An improved YOLOv8n model applied to banana leaf disease detection

To effectively address the issue of low recognition accuracy for small-target and multi-scale diseases on banana leaves, this study presents an enhanced YOLOv8n model specifically designed for banana leaf disease detection. In the context of network architecture optimization, CA attention mechanism is introduced after the P5 layer of the backbone network, and an Efficient Multi-scale EMA Attention Mechanism is incorporated after each C2f module in the neck network. These architectural improvements enhance the interaction of information across spatial and channel dimensions within feature maps, thereby enabling more accurate detection of disease targets at various scales. With respect to loss function design, the conventional detection loss is substituted with the Wise-IoU loss function. Through the introduction of a dynamic weight adjustment mechanism, this approach significantly improves the feature extraction capability of the backbone network and expedites convergence during model training. Moreover, hyperparameters are systematically optimized using a combination of grid search and adaptive optimization strategies, thereby fully exploiting the synergistic potential between the attention mechanisms and the Wise-IoU loss function. Experimental results demonstrate that the proposed model achieves improvements of 4.24%, 3.86%, 3.74%, and 4.16% in accuracy, recall, mAP@50, and mAP@50–95, respectively, confirming the effectiveness and superiority of the proposed method for banana leaf disease detection.