Abstract:
Crop diseases pose a significant threat to global food security, requiring efficient and
accurate detection methods to support sustainable agriculture. This study focuses on
evaluating YOLO-based deep learning models for detecting potato leaf diseases,
leveraging the Potato Leaves Disease Detection Dataset from publicly available
repository. The dataset contains 1,398 images categorized into Fungus, Pest, and
Healthy classes. Four YOLO model configurations YOLOv8 (large and medium) and
YOLOv9 (medium and small) were experimented onto the Potato leaf disease dataset.
The YOLOv8 (large) model outperformed all other configurations, achieving the
highest mAP50 (97.2%) and mAP50-95 (92.4%), along with the lowest training and
validation losses, demonstrating robust generalization and precise object detection
capabilities. YOLOv9 models, particularly YOLOv9 (small), exhibited lower and more
variable performance, highlighting challenges in distinguishing complex or
overlapping disease features despite their compact design. This study concludes that
YOLOv8 (large) is the most reliable model for applications requiring high precision
and robust performance in real-time agricultural disease detection. The findings
underscore the potential of YOLO-based models to transform crop disease management
by improving early detection and intervention, contributing to enhanced productivity
and sustainable farming practices.
