基于GPU的Hα全日面云污染实时改正系统研究

发布时间：2018-05-28 04:39

本文选题：并行计算 + 图像修复　；参考：《昆明理工大学》2016年硕士论文

【摘要】：Ha全日面太阳图像观测是目前监测太阳活动的主要手段。全球拥有众多太阳Hα像的观测站点,虽然各站点在选址时筛选严格,但是实际观测中仍会出现大量的有云天气。高空云层导致所观测的Ha全日面像上覆盖有一层云污染,使得图像上的太阳活动细节变得模糊不清。对云污染图像的实时检测和修复,将非常有利于观测的进行和对实时观测质量的判断。为此,本文主要进行了如下几个方面的研究：(一)在前人Ha全日面太阳图像质量评价和畸变Ha全日面太阳图像修复方法的基础上,进一步研究了适合实时改正系统云污染检测和修复的算法。(二)利用GPU,在CUDA环境下实现了Ha全日面云污染检测的相关并行处理算法。包括：在判断重度云污染图像时,采用均值法并行求得太阳区域的中心和半径,然后并行求得二值区域长短轴比值；在判断图像上是否有可修复云污染时,并行平移实现太阳区域中心化、并行实现双线性插值法把图像从直角坐标系转换到极坐标系、并行双调排序法对极坐标系下图像等分四部分区域每行分别排序、并行计算四部分区域中值临边昏暗曲线的相关系数。(三)对于修复中的关键性步骤：滤波,本文深入研究了中值滤波、形态学滤波及频域巴特沃斯低通滤波在CUDA中的并行实现方法。云污染修复即通过和一个标准全日面像模板对比,并通过滤波得到云层透过率,最后用云污染图像除以这个云层透过率。使用SSIM算法对三种滤波方法的修复结果做质量评价后发现,三者都能有效的滤除太阳上的活动细节。鉴于二阶巴特沃斯低通滤波在处理速度上的突出优势,系统将其作为最优选择。(四)在微软MFC框架下,通过调用在GPU中并行执行的CUDA函数成功开发出一套完整的、具有可视化界面的Ha全日面图像云污染实时检测和修复的软件系统。经测试,本系统可以有效的区分出重度云污染、可修复云污染和无云污染的图像,并能对云污染的程度进行量化。对于可修复的云污染,可以进行有效的实时修复。在我们的实验环境下,1分钟的观测图像,可以在0.7秒内完成一幅云污染图像的处理,完全符合实时要求。
[Abstract]:Ha solar image observation is the main method to monitor solar activity at present. There are many observational stations of solar H 伪 image all over the world. Although the site selection is strict, there will still be a large number of cloud weather in the actual observation. High-altitude clouds cause a layer of cloud pollution over the observed Ha-Sun image, blurring the details of solar activity on the image. The real-time detection and restoration of cloud pollution images will be very helpful to the observation and the judgment of the real-time observation quality. Therefore, in this paper, we mainly study the following aspects: (1) on the basis of the quality evaluation of Ha solar image and the distorted Ha solar image restoration method, Furthermore, the algorithms suitable for cloud pollution detection and repair in real time correction system are studied. (2) using GPU to realize the parallel processing algorithm of Ha sun-surface cloud pollution detection in CUDA environment. It includes: when judging the image of heavy cloud pollution, the center and radius of the solar region can be obtained by means of the mean method, and the ratio of the long and short axis of the binary region can be obtained in parallel; when judging whether there is any remediable cloud pollution on the image, The parallel translation realizes the center of the solar region, the bilinear interpolation method transforms the image from the right-angle coordinate system to the polar coordinate system, and the parallel bi-tone sorting method sorts the image in four parts of the polar coordinate system. The correlation coefficients of the four region median edge darkening curves are calculated in parallel. (3) for the key steps of restoration: filtering, the parallel implementation methods of median filter, morphological filter and Butterworth low-pass filter in frequency domain in CUDA are studied in this paper. Cloud pollution remediation is achieved by comparing it with a standard full-day image template and filtering to obtain cloud transmittance. Finally the cloud contamination image is divided by this cloud transmittance. The SSIM algorithm is used to evaluate the quality of the restoration results of the three filtering methods. It is found that all of them can filter the moving details of the sun effectively. In view of the outstanding advantage of second-order Butterworth low-pass filter in processing speed, the system regards it as the best choice. (4) under the framework of Microsoft MFC, by calling the CUDA function executed in parallel in GPU, a complete real-time detection and repair system for cloud contamination of Ha sun-plane images with visual interface has been developed. After testing, the system can effectively distinguish the severe cloud pollution, can repair the cloud pollution and no cloud pollution images, and can quantify the degree of cloud pollution. For repairable cloud pollution, it can be effectively repaired in real time. In our experimental environment, one minute observation image can complete a cloud pollution image processing in 0.7 seconds, which fully meets the real-time requirements.
【学位授予单位】：昆明理工大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：P182.2;TP391.41

【相似文献】