To effectively prevent corn diseases and monitor corn leaf growth, the DBG-YOLO detection algorithm was proposed. Based on YOLOv8, the C2f module in the backbone was replaced with DynamicConv, enhancing the model’s capability without increasing its depth or width. In the neck network, the global-local self-attention(GLSA) mechanism was used to capture global context while retaining local details. The bidirectional feature pyramid network(BiFPN) module was introduced to reduce parameters and improve multi-scale feature fusion, thus enhancing the detection of corn leaf diseases. To accelerate convergence, the exponential moving average(EMA) was introduced to the IoU threshold in SlideLoss, improving robustness while reducing false detections and missed detectinos. The results showed that compared with YOLOv8n, the precision and mAP@50 of the DBG-YOLO algorithm were increased by 5.8 percentage points and 6.6 percentage points, respectively. Meanwhile, the floating-point operations(FLOPs) and frame rate of the algorithm were reduced by 11.5% and 43.5%, respectively. In summary, the model proposed in this study comprehensively improves the accuracy of corn leaf disease detection and has good robustness, providing a reference for the deployment and application of corn leaf disease detection model in mobile detection devices.