Hyperspectral Anomaly Dectection on Multicore DSPs

As one of the major technological breakthroughs made by human beings in earth observation since the 1980s, the good spectral diagnostic ability of hyperspectral images makes it very suitable for the discovery of artificial targets against the natural background and therefore receives more and more attention. Hyperspectral images are characterized by their high spectral resolution and large band. As they provide detailed observation information in more fields, they also bring about an increase in the amount of data redundancy, which brings about a great deal of difficulty corresponding transmission, storage, processing and application. In this paper, the multi-core DSP is applied to realize the hyperspectral images anomaly detection. Firstly, the hyperspectral image is split into several blocks. And then background spectral information in each blocks is extracted by Sherman-Morrison formula sequentially. Finally, the parallelization of multi-core DSP with high-speed computing performance can realize the realtime required in the application with RX detection algorithm. The real hyperspectral dataset is applied for hyperspectral image anomaly detection to verify the validity of the proposed method. Furthermore, comparing with MATLAB and CPU experimental results, DSP parallel detection system has better detection performance and high-efficient.