CTC-YOLO: an improved YOLOv11 algorithm for steel surface defect detection

Abstract During the detection of steel surface defects, issues such as background noise interference and the varying sizes of defect features often lead to challenges in improving detection efficiency. To address this issue, we propose a steel surface defect detection model based on YOLOv11, named CTC-YOLO. Firstly, in the backbone network, we designed a Cross Stage Partial-Partially Transformer Block (CSP-PTB) module, which combines the advantages of both Transformer and Convolutional Neural Networks (CNN), effectively enhancing the extraction of local and global features. Secondly, the Triplet-Branch Attention Mechanism (TBAM) is incorporated to enhance the model’s ability to capture and integrate features across multiple dimensions, thereby extracting more effective semantic information. Additionally, we introduce the CNN-based Cross-scale Feature Fusion (CCFF) module into the neck network to enhance the model’s ability to fuse multi-scale features. Finally, a series of experimental results on the Northeast University - Defect Detection (NEU-DET) dataset indicate that CTC-YOLO achieves an accuracy of 79.7% mAP, representing a 1.8% improvement over YOLOv11n while also reducing the model’s computational complexity. To further evaluate the generalization ability and robustness of the proposed model across different datasets, testing experiments were conducted on the General Class 10 - Defect Detection (GC10-DET) dataset. The results demonstrate that the proposed model achieves an mAP of 67.7%, surpassing the baseline model by 1.6%. The CTC-YOLO model proposed in this paper not only improves the accuracy of defect detection but also reduces computational costs.