Abstract: In order to solve the problem that it is so difficult to detect mechanized harvesting beets accurately and comprehensively that the calculation of breakage rate of mechanized harvesting beets becomes inaccurate. A breakage detection method of mechanized harvesting beets based on YOLOv4 was proposed in this paper. Beet photos from different distances, different angles and different degrees of occlusion were used to make data sets, and the breakage detection model was trained and tested. The test results showed that the accuracy and recall rates which the breakage detection model based on YOLOv4 detects whole beets were 94. 02% and 91. 13%,the accuracy and recall rates which the breakage detection model based on YOLOv4 detects damaged beets were 96. 68% and 95. 21%,and the mAP of the breakage detection model based on YOLOv4 was 96. 44%. The mAP of the breakage detection model based on YOLOv4 were higher than mAP of the breakage detection model based on Faster R-CNN or SSD. It can be concluded that the breakage detection model based on YOLOv4 proposed in this paper can more accurately detect the whole and damaged beets and meet the requirement of calculating the damage rate of beets.