GPU implementation for real-time hyperspectral anomaly detection

Hyperspectral anomaly detection which has been widely used to find targets on a timely basis requires high computing performance. In this paper we further study the real-time processing of anomaly target detection algorithm (RT-CR-RXD) and propose a new implementation on graphics processing units (GPUs). In the implementation, a real-time process of target detection is simulated, where the pixel data can be sent into detector firstly after collected and then the detector gives result for RT-CR-RXD immediately. And we have taken advantage of graphics processing units (GPUs) in parallel to accelerate the complex calculation. We achieve RT-CT-RXD on the parallel structure and it can be easily further introduced into embedded systems. The presented developments are tested in different scenarios (synthetic and real hyperspectral data) using two different GPU architectures by NVIDIA: GeForce GTX750Ti and GTX610. The results reveal significant speedup compared to CPU implementation at same detection accuracy.