Kernel Sparse Representation for Anomaly Detection in Hyperspectral Imagery

Aiming at the problem of anomaly detection of hyperspectral imagery sub-pixel targets, this paper proposes a novel algorithm of kernel sparse representation. The key idea is that background pixels can be sparsely represented by their neighborhoods, whereas anomalous pixels cannot, and the algorithm can be solved by orthogonal matching pursuit algorithm. The algorithm introduced the kernel method to map the original data to high-dimensional space, and derived the kernel orthogonal matching pursuit(KOMP) algorithm to solve the pixel non-linear mixed problem. Through the introduction of "sum to one" constraint, the detection value of anomalous pixels and background pixels is increased, which is more conducive to detection. The regularization model is added to make the solution of kernel orthogonal matching pursuit algorithm more stable. Finally, this algorithm is compared with other algorithms on the actual hyperspectral data set, and achieved good results.