Abstract: To address the challenges of high computational demands and resource constraints in real-time defect detection of aerospace carbon fiber components,a lightweight object detection framework was developed.An improved model based on the YOLOv11n framework was constructed by integrating Deformable Convolution (DCNv4) into the backbone network to enhance defect shape perception.A DualConv structure was employed to improve feature extraction efficiency,while the C3K2_MCA module incorporating a multi-dimensional collaborative attention mechanism was designed to strengthen multi-scale feature correlation and representation.Experimental results demonstrated that the improved model had achieved enhancements in both mAP and precision,along with a 20.6% increase in detection speed and a 23% reduction in computational resource consumption.This method effectively balanced detection accuracy,efficiency and computational cost,making it well-suited for real-time defect detection of carbon fiber components.

Key words: carbon fiber components, defect detection, YOLOv11n framework, multi-dimensional collaborative attention mechanism, lightweight model

摘要： 针对航空航天领域碳纤维构件缺陷实时检测高算力需求与资源受限难题,提出基于多维协同注意力优化机制的轻量化目标检测框架。基于YOLOv11n框架构建改进模型,在主干网络引入DCNv4可变形卷积增强缺陷形态的感知能力,采用DualConv卷积核结构提升特征提取效率,并设计融合多维协作注意力机制的C3K2_MCA模块,以强化多尺度特征的关联与表达能力。实验结果表明,改进模型的准确率与精度均得到了提升,并实现了20.6%的检测速度提升与23%的算力资源压缩。该方法在精度、检测效率与计算成本之间达成了最优平衡,有效满足了碳纤维构件缺陷实时检测的工程需求。针对航空航天领域碳纤维构件缺陷实时检测高算力需求与资源受限难题,提出基于多维协同注意力优化机制的轻量化目标检测框架。基于YOLOv11n框架构建改进模型,在主干网络引入DCNv4可变形卷积增强缺陷形态的感知能力,采用DualConv卷积核结构提升特征提取效率,并设计融合多维协作注意力机制的C3K2_MCA模块,以强化多尺度特征的关联与表达能力。实验结果表明,改进模型的准确率与精度均得到了提升,并实现了20.6%的检测速度提升与23%的算力资源压缩。该方法在精度、检测效率与计算成本之间达成了最优平衡,有效满足了碳纤维构件缺陷实时检测的工程需求。

关键词: 碳纤维构件, 缺陷检测, YOLOv11n框架, 多维协作注意力机制, 轻量化模型