Tomato Fruit Detection and Tracking in Real-Time Using YOLOv8

As automation and robotics are being integrated into all fields, the employment of AI has improved the precision and efficiency of robotic systems, there have been multiple recent papers showing improvement in navigation using computer vision and visual servoing. In this paper we studied the performance of the YOLOv8 algorithm on tomato fruits with different degrees of ripeness, performing a 2-class classification (red and green), and verifying the results as the algorithm recognizes the fruits, in different lighting conditions, and also in real-time. In our study, we also emphasized the importance of data quality and data preprocessing and enhancement, as we took two different datasets that performed poorly with the model and merged them into one that delivers great results when it comes to Precision and Accuracy, and that was achieved by applying image processing techniques to resize the images, which in turn gave us a total of more than 2400 images in our final dataset. In the end, our model reached a mean Average Precision of 86.7 % which is good and consistent for our intended applications and shows that the dataset we produced in our study is good for training, the model also shows great performance in real-time detection speeds, as it the average inference time was 5.02 ms meaning it generally takes the model as little as 5 ms to detect a tomato fruit. Finally, we are looking to improve upon this approach, as YOLOv8 is a state-of-the-art algorithm, and using AI in farming and agriculture has a very high potential, especially in Algeria, we believe that it can be used to improve the yield of our farms if used in the correct system.