摘要
针对遥感影像提取不透水面通常会受到自然环境因素影响的问题,该文采用改进的U2-Net模型对北京市五区和南京市五区的Landsat 8 OLI影像进行季节性不透水面提取,探索深度学习提取不透水面时的自然环境影响因素及其影响机制。选择植被、水、裸土及地表温度作为影响因素,通过地理探测器研究以上因素对不透水面提取的影响机制。基于改进的U2-Net模型提取不透水面精度较高,其中北京市研究区域的提取精度为93.81%,南京市研究区域的提取精度为94.04%;4项自然环境因素对不透水面提取精度均有影响,单因素分析中地表温度影响最大,交互作用分析中地表温度与植被覆盖影响最大。研究结果表明:夏季不透水面提取精度最高,受自然因素及交互作用影响最小;提取不透水面建议采用夏季影像。本文探究了自然因素对不透水面提取的影响机制,为进一步不透水面遥感提取和动态差异分析提供有力支撑。
In view of the problem that the extraction of impervious surface from remote sensing images is usually affected by natural environmental factors,this paper uses the improved U~2-Net model to extract seasonal impervious surface from Landsat 8 OLI images in five districts of Beijing and five districts of Nanjing,and explores the influencing factors of natural environment and its influencing mechanism in the extraction of impervious surface by deep learning.Vegetation,water,bare soil and land surface temperature(LST)are selected as the environmental influencing factors,and geographical detector is applied for researching the influence mechanism of the above factors on impervious surface extraction.The results show that the impervious surface based on the improved U~2-Net model is of high accuracy,with the accuracy of the distracts in Beijing of 93.81%,and that in Nanjing of 94.04%.The four environmental factors all have an impact on the extraction accuracy of impervious surface,and in single factor analysis,LST has the greatest influence,while in interaction analysis,LST and vegetation cover have the greatest influence.The results show that the extraction accuracy of impervious surface is the highest in summer,and it is least affected by natural factors and interaction.Therefore,it is recommended to use summer images for impervious surface extraction.This paper explores the influence mechanism of natural factors on impervious surface extraction,which provides strong support for further impervious surface remote sensing extraction and dynamic difference analysis.
作者
侯幸幸
张新长
赵怡
孙颖
阮永俭
HOU Xingxing;ZHANG Xinchang;ZHAO Yi;SUN Ying;RUAN Yongjian(School of Geography and Remote Sensing,Guangzhou University,Guangzhou 510006,China;Department of Geography and Planning,SunYat-Sen University,Guangzhou 510275,China)
出处
《测绘科学》
CSCD
北大核心
2022年第11期73-84,共12页
Science of Surveying and Mapping
基金
国家自然科学基金面上项目(42071441)
作者简介
侯幸幸(1997—),女,广西岑溪人,硕士研究生,主要研究方向为基于深度学习的遥感影像地物识别;通信作者:张新长,教授,E-mail:eeszxc@mail.sysu.edu.cn
