YOLO11-DCB: an improved YOLO11-based online detection model for surface defects of automotive oil seal rings
Abstract Online detection of surface defects on automotive oil seal rings has become a critical process for improving their production efficiency. To address the problems of inaccurate feature representation and low detection efficiency for small defects in oil seal ring surface inspection, this study proposes an enhanced defect detection method, YOLO11-DCB, based on the YOLO11n architecture. First, a dynamic upsampling module is embedded into the backbone network to enhance spatial feature transformation and strengthen the model’s ability to perceive various defect characteristics. Second, the local window multi-head attention module (C2CGA) is improved to alleviate the information loss caused by spatial pyramid pooling, while reducing computational complexity and guiding the network to focus on key features. Third, a weighted bidirectional feature fusion network is developed by integrating the design principles of BiFPN into the neck of YOLO11, which enables simple and efficient multi-scale feature representation and cross-scale information interaction. Finally, the detection head is improved by designing a module-aided detection head module (DetectAux), which enhances the supervision signal during the training phase, reduces the number of parameters via depthwise separable convolution, and outputs the prediction information for various categories. Ablation experiments were conducted on the self-constructed dataset, and the results reveal that compared with the YOLO11n algorithm, the modified model increases the Mean Average Precision at an Intersection over Union (IoU) threshold of 0.5 (mAP@0.5) by 6.8%, the recall by 4.7%, and the precision by 15.8%. Compared with mainstream detection algorithms, the results demonstrate that the proposed model outperforms Fast R-CNN, YOLOv8n, RT-DETR, and MobileViT-v2 by 26.2%, 4.6%, 0.2%, and 8.3% in terms of mAP@0.5, respectively. In addition, it exhibits distinct advantages in parameter count, real-time performance and model size, which further verifies the effectiveness and superiority of the proposed online detection algorithm for oil seal ring surface defects in industrial scenarios.
Read more