<p>Accurate identification of maize leaf diseases is essential for ensuring food security, yet existing approaches struggle with complex field backgrounds, inter-class similarity, and intra-class variation. To address these challenges, we propose a collaborative maize leaf disease recognition framework that integrates lesion localization, semantic enhancement, and multi-center modeling. The proposed method effectively focuses on discriminative disease features, enhances semantic representation between categories, and models intra-class diversity more adaptively. Experimental results demonstrate that our approach achieves a high recognition accuracy of 96.90%, outperforming existing methods and significantly mitigating the effects of complex backgrounds and healthy leaf interference. This work provides a novel and effective strategy for intelligent and precise agricultural disease diagnosis.</p>