SAC-YOLO11: a channel-pruned architecture for efficient tire defect detection in manufacturing

To address challenges such as complex texture backgrounds in tire X-ray images, low detection accuracy for micro-defects, and real-time requirements, this paper proposes an improved SAC-YOLO11 detection model. Firstly, we replace the backbone network with a lightweight StarNet that enhances high-dimensional nonlinear feature extraction through star operations, achieving a 23.3% reduction in parameter count while improving small target detection performance. Secondly, we design the C3K2_Star module to integrate local and global features, optimizing multi-scale defect feature modeling capability. Finally, an Adaptive Fine-Grained Channel Attention mechanism (AFGC) is introduced in the detection head, which suppresses background interference and enhances key feature attention through interaction modeling between banded and diagonal matrices. Experimental results demonstrate that the improved model achieves 94.6% mAP, representing a 2.1% improvement over the original YOLO11 with 24.0% fewer parameters, significantly outperforming comparative models like YOLOv8n and RTDetr-L. Ablation experiments validate the effectiveness of each module. This model provides a high-precision, lightweight solution for real-time tire defect detection, demonstrating substantial value for industrial deployment.