Detection For Coronal Mass Ejection Based On Gaussian Mixture Models

With the construction of the coronagraph and the maturity of the observation technology, to detect and extract the coronal mass ejection accurating has a very important significance. The outbreak of CME can be seen as a kind of moving object detection in dynamic background, and background modeling is the key. The traditional methods such as median and mean can only extract stable background, and the accuracy rate of the frame difference that extracted background is relatively low.The adaptive Mixed Gaussian background subtraction extracted solar moving target coronal mass ejection (CME) is proposed in this paper. Therefore, we developed a dynamic new background estimation algorithm under the polar coordinate of heliocentric, The CME can be detected as a foreground object after subtracting the background that is estimated by the adaptive Mixture Gaussian Models from the original image extraction. Two image sequences of coronal observed by the Large Angle Spectroscopic Coronagraph (LASCO) in the SOHO satellite were used, We have done the following work:(1)Study of pretreatment methods:it includes noise reduction, the standardization of the sequence of the coronal image, transform the descartes coordinates to polar coordinates.(2)Improve the traditional Mixed Gaussian background Model:Compared with the traditional Mixed Gaussian model, An expectation-maximization algorithm is applied to improve the initialization of the model, and assigned a learning rate for each pixel in the sequence of image is adaptively for updating the background of coronal sequential images. The method that improved is accurate for detecting the CME. We can extract the CME through analyzing Unicom、hole filling and Otsu threshold value of segmentation.(3)Comparison and discussion of experiments:In order to verify the accuracy and effectiveness of the algorithm, based on the manual detection of listed on the CDAW CME catalog, comparison between the proposed and other CME automatic detection methods of the CACTus and the SEEDS.The method that the adaptive Mixed Gaussian background subtraction is practicable and effective for detecting the CME. It not only can detect all of the CMEs listed on the CDAW CME catalog, but also the CME with weaker intensity and smaller angle, (such as streamers etc.), and has a better performance than the CACTus and SEEDS. Compared with the manual detection of moving target detection, such as CDAW, the automatic CME detection method is more rapid and powerful, and the adaptive Mixed Gaussian background subtraction can realize the detection of moving target effectively. But the method also has a certain error, will see the crown flow as the CME detected.