An Improved YOLOv8 Four-Season Honey Longan Recognition and Location Method

Aiming at the problems of high localization difficulty and low recall rate of existing longan identification methods, In this paper, a Longan recognition and location method based on the improved single-stage target detection algorithm (you only look once v8,YOLOv8) and MPDIoU (Minimum Point Distance based IoU) loss function is proposed. Through the optimization of YOLOv8, the center point and aspect ratio information of the boundary box are introduced, and the coincidence degree of the boundary box is more accurately measured, and the prediction boundary box under the same aspect ratio is constrained, so as to provide better model recognition effect. The test results show that compared with the original model, the improved YOLO v8-MPDIoU model can identify recall (R), mean average precision (R), mAP and mean average precision (mean average precision -50% -95%, mAP50-95(B)) improved by 1.34 percentage points, 0.57 percentage points and 0.45 percentage points, respectively, and had better accuracy in 0.2-0.8 confidence intervals.