摘要
针对工业生产中铝合金型材表面缺陷在实际检测中出现漏检和误检的情况,提出一种YOLOv5-Ghost-CBAM-BiFPN模型对铝型材缺陷进行更加精确的检测。首先在YOLOv5 Backbone网中引入了一个轻量级Ghost模块,在保证准确性的前提下显著提高了检测速度。其次,将卷积块注意机制(CBAM)模块添加到Backbone网络的卷积层,以增强特征提取,进一步提高检测精度。此外,考虑到铝型材缺陷尺寸差异,在Neck模块中使用了用于多尺度特征融合的双向特征金字塔网络(Bi-FPN)来聚合不同缺陷类型的特征。实验表明:优化后的模型mAP、精确率P、召回率R都有明显提高。
Aiming at the situation of missed detection and false detection of the surface defects of aluminum alloy profiles in industrial production,a YOLOv5-Ghost-CBAM-BiFPN model is constructed to detect the defects of aluminum profiles more accurately.First,a lightweight Ghost module is introduced into the YOLOv5 Backbone network,which significantly improves the detection speed under the premise of ensuring accuracy.Secondly,the Convolutional Block Attention Mechanism(CBAM)module is added to the convolutional layer of the Backbone network to enhance feature extraction and further improve the detection accuracy.In addition,considering the difference in the defect size of aluminum profiles,a bidirectional feature pyramid network(Bi-FPN)for multi-scale feature fusion is used in the Neck module to aggregate the features of different defect types.Experiments show that the optimized model mAP,precision P,and recall R are significantly improved.
作者
郭北涛
张颢严
GUO Beitao;ZHANG Haoyan(School of Mechanical and Power Engineering,Shenyang University of Chemical Technology,Shenyang 110142,China)
出处
《机械工程师》
2024年第6期22-26,共5页
Mechanical Engineer
