A fast source separation algorithm for hyperspectral image processing

This paper describes a new algorithm for feature extraction in hyperspectral images based on independent component analysis (ICA). The improvement introduced aims at reducing the computation times without decreasing the accuracy. Instead of using the entire image, we perform ICA processing on a subset of representative pixel vectors obtained through spectral screening. Spectral screening is a technique that measures the similarity between pixel vectors by calculating the angle between them. In multispectral/hyperspectral imagery, the independent components can be associated with features present in the image. ICA projects them in different image frames. The features are separated using an algorithm involving gradient descent minimization of the mutual information between frames. The effectiveness of the proposed algorithm (SSICA) has been tested by performing target detection on data from the Hyperspectral Digital Imagery Collection Experiment (HYDICE). Small targets present in the image are separated from the background in different frames and the information pertaining to them is concentrated in these frames. Further selection using kurtosis, skewness and histogram thresholding lead to automated detection of the targets allowing a quantitative assessment of the results. When compared with a target detection ICA algorithm previously introduced by the authors, SSICA achieves similar accuracy, and, at the same time, considerable speedup is obtained.