Utilizing YOLOv10 and YOLO11 for Tomato Ripeness Detection in Vertical Farming

Introduction: With the increase in urbanization, the amount of available agricultural land is constantly decreasing. Vertical farming presents one such solution to this problem, combined with new technologies, making it a more reliable and profitable approach. Objectives: This study is set to evaluate the performances of YOLOv10 and YOLOv11 models for the ripeness detection process in the harvesting of tomatoes with the aim of performing efficient and automatic harvesting. Methods: The research used an open dataset containing 667 images, divided into three classes: ripe, rotten, and unripe tomatoes. The YOLOv10 and YOLOv11 models were trained and tested, and their performances were evaluated with the metrics [email protected], precision, and classification accuracy. Results: The YOLOv10 model had an [email protected] of 0.706, and YOLOv11 had 0.695. The rotten class had the highest precision at 0.8, and the ripe class was the hardest to classify, being most frequently confused with background. YOLOv11n was faster but slightly less accurate compared to YOLOv10b. Conclusions: YOLOv11n is faster but less accurate compared to YOLOv10b, and is more suitable for applications requiring precise image analysis in vertical farming. The choice of the model depends on the needs of the application, and the predictions can be used in automating harvesting with a robotic system. Also, if a dataset of large size having variety of images is trained using high GPU, then high increase in accuracy and precision can be seen.