Smart Crop Health Monitoring using YOLO-based Leaf Disease Classification and Pesticide Advisory System using Deep Learning

Abstract

Plant diseases are among the most significant challenges in agriculture because they reduce crop productivity and cause economic losses to farmers. Early monitoring of plant diseases is important for controlling infection spread and improving crop management. Traditional plant disease detection techniques depend heavily on manual inspection by agronomists; therefore, they are time-consuming, labor-intensive, subjective, and often inaccessible to farmers in remote locations. In recent years, plant disease classification has been performed using deep learning architectures such as CNN, ResNet, MobileNet, and EfficientNet. Although these models can be accurate, many of them require higher computational resources and usually focus only on disease detection without providing a practical treatment recommendation. This paper proposes a smart crop health monitoring system for plant leaf disease classification and pesticide advisory using the Ultralytics YOLO26s-cls model. The proposed system is trained on the New Plant Diseases Dataset containing 87,867 plant leaf images from 14 crop species and 38 disease/healthy classes. The pipeline includes image preprocessing, augmentation, YOLO26s-cls-based feature extraction, disease classification, and a validated rule-based pesticide recommendation module. The model architecture, model selection rationale, overfitting-control strategy, dataset validation procedure, and pesticide-rule validation process are explicitly described to support reproducibility. The proposed classifier achieved 99.02% validation accuracy, 99.01% macro precision, 99.00% macro recall, and 99.00% macro F1-score. The system can therefore serve as a decision-support tool for early disease identification and pesticide selection in precision agriculture.