experimental investigation demonstrating that sand particle size significantly impacts bacterial migration and retention, with sands smaller than 0.25 mm inhibiting migration and sands larger than 1.18 mm being unfavorable for retention, thereby affecting the uniformity and efficiency of calcium carbonate precipitation and strength enhancement.
89. Hydraulic fracturing of reservoirs containing rough discrete fracture networks: FDEM-UPM approach
Core Problem: Hydraulic fracturing mechanisms under the influence of fracture morphology in rough discrete fracture networks remain largely unexplored.
Key Innovation: Proposed a new hydro-mechanical (HM) coupling approach (FDEM-UPM) for modeling hydraulic fracturing, incorporating a Fourier-based methodology for reconstructing non-planar fractures. Showed that fracture morphology affects macroscopic fracture networks and micro-interactions, influencing initiation azimuth, propagation direction, and microcracking.
90. A bio-healing method for underground long rock fractures with high bridging rate
Core Problem: Ensuring prevention of unwanted fluid leakage through rock fracture networks in underground rock reservoirs, particularly for long fractures, with existing MICP methods having limitations in bridging rate, mechanical strength, and homogeneity.
Key Innovation: Proposed a new 'three-stage' injection strategy-based MICP (TS-MICP) bio-healing method. Achieved significantly improved bridging rate (32.1%–89.5%), mechanical properties (0.138–1.023 MPa), and homogeneity of CaCO3 precipitation, along with higher material utilization and reduced cementation solution consumption.
91. Influence of bacterial concentration and fissure aperture on improving dynamic mechanical properties of MICP repaired fissured sandstone
Core Problem: Fissured rocks compromise the stability of engineering structures, and optimizing eco-friendly repair methods like Microbial-induced carbonate precipitation (MICP) is needed.
Key Innovation: Investigated the optimal MICP scheme and the effects of bacterial concentration and fissure aperture on the dynamic mechanical properties of repaired fissured sandstone, showing that lower bacterial concentrations and smaller fissure apertures lead to more stable calcite crystals, uniform calcium carbonate distribution, and enhanced cementitious properties.
92. Characterizing the dynamic behavior and progressive damage evolution of carbonaceous slate under cyclic impact loading
Core Problem: Understanding the dynamic characteristics, energy evolution, and damage progression of tunnel rock (specifically carbonaceous slate) under repeated blast loading is crucial for tunnel stability.
Key Innovation: Conducted cyclic dynamic tests on carbonaceous slate using a triaxial split Hopkinson pressure bar, revealing how impact intensity and number influence peak stress/strain, energy evolution, and macro-to-micro failure characteristics, providing an experimental basis for soft rock tunnel stability analysis.
93. DMamba: Decomposition-enhanced Mamba for Time Series Forecasting
Core Problem: Existing Mamba-based architectures for long-term time series forecasting struggle with datasets characterized by non-stationary patterns because they do not differentiate between the statistical nature of inter-variable relationships in trend and seasonal components.
Key Innovation: Proposes DMamba, a novel forecasting model that employs seasonal-trend decomposition and processes components with specialized, differentially complex modules: a variable-direction Mamba encoder for seasonal dynamics and an MLP for trend relationships, achieving new state-of-the-art performance.
94. Importance inversion transfer identifies shared principles for cross-domain learning
Core Problem: Existing transfer-learning methodologies struggle to bridge radically heterogeneous scientific systems, especially under severe data scarcity or stochastic noise, hindering the identification of shared organizational principles for cross-domain knowledge transfer.
Key Innovation: Formalizes Explainable Cross-Domain Transfer Learning (X-CDTL) and introduces the Importance Inversion Transfer (IIT) mechanism, which identifies domain-invariant structural anchors. This framework achieves significant performance gains in anomaly detection tasks across diverse networks, providing a principled paradigm for cross-disciplinary knowledge propagation.
95. Generalizing GNNs with Tokenized Mixture of Experts
Core Problem: Deployed graph neural networks (GNNs) face a fundamental tradeoff between improving stability and reducing reliance on shift-sensitive features, leading to an irreducible worst-case generalization floor under distribution shifts and perturbations.
Key Innovation: STEM-GNN, a pretrain-then-finetune framework with a mixture-of-experts encoder for diverse computation paths, a vector-quantized token interface to stabilize encoder-to-head signals, and a Lipschitz-regularized head to bound output amplification, achieving a stronger balance of robustness to shifts and corruptions.
96. Stabilizing Physics-Informed Consistency Models via Structure-Preserving Training
Core Problem: A key stability challenge in physics-constrained consistency training for solving partial differential equations (PDEs), where PDE residuals can drive the model toward trivial or degenerate solutions, degrading the learned data distribution.
Key Innovation: Introduces a structure-preserving two-stage training strategy that decouples distribution learning from physics enforcement by freezing the coefficient decoder during physics-informed fine-tuning. It also proposes a two-step residual objective that enforces physical consistency on refined, structurally valid generative trajectories, enabling stable, high-fidelity inference for PDEs.
97. Bridging the Modality Gap in Roadside LiDAR: A Training-Free Vision-Language Model Framework for Vehicle Classification
Core Problem: Current LiDAR-based vehicle classification methods for intelligent transportation systems (ITS) face scalability challenges due to reliance on supervised deep learning and manual annotation, and Vision-Language Models (VLMs) are limited by a modality gap between sparse 3D point clouds and dense 2D imagery.
Key Innovation: Proposes a training-free VLM framework that bridges the modality gap by adapting off-the-shelf VLMs for fine-grained truck classification, using a novel depth-aware image generation pipeline to transform sparse LiDAR scans into depth-encoded 2D visual proxies, achieving competitive accuracy with few examples.
98. FD-DB: Frequency-Decoupled Dual-Branch Network for Unpaired Synthetic-to-Real Domain Translation
Core Problem: Synthetic data for vision tasks suffer from domain shift when applied to real-world scenarios, leading to a trade-off between photorealism and structural stability in unpaired synthetic-to-real translation.
Key Innovation: FD-DB, a frequency-decoupled dual-branch model, separates appearance transfer into low-frequency interpretable editing and high-frequency residual compensation, improving real-domain appearance consistency and boosting downstream semantic segmentation while preserving geometric and semantic structures.
99. Robust Depth Super-Resolution via Adaptive Diffusion Sampling
Core Problem: Conventional depth super-resolution methods directly regress depth values, leading to artifacts under severe or unknown degradation in low-resolution inputs.
Key Innovation: Proposes AdaDS, a generalizable framework for robust depth super-resolution that uses adaptive diffusion sampling. It adaptively selects a starting timestep in the reverse diffusion trajectory and injects tailored noise, ensuring robustness and superior zero-shot generalization.
100. RAD: Retrieval-Augmented Monocular Metric Depth Estimation for Underrepresented Classes
Core Problem: Accurate Monocular Metric Depth Estimation (MMDE) for underrepresented classes in complex scenes remains a persistent challenge, limiting physically intelligent systems.
Key Innovation: RAD, a retrieval-augmented framework, approximates multi-view stereo benefits by utilizing retrieved neighbors as structural geometric proxies, employing an uncertainty-aware retrieval mechanism, a dual-stream network, and a matched cross-attention module for geometric information transfer.
101. Reason-IAD: Knowledge-Guided Dynamic Latent Reasoning for Explainable Industrial Anomaly Detection
Core Problem: Industrial anomaly detection demands precise reasoning over fine-grained defect patterns, but existing multimodal large language models (MLLMs) pretrained on general-domain data often struggle to capture category-specific anomalies, limiting accuracy and interpretability.
Key Innovation: Reason-IAD, a knowledge-guided dynamic latent reasoning framework for explainable industrial anomaly detection, comprising a retrieval-augmented knowledge module for context-aware reasoning and an entropy-driven latent reasoning mechanism with dynamic visual injection for confident and stable predictions.
102. Position: Message-passing and spectral GNNs are two sides of the same coin
Core Problem: The common division of Graph Neural Networks (GNNs) into message-passing neural networks (MPNNs) and spectral GNNs is artificial and hinders progress, as it obscures their fundamental similarities and complementary strengths.
Key Innovation: The paper proposes a unifying viewpoint where both MPNNs and spectral GNNs are understood as different parametrizations of permutation-equivariant operators. It argues that many popular architectures are equivalent in expressive power and that progress will be accelerated by understanding their similarities and differences, working towards a common theoretical and conceptual framework.
103. Conformal Prediction Sets for Instance Segmentation
Core Problem: Current instance segmentation models achieve high average performance but lack principled uncertainty quantification, meaning their outputs are not calibrated and lack guarantees about closeness to ground truth.
Key Innovation: Proposes a conformal prediction algorithm to generate adaptive confidence sets for instance segmentation, providing a provable guarantee for the probability that at least one prediction has high Intersection-Over-Union (IoU) with the true object instance mask. It shows application in agricultural field delineation, cell segmentation, and vehicle detection.
104. Vendi Novelty Scores for Out-of-Distribution Detection
Core Problem: Existing post-hoc Out-of-Distribution (OOD) detectors typically rely on model confidence scores or likelihood estimates under restrictive distributional assumptions, limiting their robustness for safe deployment of machine learning systems.
Key Innovation: Introduces the Vendi Novelty Score (VNS), an OOD detector based on Vendi Scores, which quantifies how much a test sample increases the diversity of the in-distribution feature set. VNS is linear-time, non-parametric, combines local and global novelty signals, and achieves state-of-the-art OOD detection performance even with limited training data.
105. Continual Learning for non-stationary regression via Memory-Efficient Replay
Core Problem: Traditional offline models become outdated in dynamic, non-stationary data environments, and most continual learning research focuses on classification, leaving a gap for regression tasks.
Key Innovation: Proposes the first prototype-based generative replay framework for online task-free continual regression, using an adaptive output-space discretization model for memory-efficient replay without raw data storage, reducing forgetting and improving stability.
106. Stabilized Maximum-Likelihood Iterative Quantum Amplitude Estimation for Structural CVaR under Correlated Random Fields
Core Problem: Accurate and computationally efficient evaluation of Conditional Value-at-Risk (CVaR) in stochastic structural mechanics, especially under high-dimensional, spatially correlated material uncertainty, which is computationally prohibitive for classical Monte Carlo methods.
Key Innovation: Develops a quantum-enhanced inference framework that casts CVaR evaluation as a statistically consistent, confidence-constrained maximum-likelihood amplitude estimation problem, extending iterative quantum amplitude estimation (IQAE) with a stabilized inference scheme to preserve quadratic oracle-complexity advantage while providing finite-sample confidence guarantees and reduced estimator variance for tail-risk quantification in stochastic continuum mechanics.
107. ConjNorm: Tractable Density Estimation for Out-of-Distribution Detection
Core Problem: Existing post-hoc out-of-distribution (OOD) detection methods may fail to accurately reflect true data density or impose impractical constraints, lacking a unified perspective on density-based score design.
Key Innovation: Proposes ConjNorm, a novel theoretical framework grounded in Bregman divergence for density-based score design, which reframes density function design as searching for an optimal norm coefficient and uses an unbiased, tractable Monte Carlo estimator for the partition function, achieving state-of-the-art OOD detection performance.
108. RS-Agent: Automating Remote Sensing Tasks through Intelligent Agent
Core Problem: Existing Multimodal Large Language Models (MLLMs) are limited to basic instruction-following and descriptive tasks, struggling with complex real-world remote sensing applications that require specialized tools and knowledge.
Key Innovation: Proposes RS-Agent, an AI agent designed to autonomously leverage specialized models for complex remote sensing tasks. It integrates a Central Controller, dynamic toolkit, Solution Space, and Knowledge Space, introducing Task-Aware Retrieval and DualRAG to improve tool selection and knowledge relevance, significantly outperforming SOTA MLLMs.
109. TabNSA: Native Sparse Attention for Efficient Tabular Data Learning
Core Problem: Effectively and efficiently modeling tabular data, which presents challenges due to heterogeneous feature types, lack of spatial structure, and often limited sample sizes.
Key Innovation: TabNSA, a novel deep learning framework that integrates Native Sparse Attention (NSA) with a TabMixer backbone to dynamically focus on relevant feature subsets, significantly reducing computational complexity and capturing complex dependencies, outperforming state-of-the-art models on tabular data.
110. Ice-FMBench: A Foundation Model Benchmark for Sea Ice Type Segmentation
Core Problem: Accurate and automated segmentation of sea ice types from Synthetic Aperture Radar (SAR) imagery is challenging due to unique sea ice characteristics and SAR artifacts, and the direct transferability of general foundation models (FMs) to polar SAR data is uncertain.
Key Innovation: Introduction of IceFMBench, a comprehensive benchmark framework for evaluating remote sensing FMs on sea ice type segmentation using Sentinel-1 SAR imagery, including a standardized dataset, diverse metrics, and representative models, along with an extensive comparative evaluation and a multi-teacher knowledge distillation approach to improve spatiotemporal transferability.
111. LD-ViCE: Latent Diffusion Model for Video Counterfactual Explanations
Core Problem: Interpreting decisions of video-based AI systems in safety-critical domains is challenging due to spatiotemporal complexity and model opacity, with existing counterfactual methods lacking temporal coherence, causal insights, and target model guidance.
Key Innovation: Proposes LD-ViCE, a Latent Diffusion Model for Video Counterfactual Explanations, which operates in latent space for computational efficiency, produces realistic and interpretable counterfactuals with an additional refinement step, and incorporates target model guidance for semantic fidelity.
112. Machine Learning Detection of Road Surface Conditions: A Generalizable Model using Traffic Cameras and Weather Data
Core Problem: Transportation agencies need better, automated, and generalizable methods to assess road conditions during hazardous weather for operational decisions and resource allocation.
Key Innovation: Developed machine learning models (CNNs and random forests) trained on roadside camera images and weather data to classify six road surface conditions, achieving 81.5% accuracy on unseen cameras, prioritizing generalizability for operational deployment.
113. OpenMonoGS-SLAM: Monocular Gaussian Splatting SLAM with Open-set Semantics
Core Problem: Existing SLAM methods that integrate semantic understanding often rely on depth sensors or closed-set semantic models, limiting their scalability and adaptability in open-world environments for intelligent perception and interaction.
Key Innovation: OpenMonoGS-SLAM, the first monocular SLAM framework that unifies 3D Gaussian Splatting (3DGS) with open-set semantic understanding. It leverages Visual Foundation Models (VFMs) like MASt3R, SAM, and CLIP for robust monocular camera tracking, mapping, and open-vocabulary semantics without depth input or 3D semantic ground truth, and includes a memory mechanism for high-dimensional semantic features.
114. Patch-Level Tokenization with CNN Encoders and Attention for Improved Transformer Time-Series Forecasting
Core Problem: The effectiveness of Transformer-based time-series forecasting models is critically dependent on the quality and structure of input representations, especially as sequence length and data scale increase, requiring better methods to model both local and global temporal dependencies.
Key Innovation: A two-stage forecasting framework that explicitly separates local temporal representation learning (using CNNs on fixed-length patches with token-level self-attention) from global dependency modeling (using a Transformer encoder), leading to improved performance and scalability for multivariate time-series forecasting.
115. UAV Coverage Path Optimization for Dynamic Wildlife Population Monitoring: A Spotted Seal Case Study
Core Problem: Existing predefined UAV coverage path strategies are inadequate for monitoring wildlife populations with dynamic group distributions, leading to inefficient data collection and unreliable detection results.
Key Innovation: Developed a novel real-time, perception-driven adaptive UAV coverage path planning approach integrating onboard target detection, multi-object tracking, and visual positioning. Introduced a spatial noise filtering (SNF) method to mitigate false detections and redetections, and combined it with a genetic algorithm for dynamic path optimization, significantly improving the reliability and efficiency of monitoring dynamic populations.
116. Field measurements of hydrodynamics and sediment transport at intertidal areas in the Dutch Wadden Sea
Core Problem: Lack of detailed field measurements to understand and quantify sediment transport and exchange mechanisms in intertidal areas, particularly regarding sediment dynamics and bed stability.
Key Innovation: Provides a comprehensive dataset of field measurements (flow velocities, wave characteristics, suspended sediment concentrations, bed level dynamics, sediment composition) collected over several weeks in the Dutch Wadden Sea, supporting research into fundamental processes controlling sediment dynamics and channel-shoal sediment exchange.
117. Source-to-sink sediment transport and geomorphic controls along large river systems
Core Problem: Understanding complex sediment transport pathways in large river systems is critical for accurately interpreting terrestrial signals preserved in marginal-sea sediments, as characteristics can differ between upstream and estuarine sediments.
Key Innovation: Used a multi-proxy provenance approach on Limpopo River sediments to show that steep middle reaches are a primary source for marginal-sea sediments. Identified two sediment-routing patterns linked to longitudinal-profile morphology (stepped vs. concave-up), providing new insights into source-to-sink relationships.
118. Investigation on new definition and proactive identification method of hazard
Core Problem: Traditional hazard identification methods treat hazards as isolated or static, relying on historical data and linear logic, making it difficult to address dynamic interaction risks in complex systems and lacking practical applicability despite modern system safety theories.
Key Innovation: Redefines hazards as 'safety structures that may lose stability' based on interactions between atomic activity and environment factors, and proposes a proactive identification method with specific criteria, demonstrated in a coal mine production system.
119. A data-driven system-theoretic Bayesian network framework for probabilistic safety assessment of passenger vessels
Core Problem: Existing risk models for passenger vessels are localized, rely on subjective expert elicitation, and fail to adequately address the paradox of decreasing incident frequency but persistent catastrophic severity.
Key Innovation: A data-driven risk assessment framework synergizing System-Theoretic Process Analysis (STPA) with a Bayesian Network (BN), using a novel database of accident reports to define causal topology, identify latent systemic precursors (e.g., Structural Failure, Defective Maintenance), and provide an objective foundation for proactive safety governance.
120. Automated identification of pilot failures in aviation accidents using a BERT-based classifier and topic modeling
Core Problem: The extensive records of aviation accident investigation reports offer valuable insights, but manual analysis is time-consuming. There's a need for efficient, automated methods to analyze these documents and identify key human factors (pilot failures) contributing to accidents.
Key Innovation: A novel methodology leveraging a BERT-based classifier combined with topic modeling to automate the labeling of accident datasets and identify specific pilot failures (e.g., skill-based errors, routine violations, perceptual errors), significantly reducing manual data preparation efforts and enhancing the precision and efficiency of aviation accident data analysis.
121. Predicting operators reliability for control room alarm management using knowledge-based Bayesian networks
Core Problem: Despite standards for industrial alarm management and existing human reliability studies, quantitative operator-centered reliability assessment within alarm management activities remains limited.
Key Innovation: A Bayesian network framework that integrates alarm response task decomposition, cognitive modeling, and contextual factors to assess alarm management reliability across perception, planning, and execution phases, identifying key drivers like operator experience and task complexity, and demonstrating how HRA principles can be extended to alarm management contexts.
122. AeroTrans: Hourly AOD retrieval over land from MSG-1/SEVIRI imagery integrating Transformer and transfer learning
Core Problem: The scarcity of publicly available, accurate hourly aerosol products over Europe, Africa, and West Asia, particularly due to limitations in satellite sensor capabilities (lack of shorter-wavelength channels on MSG).
Key Innovation: Developing 'AeroTrans,' a novel deep learning framework integrating a time-sequence Transformer architecture with transfer learning, to retrieve accurate hourly AOD at 550 nm over land from MSG-1/SEVIRI imagery, enabling the tracking of diurnal aerosol variations and rapid dispersion during highly polluted events like dust storms and wildfires.
123. Mapping soil total carbon using multisource remote sensing at the continental scale
Core Problem: The challenge of mapping soil total carbon (STC) at a continental scale due to significant spatial and methodological constraints, particularly the underutilization of diverse remote sensing data types beyond optical sensors.
Key Innovation: Developed a novel approach integrating optical, thermal infrared (TIR), and radar data from multiple satellite platforms with the WoSIS soil profile database and Random Forest algorithm on Google Earth Engine to map STC across Europe, achieving significantly enhanced model performance (R2 = 0.63) by synergistically combining these multisource remote sensing data with conventional environmental covariates.
124. Rainfall patterns and catchment characteristics interactively regulate nitrogen and phosphorus dynamics in agricultural ponds
Core Problem: How rainfall patterns interact with catchment characteristics to regulate nitrogen and phosphorus dynamics and transport into agricultural ponds, contributing to water quality threats.
Key Innovation: Demonstrated that rainfall amount and intensity are primary drivers of increased nutrient concentrations (especially particulate N and P) in agricultural ponds, with pond morphology, catchment slope, soil nutrients, and landscape patterns also playing significant interactive roles.
125. An intelligent recognition and classification method for TBM tunnel surrounding rock based on cross-attention transformer and multi-source data fusion
Core Problem: Existing methods for TBM tunnel surrounding rock classification often overlook ascending-stage dynamics, struggle with multi-source data fusion, and lack interpretability, which is crucial for ensuring TBM operation safety and efficiency.
Key Innovation: Proposes CA-Trans-XGBoost, a Cross-Attention Transformer with XGBoost, which extracts dynamic features using a Transformer encoder, models structured features with MLP, enhances feature interaction via cross-attention, and uses XGBoost for classification. This method achieves high accuracy (95.0%) and interpretability, providing support for intelligent TBM excavation and parameter optimization.
126. Monotonic triaxial testing and hypoplastic modelling of calcareous sand: A focus on particle breakage and initial relative density
Core Problem: The stress-strain behavior of calcareous sand is significantly influenced by particle breakage and initial relative density, but existing constitutive models often fail to consider their combined effects comprehensively.
Key Innovation: Development of a unified breakage evolution model and a hypoplastic model within the Critical State Soil Mechanics framework that successfully incorporates the combined effects of particle breakage and initial relative density on the mechanical behavior of calcareous sand, validated against experimental results.
127. Stress-path dependency of rock shear strength influenced by shear surface integrity: Experimental and numerical investigations
Core Problem: The shear strength of rock, crucial for stability, varies significantly with stress path and is reduced by defects, making accurate constitutive models challenging.
Key Innovation: Experimental and numerical investigations reveal that shear strength is stress-path dependent, with this dependence decreasing as shear surface integrity weakens. It provides insights into crack propagation and mechanical behavior under various stress paths, offering a basis for improved constitutive models.
128. Palaeoglacier reconstruction and dynamics of Cordillera Vilcanota in the tropical high Peruvian Andes
Core Problem: Uncertainty in the behaviors and drivers of tropical palaeoglaciers in the Cordillera Vilcanota due to limited detailed geomorphological studies.
Key Innovation: Provided a detailed geomorphological analysis and morphostratigraphic reconstruction of the Cordillera Vilcanota, mapping ~23,000 features and identifying seven clear ice margins reflecting at least seven palaeoglacier advances.
129. SemanticMoments: Training-Free Motion Similarity via Third Moment Features
Core Problem: Retrieving videos based on semantic motion is challenging because existing video representation approaches overly rely on static appearance and scene context rather than motion dynamics.
Key Innovation: SemanticMoments, a simple, training-free method that computes temporal statistics (higher-order moments) over features from pre-trained semantic models, consistently outperforms existing methods for motion-centric video understanding.
130. Wearable environmental sensing to forecast how legged systems will interact with upcoming terrain
Core Problem: The ability to predict how a foot will contact a changing environment (e.g., level ground to stairs) prior to foot-strike is underexplored, which is crucial for anticipatory control in assistive legged systems.
Key Innovation: Demonstrates the feasibility of forecasting anterior-posterior foot center-of-pressure (COP) and time-of-impact (TOI) using a lightweight CNN-RNN model with wearable RGB-D camera data, achieving reasonable accuracy for anticipatory control in assistive systems.
131. VLM-Guided Iterative Refinement for Surgical Image Segmentation with Foundation Models
Core Problem: Existing surgical image segmentation methods are constrained to predefined categories, produce one-shot predictions without adaptive refinement, and lack mechanisms for clinician interaction.
Key Innovation: IR-SIS, an iterative refinement system for surgical image segmentation that accepts natural language descriptions, leverages a fine-tuned SAM3, employs a Vision-Language Model to assess quality, and applies an agentic workflow for adaptive self-refinement, supporting clinician-in-the-loop interaction.
132. SnareNet: Flexible Repair Layers for Neural Networks with Hard Constraints
Core Problem: Neural networks, when used as surrogate solvers or control policies, can produce unconstrained predictions that violate critical physical, operational, or safety requirements, limiting their reliability in sensitive applications.
Key Innovation: SnareNet is introduced, a feasibility-controlled architecture that appends a differentiable repair layer to neural networks. This layer navigates the constraint map's range space to steer outputs toward feasibility, satisfying input-dependent nonlinear constraints to a user-specified tolerance. Adaptive relaxation is also introduced for stable end-to-end training.
133. MacrOData: New Benchmarks of Thousands of Datasets for Tabular Outlier Detection
Core Problem: Existing benchmarks for tabular outlier detection (OD), such as AdBench, are limited in scale (e.g., 57 datasets), severely restricting diversity and statistical power for fairly and accurately evaluating OD methods and informing methodological choices.
Key Innovation: MacrOData is introduced, a large-scale benchmark suite for tabular OD comprising three curated components: OddBench (790 datasets with semantic anomalies), OvrBench (856 datasets with statistical outliers), and SynBench (800 synthetic datasets). This suite provides standardized train/test splits, public/private partitions with held-out test labels, semantic metadata, and extensive evaluations of diverse OD methods.
134. Single-Slice-to-3D Reconstruction in Medical Imaging and Natural Objects: A Comparative Benchmark with SAM 3D
Core Problem: The high cost and long wait times for volumetric medical imaging, and the uncertainty of whether geometric priors learned by image-to-3D foundation models from natural images transfer effectively to medical data for single-slice-to-3D reconstruction.
Key Innovation: Presents a controlled zero-shot benchmark of five state-of-the-art single-slice medical image-to-3D reconstruction models across six medical and two natural datasets, quantifying the limits of single-slice medical reconstruction and highlighting SAM3D's strongest overall topological similarity to ground truth medical 3D data.
135. A Scoping Review of Deep Learning for Urban Visual Pollution and Proposal of a Real-Time Monitoring Framework with a Visual Pollution Index
Core Problem: Research on automatic detection and application of Urban Visual Pollution (UVP) using deep learning is fragmented, lacking standardized taxonomies, cross-city benchmark datasets, and integrated real-time application systems.
Key Innovation: Provides a scoping review of deep learning-based approaches for UVP, highlighting current limitations, and proposes a real-time monitoring framework that integrates a visual pollution index to assess severity, advocating for a unified UVP management system.
136. Beyond Next-Token Alignment: Distilling Multimodal Large Language Models via Token Interactions
Core Problem: Multimodal Large Language Models (MLLMs) are substantial in size, posing deployment challenges, and existing knowledge distillation methods overlook dynamic token interactions crucial for multimodal understanding.
Key Innovation: Align-TI, a novel knowledge distillation framework, compresses MLLMs by focusing on vision-instruction and intra-response token interactions, leading to more parameter-efficient MLLMs with superior performance compared to vanilla KD and even larger models.
137. Equilibrium contrastive learning for imbalanced image classification
Core Problem: Existing supervised contrastive learning methods for imbalanced datasets suffer from poor generalization due to a lack of alignment between class means/prototypes and classifiers, and unbalanced prototype contributions.
Key Innovation: Proposes Equilibrium Contrastive Learning (ECL), a supervised CL framework that promotes geometric equilibrium by balancing class features, means, and classifiers, and aligning classifier weights with class prototypes, demonstrating superior performance on imbalanced datasets.
138. Time2General: Learning Spatiotemporal Invariant Representations for Domain-Generalization Video Semantic Segmentation
Core Problem: Domain shift and temporal-sampling shift in Domain Generalized Video Semantic Segmentation (DGVSS) lead to inconsistent and flickering predictions across unseen domains and varying temporal sampling rates.
Key Innovation: Proposes Time2General, a DGVSS framework with a Spatio-Temporal Memory Decoder for consistent per-frame masks and a Masked Temporal Consistency Loss to regularize temporal prediction discrepancies, improving cross-domain accuracy and temporal stability.
139. GenSeg-R1: RL-Driven Vision-Language Grounding for Fine-Grained Referring Segmentation
Core Problem: Fine-grained referring image segmentation requires robust vision-language grounding to accurately identify and segment objects based on natural language queries, often lacking supervised reasoning-chain annotations.
Key Innovation: Proposes GenSeg-R1, an RL-driven framework that fine-tunes vision-language models to emit structured spatial prompts for a promptable segmenter, achieving high-quality fine-grained referring segmentation without supervised reasoning-chain annotations and demonstrating strong performance on various benchmarks.
140. Physics-informed diffusion models in spectral space
Core Problem: Existing diffusion-based PDE solvers struggle with accuracy and computational efficiency, especially for sparse observations, and lack a robust way to incorporate physics-informed constraints and ensure solution regularity.
Key Innovation: Proposes a methodology combining latent diffusion models with physics-informed machine learning in a spectral space, enabling significant dimensionality reduction, ensuring solution regularity, and enforcing physics-informed constraints during inference. This approach achieves improved accuracy and computational efficiency for various PDEs.
141. BRAVA-GNN: Betweenness Ranking Approximation Via Degree MAss Inspired Graph Neural Network
Core Problem: Computing betweenness centrality, a key measure of node importance, is computationally prohibitive for large-scale networks, and existing GNN-based methods for predicting node rankings fail to generalize to high-diameter graphs like road networks.
Key Innovation: Proposes BRAVA-GNN, a lightweight GNN architecture that approximates betweenness centrality by leveraging multi-hop degree mass as size-invariant node features and using hyperbolic random graph models for training. This design enables generalization across diverse graph families, achieving significant improvements in accuracy and speedup, particularly on challenging road networks, with significantly fewer parameters.
142. CompSplat: Compression-aware 3D Gaussian Splatting for Real-world Video
Core Problem: High-quality novel view synthesis from real-world videos is challenged by long sequences, irregular camera trajectories, unknown poses leading to pose drift, feature misalignment, geometric distortion, and amplified issues from lossy compression.
Key Innovation: CompSplat, a compression-aware training framework that explicitly models frame-wise compression characteristics to mitigate inter-frame inconsistency and accumulated geometric errors, incorporating compression-aware frame weighting and adaptive pruning to enhance robustness and geometric consistency.
143. Statistical benchmarking of transformer models in low signal-to-noise time-series forecasting
Core Problem: Evaluating and improving transformer architectures for multivariate time-series forecasting in challenging low-data and low signal-to-noise environments.
Key Innovation: Statistical benchmarking demonstrating that two-way attention transformers can outperform standard baselines in low signal-to-noise regimes, and introducing a dynamic sparsification procedure for attention matrices that is effective in noisy environments.
144. Monocular Normal Estimation via Shading Sequence Estimation
Core Problem: Existing monocular normal estimation methods suffer from 3D misalignment because they directly predict normal maps, struggling to reconstruct varying geometry from subtle color variations in a single RGB image.
Key Innovation: RoSE, a new paradigm that reformulates monocular normal estimation as shading sequence estimation, leveraging image-to-video generative models and a synthetic dataset (MultiShade) to predict shading sequences, which are then converted into accurate normal maps, achieving state-of-the-art performance.
145. Supervised Metric Regularization Through Alternating Optimization for Multi-Regime Physics-Informed Neural Networks
Core Problem: Standard Physics-Informed Neural Networks (PINNs) struggle to model parameterized dynamical systems with sharp regime transitions (e.g., bifurcations), leading to spectral bias or 'mode collapse' where the network averages distinct physical behaviors.
Key Innovation: Topology-Aware PINN (TAPINN) that mitigates spectral bias by structuring the latent space via Supervised Metric Regularization, conditioning the solver on a latent state optimized to reflect metric-based separation between regimes, and trained using a phase-based Alternating Optimization schedule to manage gradient conflicts.
146. Faster-GS: Analyzing and Improving Gaussian Splatting Optimization
Core Problem: Accelerating the optimization process of 3D Gaussian Splatting (3DGS) while maintaining or improving reconstruction quality, as the current research landscape is fragmented with entangled improvements and performance/fidelity tradeoffs.
Key Innovation: The paper consolidates and augments effective strategies from prior 3DGS research, introduces novel optimizations, and investigates underexplored aspects. The resulting system, Faster-GS, achieves up to 5x faster training while maintaining visual quality and can be applied to 4D Gaussian reconstruction, establishing a new cost-effective baseline.
147. Towards Explainable Federated Learning: Understanding the Impact of Differential Privacy
Core Problem: Achieving a machine learning model that combines enhanced data privacy (Federated Learning with Differential Privacy) with explainability, especially given that DP can harm system explainability.
Key Innovation: FEXT-DP (Federated EXplainable Trees with Differential Privacy), a FL solution based on Decision Trees that incorporates DP, demonstrating improvements in training speed and MSE, and providing insights into DP's impact on explainability.
148. Quantum Multiple Rotation Averaging
Core Problem: Established classical Multiple Rotation Averaging (MRA) methods face limitations including local minima susceptibility and reliance on convex relaxations that fail to preserve exact manifold geometry, leading to reduced accuracy in high-noise scenarios.
Key Innovation: IQARS (Iterative Quantum Annealing for Rotation Synchronization), the first algorithm to reformulate MRA as a sequence of local quadratic non-convex sub-problems executable on quantum annealers, removing convex relaxation dependence and better preserving non-Euclidean rotation manifold geometry.
149. Toeplitz Based Spectral Methods for Data-driven Dynamical Systems
Core Problem: Accurately performing data-driven spectral estimation of linear evolution operators in dynamical systems, especially from equilibrium trajectories without access to the underlying equations of motion.
Key Innovation: Introduces a Toeplitz-based framework that applies Toeplitz filters to the infinitesimal generator to extract eigenvalues, eigenfunctions, and spectral measures, incorporating structural prior knowledge and demonstrating improved recovery of spectral properties compared to standard data-driven methods.
150. Deep Learning-Based Object Pose Estimation: A Comprehensive Survey
Core Problem: A comprehensive and up-to-date survey discussing recent progress, outstanding challenges, and promising future directions in deep learning-based object pose estimation is missing.
Key Innovation: Provides a comprehensive survey of recent advances in deep learning-based object pose estimation, covering different problem formulations, input modalities, training paradigms, and applications, while identifying key challenges and promising future research directions.
151. BiSSL: Enhancing the Alignment Between Self-Supervised Pretraining and Downstream Fine-Tuning via Bilevel Optimization
Core Problem: Models initialized from self-supervised pretraining often suffer from poor alignment with downstream tasks, limiting the effectiveness of subsequent fine-tuning in adapting pretrained features.
Key Innovation: Introduces BiSSL, a novel bilevel training framework that acts as an intermediate stage after conventional self-supervised pretraining. It solves a bilevel optimization problem that explicitly models the interdependence between pretext and downstream objectives, facilitating enhanced information sharing and better alignment with downstream tasks.
152. Constant Rate Scheduling: A General Framework for Optimizing Diffusion Noise Schedule via Distributional Change
Core Problem: Optimizing noise schedules in diffusion models for both training and sampling is crucial for performance, but a general and effective framework is needed.
Key Innovation: Proposes a general framework for optimizing diffusion noise schedules by enforcing a constant rate of change in the probability distribution of diffused data, quantified using user-defined discrepancy measures. This approach consistently improves the performance of both pixel-space and latent-space diffusion models across various datasets and samplers.
153. Wandering around: A bioinspired approach to visual attention through object motion sensitivity
Core Problem: Developing an efficient, real-time, and robust visual attention system for dynamic environments, particularly for distinguishing moving objects while the camera itself is in motion, to reduce computational demand.
Key Innovation: A bioinspired attention system using a Spiking Convolutional Neural Network with event-based cameras and neuromorphic hardware to achieve selective attention through object motion sensitivity, demonstrating high accuracy in multi-object motion segmentation and real-time response for robotic applications.
154. UFM: A Simple Path towards Unified Dense Correspondence with Flow
Core Problem: Dense image correspondence, crucial for applications like 3D reconstruction, is inefficiently tackled by separate approaches for wide-baseline scenarios and optical flow estimation, despite their common goal.
Key Innovation: Develops UFM, a Unified Flow & Matching model, which uses a simple generic transformer architecture trained on unified data to directly regress (u,v) flow, significantly outperforming specialized state-of-the-art methods in both optical flow and wide-baseline matching.
155. Scalable Dynamic Origin-Destination Demand Estimation Enhanced by High-Resolution Satellite Imagery Data
Core Problem: Dynamic origin-destination demand estimation (DODE) in multi-class mesoscopic network models is limited by the sparse availability of conventional traffic data from local sensors.
Key Innovation: Presents an integrated framework for DODE that incorporates high-resolution satellite imagery (via a computer vision pipeline for vehicle detection and map matching) to provide consistent, city-wide road and traffic density observations, significantly improving estimation performance, especially for links without local sensors.
156. AFABench: A Generic Framework for Benchmarking Active Feature Acquisition
Core Problem: Lack of standardized benchmarks hinders fair and systematic evaluation of various Active Feature Acquisition (AFA) methods, which are crucial for scenarios where acquiring all data features is expensive.
Key Innovation: Introduces AFABench, the first generic benchmark framework for AFA, including diverse datasets, support for various acquisition policies, and a modular design, along with a novel synthetic dataset (CUBE-NM) to test lookahead capabilities.
157. ERTACache: Error Rectification and Timesteps Adjustment for Efficient Diffusion
Core Problem: Diffusion models suffer from substantial computational overhead during inference due to iterative processes, and naive feature caching often degrades quality.
Key Innovation: Introduces ERTACache, a principled caching framework that rectifies cumulative errors by employing offline residual profiling, dynamically adjusting integration intervals, and analytically approximating cache-induced errors, achieving up to 2x speedup with preserved quality.
158. GenTrack2: An Improved Hybrid Approach for Multi-Object Tracking
Core Problem: Ensuring identifier consistency for unknown and time-varying target numbers under nonlinear dynamics in visual multi-object tracking remains a challenge.
Key Innovation: Proposes GenTrack2, a hybrid multi-object tracking method combining a stochastic particle filter (supported by PSO) and deterministic association, with novel schemes for state updating and velocity regression, achieving superior performance for both pre-recorded videos and live streams.
159. Q-DiT4SR: Exploration of Detail-Preserving Diffusion Transformer Quantization for Real-World Image Super-Resolution
Core Problem: Diffusion Transformers (DiTs) for Real-World Image Super-Resolution (Real-ISR) have a heavy inference burden, and existing quantization methods for super-resolution or generic DiTs lead to severe degradation of local textures when applied to DiT-based Real-ISR.
Key Innovation: Proposes Q-DiT4SR, the first Post-Training Quantization (PTQ) framework specifically tailored for DiT-based Real-ISR, which integrates Hierarchical SVD (H-SVD) and Variance-aware Spatio-Temporal Mixed Precision (VaSMP/VaTMP) to achieve state-of-the-art performance while significantly reducing model size and computational operations.
160. Aggregation Models with Optimal Weights for Distributed Gaussian Processes
Core Problem: The computational burdens of Gaussian Process (GP) models for large-scale datasets and the inefficiency of current aggregation models for distributed GPs in effectively incorporating correlations between GP experts.
Key Innovation: A novel approach for aggregated prediction in distributed GPs that incorporates correlations among experts, leading to better prediction accuracy and more stable predictions in less time compared to state-of-the-art consistent aggregation models, suitable for both exact and sparse variational GPs.
161. A Physics-Informed Spatiotemporal Deep Learning Framework for Turbulent Systems
Core Problem: Direct numerical simulations (DNS) of fluid thermodynamic systems, such as Rayleigh-Benard convection, are computationally prohibitive for long-term simulations.
Key Innovation: Presents a novel physics-informed spatiotemporal deep learning surrogate model combining convolutional neural networks and a recurrent architecture, penalized by governing partial differential equations, to replicate turbulent dynamics with significantly reduced computational cost and quantified uncertainty.
162. A Nonparametric Discrete Hawkes Model with a Collapsed Gaussian-Process Prior
Core Problem: Existing discrete-time Hawkes models are often constrained by fixed-form baselines and excitation kernels, lacking flexible, nonparametric treatments for both components, making it difficult to capture diverse event dynamics without prespecifying trends or decay shapes.
Key Innovation: Proposes the Gaussian Process Discrete Hawkes Process (GP-DHP), a nonparametric framework that places Gaussian process priors on both the baseline and excitation. It performs inference through a collapsed latent representation, yielding smooth, data-adaptive structure, near-linear-time MAP estimation, and interpretable baseline/excitation functions, improving predictive log-likelihood in case studies.
163. SpikySpace: A Spiking State Space Model for Energy-Efficient Time Series Forecasting
Core Problem: Time-series forecasting on resource-limited edge devices requires real-time, energy-efficient processing, and existing Spiking Neural Network (SNN) based forecasters often use computationally complex transformer blocks.
Key Innovation: SpikySpace, the first fully spiking state-space model, which reduces quadratic computational cost to linear time via spiking selective scanning and introduces efficient bit-shift approximations (PTsoftplus, PTSiLU) for activation functions, achieving higher accuracy and over 96.1% energy reduction.
164. Spark: Modular Spiking Neural Networks
Core Problem: Present neural network models are inefficient in terms of data and energy, and effective learning algorithms for spiking neural networks (SNNs) remain elusive, hindering their adoption despite their potential for efficient hardware implementations.
Key Innovation: Spark, a new framework for spiking neural networks built upon modular design, providing an efficient and streamlined pipeline compatible with traditional ML pipelines, demonstrated by solving the sparse-reward cartpole problem with simple plasticity mechanisms.
165. Formation mechanism and earth pressure quantification of internal soil arching during caisson penetration
Core Problem: Accurate evaluation of penetration resistance and understanding the soil arching effect during caisson penetration is crucial to avoid installation failure, but the mechanism and its impact on earth pressure are not fully quantified.
Key Innovation: Investigates the internal soil arching mechanism through experiments and simulations, proposing a calculation method for internal earth pressure that explicitly incorporates this effect and establishing a criterion for critical arching initiation depth.
166. A hybrid non-linear observer for maritime surface vessel state estimation using physics-based and data-driven modelling
Core Problem: Improving the accuracy of state estimation for maritime surface vessels, which is crucial for overall control system performance, beyond what conventional physics-based non-linear observers can achieve.
Key Innovation: Proposal of a hybrid state observer combining a physics-based control model with a data-driven deep neural network correction model, demonstrating a 30% reduction in estimation error compared to a fixed-gain physics-based observer, thereby enhancing state estimation accuracy for maritime vessels.
167. Tree rings and salt lakes give clues about ancient rainfall
Core Problem: Reconstructing past climate changes, specifically ancient rainfall patterns, to understand historical climate variability.
Key Innovation: Utilizing tree rings and salt lakes as proxies to provide valuable clues about ancient rainfall and broader past climate changes.
168. Correction: Sedimentary architecture of gravity flow deposits in the liushagang formation of the paleogene, weixinan sag, beibu gulf basin, South China sea
Core Problem: Understanding the sedimentary architecture and formation of gravity flow deposits in the Liushagang Formation.
Key Innovation: A correction to previous findings regarding the detailed sedimentary architecture of gravity flow deposits, refining the understanding of their formation and distribution in the specified geological context.
169. Competition between thermoelastic process and mineral reaction on fracture flow channeling: Implications for long-term thermal performance of EGS reservoirs
Core Problem: The dynamic evolution of fracture flow channeling in Enhanced Geothermal Systems (EGSs) under complex thermo-hydro-mechanical-chemical (THMC) coupled processes remains underexplored, impacting thermal extraction performance.
Key Innovation: Developed a 3D field-scale THMC coupled EGS model to systematically investigate how thermoelastic processes (intensifying channeling) and mineral dissolution (leading to flow dispersion) compete and interact, providing insights for optimizing EGS reservoir selection and injection strategies.
170. Stochastic modeling of crack branching under uncertainties: A degradation branching framework
Core Problem: Existing degradation branching models often understate total degradation by assuming a single branch initiation per event and lack a comprehensive framework for dynamic fracture branching propagation under various uncertainties.
Key Innovation: Proposes a stochastic model (DFBPU) for dynamic fracture branching propagation under uncertainties, generalizing existing models by allowing random numbers of initial and offspring branches, generation-dependent crack growth rates, and random branching times, and deriving statistical properties of total degradation.
171. A privacy-enhanced multi-party industrial control systems collaboration framework
Core Problem: Industrial control systems (ICS) generate vast operational data, but effective utilization for intelligent analysis is hindered by resource limitations of devices and significant privacy concerns, preventing secure collaboration across multiple ICS clients.
Key Innovation: A distributed, reliability-enhanced collaborative training framework (IEEB) integrating federated learning, edge computing, and blockchain to enable secure, privacy-enhanced multi-party collaboration in ICS, improving model accuracy, reducing training time, and maintaining robustness against malicious clients.
172. Quantitative kinematic reconstruction of the Tibetan-Himalayan Orogen since 130 Ma
Core Problem: Understanding the kinematic evolution of the Tibetan-Himalayan Orogen, the largest region of active continental deformation, to comprehend convergence dynamics, topography development, and climate changes.
Key Innovation: A new quantitative tectonic reconstruction of the Tibetan-Himalayan Orogen since the Cretaceous, using high-resolution oceanic spreading records and a thorough review of structural geology, sedimentary provenance, and paleomagnetism data, explicitly resolving continental deformation.
173. Land-use change alters soil nitrogen supply potential and nitrogen-cycling functional genes on China's Loess Plateau
Core Problem: The mechanistic links between microbial functional potential and soil nitrogen supply following land-use change (grassland to afforested/abandoned cropland) on the Loess Plateau are poorly quantified.
Key Innovation: Showed that land-use change significantly alters soil N supply potential and N-cycling functional genes, with afforestation enhancing Nmin and promoting genes for nitrate reduction, nitrification, and N assimilation, driven by both abiotic conditions and fungal communities.
174. Heatwaves only partially offset increased water consumption from earlier greening: the case of the Kashi Basin in Central Asia
Core Problem: The relative contributions of earlier green-up dates (GUD) driven by climate warming and reduced evapotranspiration (ET) during summer heatwaves to overall water consumption and streamflow in semi-arid basins remain unclear.
Key Innovation: Integrates satellite phenology, reanalysis heatwave indices, and a physically based eco-hydrological model to disentangle the effects of earlier GUD and heatwaves on ET and streamflow. Demonstrates that heatwave-related ET savings only partially compensate (approx. 10%) for the increased water consumption linked to earlier green-up, leading to overall water depletion.
175. Permeability prediction of porous media with non-circular pores using a modified Kozeny–Carman equation enhanced by fractal theory
Core Problem: The accuracy of the Kozeny–Carman (K–C) equation for permeability prediction in geotechnical engineering is compromised by its assumption of circular pore shapes, and the influence of structural parameters needs clarification.
Key Innovation: Derives a novel fractal analytical model for permeability and the K–C constant that incorporates irregular pore shapes and contains no empirical constants, enhancing prediction accuracy and revealing fluid transport mechanisms by relating permeability to key structural parameters.
176. A similarity model for subgrade compaction from collaborative laboratory-field tests
Core Problem: Intelligent Compaction (IC) model development for accurate quality evaluation is challenged by the absence of sufficient field data.
Key Innovation: This study proposed a similarity model based on coordinated laboratory and field tests, establishing a spatio-temporal equivalence principle to correlate laboratory compaction time and rolling passes. This method effectively expands IC datasets, leading to significantly reduced prediction error and improved generalization for compaction quality evaluation, enhancing assessment reliability.
177. An enhanced segmentation method for 3D point cloud of tunnel support system using PointNet++ and coverage-voted strategy algorithms
Core Problem: Segmenting large 3D point cloud datasets of tunnel support systems, especially with multi-scale targets in coal mine roadways, remains challenging for accurate structural evaluation.
Key Innovation: Proposed an enhanced segmentation method integrating improved PointNet++ with a coverage-voted strategy, achieving 94% accuracy for common supporting components and significantly improving IoU for bearing plate segmentation.
178. Damage behavior of soaked sandstone subjected to cyclic loadings
Core Problem: Ensuring the stability and longevity of geological energy storage systems requires a thorough understanding of the damage characteristics of soaked sandstone under cyclic loading.
Key Innovation: Explored the damage evolution of soaked sandstone under multi-stage cyclic loading through experimental tests, revealing relationships between plastic hysteresis, energy density, AE characteristics, pore structure changes, and damage variables, providing guidance for designing operational parameters for energy storage geological bodies.
179. Numerical study of material contrast effect on damage and instability in wellbores under repeated drill string impacts
Core Problem: Understanding how material contrasts between geological layers affect damage accumulation and instability in wellbores subjected to repeated drill string impacts, potentially leading to borehole collapse.
Key Innovation: Employs an elastic-plastic damage model to numerically study the effects of repeated mechanical impacts and material contrasts on wellbore stability. It identifies four damage patterns, quantifies the impact of elastic and plastic parameter contrasts on critical damage, and shows that damage localizes at material boundaries.