Multiclass Detection of Citrus Leaf Diseases Using an Enhanced YOLOv11 Architecture

Citrus crops are highly susceptible to multiple leaf diseases, which can significantly reduce productivity and quality. Accurate identification and classification of these diseases are crucial for effective crop management. This study presents a multiclass detection framework for citrus leaf diseases using an enhanced YOLOv11 architecture integrated with attention mechanisms. Two advanced modules, Cross-Scale Attention Module (CSAM) and Attention-Guided Feature Fusion (AGFF), are incorporated to improve feature extraction and enhance detection accuracy for challenging disease patterns. The system is trained and evaluated on a dataset containing six categories: Canker, Greasy_spot, HLB, Healthy, Melanose, and Sooty_mold. Experimental results demonstrate that the proposed YOLOv11-CSAM and YOLOv11-AGFF models achieve high precision and recall, effectively distinguishing between subtle disease variations and healthy leaves. The framework is implemented using Python and Flask for back-end processing, with HTML, CSS, and JavaScript for the front-end interface, enabling a user-friendly platform for leaf disease prediction. This approach provides an efficient solution for automated disease classification, reducing the dependency on manual inspection and aiding in informed agricultural decisions. The proposed models offer robustness and adaptability, ensuring reliable performance across diverse leaf images and disease manifestations.

Keywords: Citrus leaf, disease detection, YOLOv11, CSAM, AGFF, deep learning, attention module, image-based prediction, Flask