<p>The detection of minor defects on steel surfaces is an essential part of industrial production. It helps reduce production costs, improve safety and compliance, and maintain sustainability and competitiveness. However, current industrial defect detection methods in China suffer from low accuracy, high false-positive rates, and high missed detection rates. To address these issues, this study proposes a lightweight steel defect detection model based on YOLOv8n, named YOLO-DBL. The model introduces improvements in the backbone network, feature extraction methods, and attention modules. For the backbone network, this study proposes a novel DWR-C2f design, which incorporates the Dilation-wise Residual (DWR) module and the C2f module from YOLOv8 to enhance multi-scale feature extraction capabilities. In the feature fusion part, the model employs optimization and fusion with the Weighted Bi-directional Feature Pyramid Network (BiFPN), achieving efficient bidirectional cross-scale connections and weighted feature fusion. Additionally, the model integrates the Large Separable Kernel Attention (LSKA) module, which not only improves the extraction of minor features but also significantly reduces the model’s computational load. This study, based on the NUE-DET public dataset, involved extensive experimentation. The experimental results indicate that the new model achieves a 2.7% improvement in mean Average Precision (mAP) and a 15.3% reduction in the number of parameters compared to the original model. The complexity of the model is significantly reduced, achieving multi-faceted performance optimization, and meeting the real-time and accuracy requirements of industrial detection.</p>