基于聚类和最佳指数的快速高光谱波段选择方法

Rapid hyperspectral band selection approach based on clustering and optimal index algorithm

最佳指数法是常用的高光谱图像数据波段选择方法,但存在运算时间过长的问题。运用K-means聚类算法,对最佳指数方法进行了改进,提出了聚类最佳指数法,并进行了一系列伪装目标识别的对比实验。实验结果表明,与最佳指数法相比,改进后的方法在保证目标分类精度的前提下,运算速度提高了数十倍;与单纯使用K-means聚类运算相比,不仅运算时间缩短,而且分类精度有所提高。利用改进算法能够在伪装环境下更加快速有效地识别目标。

Optimal index algorithm is a frequently-used hyperspectral band selection approach, but it takes too much computation time. Using the concept of K-means clustering algorithm, the clustering optimal index algorithm is presented to improve the optimal index algorithm and a series of contrast experiments for camouflage target recognition are conducted. The results indicate that, compared with optimal index method, in guarantee the premise of target classification accuracy, computational speed of the improved method increases dozens of times. Meanwhile, compared with K-means clustering algorithm, the operation time is shortened, the classification accuracy is improved. In camouflage environment, the target can be identified more quickly and effectively by the modified algorithm.