Research On The Near Real-time Frame Selection Technology Of ONSET Telescope

The Optical and Near-infrared Solar Eruption Tracer(ONSET)is an important device in China’s solar physics research.The telescope can obtain a large amount of solar image data every day,which brings tremendous pressure to the entire data processing and storage.The research on real-time frame selection of observation images is helpful to quickly select effective data,reduce storage pressure,and improve the efficiency and quality of ONSET observations.The research work has important practical significance and application value.This thesis starts from the current common subjective and objective evaluation of image quality research,and closely combines GPU parallel technology to solve the problem of ONSET real-time data processing.It focuses on the subjective and objective consistency of the execution time of frame selection processing and the frame selection results based on the characteristics of ONSET observation data.The study includes the following four parts:First,a set of real-time processing modules for frame selection was designed and implemented,and 20 kinds of image quality evaluation algorithms were integrated and implemented under the module.Among them,the main research was based on the image grayscale gradient distribution class,the image-based grayscale probability distribution class,and the Realization of three kinds of frame selection algorithms in Fourier transform frequency domain method of image,the three kinds of algorithms are discussed from the perspective of improving the original algorithm or parallel optimization.Secondly,the experiment uses five sets of data whose average image quality is above the ONSET observation to verify the effectiveness of the three types of improved methods.The stability of the real-time processing module for frame selection is demonstrated,and the frequency domain method under local surface data is demonstrated.The execution time is optimal and the average time is less than one second,which meets the real-time requirements of ONSET frame selection.Thirdly,we select ten sets of test data for ONSET observed white light and Hα wavebands of good,middle and poor levels,and define the three criteria of subjective evaluation of various frame selection methods based on the principle of subjective and objective consistency.The method of selecting frame has superiority;in the experiment of verifying the frame selection effect by calculating the SNR evaluation value for various frame selection methods,it is considered that the spectrum ratio method integrated in the frame selection real-time processing module can satisfy the real-time performance of frame selection.Fourth,with the acquisition and storage system applied to the actual ONSET observation,the real-time performance of the selected frame in the serial environment is further studied.Referring to the frame selection results of spectral ratio method,combined with the file overlap ratio and frame execution time of the frame selection results and spectrum ratio methods of various methods,it is demonstrated that the average gradient improved algorithm can achieve good frame selection accuracy under the serial environment.Based on the average gradient improvement method,the influence of the calculation region selection size and position factor on the accuracy and speed of the selected frame is analyzed and the conclusion is drawn that the calculation window is selected in the solar activity region and the solar activity region for the white light band and the Hα band image respectively.In addition,the window size is 400?400,which can achieve high frame selection accuracy and speed and meet the real-time frame selection requirements.Taken together,the work of this paper has studied the frame selection time and effect of ONSET observation images,covering the frame selection processing of ONSET observation system in different operating environments.In the environment with NVIDIA GPU devices,using the frame selection and real-time processing module to accelerate the frame selection algorithm,the parallel spectrum ratio method can achieve the performance of real-time frame selection;under the conventional serial environment,the appropriate size and position calculations are selected based on the existing conclusions.Area,average gradient improvement method can meet the needs of real-time frame selection.