基于Transformer语义分割模型的露天矿场识别
Open-pit mine recognition based on Transformer model
陈佳晟 1游翔 2沈盛彧 3廖梓凯 2张彤1
作者信息
- 1. 武汉大学 测绘遥感信息工程国家重点实验室,湖北 武汉 430079
- 2. 四川省水土保持生态环境监测总站,四川 成都 610074
- 3. 长江科学院 水土保持研究所,湖北 武汉 430010
- 折叠
摘要
露天矿场是生产建设项目水土保持信息化监管的重要对象,对其范围的高效精准识别对于监测非法违规开采行为,加强开采过程中的水土流失预防与治理具有重要意义.基于Transformer深度学习模型提出了露天矿场的遥感影像智能识别方法,并在四川省宜宾市的露天矿场影像数据集上与常用的基于卷积神经网络的深度学习识别方法进行了实验对比.结果表明:该方法对露天矿场范围识别的精确率、召回率、F1-score和IoU指标分别达到91.25%,90.66%,90.95%和83.41%,能够满足水土保持遥感监管的精确度要求;在识别精确度和识别效果上优于对比方法,在运行效率上与对比方法保持在同一数量级,表现出较强的应用价值.该方法在大区域露天矿场范围快速准确识别方面有推广应用潜力.
Abstract
Open-pit mine is an important object of water and soil conservation information supervision in production and con-struction projects.The efficient and accurate identification of its scope is of great significance for monitoring illegal mining behav-iors and strengthening the prevention and control of soil and water loss in the mining process.We introduced an intelligent recog-nition method utilizing a Transformer-based deep learning model for analyzing remote sensing images of open-pit mining areas.Comparative experiments were conducted on the open-pit mine dataset in Yibin City,Sichuan Province,using widely adopted deep learning recognition methods based on convolutional neural networks.The results indicated that the reveal precision,recall,F1-score,and IoU values of this method for identifying the scope of open-pit mines were 91.25%,90.66%,90.95%and 83.41%,respectively,which can meet the accuracy requirements of remote sensing supervision for water and soil conservation.Additionally,the efficiency and accuracy of our method remained superior to the contrasted methods while it shows equivalent run-ning efficiency,indicating significant practical utility.The method introduced in this paper holds substantial potential for wide-spread application,enabling swift and accurate recognition of open-pit mines across extensive regions.
关键词
水土保持/遥感监管/露天矿场/深度学习/Transformer模型/语义分割/宜宾市Key words
water and soil conservation/remote sensing supervision/open-pit mine/deep learning/Transformer model/se-mantic segmentation/Yibin City引用本文复制引用
出版年
2024
国家科技期刊平台
NSTL
万方数据