An efficient YOLO-based framework for multi-class plant disease detection

Abstract

Plant health plays a critical role in agriculture, climate balance, and economic stability. However, plant diseases caused by bacteria, fungi, and viruses can significantly reduce crop productivity if not detected early. Traditional manual inspection methods are time-consuming, labor-intensive, and prone to human error, especially in large-scale farming. To address these challenges, this study proposes an automated and accurate plant disease detection system using deep learning-based object detection models for early disease diagnosis in agriculture. A publicly available dataset containing 38 different plant leaf diseases annotated in You Only Look Once (YOLO) format is used, along with a standardized preprocessing pipeline to ensure data quality and consistency. Three modern architectures: YOLOv8, YOLOv11, and YOLOv26 were trained and evaluated under identical conditions using the Ultralytics framework on Google Colab. Experimental results show that YOLOv11 achieves the highest accuracy in terms of precision, recall, and mean Average Precision (mAP), while YOLOv8 provides the fastest inference speed with lower computational complexity. Based on the results, the study concludes that YOLO-based models show great potential for plant disease detection, with YOLOv11 offering superior detection accuracy among the evaluated models. The practical implications of these findings lie in the potential for precision agriculture to monitor diseases in real-time, minimize crop losses, and aid in timely decision-making for farmers and agricultural stakeholders.