基于改进YOLOv7-Tiny的树叶遮挡环境下红心李识别OA

Identifying accurately red-hearted plums was the priority task in red-hearted plum detection.However,the luxuriant branches and overlapping fruits increased the difficulty of recognizing red-hearted plums.Based on this,the backbone of the YOLOv7-Tiny model was modified to improve the accuracy of fruit detection in occluded environments.First,the SE attention mechanism was modified to the NAM attention mechanism in the MobileBottleneck(Bneck)mo dule of the MobileNetV3 backbone,focusing on the information of adjusting the weights during the training process to im-prove the detection of critical features of the fruits and the new NBneck module was constructed.In addition,the Diverse Branch Block(DBB)module was added to the MobileNetV3 backbone to enhance the ability to extract features that were not obvious to the occluded fruits,and ultimately,a new backbone DN-MBV3 was constructed to replace the original backbone network of the YOLOv7-Tiny and to reduce the number of parameter in the network.Under the same experi-mental conditions,the improved YOLOv7-Tiny accuracy was better than that of other networks,including SSD,YOLOv4-Tiny,EfficientDet,Faster-RCNN,and other models.Compared with YOLOv7-Tiny,the mean average accura-cy(mAP)of the red-hearted plum was improved by 4.89 percentage points,the recall rate was increased by 11.59 percentage points,and the model volume was reduced by 4.3 MB,which had the advantages of high detection accuracy and small size.