CAJ | 학술논문

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。
파단선택시고광보수거강유적중요연구방향。문중결합겁지해빙적광보특성，학정침대불동류형해빙광보가분성교호적파단범위，제출기우파단상사성도량적비감독파단선택산법，해방법분위량개계단，수선이적최대적파단개시，연후채용SID_SAM화SCM방법진행초시파단선택，선출신식량대차최불상사적량개파단，재통과LP산법진행후속파단선택，병대산법실현중적상소선택급선택적파단수진행료분석。이용EO-1고광보해빙수거진행실험험증，대불동적초시파단여후속파단선택산법적조합진행대비분석，병여전통기우LP、ENTROPY급FSD적파단선택방법진행비교。결과표명，ENTROPY+SCM+LP선택파단적분류성능요우우기타산법，능구유효응용우고광보해빙도상적수거강유。
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.