Deep Learning Approaches for Crop Disease Detection Using Yolov11 and Yolov10

Good agricultural practices require intensive monitoring of crops from seedling to maturity. The process involves correcting nutrient imbalances, regularly monitoring for plant disease, water regulation through irrigation processes, and controlling the application of fertilizers and pesticides. Crop health is also hampered by poor natural defense mechanisms and climate change impacts, leading to disease outbreaks and slowing agricultural development. A major obstacle in crop management is the correct identification and disease diagnosis, an activity that is commonly impaired by low diagnostic efficiency, overfitting in analysis models, and high error rates. This study investigates the ability of You Only Look Once (YOLO) models, YOLOv11 and YOLOv10, in plant disease detection. A custom dataset was created, including images obtained from the dataset (Plant Village) on Kaggle, Roboflow, and images manually annotated for training, testing, and validation. The models were tested and evaluated against major performance metrics, including mAP, precision, F1 score, and recall. The findings showed that YOLOv11 detected diseases slightly better than YOLOv10. Yolo was used as the reference, with significant performance metrics for YOLOv11 being different from YOLOv10 being mAP@ 0.5 (0.986 vs. 0.985), precision (0.970 vs. 0.967), and recall (0.982 vs. 0.957). In addition, YOLOv11 was more efficient, consuming less computational resources and processing time, making it a more convenient option to detect diseases in the crops.