关键词: 乌梅, 成熟度检测, 深度学习, 自然环境

Abstract: Black plum with different maturity has different pharmacological effects. To judge the maturity of most fruits in the orchard, a series of improvements are carried out based on the YOLOv7 target detection algorithm. The Vision Transformer with Bi-Level Routing Attention (BiFormer) module is added to the Backbone of YOLOv7 model to improve the feature expression ability of the network. The fruit maturity refinement module is designed to improve the correct rate of fruit maturity detection. The study shows that the improved YOLOv7—1 model has a Mean Average Precision (mAP) of 0.805, and is higher by 4.8, 12.4, 0.9, 0.7, 12.6, 1.7, 5.8 and 12.3 percentage points, respectively, compared with the improved YOLOv7—2 model, Faster R—CNN model, YOLOv3 model, Mask R—CNN model, YOLOv5s model, YOLOv5l model, YOLOv7 model, and YOLOv8 model.  The improved YOLOv7—1 model can improve the accuracy of identifying the maturity of black plum.

Key words: black plum, maturity detection, deep learning, natural environment