Abstract Aiming at the low efficiency of manual interpretation of gravity change anomaly feature region, we propose an automatic identification method using YOLOv5s algorithm to identify the four-quadrant feature area of gravity change, and conducts tests and verification based on the measured gravity change data of the north-south seismic belt. The results show that: 1) our image recognition model can effectively identify the relatively standard four-quadrant distribution feature region, and the accuracy, recall rate and average accuracy of the model prediction results are all in a reasonable range; 2) The model can accurately identify the four-quadrant feature area in the gravity change image of the north-south seismic belt from September 2021 to May 2022, and the Lushan M6.1 earthquake in June 2022 and Luding M6.8 earthquake in September 2022 in the feature area occurred successively. The results show that the proposed method has a good application potential for the screening of gravity anomaly regions and the study of potential seismic risks. 3) The recognition ability of the model for the non-standard four-quadrant feature area with obvious distortion is still insufficient, and it is still necessary to build more data training sets with more reasonable labeling to further improve the universality of the model for the recognition of the four-quadrant feature area.