Industrial Surface Defect Detection Method Based on Improved YOLOv11

Surface imperfection detection in industrial settings is vital for maintaining product quality and ensuring manufacturing safety. However, detecting tiny defects under complex background interference remains a significant challenge to detection accuracy. In order to greatly improve the model’s localization and classification capabilities for small defect targets, this paper suggests an enhanced approach based on YOLOv11 that incorporates a newly designed Detect+ module and a C2BI module. Specifically, the Detect+ module integrates an Inner-FocalLoss function to optimize the decoupled detection head by reducing the weight of easily classified samples, thereby lowering missed detection rates. Additionally, it helps to solve the issue of class imbalance and enhances bounding box regression’s accuracy. To improve the model’s perception of important areas, reduce background noise, and facilitate the extraction of minute defect features in intricate scenes, the C2BI module adds a BiFormer dual-path routing attention technique. Experiments conducted on the NEU-DET, GC10-DET, and Severstal Steel datasets demonstrate that the proposed method achieves mAP scores of 80.3, 68.3, and 43.9%, respectively, outperforming Faster R-CNN (75.73%) and the original YOLOv11 (75.4%), confirming that the suggested approach is effective.