An Adaptive Identification Method for Potential Landslide Hazards Based on Multisource Data

The effectiveness of landslide disaster prevention depends largely on the quality early identification potential hazards, and how to comprehensively, deeply, accurately identify such hazards has become a major difficulty in management. Existing deep learning methods for hazard often use fixed-size window modeling ignore different sizes required by landslides scales. To address this problem, we propose an adaptive method based multisource data. Taking Yongping County, China, as study area, create factor dataset background terms topography, geology, human activities, hydrology, vegetation sample model after processing. Moreover, combine differential interferometric synthetic aperture radar (D-InSAR) multitemporal InSAR (MT-InSAR) process surface deformation measure richness average pixel difference within current point image. Therefore, convolutional neural networks (CNNs) with are adaptively selected. results show that precision area is 85.30%, recall 83.03%, F1 score 84.15%. recognition rate reaches 80%, which better than fixed-window result proves proposed method. This can help improve intelligent systems also contribute other geological mudslides collapses.