This algorithm is proposed for sparse representation anomaly target detection of hyperspectral image based on the spatial-spectral combination,which makes full use of t he spatial and spectral properties of hyperspectral image.Firstly,the multi-scale Gauss filter is used to filt er the original hyperspectral image,which will reduce the influence of the noise on anomaly targets.Second ly,the subspace is divided through the correlation coefficients between adjacent bands of hyperspectral images.Then,the subspaces are detected for the anoma ly objects to use the sparsity divergence index.Finally, we get the result of final target detection to superimpose subspace detection results.The simulation results show that this algorithm has high detection accuracy,and low false alarm rate,which improves the ability of hyperspectral anomaly target detection based on sparsity representation theo ry.