基于“高分五号”遥感图像的地物分类方法  

Method of terrain classification based on GF⁃5 satellite remote sensing images

作　　者：魏友华[1,2] 王瑶 何雪梅 郭科[1,2,3] 常睿春 WEI Youhua;WANG Yao;HE Xuemei;GUO Ke;CHANG Ruichun(Geomathematics Key Laboratory of Sichuan Province,Chengdu University of Technology,Chengdu 610059,China;Institute of Digital Hu Line,Chengdu University of Technology,Chengdu 610059,China;Department of Applied Mathematics,Chengdu University of Technology,Chengdu 610059,China)

机构地区：[1]成都理工大学数学地质四川省重点实验室,四川成都610059 [2]成都理工大学数字胡焕庸线研究院,四川成都610059 [3]成都理工大学应用数学系,四川成都610059

出　　处：《现代电子技术》2020年第18期85-88,共4页Modern Electronics Technique

基　　金：国家自然科学基金面上项目(41672325);数学地质四川省重点实验室开放基金项目(scsxdz2019zd03)。

摘　　要：作为全波谱段高光谱卫星,"高分五号"卫星可提供丰富的遥感数据,研究利用"高分五号"卫星遥感图像实现精确的地物分类方法,这是实现高光谱遥感信息资源所有权的重要组成部分。文章立足于"高分五号"卫星数据,先进行遥感图像预处理,有效地消除大气干扰和噪声,优选核主成分分析方法进行"高分五号"卫星高光谱图像的特征提取,用最大似然法(MLE)、卷积神经网络(CNN)、支持向量机(SVM)3种分类器对高光谱图像数据分类。结果表明,RBF核的支持向量机分类器精度可达到97.889 7%,Kappa系数为0.966 7,比传统的最大似然法、神经网络分类方法分类精度分别高出15.478%,8.670 8%。由此可以实现"高分五号"卫星数据预处理,及其在土地利用分类应用上的可行性。As a full spectrum hyperspectral satellite,GF⁃5 satellite can provide a rich remote sensing data.A method is studied in this paper to realize precise terrain classification by GF⁃5 satellite remote sensing image.It is an important component of realizing the ownership of the hyperspectral remote sensing information resource.The remote sensing image is preprocessed according to the data of GF⁃5 satellite,by which the atmospheric interference and noise is effectively eliminated.The kernel principal component analysis method is optimally selected to extract the features of the hyperspectral image of GF⁃5 satellite.The three classifiers of maximum likelihood method(MLE),convolutional neural network(CNN)and support vector machine(SVM)are utilized to classify the hyperspectral image data.The results show that the accuracy of SVM classifier based on RBF kernel function can reach 97.8897%,which is 15.478%and 8.6708%higher than that of traditional MLE and CNN respectively,and its Kappa coefficient is 0.9667.It can realize the data preprocessing of GF⁃5 satellite and the feasibility for application in land use classification.

关 键 词：地物分类 遥感图像 图像预处理 大气干扰消除 特征提取 数据分类 

分 类 号：TN965-34[电子电信—信号与信息处理] TP79[电子电信—信息与通信工程]