基于残差网络的泥石流孕灾沟谷快速识别

Rapid Identification of Hazard-formative Valley Based on Residual Network

本研究选取怒江州怒江流域,提取沟谷数字高程模型(DEM)图训练深度残差网络,对泥石流的灾害区域进行快速识别。首先,使用增强后的DEM图训练残差网络,然后用该训练后的模型对沟谷DEM图进行分类,并使用准确率、召回率、AUC值等对模型进行评价。测试结果表明,残差网络结合DEM图可以达到泥石流沟谷最高73%的识别率以及76%的召回率,AUC值约为0.7,模型性能较为良好,为泥石流孕灾沟谷的识别提供了新思路。

In this study,the Nujiang River basin in Nujiang Prefecture was selected,combined with Digital Elevation Model(DEM)images and deep residual networks,to quickly identify the disaster areas of mud-rock flow.First,use the enhanced DEM image to train the residual network,then use the trained model to classify the valley DEM,and use the accuracy rate,recall rate,AUC value,etc.to evaluate the model.The test results show that the residual network combined with the DEM data can achieve a maximum accuracy of 73%and a recall rate of 76%for debris flow valleys.The AUC value is about 0.7.The model performance is relatively good.It provides a new method for the identification of mud flow.