摘要
针对印刷电路板表面缺陷目标小,检测精度低问题,设计了印刷电路板表面缺陷检测网络Multi-CR YOLO,满足实时检测速度的前提下,有效提高了检测精度。首先,由3个Multi-CR块组成的主干特征提取网络Multi-CR backbone对印刷电路板小目标缺陷进行特征提取。其次,SDDT-FPN特征融合模块使层级高的特征层向层级低的特征层进行特征融合,同时为小目标预测头YOLO Head-P3所在特征融合层加强特征融合,进一步增强低层特征层的表达能力。PCR模块加强主干特征提取网络与SDDT-FPN特征融合模块不同尺度的特征层的特征融合机制,且防止模块之间进行特征融合时信息丢失。C_(5)ECA模块负责自适应调节特征权重和自适应注意小目标缺陷信息的要求,进一步提高了特征融合模块的自适应特征提取能力。最后,3个YOLO-Head负责针对不同尺度的小目标缺陷进行预测。实验表明,Multi-CR YOLO网络模型检测mAP达到98.55%,模型大小为8.90 MB,达到轻量化要求,检测速度达到了95.85 fps,满足小目标缺陷实时检测的应用需求。
Aiming at the problem of small target and low detection accuracy of printed circuit board surface defects,Multi-CR YOLO,a printed circuit board surface defect detection network,is designed to meet the premise of real-time detection speed and effectively improve the detection accuracy.Firstly,the backbone feature extraction network Multi-CR backbone,which consists of three Multi-CR residual blocks,performs feature extraction for small target defects on printed circuit boards.Secondly,the SDDT-FPN feature fusion module enables the feature fusion from the high level feature layer to the low level feature layer,and at the same time strengthens the feature fusion for the feature fusion layer where the small target prediction head YOLO Head-P3 is located,to further enhance the expressive ability of the low level feature layer.The PCR module strengthens the feature fusion mechanism of the different scales of the backbone feature extraction network and the feature layer of the SDDT-FPN feature fusion module,and prevents the fusion mechanism between the modules.The C_(5)ECA module is responsible for adaptively adjusting the feature weights and adaptively paying attention to the requirement of small target defect information,which further improves the adaptive feature extraction capability of the feature fusion module.Finally,the three YOLO-Head are responsible for predicting small target defects for different scales.The experiments show that the detection mAP of the Multi-CR YOLO network model reaches 98.55%,the model size is 8.90 MB,which meets the lightweight requirement,and the detection speed reaches 95.85 fps,which meets the application requirements of real-time detection of small-target defects.
作者
姜媛媛
蔡梦南
Jiang Yuanyuan;Cai Mengnan(School of Electrical and Information Engineering,Anhui University of Science and Technology,Huainan 232001,China;Institute of Environment-Friendly Materials and Occupational Health,Anhui University of Science and Technology,Wuhu 241003,China)
出处
《电子测量与仪器学报》
CSCD
北大核心
2023年第11期217-224,共8页
Journal of Electronic Measurement and Instrumentation
基金
安徽省重点研究与开发计划(202104g01020012)
安徽理工大学环境友好材料与职业健康研究院研发专项基金(ALW2020YF18)项目资助
