A YOLO-Assisted Framework for Potato Leaf Disease Detection and Classification using CNN and Multi-Head Attention Mechanism

Potato leaf diseases create a serious threat to the production of agriculture, leading to large yield losses and economic challenges. Accurate and timely detection has become crucial for sustainable farming and effective management of diseases. Our study proposes a hybrid deep neural network model that combines EfficientNet variants (B0, B3, B5), an attention mechanism, and YOLO-based Region Proposal Networks (RPN) for both classification and localization of potato leaf diseases. EfficientNet serves as the backbone for multiscale feature extraction, while the attention mechanism selectively focuses on critical regions to improve classification accuracy. YOLO-based RPN refines the localization of disease-affected areas, enabling precise identification of specific diseases. We evaluate the model on a balanced dataset of 15,384 images across seven disease categories, including bacterial, fungal, and viral infections, as well as healthy leaves. Using stratified k-fold cross-validation, our model achieves an overall accuracy of 98%, with individual precision and recall metrics demonstrating strong performance across disease classes. Specifically, bacterial and fungal diseases reach F1-scores of 0.99 and 0.98, respectively. The experimental evidence proves the effectiveness of using the proposed method performs well for automated identification of diseases in potato leafs. By providing a scalable and efficient solution, this study contributes to enhancement of agricultural automation and precision farming, helping farmers reduce crop losses and improve yield quality.