Abstract: To address the problems of complex architecture and poor cross-scale target detection performance in existing deep learning surface defect detection models, a defect detection model based on the YOLOv8-Seg improvement is proposed, with a focus on detecting surface defects in cathode copper plates. Firstly, the CBAM attention mechanism is introduced to reduce the interference of channel and spatial attention on irrelevant background information. Next, the GELAN module is used to replace the C2f module in the backbone structure of the model, enhancing the model’s ability to detect small targets. Then, to solve the gradient vanishing problem, the activation function is modified to GELU, improving the model’s detection accuracy and recall rate, while accelerating convergence. Finally, to meet the visualization needs of industrial production and assess scattered defect information, local images of defects are sequentially processed using grayscale transformation, Gaussian filtering for denoising, inverse binarization, and defect contour labeling, followed by the calculation of the defect area. Experimental results show that the improved model achieved a 6.5% and 11.5% increase in detection accuracy and recall rate, respectively, compared to the original model on the self-built cathode copper plate surface defect dataset. The detection accuracy reached 92.8% and 74.3% on the Northeastern University steel plate surface defect dataset and the Tianchi aluminum profile surface defect dataset, respectively, validating the effectiveness of the proposed improvement.