面向对象结合卷积神经网络的GF-1影像遥感分类  

Remote sensing image of GF-1 classification using object-oriented method and convolutional neural network

作　　者：蒋治浩 林辉[1,2,3] 张怀清 蒋馥根[1,2,3] JIANG Zhihao;LIN Hui;ZHANG Huaiqing;JIANG Fugen(Research Center of Forestry Remote Sensing&Information Engineering,Central South University of Forest&Technology,Changsha 410004,Hunan,China;Key Laboratory of Forestry Remote Sensing Based Big Data&Ecological Security for Hunan Province,Central South University of Forest&Technology,Changsha 410004,Hunan,China;Key Laboratory of State Forestry&Grassland Administration on Forest Resources Management and Monitoring in Southern Area,Central South University of Forest&Technology,Changsha 410004,Hunan,China;Research Institute of Forest Resources Information Techniques,Chinese Academy of Forestry,Beijing 100091,China)

机构地区：[1]中南林业科技大学林业遥感信息工程研究中心,湖南长沙410004 [2]中南林业科技大学林业遥感大数据与生态安全湖南省重点实验室,湖南长沙410004 [3]中南林业科技大学南方森林资源经营与监测国家林业与草原局重点实验室,湖南长沙410004 [4]中国林业科学研究院资源信息研究所,北京100091

出　　处：《中南林业科技大学学报》2021年第8期45-55,67,共12页Journal of Central South University of Forestry & Technology

基　　金：“十三五”国家重点研发计划子课题“人工林三维可视化模拟技术及监测系统集成”(2017YFD0600905);中央级科研院所基本科研业务费专项资金项目“基于虚拟环境的人工林经营模拟技术”(CAFYBB2019SZ004)。

摘　　要：【目的】近年来,越来越多高时间分辨率、高空间分辨率卫星相继出现,为我们的生产生活提供了很大的便利,如何利用好这些数据庞大、信息丰富的遥感影像一直以来都是国内外研究的热点问题。其中遥感影像的分类是将大量的遥感影像应用于各个领域的基础,针对传统方法对于高分辨率影像分类精度提高难的问题,提出一种面向对象结合卷积神经网络的遥感分类方法。【方法】首先利用构建moran’s I指数与地理探测器q统计量的二维空间的方法,确定最佳分割尺度,以最大面积法确定均质因子权重,对预处理后的GF-1影像进行分割,利用分割后的对象的特征作为分类模型的输入变量,建立一维卷积神经网络(1D-CNN)的分类模型,构建了基于像元的支持向量机,面向对象的支持向量机分类模型,对研究区进行了分类。【结果】利用面向对象的一维卷积神经网络方法进行分类,分类结果总体精度为93.10%,Kappa系数为0.9167,同基于像元支持向量机方法相比,总体精度提高了24.35%,Kappa系数提高了0.2923;同面向对象的支持向量机方法相比,总体精度提高了6.2%,Kappa系数提高了0.0746。【结论】利用构建的moran’s I指数与地理探测器q统计量的二维空间和最大面积法确定最佳分割参数,建立一维卷积神经网络结合面向对象的方法对遥感影像进行分类,与传统模型相比得到的分类结果精度较高,是一种快速有效的分类方法。【Objective】In recent years,more and more satellites with high temporal resolution and high spatial resolution have appeared one after another,which provides a lot of convenience for our lives.How to make good use of these huge amounts of data and information-rich remote sensing images has always been a hot issue for domestic and foreign research.Among them,the classification of remote sensing images is the basis for applying a large number of remote sensing images to various fields.Aiming at the difficulty of improving the classification accuracy of high-resolution images by traditional methods,an object-oriented remote sensing classification method combined with convolutional neural network is proposed.【Method】First,the optimal segmentation scale was determined by constructing a two-dimensional space of moran’s I index and geographic detector q statistics.Theweight of the homogeneous factorswas determined by the maximum area method.After segmenting the preproccessed image,the characteristics of the segmented objects were used as input variables of the classification model to establish a one-dimensional convolutional neural network(1D-CNN)classification model.And pixel-based support vector machine,object-oriented support vector machine classification model were built for comparison.【Result】Using the object-oriented one-dimensional convolutional neural network method for classification,the overall accuracy of the classification result is 93.10%,and the Kappa coefficient is 0.9167.Compared with the pixel-based support vector machine,the overall accuracy is improved by 24.35%,and the Kappa coefficient is increased by 0.2923.Compared with the object-oriented support vector machine,the overall accuracy is improved by 6.2%,and the Kappa coefficient is increased by 0.0746.【Conclusion】Using the two-dimensional space of moran’s I index and geographic detector q statistics and maximum area method to determine the optimal segmentation parameters,a 1D-CNN combined oriented-objected approach was built to c

关 键 词：遥感分类 面向对象 最佳分割参数 卷积神经网络 

分 类 号：S771.8[农业科学—森林工程]