A joint kernel collaborative-representation-based approach for anomaly target detection of hyperspectral images

An anomaly target detection algorithm based on sparsity divergence index of joint kernel collaborative-representation is proposed in order to further improve the detection effect of abnormal target in the hyperspectral image.Firstly,the kernel collaborative-representation-based model is introduced;Secondly,in addition to the high spectral correlation,the spatial correlation is taken into full consideration at the same time.By combining with the representation method of the sparsity divergence index,the sparsity divergence index models based on spectral and spatial kernel collaborative-representations are proposed respectively.Then,a new sparsity divergence index representation model based on joint kernel collaborative-representation is proposed.At last,by the experiment with simulated hyperspectral image data,we discuss the effects of dual window design on the detection results of the proposed algorithm.In the simulation experiments carried out with real AVIRIS hyperspectral image data,we analyze the effects of the selection of different bands and principal component analysis on the detection results.The proposed algorithm is compared with the local RX algorithm,local kernel RX algorithm,the algorithm based on collaborative representation,the algorithm based on local sparsity divergence and the algorithm based on sparsity divergence index weighting.The experimental results show that the proposed algorithm has higher precision and lower false alarm probability.