A method and device for hot melt sliding collapse susceptibility assessment based on multi-spectral and digital elevation model state space network

Abstract

The present application belongs to the technical field of geoscience and geological disaster remote sensing prevention and treatment, and particularly relates to a hot melt collapse susceptibility assessment method and device based on a multi-spectral and digital elevation model state space network. The method first acquires high-resolution multi-spectral images and a digital elevation model constructed from stereo pairs in a study area; generates a pixel-level label based on an existing hot melt collapse vector list, and cuts training samples on source domain images in a sliding window manner; constructs a feature extraction backbone based on a state space network, encodes multi-spectral and elevation grid data, expands local features into sequences, models long-distance spatial dependency in a state space unit, and highlights spectral and topographic features related to hot melt collapse using an attention enhancement module; simultaneously inputs labeled source domain samples and unlabeled target domain samples into a shared network, extracts task-related features and domain-related features through a feature decoupling module, and introduces a correlation alignment loss based on covariance alignment between domain-related features to realize cross-regional feature distribution alignment; after model training converges, the whole image of the target area is predicted in a sliding window manner, and overlapping areas are weighted and fused to generate a continuous hot melt collapse susceptibility probability map. The present application uses multi-source remote sensing information, introduces a state space network and a cross-regional feature alignment mechanism, and significantly improves the precision, spatial continuity and cross-regional generalization ability of hot melt collapse susceptibility assessment in permafrost regions.

Images