An Efficient Dimensionality Reduction Method for the Classification of Satellite Remote Sensing Hyperspectral Images

Finding an informative subset of features from the original hyperspectral images has become essential because of its wide applications in ground object identification. However, information extraction from hyperspectral images is becoming challenging because of its high correlation among the image bands and spectral and spatial redundancy. This paper proposed a feature reduction approach, combining both feature extraction and feature selection. A combination of Minimum Noise Fraction (MNF) and information-based measure, cross cumulative residual entropy (CCRE), is proposed to select the subset of features from the original image to obtain improved classification accuracy. In the proposed method, feature ranking is improved by scaling the CCRE to a specific range to avoid redundant features. The proposed technique (MNF-nCCRE) is tested on two…