Enhanced apple target detection and ripeness assessment utilizing an improved YOLOV8 algorithm

This paper addresses the inefficiencies and economic burdens inherent in traditional apple harvesting methodologies, which are predominantly labor-intensive. To mitigate these challenges, we propose an apple object detection and maturity assessment approach that leverages an enhanced YOLOv8 model. This methodology integrates deep learning, image processing, and heat calculation techniques, thereby enabling rapid and precise identification and localization of apples irrespective of image resolution. The model's accuracy, after 100 epochs of training, has stabilized at 96.6%, with a recall rate of 90%. The mean Average Precision (mAP) has been recorded at 53.9, and the inference time per sample on a GPU is an average of 0.3ms. In comparison to the conventional YOLOv5 object detection model, our approach has achieved a mAP improvement of 0.7, and has significantly enhanced inference speeds on both CPU (by 284.1ms/sample) and GPU (by 0.28ms/sample), while concurrently providing an effective assessment of apple maturity. Consequently, the apple object detection method, grounded in the refined YOLOv8 model, demonstrates the ability to accurately discern apple images across varying resolutions and to evaluate apple maturity through heat calculation. This innovation not only addresses the drawbacks of time-consuming manual measurements and suboptimal machine vision recognition accuracy but also lays a robust theoretical foundation for practical applications in the agricultural sector.