基于类间可分性的波段选择算法  

Band Selection Algorithm Based on Inter-Class Separability

作　　者：张立国[1] 孙胜春 王磊[1] 金梅[1] 张勇[1] 刘博[1] Zhang Liguo;Sun Shengchun;Wang Lei;Jin Mei;Zhang Yong;Liu Bo(Hebei Key Laboratory of Measurement Technology and Instrument,Yanshan University,Qinhuangdao,Hebei 066004,China)

机构地区：[1]燕山大学河北省测试计量技术与仪器重点实验室,河北秦皇岛066004

出　　处：《激光与光电子学进展》2022年第4期409-416,共8页Laser & Optoelectronics Progress

基　　金：河北省中央引导地方科技发展专项(199477141G);河北省引智项目。

摘　　要：如何从图像中选择出分类效果好的波段组合是高光谱图像分类任务的关键问题。针对上述问题,提出一种基于单波段图像类间可分性和波段间相关性的波段选择算法。根据类间可分性原则,采用单波段图像中各类样本点矩阵的均值和标准差来衡量单波段图像的类间可分性,结合波段间的相关系数来选择出类间可分性好、波段间相关性低的波段组合。最后对所提算法波段选择前后的图像和自适应波段选择算法波段选择后的图像进行支持向量机分类。在Indian Pines和Salinas数据集上的分类结果表明,当波段选择的光谱波段数目为20个,分类训练集为每类地物随机抽取20个样本点时,所提算法的总体分类精度较自适应波段选择算法分别提高了7.34个百分点和2.96个百分点。How to select a combination of bands with a good classification effect from an image is a key issue in the task of hyperspectral image classification. Aiming at the above problems, a band selection algorithm based on the separability of single-band image categories and the correlation between bands is proposed. According to the principle of inter-class separability, the mean and standard deviation of all kinds of sample point matrices in single-band images are used to measure the inter-class separability of single-band images. Combined with the correlation coefficient between bands, the band combinations with good inter-class separability and low inter-band correlation are selected.Finally, the images before and after band selection of the proposed algorithm and the images after band selection of the adaptive band selection algorithm are classified by support vector machine. The classification results on Indian Pines and Salinas datasets show that when the number of spectral bands selected is 20 and the classification training set is randomly selected 20 sample points for each type of ground objects, the overall classification accuracy of the proposed algorithm is improved by 7. 34 percentage points and 2. 96 percentage points respectively compared with the adaptive band selection algorithm.

关 键 词：遥感 高光谱图像 波段选择 类间可分性 

分 类 号：TP751.1[自动化与计算机技术—检测技术与自动化装置]