融合GhostBottleneck及注意力机制的铝型材表面缺陷检测算法研究∗  

Research on an aluminum profile surface defect detection algorithm integrating GhostBottleneck and attention mechanism

作　　者：李季村 郑鹏[1] 李岩[1] 何青泽 LI Jicun;ZHENG Peng;LI Yan;HE Qingze(School of Mechanical and Power Engineering,Zhengzhou University,Zhengzhou 450001,China)

机构地区：[1]郑州大学机械与动力工程学院,郑州450001

出　　处：《现代制造工程》2025年第3期115-123,共9页Modern Manufacturing Engineering

基　　金：国家自然科学基金资助项目(51775515)。

摘　　要：在铝型材制造过程中,铝型材表面会因受到材料或加工工艺等因素的影响而产生擦花、脏点等缺陷,直接影响铝型材使用性能。分析了铝型材表面缺陷特点并对比现有深度学习目标检测算法,基于YOLOv8网络模型提出了一种融合GhostBottleneck及注意力机制的铝型材表面缺陷检测算法。首先,将Ghost卷积引入Bottleneck层,并用DWConv替换骨干网络中部分卷积结构,在保证检测精度的同时,降低模型复杂程度;然后,在此基础上,将注意力机制添加到YOLOv8检测头模块中,用于提高该模型的检测精度;最后,开展了实验验证,实验结果表明,融合GhostBottleneck及注意力机制的铝型材表面缺陷检测算法精度达到了0.932,相比基础YOLOv8算法,精度提升了5.9%,且模型运算参数量减少了24%,整体性能可满足工业上对铝型材缺陷检测的精度及速度要求。During the manufacturing process of aluminum profiles,defects such as scratches and dirt spots may occur on the surface due to factors like materials or processing techniques,directly affecting the usability of aluminum profiles.It analyzes the characteristics of surface defects in aluminum profiles and compares existing deep learning-based object detection algorithms.Based on the YOLOv8 network model,an aluminum profile surface defect detection algorithm integrating GhostBottleneck and attention mechanism is proposed.By introducing Ghost into the Bottleneck layer and replacing some of the convolutional structures in the backbone network with DWConv,the complexity of the model is reduced while ensuring detection accuracy.Furthermore,the ECA attention mechanism is added to the YOLOv8 detection head module to enhance the detection accuracy of the model.It conducts experimental verification,and the experimental results show that the accuracy of the improved algorithm reaches 0.932,representing a 5.9%improvement compared to the basic YOLOv8 algorithm.Moreover,the number of model operation parameters is reduced by 24%.The overall performance meets the industrial requirements for the accuracy and speed of defect detection in aluminum profiles.

关 键 词：铝型材 表面缺陷检测 注意力机制 YOLOv8 模型轻量化 GhostBottleneck 

分 类 号：TH164[机械工程—机械制造及自动化] TP391.4[自动化与计算机技术—计算机应用技术]