Performance Evaluation of YOLOv8, YOLOv9 and YOLOv10 Models in Detecting Fusarium Wilt Disease in Banana Leaf Plants

Agriculture is a critical component of the socio-economic landscape in the Philippines, with banana plantations playing a significant role. However, diseases such as Fusarium Wilt pose a major threat, causing substantial economic losses. Computer vision, particularly deep learning models like the You Only Look Once (YOLO) series, offers promising solutions for early disease detection. In this study, the researchers evaluate the performance of the YOLOv8, YOLOv9, and YOLOv10 models in detecting Fusarium Wilt disease in banana leaf plants. The models were trained and tested using a dataset of 450 images, including 331 infected and 152 healthy leaves. The YOLO models were then evaluated on metrics including precision, recall, and mean average precision (mAP0.5-0.95). Hyperparameters were fine-tuned with batch sizes of 16, 32, and 64, and learning rates of 0.01, 0.001, and 0.0001. YOLOv8 and YOLOv9 showed inconsistent performance with varying batch sizes and learning rates, while YOLOv10 exhibited more predictable improvements, favoring lower batch sizes. Lower learning rates generally enhanced mAP50-95, though precision and recall varied. YOLOv8 and YOLOv9 achieved peak mAP50-95 with a batch size of 32 and a learning rate of 0.001, while YOLOv10 performed best with a batch size of 16 and a learning rate of 0.01. Among the models, YOLOv9 with a batch size of 32 and a learning rate of 0.001 achieved the highest mAP50-95, indicating high precision and recall in detecting Fusarium Wilt. These findings emphasize the need for systematic hyperparameter tuning to optimize YOLO models for specific tasks and resources.