Real-Time Detection of Coffee Cherry Ripeness Using YOLOv11

This study aims to develop a real-time coffee fruit ripeness detection system using the YOLOv11 algorithm to assist farmers in determining the optimal harvest time. The dataset comprises 302 images categorized into three ripeness levels: ripe, semi-ripe, and unripe. Model training was conducted on Google Colab with data augmentation to enhance dataset variability and prevent overfitting. After 20 epochs, the model demonstrated strong performance in the ripe category (mAP50: 0.774, Precision: 0.645, Recall: 0.812) and satisfactory results for semi-ripe fruits (mAP50: 0.695, Precision: 0.624, Recall: 0.679). However, detection performance for unripe fruits was lower (mAP50: 0.4). The system achieved an inference time of 183.4 ms per image, with fast preprocessing and postprocessing (0.5 ms each), indicating its suitability for real-time applications. While the model performs well overall, further improvement is needed in detecting unripe coffee fruits for enhanced system effectiveness.