<![CDATA[Manual detection of grape ripeness is inefficient and prone to subjective errors. This study developed a web-based automatic classification system using the YOLOv11 algorithm with the YOLO11s.pt model and the Roboflow platform. A deep learning approach was applied to automate the classification of grapes into four ripeness categories: unripe, semi-ripe, ripe, and rotten. The dataset used consisted of 897 images obtained directly from the vineyard, then expanded to 6,135 images through preprocessing and augmentation. The labeling process was carried out using Roboflow, and model training was carried out on Google Colab for 200 epochs. The training results showed high performance, with a recall value of 0.95, a precision of 0.98, and a mean Average Precision (mAP) of 0.84. The system was able to distinguish multi-class objects with an average detection time of 1,02 seconds per image, thus supporting semi real-time operations. However, the accuracy of the semi-ripe class classification is still a challenge due to visual similarities with other classes. This system has been integrated into a web application that displays classification results in semi real-time, and has the potential to be applied in a digital agricultural system. For further research, it is recommended to optimize the dataset, especially by adding the amount of training data on the rotten and half-ripe grape classes. In addition, the development of the application into a mobile application is recommended to increase accessibility and flexibility of use.]]>
Copyrights © 2025
