Thermal melt slump recognition method and device, system, storage medium
By combining multi-source remote sensing data and SBAS-InSAR technology, the SAMLoRA-Transformer model was constructed, which solved the problems of traditional thermal fusion landslide identification relying on manual interpretation and deep learning misjudgment, and achieved high-precision landslide identification and dynamic monitoring.
CN122391856APending Publication Date: 2026-07-14WUHAN INST OF TECH +1
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- WUHAN INST OF TECH
- Filing Date
- 2026-04-15
- Publication Date
- 2026-07-14
AI Technical Summary
Technical Problem
Traditional thermal collapse identification relies on manual interpretation, which is costly and has low accuracy. Deep learning methods are prone to misjudgment in complex scenarios and have poor robustness.
Method used
By combining multi-source remote sensing data with SBAS-InSAR technology, a SAMLoRA-Transformer multi-scale feature fusion model was constructed. The self-attention mechanism was used to capture long-distance dependencies and identify thermal slump.
Benefits of technology
It improves the accuracy and efficiency of thermal slump identification, provides reliable technical support for large-scale dynamic monitoring and disaster early warning, and helps ecological protection and disaster prevention in permafrost areas.
✦ Generated by Eureka AI based on patent content.
Smart Images
Figure CN122391856A_ABST
Abstract
The application discloses a hot melt collapse recognition method, device and system and a storage medium, and comprises the following steps: acquiring multi-source remote sensing data containing multispectral images, high-resolution remote sensing images, field survey data and DEM data; extracting surface time sequence deformation characteristics and hot melt collapse deformation characteristics by adopting SBAS-InSAR technology; constructing and training a SAMLoRA-Transformer multi-scale feature fusion model, the model extracts image basic semantic features by taking a pre-trained SAM as a backbone network, introduces a Low-Rank Adaptation low-rank fine-tuning mechanism, constructs a multi-scale fusion module through a Transformer architecture, maps geographic space auxiliary data into a high-dimensional feature vector, and realizes cross-modal interaction and weighted fusion of the image features; and the trained model is used for hot melt collapse recognition, and spatial distribution vector data is output. The application can improve the recognition efficiency and accuracy in a complex scene.
Need to check novelty before this filing date? Find Prior Art