Abstract
The prevalence of diseases in wheat crops poses a significant threat to global food security, as it reduces yield and quality. Addressing these challenges is critical for sustainable agriculture. This study proposes and evaluates a hybrid deep learning (DL) model, EffiXB3, which combines Xception and EfficientNetB3 architectures, enhanced with edge-aware features, to improve disease classification in wheat crops. EffiXB3 employs a dual-input stream architecture, where one stream processes structural features, while the other incorporates textural features through Canny edge detection. The performance of individual models, Xception and EfficientNetB3, was assessed alongside the hybrid EffiXB3 model in a multi-class classification task involving five wheat leaf categories: Blast, Brown Rust, Healthy, Leaf Blight, and Septoria. Xception and EfficientNetB3 achieved classification accuracies of 95% and 93%, respectively. The proposed EffiXB3 model outperformed both, achieving an accuracy of 98.5%. The integration of edge-aware features substantially improved robustness and classification performance, particularly in differentiating visually similar disease patterns. The findings demonstrate the effectiveness of hybrid DL models with edge feature integration in diagnosing agricultural diseases. EffiXB3 offers a promising approach for enhancing disease detection in wheat, contributing to improved crop management and food security.