This study proposed a novel lightweight crop leaf disease detection algorithm, EMD-YOLO, based on YOLOv8n, aimed at accurately and rapidly identifying diseases, reducing manual diagnostic costs, and minimizing the impact of leaf diseases on crop production and quality. The algorithm integrates the multi-scale dilated attention (MSDA), EffectiveSE attention mechanism, DySample upsampling, and Wise-IoU loss function. MSDA combines multi-scale spatial convolution with attention mechanisms to improve multi-scale feature efficiency; EffectiveSE strengthens feature selection, enhancing the model’s representational performance; DySample upsampling preserves important features, enhancing feature map resolution and detection performance; Wise-IoU loss function optimizes Intersection over Union (IoU) computation, improving model localization accuracy. Experiment results showed that EMD-YOLO outperformed the baseline model YOLOv8n, achieving improvements of 3% in precision and 3.6% in mean Average Precision (mAP), with reductions of 0.8M in parameters and 1.4MB in weights, while demonstrating good generalization and suitability for mobile crop leaf disease detection devices.