Addressing the issues of diverse defect types, significant size variations on steel surfaces, as well as the complexity and limited accuracy of existing models, an improved steel surface defect detection algorithm based on YOLOv8n was proposed. Initially, Omni-dimensional Dynamic Convolution (ODConv) was introduced in the feature extraction module to efficiently capture multi-dimensional features, thereby reducing information loss. Subsequently, an Asymptotic Feature Pyramid Network (AFPN) structure was embedded in the feature fusion module to facilitate direct interaction between non-adjacent level features, alleviating semantic gaps and enhancing the quality of feature integration. Lastly, the original CIoU loss was supplanted by Wise-IoUv3, which employs a dynamic non-monotonic focusing mechanism in the loss function, to optimize bounding box regression. This adjustment expedited network convergence while boosting detection precision. Extensive experiments conducted on the NEU-DET dataset demonstrated that the refined YOLOv8-ODAW network model achieved a 7.3% increase in mAP50%, a 21.95% decrease in GFLOPs compared to the original model, exhibiting superior localization capability and detection accuracy for steel surface defects. The detection speed met the requirements of industrial inspection standards.