Loess Landslide Detection Using Object Detection Algorithms in Northwest China

Regional landslide identification is important for the risk management of hazards. The traditional methods regional were mainly conducted by a human being. In previous studies, automatic recognition focused on new landslides distinct from environment induced rainfall or earthquake, using image classification method and semantic segmentation deep learning. However, there lack research old loess landslides, which are difficult to distinguish environment. Therefore, this study uses object detection learning identify with Google Earth images. At first, database historical samples was established based A total 6111 interpreted in three areas Gansu Province, China. Second, algorithms including one-stage algorithm RetinaNet YOLO v3 two-stage Mask R-CNN, chosen identification. R-CNN achieved greatest accuracy, an AP 18.9% F1-score 55.31%. Among areas, order accuracy high low Site 1, 2, 3, F1-scores 62.05%, 61.04% 50.88%, respectively, positively related their difficulty. results proved that can be employed