融合RCS-OSA结构的PCB缺陷检测算法
PCB Board Defect Detection Algorithm Integrating RCS-OSA Structure
王超 1梁根2
作者信息
- 1. 吉林化工学院信息与控制工程学院,吉林吉林 132022;广东石油化工学院理学院,广东茂名 525000
- 2. 广东石油化工学院理学院,广东茂名 525000
- 折叠
摘要
针对当前印制电路板(Printed Circuit Board,PCB)检测参数庞大、人工误检率高等问题,文章研究了一种改进的YOLOv5检测模型.首先使用实时控制系统(Real-Time Control Systems,RCS)模块替换三卷积跨阶段局部瓶颈模块(Cross Stage Partial Bottle Neck Mudule,C3)结构,通过重参数化卷积增强网络的特征提取能力;其次添加One-Shot聚合(One-Shot Aggregation,OSA)结构,一次性聚合多个特征级联;最后堆叠RCS-OSA模块,确保特征复用并加强不同层之间的信息流通.实验结果表明,改进后的算法检测精度达到94.6％,比原模型提高了 2.2个百分点.该算法能够高效检测PCB的多种缺陷类型,对PCB质量检测有实际应用价值.
Abstract
Aiming at the problems of huge Printed Circuit Board(PCB)detection parameters and high manual error detection rate,this paper proposes an improved YOLOv5 detection model.Firstly,the Cross Stage Partial Bottle Neck Mudule(C3)structure is replaced by Real-Time Control Systems(RCS)module,and the feature extraction ability of the network is enhanced by re parameterized convolution;Then,add One-Shot Aggregation(OSA)structure to aggregate multiple feature cascades at one time;Finally,RCS-OSA modules are stacked to ensure feature reuse and strengthen information flow between different layers.Experimental results show that the detection accuracy of the improved algorithm reaches 94.6％,which is 2.2％higher than the original model.The algorithm can efficiently detect a variety of defect types of PCB,and has practical application value for PCB quality detection.
关键词
缺陷检测/YOLOv5/印制电路板(PCB)/特征级联Key words
defect detection/YOLOv5/Printed Circuit Board(PCB)/feature cascade引用本文复制引用
基金项目
广东省基础与应用基础研究基金项目(2023A1515012894)
出版年
2024
NETL
NSTL
万方数据