Supervised detection for hyperspectral imagery based on high-dimensional multiscale autoregression

A supervised detection algorithm is presented to detect the target region in hyperspectral imagery. In order to utilize the spatial scale information in hyperspectral data, the multiscale observation of hyperspectral imagery of different connected nodes at different scales are described by a high-dimensional autoregressive model. Then, a high-dimensional multiscale autoregression based detector to detect target region is constructed, utilizing the equality between joint distribution of various multiscale observations and that of the regression noise, and the multivariate t distribution statistics of the regression noise. Theoretical analysis and the experiment involving five performance indexes show that our detector is effective to detect target region in hyperspectral imagery.