基于斜坡单元和语义分割的皖南地区滑坡灾害易发性评估

滑坡灾害严重影响着人们的生命和财产安全，对自然环境造成重大破坏。以斜坡单元为单位进行滑坡易发性评估能够更加接近真实的滑坡地形，为滑坡灾害的防治提供更加科学的理论支持。本文以安徽省皖南地区为研究区，基于黄山、宣城、池州市滑坡点数据以及皖南地区基础地理数据，利用主成分分析和多重共线性分析方法筛选滑坡评价因子，提出将斜坡单元几何形状信息和语义分割方法相结合的创新方法，构建滑坡灾害易发性评估模型，对皖南地区的滑坡灾害易发性进行评估，揭示其空间分布规律。结果表明：结合斜坡单元和语义分割方法构建的滑坡易发性评估模型具有较高的预测精度，能够充分考虑斜坡单元的几何形状信息对滑坡易发性的影响，较为准确地评估皖南地区的滑坡易发性。评估结果符合滑坡形成机理，其中62.19%的滑坡单元分布在滑坡易发性等级中—高的斜坡单元上，模型预测AUC值为0.878，与缺少几何形状信息的CNN模型进行对比，预测精度明显提高。

Landslide susceptibility assessment in southern Anhui Province based on slope units and semantic segmentation

Landslide disasters seriously affect people's lives and property safety and cause significant damage to the natural environment. Landslide susceptibility assessment based on slope units can allow for a more accurate representation of the actual terrain and provide more scientific theoretical support for the prevention and control of landslide disasters. Based on the data of landslide points in Huangshan, Xuancheng, Chizhou cities and the basic geographical data of southern Anhui Province, this paper selects landslide evaluation factors by using principal component analysis and multicollinearity analysis, proposes an innovative method that combines the geometric shape information of slope units and semantic segmentation, builds a landslide disaster vulnerability assessment model to evaluate the vulnerability of landslide disasters in southern Anhui Province and reveal its spatial distribution pattern. The results show that the landslide susceptibility assessment model constructed by combining slope units and semantic segmentation has high prediction accuracy and can fully consider the influence of geometric shape information of slope units on landslide susceptibility, and accurately assess the landslide susceptibility in southern Anhui Province. The evaluation results are consistent with the formation mechanism of landslides, with 62.19% of landslide units distributed on slope units with medium to high landslide susceptibility levels. The model predicts an AUC value of 0.878, which is significantly improved in prediction accuracy compared to CNN models lacking geometric shape information.