Research On Real - Time Correction System Of Hα Full - Day Cloud Pollution Based On

At present, Ha full-disk solar image observation is the primary way of monitoring solar activity. There are numerous astronomical observation stations around the world, keeping the monitoring of Hα full-disk solar image observation all the time. Although strict standards have been demanded in selecting sites, it is inevitable that there are still lots of days with clouds in routine observation. Sky clouds lead to solar images covered with shadows, which obscure the details of solar features.If we can achieve the function of detecting and removing cloud shadows, it will be very beneficial in the process of observation and judging the quality of real-time images. To meet those demands, this paper mainly studied the following aspects:First, by summarizing predecessors’achievements in quality assessment and correcting distortions of Ha full-disk solar image, we further studied the algorithm of detecting and removing cloud shadows which can better apply in our real-time correcting system.Second, for detecting cloud shadows, graphic processing unit (GPU) with compute unified device architecture (CUDA) parallel programming environment is successfully employed to the related algorithm. The parallel computing steps include: in identifying heavy cloud-covered images procedure, we first obtain the location of the center and the length of the radius, then calculating the ratio of major and minor axes of a fitted ellipse; in recognizing recoverable cloud-covered images procedure, a four-step approach is applied:(1) centering a full-disk image by parallel moving; (2) translating the image from Cartesian coordinates to polar coordinates by bilinear interpolation; (3)sorting the data in every row of four quadrants of polar coordinates by Bitonic sort; (4)calculating the cross-correlation coefficients of four median limb-darkening profiles.Third, to find out the proper filtering method which is a critical step in removing cloud shadows, this paper further studied median filtering, morphological filtering and Butterworth low-pass filter in frequency domain, leading them parallel implement in CUDA programming environment. In the removal procedure, we first compare the cloud-covered image with a "Quiet Sun", then filter out the details of solar features, at last compare the cloud-covered image with the transmittance of clouds. We also utilize the Structural Similarity Index Measurement(SSIM) algorithm for evaluating the performance of restoration. The results demonstrate that three methods are significant effective. In view of the outstanding advantages of second-order Butterworth low-pass filter in frequency domain in processing speed, we consider it as the best choice.Fourth, in Microsoft’s MFC framework, by calling the CUDA functions which can been parallel implement in GPU, we successfully develop a complete software system with visual interface. The system can achieve the function of detecting and removing cloud shadows in real-time.After testing, this system can’t only effectively distinguish the heavy cloud-covered images, the recoverable cloud-covered images and the none cloud-covered images, but the cover level can be quantified. For the recoverable cloud-covered image, the system can get it repaired, In the experiment, the total processing time is about 0.7s for one cloud-covered image. Compared to one-minute intervals, the system can fully meet the need of displaying in real-time.