摘要
针对现有的轮毂内部缺陷检测准确度有限,模型复杂度高的问题,提出一种基于YOLOv8的轻量化轮毂缺陷检测算法SNCL-YOLO。首先,采用StarNet网络架构构造C2f STA模块,取代原骨干网络中的C2f模块,降低存储需求并且提升计算速度,以提高检测效率;其次,在颈部网络中引入卷积注意力融合模块(convolution and attention fusion module,CAFM),增强全局和局部的特征建模;最后,将原检测头替换,采用LSCD Head轻量级共享卷积检测头,减少网络参数量和计算量。实验结果表明,改进后的模型参数量减少11.9%,计算量减少9.8%,平均精度均值(mAP)从90.7%提升到了94.1%,增加3.4百分点。该模型在计算量以及参数降低情况下,有效增强了铝合金轮毂缺陷检测模型的检测性能。
In order to solve the problems of limited accuracy and high complexity of model existed in the internal defect detection of wheel hubs,a lightweight wheel defect detection algorithm SNCL-YOLO based on YOLOv8 is proposed.Firstly,the StarNet network architecture is used to construct the C2f STA module to replace the C2f module in the original backbone network,which reduces the storage requirement and improves the computing speed to improve the detection efficiency.Secondly,the convolution and attention fusion module(CAFM)is introduced into the neck network to enhance the global and local feature modeling.Finally,the original detection head is replaced by LSCD Head lightweight shared convolutional detection head,which reduces the number of network parameters and the amount of computation.The experimental results show that the number of parameters of the improved model is reduced by 11.9%,the amount of calculation is reduced by 9.8%,and the mean average precision(mAP)is increased from 90.7%to 94.1%,with an increased mAP of 3.4 percentage points.The model effectively strengthens the detection performance of the aluminum alloy wheel defect detection model under the reduction of computational cost and parameter.
作者
曹振锋
王明泉
路宇鹏
吴志成
王晋华
CAO Zhenfeng;WANG Mingquan;LU Yupeng;WU Zhicheng;WANG Jinhua(School of Information and Communication Engineering,North University of China,Taiyuan 030051,China)
出处
《机械与电子》
2025年第6期31-36,共6页
Machinery & Electronics
基金
国家自然科学基金资助项目(61171177)。
