Landslide Detection Methods Based on Deep Learning in Remote Sensing Images

Abstract

Remote sensing image target detection algorithm based on deep learning has developed rapidly in recent years. Landslide detection and recognition in remote sensing images is an effective means to quickly identify and locate the landslides after disasters, but so far target detection based on deep learning is rarely applied to landslide. In this paper, representative algorithms (Faster R-CNN” YOLOv3, SSD) were selected to process a landslide remote sensing image data set of 2000 orders of magnitude we produced in Pascal VOC 2007 format. The technical advantages and functional characteristics of three different target detection algorithms were discussed and compared. It is experimentally verified that the Faster R-CNN algorithm based on candidate regions has higher detection accuracy, and the two regression-based methods, YOLOv3 and SSD, with faster detection speeds, are more suitable for real-time monitoring and practical production applications.