Cross Correlation Based Clustering for Feature Selection in Hyperspectral Imagery

Abstract

One of the main problems with hyperspectral image processing is to be contained large amount of data. Furthermore, pattern recognition methods are highly sensitive to problems related to high dimensional feature spaces. Therefore, feature selection in hyperspectral remote sensing data is investigated by researchers. This paper propose a clustering strategy that divides a feature set into subsets within which features are closely related to each other by means of cross correlation between all spectral bands. After that a band selection strategy based on Minimum Redundancy Maximum Relevance (mRMR) eliminates redundant bands into band clusters. The effectiveness of the proposed method is carried out on a real hyperspectral data set. The obtained results clearly affirm the superiority of the proposed method.