基于多时相特征对比的滑坡遥感解译及风险性评价OA

To address the insufficient interpretation accuracy of disaster data and the absence of risk assessment frameworks for rainfall-induced cluster landslides,this study takes Jiangwan Town,Shaoguan City,Guangdong Province,as a typical case to perform multi-temporal remote sensing interpretation and landslide risk assessment.The research established a refined identification process using high-resolution remote sensing imagery.By employing MATLAB for small-size image segmentation,the study initially screened potential landslide points based on color features.Subsequently,by integrating morphological attributes,spatial locations,and multi-temporal characteristics,it manually filtered out pre-existing landslides to isolate those specifically triggered by the target rainfall event.Field investigations verified the precision of this approach,identifying a total of 2 123 landslide disaster points.Following data acquisition,an assessment index system incorporating nine environmental factors—elevation,slope,aspect,profile curvature,plan curvature,lithology,distance to roads,distance to water systems,and land use type—served as the foundation for susceptibility assessment modeling.The study compared the performance of four machine learning models:support vector machine(SVM),linear support vector machine(LSVM),logistic regression(LR),and Bayesian network(BN).Finally,the research conducted a comprehensive landslide risk assessment using matrix analysis,integrating susceptibility assessment results with vulnerability data derived from population,building,and road densities.The results demonstrate that the SVM model achieves the highest predictive accuracy with an area under the curve(AUC)of 0.816.The high-susceptibility zones concentrate around rivers and main roads,aligning with field observations.High-risk and medium-risk areas primarily encompass the northwestern part of Huyang Village,the southeastern part of Liangwu Village,the central part of Weiping Village,and the northwestern part of Guoxi Village,effectively covering all major disaster sites identified during field surveys.This research framework provides a robust technical reference for the fine-grained prevention and control of cluster landslide disasters.