Citation

Ghorbanzadeh, O., Xu, Y., Ghamisi, P., Kopp, M., Kreil, D. (2022). Landslide4Sense: Reference Benchmark Data and Deep Learning Models for Landslide Detection. IEEE Transactions on Geoscience and Remote Sensing, 60: 1-17. Link to paper

Abstract

This work presents a reference benchmark dataset for identifying landslides using remote sensing technology. The dataset comprises 3,799 image patches combining optical data from Sentinel-2 with topographic information including digital elevation models and slope data derived from ALOS PALSAR. The incorporation of topographic layers enhances the detection of landslide boundaries, addressing limitations of optical-only approaches. Data was collected across four regions at different times: Iburi, Japan (2018), Kodagu, India (2018), Gorkha, Nepal (2015), and Taiwan (2009). Each pixel received binary classification as landslide or non-landslide through manual annotation. The researchers evaluated 11 state-of-the-art deep learning segmentation architectures, finding ResU-Net achieved superior performance. The benchmark dataset and trained models are made publicly available as a resource for remote sensing and computer vision research communities. This standardized dataset enables fair comparison of different deep learning approaches and facilitates the development of more robust landslide detection methods applicable across diverse geographical and environmental conditions.

AI Visual Interpretation (Beta)