Fast high-order matched filter for hyperspectral image target detection

Hyperspectral image target detection is an important application for both of its civil and military uses. Traditional hyperspectral image target detection algorithms are usually designed based on the second-order statistics of the data where it is assumed that the target follows a Gaussian distribution. However, due to the low spatial resolution of the hyperspectral sensor, sometimes the targets of interest only occupy a few pixels. In this case, targets are more suitable to be described by high-order statistics. In this paper, we propose a fast high-order matched filter (FHMF) algorithm which describes the essence of the data more properly and then formulate the detection problem as an optimization problem. To solve the optimization problem efficiently, a fixed-point algorithm is proposed inspired by the FastICA algorithm of the signal processing field. The experiments are conducted with a synthetic hyperspectral image and a real hyperspectral image. The experiment results show that FHMF has better detection performance than other classical detection algorithms. In addition, the fast convergence speed also demonstrates the effectiveness of the proposed fixed-point algorithm.