A sparse representation and Cauchy distance combination graph for hyperspectral target detection

Hyperspectral target detection under complex background is a challenging and difficult task in remote-sensing earth observation. However, most existing algorithms assume that the background obeys the multivariate Gaussian model and ignores the complex spatial distribution. In this work, a hyperspectral target detection method based on sparse representation and Cauchy distance combined graph (SRCG) model is proposed. Firstly, pure dictionary sparse representation is used to obtain the similarity of the prior target pixel and test pixels. Secondly, the pixel-to-pixel Cauchy distance of the hyperspectral image is evaluated. Finally, the vertex edge graph pixel selection model is constructed to obtain the desired target pixels. The experimental results demonstrate the priority of the SRCG on six public and our collected hyperspectral datasets.