基于相似性的高光谱海冰图像的波段选择研究  

Similarity-based band selection for hyperspectral image of sea ice

作　　者：张龙[1,2] 韩彦岭[2] 张云[2] 袁国良[1] 

机构地区：[1]上海海事大学,上海201306 [2]上海海洋大学,上海201306

出　　处：《电子设计工程》2015年第21期155-159,共5页Electronic Design Engineering

基　　金：国家自然科学基金资助项目(41376178);上海市科学技术委员会资助课题(11510501300);上海市青年教师培养计划(ZZHY13033);上海海洋大学科技发展专项基金(A2-0209-14-200070);上海海洋大学博士科研启动基金(A2-0302-14-300078)

摘　　要：波段选择是高光谱数据降维的重要研究方向。文中结合极地海冰的光谱特性,确定针对不同类型海冰光谱可分性较好的波段范围,提出基于波段相似性度量的非监督波段选择算法,该方法分为两个阶段,首先以熵最大的波段开始,然后采用SID_SAM和SCM方法进行初始波段选择,选出信息量大且最不相似的两个波段,再通过LP算法进行后续波段选择,并对算法实现中的像素选择及选择的波段数进行了分析。利用EO-1高光谱海冰数据进行实验验证,对不同的初始波段与后续波段选择算法的组合进行对比分析,并与传统基于LP、ENTROPY及FSD的波段选择方法进行比较。结果表明,ENTROPY+SCM+LP选择波段的分类性能要优于其他算法,能够有效应用于高光谱海冰图像的数据降维。Band selection has become an important research direction of hyperspectral data dimensionality. In this paper, we combine the spectral characteristics of the polar sea ice, to determine the spectral bands with better separability for different types of ice. We propose similarity-based unsupervised band selection algorithm. It includes two steps：first start with the band of biggest entropy, then we adopt the SCM and the SID_SAM methods to select two initial bands with the most distinctivity and information, finally use the LP-based algorithm to select the subsequent bands. And pixel selection and the numbers of selected bands in the algorithm were analyzed. In the experiment, we use the EO-1 hyperspectral data for analysis of combinations of different algorithms. Comparison with the conventional LP-based, ENTROPY-based and FSD-based method, the ENTROPY＋SCM＋LP-based method shows a better result in the way of classification performance. It can be effectively applied to dimensionality reduction of hyperspectral image of sea ice.

关 键 词：波段选择 相似性度量 高光谱海冰图像 分类 像素选择 

分 类 号：TN927.23[电子电信—通信与信息系统]