基于混合网络与数字高程模型的遥感图像滑坡识别  

Landslide Identification from Remote Sensing Images Based on Hybrid Network and Digital Elevation Model

作　　者：张敏 朱雨涵 王承心 陈玮容 陈新房 ZHANG Min;ZHUYu-han;WANG Cheng-xin;CHEN Wei-rong;CHEN Xin-fang(College of Information Engineering,Institute of Disaster Prevention,Sanhe 065201,China;Hebei Province University Smart Emergency Application Technology Research and Development Center,Sanhe 065201,China)

机构地区：[1]防灾科技学院信息工程学院,河北三河065201 [2]河北省高校智慧应急应用技术研发中心,河北三河065201

出　　处：《电脑与电信》2024年第11期12-16,共5页Computer & Telecommunication

基　　金：防灾科技学院大学生创新创业训练计划项目(X202411775096);河北省高等学校科学技术研究项目(ZC2023108);廊坊市科学技术研究与发展计划自筹经费项目(20240111018)。

摘　　要：滑坡识别是确保地质安全、及时应对自然灾害的重要任务,其准确性直接关系到灾害预警的及时性与应急响应的效率。鉴于滑坡灾害的复杂多变与高频发生特性,本研究专注于将RGB图像和DEM数据的组合作为输入,并利用ResNet优化与瓶颈注意力机制(BAM)相结合,对遥感图像进行滑坡识别。通过ResNet50对滑坡数据进行深度优化与特征提取,有效过滤冗余信息,提取关键特征向量并作为BAM的输入,增强模型对滑坡特征的识别与关注能力。实验结果显示,ResNet50-BAM的混合模型展现出更强的泛化能力与更高的检测准确率,为地质灾害的预警与应急响应提供了更加可靠的技术支持。Landslide identification is a crucial task for ensuring geological safety and timely responding to natural disasters,and its accuracy directly relates to the timeliness of disaster warnings and the efficiency of emergency responses.In light of the complex and frequent nature of landslide disasters,this study focuses on combining RGB images and DEM(Digital Elevation Model)data as inputs and leveraging the integration of ResNet optimization with the Bottleneck Attention Module(BAM)to identify landslides in remote sensing images.By deeply optimizing and extracting landslide data features through ResNet5o,redundant information is effectively filtered,and key feature vectors are extracted as inputs for BAM,thereby enhancing the model's recognition and focus capabilities for landslide features.Experimental results demonstrate that the hybrid model of ResNet50-BAM exhibits stronger generalization ability and higher detection accuracy,providing more reliable technical support for geological disaster early warning and emergency response.

关 键 词：滑坡检测 DEM BAM ResNet50 ResNet50-BAM 

分 类 号：TP751[自动化与计算机技术—检测技术与自动化装置] P642.22[自动化与计算机技术—控制科学与工程] P237[天文地球—工程地质学]