There are often demands for region target detection such as air pollution detection and oil spills monitoring, even though small target detection has gained much attention in the field of hyper spectral detection. In this paper, we present a multiscale-multivariate autoregressive (MMA) method to handle such region targets, considering the spatial correlations among different band and scale. Some preliminary deductions related to MMA model based detection and the corresponding stability computation method has been introduced. Experimental results on EPS-A data show that our approach is effective in detecting target region for hyper spectral imagery.