CFAR target detection in unknown background based on subspace projection in aerial hyperspectral imagery

A detection algorithm is presented to detect the targets in unkown background in aerial hyperspectral imagery. Spectral signatures of background endmembers can be obtained by fuzzy clustering because of targets' low probabilities. Then, in order to suppress the background spectral signature and noise, the hyper-spectral data cube are projected onto the orthogonal subspace of background spectral signatures and targets spectral signatures subspace. Finally, a constant false alarm rate (CFAR) detector is constructed by means of generalized likelihood ratio test (GLRT). Theoretic analysis and experimental results verify the effectiveness of the algorithm.