太阳高分辨高速重建算法的研究

发布时间：2018-03-17 02:24

本文选题：新真空太阳望远镜　切入点：太阳图像重建　出处：《中国科学院研究生院(云南天文台)》2016年博士论文　论文类型：学位论文

【摘要】：深入研究太阳大气活动现象需要大视场、长时间和多波段的高时空分辨率数据。目前地基大口径太阳望远镜都借助于自适应光学技术和高分辨图像重建技术实现太阳高空间分辨率成像观测。高分辨图像重建技术能有效克服湍流影响而实现目标接近接近望远镜衍射极限分辨率成像,相比自适应光学技术,它容易操作和实现,但也因重建算法运算耗时、重建流程繁琐复杂而缺乏实时性。随着一米新真空太阳望远镜NVST在2010年建成和投入使用,国内太阳观测迈入高空间分辨率时代。NVST采用斑点重构方法来处理每日观测数据,为能及时快速的处理海量观测数据,需有一套高效的重建策略和高速运算的重建算法。本论文结合NVST实际观测,开展太阳高分辨高速重建算法的研究,研究内容包含NVST数据重建模式、重建算法优化以及算法并行运算等几个方面。具体的研究细节和取得的相应进展如下:根据NVST实际观测和数据处理现状,确立了两级数据处理模式Level1和Level1+,在斑点重构算法比较耗时的大背景下,两级处理模式让NVST做到了数据及时发布。此外,在保证数据处理精度的前提下,对位移叠加法和斑点掩模法进行了深入分析和优化,对其中关键参数进行了实验分析,最大限度地精简了运算量,提高了观测数据重建效率。在算法优化的基础上,采用硬件加速方式提高了高分辨重建算法的运算速度。论述了高分辨重建算法中能够并行计算的环节,就高分辨重建算法在GPU和集群中并行加速进行了实验,在GPU中实现了准实时的Level1级重建并实现了子块图像四维重谱并行计算;另外在集群中也实现了Level1和Level1+级快速重建。对NVST多通道数据的实际处理展开Level1和Level1+处理方法的应用研究,并针对实际观测和数据重建中遇到的问题给出解决方案。提出了一种重建中抑制光球通道s CMOS相机固定图形噪声的方法,通过对月亮Level1+级重建的结果看出,“二次平场”能有效抑制斑点掩模法中的固定图形噪声放大。在日面边缘目标的重建实验中,提出了一种改进的互相关方法,该方法有效避免了相关极值误差,保障了边缘目标高分辨重建的顺利进行。为提高色球数据重建结果的信噪比,采用并改进了Level1重建算法,在改进方法后,重建结果的分辨率大幅提高。
[Abstract]:A deep study of solar atmospheric activity requires a large field of view. Long time and multi-band high spatial and temporal resolution data. At present, the large aperture solar telescopes are based on adaptive optics technology and high resolution image reconstruction technology to realize the solar high spatial resolution imaging observation. High resolution image. The reconstruction technique can effectively overcome the influence of turbulence and achieve the target close to the telescope diffraction limit resolution imaging. Compared with adaptive optics, it is easy to operate and implement, but also because the reconstruction algorithm is time-consuming, the reconstruction process is complex and lack of real-time. With the new vacuum solar telescope NVST completed and put into use in 2010, In order to process large amount of observational data in time, NVST adopts speckle reconstruction method to process daily observation data in the era of high spatial resolution. It is necessary to have a set of high efficient reconstruction strategy and high speed reconstruction algorithm. This paper combines the actual observation of NVST and carries out the research of high resolution and high speed reconstruction algorithm of the sun. The research content includes the reconstruction mode of NVST data. The research details and corresponding progress are as follows: according to the actual observation and data processing status of NVST, the two-level data processing mode Level1 and Level1 are established. Under the background that the speckle reconstruction algorithm is time-consuming, the two-level processing mode enables NVST to release the data in time. In addition, the displacement superposition method and the speckle mask method are deeply analyzed and optimized under the premise of ensuring the accuracy of the data processing. The key parameters are analyzed experimentally, the computation is simplified to the maximum extent, and the efficiency of reconstruction of observation data is improved. On the basis of the optimization of the algorithm, Hardware acceleration is used to improve the speed of high resolution reconstruction algorithm. The parallel computation of high resolution reconstruction algorithm is discussed, and the parallel acceleration of high resolution reconstruction algorithm in GPU and cluster is tested. Quasi-real-time Level1 level reconstruction is realized in GPU and sub-block image four-dimensional respectral parallel computation is realized. In addition, Level1 and Level1 level fast reconstruction are also realized in cluster. The application of Level1 and Level1 processing methods to the practical processing of NVST multi-channel data is studied. According to the problems encountered in actual observation and data reconstruction, a method to suppress the fixed figure noise of the optical sphere channel s CMOS camera in reconstruction is proposed. The results of the Level1 reconstruction of the moon show that the "quadratic flat field" can effectively suppress the noise amplification of the fixed pattern in the speckle mask method. An improved cross-correlation method is proposed in the experiment of the reconstruction of the target on the edge of the solar plane. In order to improve the SNR of color sphere data reconstruction results, Level1 reconstruction algorithm is adopted and improved. The resolution of the reconstruction results is greatly improved.
【学位授予单位】：中国科学院研究生院(云南天文台)
【学位级别】：博士
【学位授予年份】：2016
【分类号】：P111.41

【参考文献】