基于多源数据融合的山体滑坡研究  

Research on Landslides Based on Multi-source Data Fusion

作　　者：刁君华 冯向萍[1] 马新春 DIAO Jun-hua;FENG Xiang-ping;MA Xin-chun(School of Computer and Information Engineering,Xinjiang Agricultural University,Urumqi 830052,China;Xinjiang Electronics Research Institute Co.,Ltd.,Urumqi 830013,China)

机构地区：[1]新疆农业大学计算机与信息工程学院,新疆乌鲁木齐830052 [2]新疆电子研究所股份有限公司,新疆乌鲁木齐830013

出　　处：《计算机技术与发展》2025年第4期214-220,共7页Computer Technology and Development

基　　金：新疆维吾尔自治区科技支疆项目计划(2022E02080)。

摘　　要：山体滑坡是一种频发的地质灾害,通过对滑坡的识别来进行灾害防治是一项重要的任务。目前,单一的数据源使得滑坡识别效率较低,仅从滑坡遥感影像上来进行滑坡研究,无法充分识别坡度、坡向和高度变化等地形特征。针对于此,实验结合滑坡遥感影像和DEM数据进行研究,设计了一个不共享权重的孪生网络,对两类多源异质数据进行特征融合。其次,使用一种基于注意力机制改进的DeeplabV3+语义分割模型进行分割研究,并将改进模型与FCN、Unet、SegNet和AED-Net等语义分割模型进行对比。实验结果表明,多源数据的融合相比于单一数据具有更好的滑坡分割效果,在精确率(Precision)、召回率(Recall)、F1分数(F1-score)和平均交并比(MIoU)上分别提高了3.5百分点、4.9百分点、4.2百分点和3.1百分点。引入注意力机制的DeepLabV3+模型的F1-score和MIoU比原模型分别提高了1.6百分点和1.1百分点。Landslide is a frequent geological disaster.It is an important task to prevent and control landslides by identifying them.At present,the efficiency of landslide identification is low due to the single data source.Landslide research based on landslide remote sensing images alone cannot fully identify terrain features such as slope,direction and height change.In this experiment,landslide remote sensing images and DEM data were combined for research,and a twin network without shared weights was designed to fuse the features of two multi-source heterogeneous data.Secondly,a DeeplabV3+semantic segmentation model based on the improved attention mechanism was used for segmentation research,and the improved model was compared with the FCN,Unet,SegNet and AED-Net and other semantic segmentation models.The experimental results show that the fusion of multi-source data has better landslide segmentation effect than single data,and the Precision,Recall,F1-score and mean intersection over union(MIoU)are improved by 3.5 percentage points,4.9 percentage points,4.2 percentage points and 3.1 percentage points,respectively.The F1-score and MIoU of the DeepLabV3+model with the introduction of the attention mechanism are respectively improved by 1.6 percentage points and 1.1 percentage points compared with the original model.

关 键 词：滑坡识别 多源数据 特征融合 孪生网络 注意力机制 语义分割 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]