GPU Implentation of Recursive Automatic Target Generation Process and Recursive Orthogonal Subspace Projection in Hyperspectral Imagery

Despite of the fact that recursive automatic target generation process (RATGP) and recursive orthogonal subspace projection (ROSP) have played a crucial role in a hyperspectral unmixing field. However, it's hard to process rapidly in terms of the amounts of data in a hyperspectral remote sensing image. With the feature of graphic processing units (GPUs), calculating a large-scale parallel computing, the paper was studied through the method of parallel and optimization to the RATGP and ROSP algorithm. Therefore, we proposed a parallel design method based on matrix operation. The experiment results showed that the mechanism of optimization has been greatly improved the efficiency of the algorithm in a real hyperspectral image under the condition of keeping precision of unmixing results.