Natural environments and complex conditions add to the difficulty of accurately detecting maize leaf diseases, with leaf cover and uneven light being key factors in this challenge. In order to achieve effective detection of maize leaf diseases, a modified YOLOv8n algorithm is examined in the paper. The improvements include: (1) incorporating a Multi-scale Efficient Local Attention (MELA) mechanism into the backbone network to construct C2f-MELA, enhancing small target detection through multi-scale feature extraction and local feature emphasis; (2) introducing RFCBAMConv receptive field attention module to strengthen spatial feature extraction capabilities; and (3) Implementing the LKSG (Large Selective Kernel Network and SpatialGroupEnhance) mechanism of attention in the cervical network to improve the fine-grained perception of disease features and enhance the spatial relationships of feature maps. Experiments on a maize leaf disease dataset show that our improved method outperforms the standard YOLOv8n model by 5.5% in accuracy, 3.3% in mAP@0.5, and 2.8% in mAP@0.5:0.95. The research provides a reference for detecting maize leaf disease in natural and complicated situations, as well as data support for agricultural robots that can autonomously identify and localize disease spots.