Landslide detection with ALOS-2/PALSAR-2 data using convolutional neural networks: a case study of 2018 Hokkaido Eastern Iburi earthquake

Landslide events are triggered by heavy rain or earthquake each year around the world. The remote sensing technique is an effective tool for disaster mapping, monitoring, and early warning. In particular, synthetic aperture radar (SAR) has great potentials due to its all-weather with day and night time imaging capabilities. Thus, rapid damage detection of landslide events using SAR data is expected. However, the technique for landslide detection has not been established. In recent years, convolutional neural networks (CNNs) have been widely developed in semantic segmentation. The CNNs based on the U-Net, in particular, has a capability of fewer training images for semantic segmentation. In this study, we demonstrate a landslide detection using the CNNs based on the U-Net with SAR data. The landslides induced by the 2018 Hokkaido Eastern Iburi earthquake was selected for a case study. ALOS-2/PALSAR-2 data were acquired on 29 March 2018 and 13 September 2018 in ascending orbit and were acquired on 23 August 2018 and 6 September 2018 in descending orbit. For pre-processing, radiometric calibration and ortho-rectification were performed in each PALSAR-2 data. Firstly, we performed a conventional method, which is the backscattering coefficient difference (BCD) for landslide detection. The highest value of the F-measure (31.9%) was obtained within a 7 x 7 window size with the slope angle in the ascending orbit dataset. Secondly, we performed a CNNs method based on the U-Net for landslide detection. The highest value of the F-measure (79.9%) was obtained using the CNNs method in the ascending orbit dataset. In conclusion, the proposed CNNs method is more effective than the threshold method for landslide detection using pre- and post-event ALOS-2/PALSAR-2 data.

[1]  Yuzo Suga,et al.  Landslide detection using COSMO-SkyMed images: a case study of a landslide event on Kii Peninsula, Japan , 2018 .

[2]  Nancy Chinchor,et al.  MUC-4 evaluation metrics , 1992, MUC.

[3]  Alexei A. Efros,et al.  Image-to-Image Translation with Conditional Adversarial Networks , 2016, 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR).

[4]  Simon Plank,et al.  Landslide Mapping in Vegetated Areas Using Change Detection Based on Optical and Polarimetric SAR Data , 2016, Remote. Sens..

[5]  Thomas Brox,et al.  U-Net: Convolutional Networks for Biomedical Image Segmentation , 2015, MICCAI.

[6]  Shota Iino,et al.  CNN-based generation of high-accuracy urban distribution maps utilising SAR satellite imagery for short-term change monitoring , 2018, International Journal of Image and Data Fusion.

[7]  Shunichi Koshimura,et al.  Towards Operational Satellite-Based Damage-Mapping Using U-Net Convolutional Network: A Case Study of 2011 Tohoku Earthquake-Tsunami , 2018, Remote. Sens..

[8]  Thomas Blaschke,et al.  Evaluation of Different Machine Learning Methods and Deep-Learning Convolutional Neural Networks for Landslide Detection , 2019, Remote. Sens..

[9]  Yu Zhou,et al.  Polarimetric SAR Image Classification Using Deep Convolutional Neural Networks , 2016, IEEE Geoscience and Remote Sensing Letters.

[10]  Karsten Spaans,et al.  A New Method for Large-Scale Landslide Classification from Satellite Radar , 2019, Remote. Sens..