Research On GPU-based Reconstruction Of Solar Observation Images

The solar telescopes used at present in China are all ground-based telescopes, which are subject to the influence of earth’s atmosphere when capturing the image of the sun, resulting in the degradation of observed images. Therefore image reconstruction is necessary. In order to achieve real time reconstruction of the observed images, an application of new computer technology in processing the observed data is urgently needed owing to the developement of observation technology.This thesis aims to accelerate the rate of image reconstruction for achieving quasi-real time processing with the help of parallel technique and by utilizing the parallel processing unit in GPU under the framework of CUDA developed by NVIDIA Corporation. The thesis consists of two parts. The first part deals with the realization of Level1 quasi-real time reconstruction of the solar chromosphere images captured by NVST. In the second part, methods of high-speed reconstruction of large field photosphere images captured by ONSET are studied and implemented.Frames-selection technique and shift-and-add technique, which have been proved efficient after years of application in Yunnan Observatory, are used to achieve NVST Leve11 image reconstruction. In this thesis, Level1 reconstruction program is designed and implemented based on GPU parallel computing, thus enabling the program to have its reconstruction speed up by nearly 30 times to the effect of quasi-real time reconstruction without changing the original way of processing.The main contents of this part are as follows:First, frame selection of the image sequence by applying spectral ratio method.Second, displacement measurement of the Lucky images.Third, aligning and overlaying the above-mentioned images.In practical observation, there is no mature scheme for the reconstruction of ONSET large field photosphere image at present. This thesis studied and implemented the GPU-based Partition-Frame Selection-Reconstruction-Mosaicking of the ONSET images. Preliminary results show that the observation quality has improved while fulfilling the requirement for quasi-real time processing.The main contents of this part are as follows:First, segmentation of one images into sub=block.Second, frame selection of each sub-block facilitated by the technique of average gradient.Third, reconstruction of each sub-block by means of shift and add.Fourth, composition of each sub-block images into full-field images while fusing the block boundary.With further improvements, these two works have the potential to boost the efficiency of data processing as well as advance the scientific output when applied to the regular image processing of the two important solar telescopes in China, NVST and ONSET.