Improved Tomato Classification in Industrial Environment Using YOLOv8 and Data Augmentation Techniques

This research investigates the utilization of the YOLOv8 model for the detection and classification of tomato ripeness levels in an industrial setting. The study assesses the influence of data augmentation methods on model efficacy, specifically in distinguishing four primary ripeness classifications: Unripe, Ripe, Half-Ripe, and Rotten. Data augmentation techniques, such as rotation, flipping, cropping, and illumination modifications, were employed to enhance the model's capacity to manage real-world differences in texture and color. The findings indicate that data augmentation substantially improves detection accuracy in all categories. The model's accuracy improved by 8% in the Half-Ripe and Rotten categories, which are often more difficult due to their intricate visual attributes. Moreover, the Unripe and Ripe categories experienced a 7% enhancement in accuracy. The enhancements render the YOLOv8 model exceptionally efficient for real-time applications in industrial settings. This study demonstrates that employing data augmentation with the YOLOv8 model markedly enhances the efficiency and accuracy of automated tomato sorting systems. The results indicate that this method is appropriate for application in the tomato processing sector, with the capacity to enhance post-harvest management according to tomato ripeness. Subsequent study may investigate supplementary methodologies, such as transfer learning, to augment model efficacy.