基于改进OGMRF-RC模型的SAR图像分类方法  

A SAR image classification method based on an improved OGMRF-RC model

作　　者：李渊 毋琳[1,2,3,5] 戚雯雯 郭拯危 李宁[1,2,3] LI Yuan;WU Lin;QI Wenwen;GUO Zhengwei;LI Ning(College of Computer and Information Engineering,Henan University,Kaifeng 475004,China;Henan Engineering Research Center of Intelligent Technology and Application,Henan University,Kaifeng 475004,China;Henan Key Laboratory of Big Data Analysis and Processing,Henan University,Kaifeng 475004,China;School of Information and Electronic Engineering,Shangqiu Institute of Technology,Shangqiu 476000,China;College of Environment and Planning,Henan University,Kaifeng 475004,China)

机构地区：[1]河南大学计算机与信息工程学院,开封475004 [2]河南大学河南省智能技术与应用工程技术研究中心,开封475004 [3]河南大学河南省大数据分析与处理重点实验室,开封475004 [4]商丘工学院信息与电子工程学院,商丘476000 [5]河南大学环境与规划学院,开封475004

出　　处：《自然资源遥感》2021年第4期98-104,共7页Remote Sensing for Natural Resources

基　　金：国家自然科学基金项目“基于高分辨星载SAR图像的丹江口库区生态红线保护区的水源涵养功能监测与评估”(编号:61871175);河南省科技攻关计划项目“众包数据辅助下基于深度学习的高分辨率土地覆盖自动分类研究”(编号:202102210175);“基于多极化SAR影像的小麦物候期监测研究”(编号:212102210093);“黄河下游游荡型河段河势雷达遥感监测研究”(编号:212102210101);河南省高等学校重点科研项目“河南地区冬小麦雷达极化特征及其遥感监测研究”(编号:19A420005);“多覆被类型地表水源涵养功能SAR反演评估研究”(编号:21A520004);河南省青年人才托举工程项目“基于多极化雷达影像的河南地区冬小麦长势监测研究”(编号:2019HYTP006)共同资助。

摘　　要：合成孔径雷达(synthetic aperture Radar,SAR)图像分类是遥感应用中的关键技术之一。针对对象高斯-马尔可夫随机场(object-based Gaussian-Markov random field,OGMRF)模型中区域类别标签对分类精度影响的问题,提出了区域类别模糊概率(regional category fuzzy probability,RCFP)标签场方法,使临界对象具有多种类别划分的可能性,避免唯一标签导致的错分类现象。该方法综合考虑区域特征与邻域特征,利用区域边缘信息和后验概率获得RCFP,并将其纳入特征场参数求解过程中,使特征场参数更加接近真实情况,从而提高SAR图像分类精度。以河南省开封市东部约1400 km 2的区域为研究区,采用Sentinel-1卫星SAR图像开展农田、建筑、水域3类地物的分类验证实验,与K-means,FCM,马尔可夫随机场和具有区域系数的OGMRF等方法相比较,所提出方法的总体分类精度达到94.16%,Kappa系数为0.8957,在5种方法中效果最好。The classification of synthetic aperture Radar(SAR)images is one of the key technologies in the field of remote sensing applications.To address the problem that regional class labels affect the classification accuracy in the object-based Markov random field(OMRF)model,this paper proposes the concept of regional category fuzzy probability(RCFP)label field,which can effectively avoid the misclassification caused by wrong class labels by fully considering the possible categories of a single object.The RCFP of every region can be obtained using the regional edge information and posterior probability according to the features of the region and its adjacent regions.Then it is included in the calculation of feature field parameters to make the feature field parameters highly close to the real conditions of objects.The study area is located in the eastern part of Kaifeng City,Henan Province,covering an area of about 1400 km 2.Sentinel-1 SAR images were used for the classification experiment of farmlands,buildings,and water in the study area,and the performance of the improved method in this study was compared with that of the method of K-means,fuzzy C-means(FCM),MRF,and OGMRF-RC.The experimental results show that the overall accuracy(OA)and the Kappa coefficient of the proposed method are 94.16%and 0.8957 respectively,which are higher than those of other methods.

关 键 词：SAR图像分类 马尔可夫随机场 特征场 区域类别模糊概率 Sentinel-1 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]