EM-YOLO: An Improved Steel Surface Defect Detection Model Based on YOLOv8n

The detection of surface defects in steel is vital for guaranteeing the material's performance and service life. However, the actual detection process often encounters challenges such as severe background interference from defective parts and significant variations in defect scales. To address these issues, this paper proposes an improved YOLOv8n model named EM-YOLO. Firstly, to tackle the issue of high background interference, the study designs the ECCA (Enhanced Channel Coordinate Attention) module based on Coordinate Attention, which enhances the capture of channel features and strengthens their interaction with spatial information, significantly improving the model's ability to focus on key features and suppress irrelevant information.Next, to address the multi-scale problem, the paper introduces the GMSConv (Group Multiscale Convolution) module and integrates it into the C2f part of the YOLOv8n network, forming the C2f_GMSConv module, to comprehensively capture and understand defects of varying sizes.Finally, during the model training process, the Inner EIoU is used as the loss function to optimize the precision of bounding box regression and to accelerate model convergence. Experimental results on the NEU-DET dataset show that the improved model achieves an [email protected] of 80.6%, which is a 4.1% increase over the original model. This model meets the requirements for accurate detection and identification of steel surface defects.