基于子空间划分的高光谱图像波段选择方法

摘要

高光谱遥感图像数据具有数据量大、波段冗余度大、波段间相关性强的特点,不利于图像的判读解译.如何从上百个波段中选出最优的波段组合对目标识别和分类是需要解决的问题.论文利用自动子空间划分法,结合相关系数矩阵"分块"的特点将所有波段进行大致划分,运用自适应波段选择法和光谱角制图算法进行波段选择.首先对所有波段利用相关系数矩阵划分子空间,再在各个子空间提取指数最大的波段,最后依据地物光谱可分性选取最佳的波段组合.最终计算论文方法与常用波段选择方法所选波段的相关系数和以及均方差,验证了论文方法的优越性.%Hyperspectral remote sensing image carries a large amount of information, large redundancy and strong band correlation, which is not conducive to the image interpretation.How to select the optimal band combination of target identification and classification from hundreds of bands is a problem to solve.In this paper, automatic subspace partition method is adopted, the correlation coefficient matrix "block" is used to divides all bands into several subspaces roughly.Then within each subspace, adaptive band selection and spectral angle plotting are conducted.Firstly, the correlation coefficient matrix is used to divide bands into several subspaces.Next, the biggest index of bands is selected.Finally, the best band combination is selected according to the spectral separability.At last, the correlation coefficient and mean square deviation of the bands which are selected from the proposed method and the conventional band selection methods are calculated and it verifies that the proposed method has certain advantages.