基于混合高斯模型的日冕物质抛射探测方法研究

发布时间：2019-04-02 04:41

【摘要】：随着日冕观测技术的不断成熟,观测图像中日冕物质抛射(Coronal Mass Ejection, CME)的准确探测与提取具有非常重要的意义。CME爆发可以看为一种动态背景下的运动目标检测,而背景建模是其关键。传统的中值与均值背景法只能提取稳定的背景,而帧间差分法提取的前景准确率比较低。本文采用自适应混合高斯背景差分法对日冕物质抛射进行提取。根据改进的自适应混合高斯算法,在较为稳定的日心极坐标下,建立日冕序列图像的动态背景,并用原始图像减去动态背景的结果做为前景,从而探测CME。本工作采用欧洲航天局的SOHO卫星搭载的“大视角分光日冕仪”观测的2组日冕序列图像做为实验对象,具体工作内容包括如下几个方面：(1)预处理方法的研究：包括对日冕序列图像的降噪、图像的标准化、将图像从笛卡尔坐标转换到极坐标等。(2)对传统混合高斯背景模型的改进：与传统的混合高斯模型相比,我们对其初始化方法进行改进,用最大期望算法对权值等参数进行优化,并且通过对方差和均值的分析,用自适应学习率对序列图像中的每一个像素动态更新。经过改进的自适应混合高斯背景模型能够准确的探测运动目标CME。通过对CME的联通性分析、孔洞填充以及用大津法阈值对二值图像进行分割,最后提取CME。(3)实验的对比和讨论：为了验证算法的准确性与有效性,以标准的人工探测目录列表为基准,用基于帧间差分的两种经典CME探测算法(CACTus和SEEDS)和我们的自适应混合高斯背景模型差分法对比。我们认为采用自适应混合高斯背景差分法探测CME是有一定优势的。它不仅能探测到标准目录列出的全部CME,还能识别CDAW探测不到的强度弱和张角小的运动物质(如冕流等),且探测数量多于CACTus和SEEDS探测算法。与对运动目标的手动探测方式相比,自动探测方法对CME的探测更为快速、准确,而自适应混合高斯背景差分法与其他的自动探测算法相比,其探测率较高、能有效的实现运动目标的检测。但这种方法也有一定的误检,会将一些冕流作为CME检出。
[Abstract]:With the development of coronal observation technology, it is very important to detect and extract coronal mass ejection (Coronal Mass Ejection, CME) accurately from observed images. CME burst can be regarded as a kind of moving target detection under dynamic background. Background modeling is the key. The traditional median and mean background method can only extract stable background, but the foreground accuracy of inter-frame difference method is low. In this paper, the adaptive mixed Gao Si background difference method is used to extract coronal mass ejections. According to the improved adaptive hybrid Gao Si algorithm, under the stable heliocentric polar coordinates, the dynamic background of the coronal sequence image is established, and the result of subtracting the dynamic background from the original image is used as the foreground to detect the CME.. In this work, two sets of coronal sequence images from the SOHO satellite of the European Space Agency (ESA) were used as experimental objects. The specific work includes the following aspects: (1) the research of preprocessing methods: including the de-noising of coronal sequence images and the standardization of images. The image is transformed from Cartesian coordinates to polar coordinates, etc. (2) the improvement of the traditional mixed Gao Si background model: compared with the traditional mixed Gao Si model, we improve its initialization method. The maximum expectation algorithm is used to optimize the weights and other parameters, and the adaptive learning rate is used to dynamically update every pixel in the sequence image through the analysis of the difference and the mean value. The improved adaptive hybrid Gao Si background model can accurately detect the moving target CME.. By analyzing the connectivity of CME, filling holes and segmenting binary images with the threshold of large-scale method, finally, comparing and discussing the experiment of extracting CME. (3): in order to verify the accuracy and validity of the algorithm, Two classical CME detection algorithms (CACTus and SEEDS) based on inter-frame difference are compared with our adaptive hybrid Gao Si background model difference method based on the standard manual probe catalog list. We think that the adaptive hybrid Gao Si background difference method has some advantages in detecting CME. It can not only detect all the CME, listed in the standard catalog, but also recognize the weak intensity and small angle of motion (such as coronal stream) that CDAW cannot detect, and the number of detection is more than that of CACTus and SEEDS detection algorithms. Compared with the manual detection of moving targets, the automatic detection method for CME detection is faster and more accurate, while the adaptive mixed Gao Si background difference method has a higher detection rate than other automatic detection algorithms. Can effectively achieve the detection of moving objects. But this method also has a certain error detection, some coronal stream will be detected as CME.
【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】：P182.62;TP391.41

【相似文献】