Anomaly detection and GPU implementation for spatial-spectral kernel spectral angle

Anomaly target detection has become a hotspot of hyperspectral remote-sensing information processing. To fully utilize the potential information of hyperspectral images and to develop an efficient detection algorithm for hyperspectral imagery anomalies, SS-KSAM, a novel algorithm that integrates spatial and spectral information, was proposed. Without assuming a background model, the proposed algorithm can obtain intermediate results by calculating the summation of the kernel spectral angle between the target pixel and its spatial neighbor pixels. Then, the final results can be obtained through corrosion in extended morphology. Results showed that noise interference is removed and the false alarm rate decreased. A parallel optimization method that is based on the graphics processing unit (GPU)/compute-unified device architecture model was further proposed to improve algorithm efficiency. Simulation results show that the utilization of the parallel characteristics of GPU can shorten detection time and ensure high detection precision.