Abstract: Aiming at the problems of strip steel surface defect detection for defects with variable scales and complex backgrounds that are prone to leakage and false detection, as well as poor detection accuracy, a strip steel surface defect detection algorithm based on YOLOv8n is proposed. Firstly, to improve the feature learning ability of the backbone network, deformable convolution DCNv2 is introduced into the backbone network, and at the same time, a dynamic convolution module is incorporated, which effectively improves the feature extraction ability of the network by enlarging the sensory field. Secondly, in the feature fusion, a more efficient Dysample upsampling method is used. The channel attention mechanism module SE (squeeze and excitation) is introduced to improve the network’s ability to extract deep feature information. Finally, DIoU combined with inner-IoU is used as a loss function to further improve the detection accuracy of the algorithm. A large number of experiments are carried out on NEU-DET data, and the experimental results show that the improved algorithm improves the average detection accuracy by 2.2%, reaching 79.5%. In addition, the experimental results on the GC-10 dataset show that the method in this paper has good robustness.