Multi-source image feature extraction and segmentation techniques for karst collapse monitoring

Introduction Karst collapse monitoring is a complex task due to data sparsity, underground dynamics, and the demand for real-time risk assessment. Traditional approaches often fall short in delivering timely and accurate evaluations of collapse risks. Methods To address these challenges, we propose the Integrated Karst Collapse Prediction Network (IKCPNet), a novel framework that combines multi-source imaging, geophysical modeling, and machine learning techniques. IKCPNet processes seismic imaging, hydrological patterns, and environmental factors using an advanced data encoding mechanism and a physics-informed module to capture subsurface changes. A dynamic risk assessment strategy is incorporated to enable real-time feedback and probabilistic mapping. Results Experimental evaluations on the OpenSARShip dataset demonstrate that IKCPNet achieves an accuracy of 94.34 ± 0.02 and an IoU of 90.23 ±0.02, outperforming the previous best model by 1.22 and 0.89 points, respectively. Discussion These results highlight the effectiveness of IKCPNet in improving prediction accuracy and risk mitigation, showcasing its potential for enhancing geological hazard monitoring through multi-source data integration.