摘要
近地球表面风况及其环境变化与沙丘特征形态演变过程息息相关,在研究过程中,较大范围的沙丘特征线提取存在效率低、成本高等问题。基于U-Net深度学习,对库布齐沙漠内存在的大范围沙丘特征线进行高精度提取,所识别的沙丘特征线为沙脊线和沙丘背风坡坡脚线。实验结果表明:基于U-Net深度学习方法提取卫星影像中的沙丘特征线精度评价指标分别为:MIoU值77.43%、MPA值80.25%、Precision值87.67%,评价指标数据均优于SegNet方法;提取出的沙脊线走向呈NW-SE分布,与气象站测得的风向基本保持一致;利用U-Net深度学习方法自动提取的沙丘特征线的准确性高,与实际观测结果较为符合,可有效地用于区域性的沙脊线走向分析,为沙丘特征演变研究提供了有利方法。
Wind regime and environmental changes near the earth’s surface are closely related to the evolution process of dune feature morphology.The extraction of a large range of dune feature lines had the problems of low efficiency and high cost in the course of the study.In this paper,based on U-Net deep learning,a large range of dune feature lines existed in the Kubuqi desert were extracted with high precision.The identified dune feature lines were the sand ridge line and the foot line of leeward slope of the dune.The experimental results showed that the accuracy evaluation indexes of dune feature lines extracted from satellite images based on U-Net deep learning method were:MIoU value was 77.43%,MPA value was 80.25%,Precision value was 87.67%.The trend of the extracted sand ridge line was NW-SE distribution,which was basically consistent with the wind direction measured by the weather station.The accuracy of dune feature lines automatically extracted by U-Net deep learning method was in line with the actual observation results,which can be effectively used for regional sand ridge line strike analysis,providing a favorable method for the study of dune feature evolution.
作者
陈明均
陈竹安
CHEN Mingjun;CHEN Zhuan(Faculty of Geomatics,East China University of Technology,330013,Nanchang,PRC;CNNC Engineering Research Center of 3D Geographic Information,330013,Nanchang,PRC;Key Laboratory of Mine Environmental Monitoring and Improving Around Poyang Lake,Ministry of Natural Resources,330013,Nanchang,PRC)
出处
《江西科学》
2023年第2期333-338,共6页
Jiangxi Science
基金
国家自然科学基金项目(51708098、52168010)
江西省教育厅课题项目(GJJ180396)。
