Potential Landslide Identification Based on Improved YOLOv8 and InSAR Phase-Gradient Stacking
Yanrong Mao, Ruiqing Niu, Bingquan Li, Jun Li
Abstract
Landslides, as a major geological hazard, caused thesignificant threats to human life and property. Therefore, the ide-ntification of potential landslides is a crucial concern. This study combines InSAR phase-gradient stacking with deep learning to achieve efficient and accurate identification of large-scale potentiallandslides. By stacking phase-gradients to highlight local surfacedeformation information and refining the YOLOv8 model based on small target features of local deformations, this study introdu-ces improvements. This involves adding a CBAM layer to the ba-ckbone, replacing the C2f modules with GhostNetV2 to suppress information loss during long-distance feature transmission, enha-ncing the network's perception and detection capabilities for sm-all targets. Additionally, a 160×160 small target detection head is added to the detection module to specifically handle small target detection tasks, improving accuracy and performance. A new loss function, FocalSIoU Loss, is introduced based on the characteri-stics of the dataset, combining SIoU with the bias option Gamma to make the model more targeted during training. The improved model achieved a maximum mAP50 value of 93.4% on the valid-ation set.Finally using factor masks to identify region deformatio-n points, this study reduces misjudgments of non-landslides, ide-ntifying 378 potential landslides in the study area with a false po-sitive rate of only 10.2%.