Kernel-based subpixel target detection for hyperspectral images

In this paper, a kernel-based subspace detection algorithm is proposed for hyperspectral subpixel target detection, which combines Kernel principal component analysis (KPCA) with Linear mixture model (LMM). The LMM is used to describe each pixel as mixture of target, background and noise. The KPCA is used to build background subspace. Finally, a normalized statistical detector maximizing Signal-to-noise (SNR) is used to detect whether each pixel includes target. The algorithm has two merits. First, high order statistics of local regions are exploited to search anomaly regions in order to reduce processing time and improve performance of detection algorithm. Second, the KPCA can better construct subspaces of target and background from nonlinear data. The numerical experiments are performed on AVIRIS data with 126 bands. The experimental results show that the algorithm has good detection performance and good ability to restrain background and can commendably overcome spectral variability of the targets.