光学技术, 2018, 44 (5): 634, 网络出版: 2018-10-08
基于高光谱基本准则的波段选择方法
Band selection method based on hyperspectral fundamental criterion
高光谱图像 波段选择 子空间划分 离散波段指数 类间可分性因子 hyperspectral image band selection subspace division discrete band index inter-class separability factor
摘要
高光谱数据具有光谱波段多、维度高、数据量庞大的特点, 为了提高高光谱数据的处理速度, 需要进行降维处理, 而波段选择是高光谱降维的基本方法之一。综合考虑, 提出基于高光谱波段选择相关性、信息量及类间可分性的方法。通过虚拟维度确定高光谱图像的本征维数, 并根据波段间的相关系数进行子空间划分; 提出利用基于信息量的离散波段指数, 在各个子空间中计算出最大的波段指数构成子集; 根据类间可分性准则在子空间中选出可分性因子最大的合适波段。利用光谱角匹配选出最适合分类的波段, 组成最后的波段子集, 从而实现波段选择的降维处理。通过实验验证,所提方法与传统的最佳指数和自适应波段选择方法相比, 在一定程度上提高了高光谱图像的分类精度。
Abstract
Hyperspectral data has the characteristics of many spectral bands, high dimensions and large amount of data. In order to improve the processing speed of hyperspectral data, it is necessary to reduce the dimension. Band selection is one of the basic methods of hyperspectral dimensionality reduction. A method based on correlation, information and inter-class separability in the hyperspectral bands is proposed. The intrinsic dimension of the hyperspectral image is determined by virtual dimension, and the subspace is divided according to the correlation coefficient between the bands. The maximum band index is calculated from each subspace using the discrete band index to form the subset. Simultaneously, in the subspace, the appropriate band of the largest separability factor is selected according to the separability criterion. The band subset is formed by the most suitable classification band that is selected by using the spectral angle match. So, the processing of reducing dimensions by band selection has been realized. The experimental results show that compared with the traditional optimal index and adaptive band selection method, the proposed method has improved the classification accuracy of hyperspectral images.
严阳, 华文深, 刘恂, 崔子浩. 基于高光谱基本准则的波段选择方法[J]. 光学技术, 2018, 44(5): 634. YAN Yang, HUA Wenshen, LIU Xun, CUI Zihao. Band selection method based on hyperspectral fundamental criterion[J]. Optical Technique, 2018, 44(5): 634.