一种自动获取端元的RMS误差迭代改进算法  

An Improved Endmember Extraction Algorithm Based on Iterative RMS Error Analysis

作　　者：郑淑倩[1] 张友静[1,2] 邓世赞[3] 

机构地区：[1]河海大学地球科学与工程学院,南京210098 [2]河海大学水文水资源与水利工程科学国家重点实验室,南京210098 [3]江苏省地质勘查技术院信息中心,南京210008

出　　处：《遥感信息》2012年第5期19-25,共7页Remote Sensing Information

基　　金：国家自然科学基金项目(40830639);国家"973"计划项目(2010CB951101)

摘　　要：端元提取是混合像元分解算法中的关键技术之一,端元的质量直接影响分解结果的精度。本文对基于均方根(RMS)误差分析迭代提取端元的算法进行了改进,提出在端元选择时,增加像元纯净指数(PPI)、光谱矢量距离以及RMS误差值作为约束条件。利用南京地区2002年TM遥感影像作为试验数据,用本文提出的方法提取各组分丰度图,结合V-I-S模型以及研究区的实际情况,分析所提取的各组分丰度空间分布合理性,参考同期IKONOS影像解译结果,对改进前后的分解算法进行精度比较。试验结果表明:基于改进法得到的各组分结果精度较好,其与实测值的回归曲线在相关系数、斜率以及截距方面均得到了较明显的改善,但对于光谱非线性混合现象较严重的地物仍存在一定局限性。Endmember extraction is the key technology of mixed pixel decomposition algorithms and the quality of endmembers directly affects the precision of the result.This paper proposed an improved iterative algorithm based on RMS error,which took the index of Pixel Purity Index（PPI）,the distance between spectral vectors and the value of RMS error into consideration.The parameter of pixel purity index was used to guarantee the purity of pixels that participate in calculating the spectral vector of endmembers.The distance between spectral vectors and the value of RMS error can improve the precision of similarity judgment,which can help to avoid mixing pixels of other ground features.Landsat TM image of Nanjing area acquired in the year of 2002 was used as experimental data to extract the component abundance map in the proposed method,and the rationality of spatial distribution of various components was then appraised based on the V-I-S model and the actual situation of the study area.The interpreted IKONOS image of the same time was used to evaluate the precision and the effectiveness of the improved algorithm.The results show that the correlation coefficient,slope and intercept of linear regression equations are significantly improved,but it still has some limitations to the ground objects whose spectrum are non-linearly mixed.

关 键 词：端元提取 PPI RMS误差分析迭代算法 光谱矢量距离 V-I-S 

分 类 号：TP79[自动化与计算机技术—检测技术与自动化装置]