Hyperspectral target detection via exploiting spatial-spectral joint sparsity

In this paper, we propose a new spatial-spectral joint sparsity algorithm for target detection in hyperspectral imagery (HSI). The proposed algorithm embeds the sparse representation (SR) into the conventional subspace target detector in hyperspectral images. This algorithm is based on such an idea that a pixel in HSI rely on a low-dimensional subspace and can be represented as a sparse linear combination of the training samples. Substituting SR for the conventional subspace method, a sparse matched subspace detector (SMSD) is developed. Moreover, 3D discrete wavelet transform (DWT) and independent component analysis (ICA) are exploited to extract the spatial and spectral distribution information in the hyperspectral imagery and capture the joint spatial-spectral sparsity structure. By integrating the structured sparsity and the SMSD, the proposed algorithm is able to carry out target detection task in the hyperspectral images. Experiments are conducted on real hyperspectral image data. The experimental results show that the proposed algorithm outperforms both the conventional matched subspace detector (MSD) and the state-of-the-arts sparse detection algorithm.