Regularization Framework for Target Detection in Hyperspectral Imagery

Target detection in hyperspectral imagery (HSI) is of great interest in the signal processing field. The key to the target detection methods lies in the proper estimation of a variety of measurement matrices that are estimated from the HSI. However, these matrices are usually ill conditioned due to the high dimension of the HSI or the inherent correlation between different image bands, which can result in inaccurate inverse matrices estimation. Therefore, how to handle the potentially inaccurate inverse calculation greatly affects the detection performance. This letter proposes a regularization framework that is suitable for the state-of-the-art measurement matrices used in target detectors. It adds a scaled identity matrix to these matrices in order to strengthen the stability of the inverse matrices, and, consequently, improve the detection performance. Extensive experiments were carried out on HSI that revealed the regularized detectors clearly outperformed the original detectors.