Multi-layer filtering approach for hyperspectral target detection

Abstract

Target detection is an important application in hyperspectral image processing field. In this paper, we propose a new target detection method incorporates the idea of using multi-layer spectral filters, which aims to boost the performance of the traditional detection methods. The proposed algorithm enhances the targets and suppresses the undesired backgrounds through a layer-by-layer filtering procedure. Several second-order algorithms for hyperspectral target detection have been proposed, such as Matched Filter (MF), Constrained Energy Minimization (CEM) and Adaptive Coherence Estimator (ACE). In this paper, a basic second-order filter detector such as CEM, MF and ACE is used to filter the spectral data. After each layer of filtering, we transform the spectral vectors with a nonlinear function based on the previous layer's filtering results. Through the layer-by-layer filtering process, we obtain the gradually increasing improvements of the detection performance. Experimental results for detecting targets in real hyperspectral image are presented with our multi-layer filtering approach. Our method suggests significant advantages on real hyperspectral data, and improves the performance of the classical second-order algorithms, such as CEM, MF and ACE.