Implementation on Image Frame Selection of GPU for ONSET Real-time Data Processing

Optical and Near-infrared Solar Eruption Tracer(ONSET) can get a lot of sun image data every day. By real-time frame selection to these observation data, on the one hand can the storage pressure be reduced and the observation time be extended, one the other hand the quality of subsequent image reconstruction can also be improved. In this paper, a set of real-time processing image frame selection module based on GPU is designed and implemented for the frame selection in real-time observation process, and it has achieved high-speed parallel processing of gradient method and spectral ratio method in the module. This paper has discussed the implementation of the module in detail, and compared the speed of the two frame selection methods by experiment. Experimental results have shown that the module is stable and reliable. In the implementation efficiency, the best overall execution time of frame selection for nearly full view image is 1.2 seconds, which is seven times better than the original serial implementation. For partial view image it takes 0.7 seconds, and that makes an average efficiency-increase of five times. Implementation and current performance of the overall module have been able to meet the requirements of real-time observation and processing.