摘要
山体滑坡是重要的地质灾害,构建快速、高精度的自动检测滑坡方法,对于减轻滑坡地质灾害的影响具有重要的意义。本文提出了一种基于注意力机制与卷积神经网络的滑坡识别方法。通过采用改进的ResU-Net作为主干网络,并引入残差连接以及注意力机制,实现不同网络层的多尺度滑坡特征融合,以促进深度卷积神经网络对滑坡区域的识别效果。实验结果表明,该方法在滑坡遥感数据集上的平均像素精度、平均交并比分别达到了88.79%、92.56%,实现了高效、精准的滑坡识别。
Landslides are important geological hazards,and constructing fast and high-precision automatic landslide detection methods is of great significance for reducing the impact of landslide geological hazards.This article proposes a landslide recognition method based on attention mechanism and convolutional neural network.By using an improved ResU-Net as the backbone network and introducing residual connections and attention mechanisms,multi-scale landslide feature fusion at different network layers is achieved to promote the recognition effect of deep convolutional neural networks on landslide areas.The experimental results show that the average pixel accuracy and average intersection to union ratio of this method on landslide remote sensing datasets reach 88.79%and 92.56%,respectively,achieving efficient and accurate landslide recognition.
作者
冯晨
FENG Chen(School of Information Technology Engineering,Fuzhou Polytechnic,Fuzhou,China,350108)
出处
《福建电脑》
2023年第12期38-43,共6页
Journal of Fujian Computer
基金
福建省自然科学基金(No.2020J01132452)
国家自然科学基金(No.62277010)
福州职业技术学院2022年校级课题学校引导计划项目(No.FZYKJZXYD202201)资助。
关键词
山体滑坡
滑坡检测
注意力机制
卷积神经网络
Mountain Landslide
Landslide Detection
Attention Mechanism
Convolutional Neural Network
