Deep Learning Method of Landslide Inventory Map with Imbalanced Samples in Optical Remote Sensing

Landslide inventory mapping (LIM) is a key prerequisite for landslide susceptibility evaluation and disaster mitigation. It aims to record the location, size, extent of landslides in each map scale. Machine learning algorithms, such as support vector machine (SVM) random forest (RF), have been increasingly applied detection using remote sensing images recent decades. However, their limitations impeded wide application. Furthermore, despite widespread use deep algorithms sensing, LIM, are limited less unbalanced samples. To this end, study, full convolution networks with focus loss (FCN-FL) were adopted historical regions imbalanced samples an improved symmetrically connected network function increase feature level reduce contribution background value. In addition, K-fold cross-validation training models (FCN-FLK) used improve data utilization model robustness. Results showed that recall rate, F1-score, mIoU by 0.08, 0.09, 0.15, respectively, compared FCN. also demonstrated advantages over U-Net SegNet. The results prove method proposed study can solve problem sample mapping. This research provides reference addressing LIM.