Objective: To improve the efficiency of cherry classification and sorting in industrial environment. Methods: An improved cherry defect recognition and sorting model based on Faster R-CNN framework was proposed. Results: By comparing VGG16, MobileNet-V2 and ResNet50 network, the effect of Resnet50 network was the best, the improved Faster R-CNN model had 97.75%, 99.77%, 98.90%, 97.56%, 96.67%, 98.80% of detection precision for cherry fissure, twinning, growth stimulation, mildew, Browning rotten and intact fruit, respectively. The average detection accuracy of the improved Faster R-CNN model was 98.24%, which was higher than other models, and the detection speed was 31.16 frames/s. Conclusion: The test method had a high identification accuracy for cherry defects.