Anomaly Detection in Hyperspectral Imagery Based on Kernel ICA Feature Extraction

Abstract

A kernel-based independent component analysis algorithm, which combines kernel principal component analysis (KPCA) and independent component analysis (ICA) is proposed for anomaly detection in hyperspectral imagery. The conventional RX anomaly detector suffers from high false alarm rates and low probability of detection. In this paper, KPCA is performed on a feature space to whiten data and fully mine the nonlinear information between spectral bands. Then, ICA seeks the projection directions in the KPCA whitened space for making the distribution of the projected data mutually independent. Finally, RX detector is performed on the projected data to locate the anomaly targets. The kernel ICA algorithm extracts the nonlinear independent components along with the dimensional reduction, and improves the performance of RX detector in hyperspectral data. Numerical experiments are conducted on real hyperspectral imagery collocted by the Airborne Visible/Infrared Imaging Spectrometer (AVIRIS). Using receiver operating characteristic (ROC) curves, the results show the improved performance and reduction in the false-alarm rate.