A lightweight and efficient yolo-based framework for uav-assisted pistachio disease detection

Pistachio diseases, despite their invaluable significance to global cuisine and economy, have received limited attention in the development of automated detection solutions. To address this gap, we introduce YOLO-PGWD, a lightweight and high-accuracy detection model based on YOLO11s, trained on a newly constructed Pistachio Leaf Disease Dataset. The proposed architecture incorporates four key enhancements: replacing conventional bottlenecks in the neck with GhostBottleneckv3 for efficient feature extraction, employing the WIoUv3 loss function to improve bounding box localization, integrating the DiNAT attention mechanism to enhance the focus level on subtle patterns, and substituting selected convolutional layers in the backbone with PConv to reduce computational complexity. To assess the effectiveness of YOLO-PGWD model, comprehensive experimental studies are conducted. The proposed model performs 72.86%, 63.01%, 67.58%, and 70.75% in terms of precision, recall, F1-score, and mAP50. Conducted generalization experiments prove the reliability of the proposed model across diverse agricultural conditions. When the results obtained are evaluated, it is confirmed that the YOLO-PGWD model provides a robust technical foundation for advancing plant disease detection systems and the development of smart agricultural scenarios.