In order to solve the problem of setting sparsity level in sparse representation-based target detection algorithms

this paper proposes a novel collaborative representation-based algorithm for hyperspectral target detection

and then extends it into a kernel version.The key idea is that a background pixel can be approximately represented as a linear combination of its surrounding neighbors (background dictionary)

while a target pixel can only be approximately represented as a linear combination of its surrounding neighbors and the prior target spectrums (union dictionary).First the unknown pixel is collaboratively represented by the background dictionary and union dictionary

respectively.Then targets can be determined by comparing the reconstruction residuals.Experimental results on real hyperspectral data set demonstrate the effectiveness of our proposed detector as well as its kernel version when compared with other algorithms.