Abstract: Coronal Mass Ejection (CME) is a strong solar eruption, which has a great impact on space weather and human life. Therefore, CME detection is of great significance in predicting CME and ensuring the safety of human production and life. The existing detection methods mostly use artificial defined features and artificial defined threshold to detect CME. Because artificially defined features cannot precisely represent weak CMEs and it is difficult to select a universal threshold, the detection effect of existing CME detection methods for weak CMEs needs to be improved. A CME detection algorithm based on Faster R-CNN is proposed in this paper. In this method, a dataset containing 9 113 coronal images was manually annotated by combining the log information of three famous CME catalog, CDAW, SEEDS and CACTus. Then, according to the characteristics of CME images with fewer features than natural images and the difference in target size from natural images, Faster R-CNN was improved in feature extraction and anchor point selection. Using the CME data in June 2007 as the test set, the algorithm detected all 22 strong CME events and 138 out of 151 weak CME events, and the detection errors of characteristic parameters such as center Angle and Angle width of CME events were within 5 degrees and 10 degrees, respectively.