Abstract: An adaptive kernel collaborative representation based anomaly detector for hyperspectral imagery was proposed in view of target detection. Although the sparse representation considers the sparsity of weighted vector, the collaborative representation puts more emphasis on each atoms contribution to the linear combination. A comparability weighted regularization matrix and sum-to-one constraint were employed to reinforce the stability and separability. Then, the kernel parameter was adaptively estimated by using local statistics to improve its own local adaption. The proposed AKCRD algorithm was applied to two hyperspectral datasets comparing with other algorithms such as RX, KRX, SVDD, CRD and KCRD. The simulation results show that the proposed algorithm has better detection performance.