Recognition Method of Tomato Ripeness Based on YOLOv7 and YOLOv8

The increasing demand for efficient and accurate agricultural practices necessitates advanced technological solutions in crop management. One critical aspect is the accurate classification of tomato ripeness, essential for optimizing harvesting processes and ensuring product quality. Traditional methods are labor-intensive and prone to errors, emphasizing the need for automated systems. This research addresses this issue by exploring the application of advanced computer vision models, specifically YOLOv7 and YOLOv8, for tomato ripeness classification. The primary objectives were to develop and train YOLOv7 and YOLOv8 models, compare their performance, and evaluate YOLOv8's robustness under various simulated real-world conditions. Trained on a dataset of unripe, ripe, and rotten tomatoes, YOLOv8 demonstrated superior performance compared to YOLOv7, with higher precision, recall, and F1-scores across all categories of tomatoes. Further analysis included assessing YOLOv8's resilience to vary brightness, blurring, and noise addition. Results indicated that YOLOv8 maintained robust performance under these conditions, though slight declines in accuracy were observed. In conclusion, YOLOv8 proved to be a more effective and robust model for tomato ripeness classification, achieving higher true positive rates and overall better performance metrics than YOLOv7. These findings suggest that YOLOv8 is well-suited for practical deployment in agricultural settings, offering significant improvements in the accuracy and reliability of tomato classification.