An Algorithm For Determination of CME Kinematic Parameters Based On Machine Learning

Coronal Mass Ejections (CMEs) are the major source of space weather events, causing severe disturbance to Sun-Earth space environment. Since there are more and more space activities and facilities, it’s becoming increasingly significant to detect and track CMEs. We develop a new algorithm to automatically detect CMEs and derive CME’s kinematic parameters based on machine learning. Our method consists of three steps: Recognition, tracking and determination of parameters. First, we train a convolutional neural network (CNN) to classify images from SOHO LASCO coronagraph observation into two categories that contains CME(s) or not. Next, we apply Principal Component Analysis (PCA) algorithm and Otsu’s method to acquire binary-labelled CME regions. Then, we employ the track-match algorithm to track CME motion in time-series image sequence and finally determine CME kinematic parameters e.g., velocity, angular width (AW), and central position angle (CPA). The algorithm is validated on several CME events with different morphological characteristics. We compare the results with a manual CME catalog and automatic CME catalogs (including Computer Aided CME Tracking (CACTus), Solar Eruptive Event Detection System (SEEDS), CORonal Image Process method (CORIMP)). Our algorithm shows some advantages in the recognition of CME structure and the accuracy of the kinematic parameters. In the future, the algorithm is capable of being applied to initialize magnetohydrodynamic simulations to study the propagation characteristics of real CME events in the interplanetary space, and provide a more efficient prediction of CMEs' geo-effectiveness.