贝叶斯方法和非参数模型支持下的遥感影像线性光谱分解  

Linear Spectral Unmixing of Remote Sensing Images Using Bayesian Method and Non-parametric Model

作　　者：李熙[1] 陈锋锐[2] 李畅[3] LI Xi CHEN Feng-rui LI Chang(State Key Laboratory of Information Engineering in Surveying, Mapping and Remote Sensing, Wuhan University, Wuhan 430079, China Key Laboratory of Geospatial Technology for the Middle and Lower Yellow River Regions, Ministry of Education, Henan University, Kaifeng 475001, Henan Province, China College of Urban and Environmental Science, Huazhong Normal University, Wuhan 430079, China)

机构地区：[1]武汉大学测绘遥感信息工程国家重点实验室,武汉430079 [2]河南大学黄河中下游数字地理技术教育部重点实验室,河南开封475001 [3]华中师范大学城市与环境科学学院,武汉430079

出　　处：《应用科学学报》2016年第6期661-669,共9页Journal of Applied Sciences

基　　金：湖北省自然科学基金(No.2014CFB726);国家自然科学基金(No.41101413,No.41401503);2015测绘地理信息公益性行业科研专项项目基金(201512026);中央高校基本科研业务费专项资金项目基金(No.2042016kf0162)资助

摘　　要：针对遥感影像混合像元分解中的地物光谱不确定性问题,提出利用非参数模型来刻画地物光谱的概率分布,并基于贝叶斯方法得到地物面积比例的后验概率分布,最后利用无偏估计和最大似然估计来估算地物面积比例.通过Landsat遥感影像不透水层制图的实验表明,所提方法的分解精度高于传统的线性光谱分解算法和硬分类方法,证明了贝叶斯方法能够较好地解决地物光谱不确定的问题.To solve the problem of spectral uncertainty in spectral unmixing of remote sensing images, a non-parametric model is proposed to describe the probability distribution of land cover spectrum. A Bayesian method is used to derive a posteriori probability distribution of the proportion of land cover types. Proportions of land cover types are calculated using unbiased estimation and maximum likelihood estimation. Experiment is carried out to map the impervious surface using a Landsat image. The results show that the proposed Bayesian method has unmixing algorithms and the method of the spectral uncertainty problem. higher accuracy than conventional linear spectral hard classification, therefore is effective in solving

关 键 词：遥感 端元 混合像元 贝叶斯 非参数模型 

分 类 号：P751.1[交通运输工程—港口、海岸及近海工程]