To address the challenges of diverse steel surface defects, large scale variations, and low detection accuracy, an efficient steel surface defect detection algorithm, termed EMC-YOLO11n, is proposed. The proposed method incorporates the EIEStem module into the backbone network to effectively extract edge and spatial information, thereby enhancing the model’s local and global feature representation. The feature interaction module MBE-C3k2 is reconstructed to efficiently handle complex background features, which reduces the overall computational cost of the model. In addition, the C2DyT module is redesigned to improve feature extraction capability and feature transmission efficiency. Experimental validation on the NEU-DET steel surface defect dataset shows that, compared with the YOLO11n algorithm, EMC-YOLO11n improves mAP@0.5 by 4 percentage points. In addition, precision, recall, and F1-score are increased by 4.6, 1.6, and 1.7 percentage points, respectively. These results indicate that the proposed method further enhances detection accuracy, achieves a better balance between precision and recall, and demonstrates strong robustness.