摘要
从高分辨率卫星遥感影像中提取水体对于水体监测和管理具有重要意义,而阴影和建筑物的干扰制约了水体提取的精度。该文分别利用卷积神经网络和Deeplabv3语义分割神经网络,开展了高分辨率卫星遥感数据水体提取研究,探讨深度学习在水体提取中的应用能力。首先,以高分辨率卫星遥感影像为数据源,分别建立水体分类数据集和水体语义分割数据集,构建并训练卷积神经网络及Deeplabv3网络,得到最优的两种水体提取模型,进一步利用同一测试集对两种模型和其他方法进行精度评价。结果表明,卷积神经网络、Deeplabv3方法精度分别达到95.09%和92.14%,均高于水体指数法、面向对象法和支持向量机法;而且该两种深度学习方法都能够有效去除阴影和建筑物的影响,说明了深度学习方法的有效性,其中,卷积神经网络的适用性更好。
Extracting water body from high resolution satellite remote sensing images is of great significance for its monitoring and management.Because of the interference of shadows and buildings,the accuracy of water body extraction is limited.In this paper,the convolution neural network and Deeplabv3 semantic segmentation neural network are used to carry out the research of water body extraction from high resolution satellite remote sensing data,and the applicability of deep learning in water body extraction is discussed.Taking high-resolution satellite remote sensing images as the data source,data sets for water classification and water body semantic segmentation were established,convolution neural network and Deeplabv3 network were then constructed and trained,and two optimal water body extraction models were obtained.The accuracy of the two models and other methods was further evaluated using the same test set.The results show that the accuracy of the convolution neural network method is 95.09%,and that of the Deeplabv3 method is 92.14%.The convolution neural network has better applicability.The precision of two deep learning methods is higher than that of the water body index method,the object-oriented method and the support vector machine method.Moreover,these two deep learning methods have demonstrated their effectiveness in removing the influence of shadows and buildings.
作者
陈前
郑利娟
李小娟
徐崇斌
吴俣
谢东海
刘亮
CHEN Qian;ZHENG Li-juan;LI Xiao-juan;XU Chong-bin;WU Yu;XIE Dong-hai;LIU Liang(College of Resource Environment and Tourism,Capital Normal University,Beijing 100048;Institute of Remote Sensing and Digital Earth,Chinese Academy of Sciences,Beijing 100101;Land Satellite Remote Sensing Application Center,Ministry of Natural Resources of P.R.China,Beijing 100048;Beijing Institute of Space Mechanics & Electricity,Beijing 100080,China)
出处
《地理与地理信息科学》
CSCD
北大核心
2019年第4期43-49,I0001,共8页
Geography and Geo-Information Science
基金
国家重点研发计划项目(2017YFC0212302)
