基于局部连续性与全局相似性的光谱保持型亚像元映射  

Spectrum preserving Sub-pixel Mapping Based on Local Connectivity and Nonlocal Similarity

作　　者：黄慧娟[1] 禹晶[1] 肖创柏[2] 孙卫东[1] 

机构地区：[1]清华大学电子工程系,北京100084 [2]北京工业大学计算机学院,北京100124

出　　处：《自动化学报》2014年第8期1612-1622,共11页Acta Automatica Sinica

基　　金：国家自然科学基金(61171117);国家科技支撑计划项目(2012BAH31B01);北京市教育委员会科技计划重点项目(kz201310028035)资助~~

摘　　要：遥感图像的像元级分类精度受混合像元的影响.亚像元映射以像元分解获得的丰度值为基础,在地物分布规律的约束下,细化估计各类地物的亚像元级分布模式.本文同时考虑了地物分布的空间与光谱信息,提出了一种基于局部连续性与全局相似性的光谱保持型亚像元映射算法.针对地物的空间分布特性,提出了利用类内离散度对局部连续性进行建模,并通过相似分布像元表示误差引入全局相似性约束项.针对地物的光谱特性,采用最小化光谱误差约束了亚像元映射过程中的光谱无失真性.模拟数据与真实数据上的实验结果表明,本文算法比其他同类算法具有更高的估计精度,且更适合于实际应用.In remote sensing images, the accuracy of land cover classification at pixel scale is affected by mixed pixels greatly. Sub-pixel mapping tries to predict land-cover map at sub-pixel scale according to spectral unmixing abundances and some constraints of land-cover distribution patterns. In this paper, using both spatial and spectral information of land- cover, we propose a new spectrum preserving sub-pixel mapping algorithm based on local connectivity and as a constraint similarity. Spatially, local dependence is re-modeled by the with-in class scatter, nonlocal similarity is introduced by minimizing the representation errors among similar pixels. Spectrally, spectrum preserving is realized by minimizing the spectra errors in sub-pixel mapping. Comparative experiments with artificial and real images show that the proposed algorithm achieves a higher accuracy than other related algorithms, thus it is more suitable for practical application.

关 键 词：亚像元映射 像元分解 局部连续性 全局相似性 光谱保持 

分 类 号：TP751[自动化与计算机技术—检测技术与自动化装置]