Maximum Relevance, Minimum Redundancy Band Selection for Hyperspectral Images

In this paper, we propose a new band selection method for hyperspectral images based on normalized mutual information. Relevance of selected band set to class labels has been measured by average of normalized mutual information between each of them and class label and Redundancy of them is measured by average of normalized mutual information between each pair of them. Based on relevance of bands and redundancy of them, we propose a cost function that maximize relevance of selected bands and simultaneously minimize redundancy between them. We use a greedy search algorithm for optimizing this cost function. We compare the results of this method with other band selection algorithms and feature extraction algorithms PCA and LDA. Mutual information accounts for higher order statistics, not just for first and second orders as PCA and LDA do. Hence mutual information is a better criterion for hyperspectral images, because they have higher order statistics than two. Our classification results for AVARIS data shows proposed method outperform usual methods.

Keywords:

Hyperspectral Image , Classification , Hughes Effect , Band Selection , Mutual Information , MaximumRelevance , Minimum Redundancy , Greedy Search Algorithm