改进YOLOv5s的PCB元器件检测技术  

Improved YOLOv5s Component Detection Technology for PCB

作　　者：张晋钊 刘新妹[1,2] 殷俊龄[1,2] ZHANG Jinzhao;LIU Xinmei;YIN Junling(State Key Laboratory of Electronic Testing Technology,North University of China,Taiyuan 030051,China;School of Information and Communication Engineering,North University of China,Taiyuan 030051,China)

机构地区：[1]中北大学电子测试技术国家重点实验室,山西太原030051 [2]中北大学信息与通信工程学院,山西太原030051

出　　处：《机械与电子》2024年第10期35-41,共7页Machinery & Electronics

基　　金：山西省重点研发项目(201903D121058)。

摘　　要：针对PCB元器件检测中硬件运算需求高、检测速度慢等问题,提出一种基于YOLOv5s改进的轻量化PCB元器件目标检测算法YOLO SCGS。该算法以ShuffleNetv2替换主干网络,在Neck层引入CA注意力机制,使用轻量级卷积方法代替标准卷积,加入SPD Conv重构CNN体系结构,代替卷积步长和卷积层,提高检测精度。实验结果表明,所提算法相较原YOLOv5s检测速度提升16.61%,浮点计算需求减少85.44%,模型权重缩小84.83%,有更低的运算需求和更优的检测速度,满足轻量化本地部署的要求。This paper proposes a lightweight PCB component target detection YOLO SCGS based on YOLOV5s to address the high hardware computational requirements and slow detection speed in PCB component detection technology.This algorithm replaces the backbone network with ShuffleNetv2,introduces CA attention mechanism in the Neck layer,and designs a cross level partial network GSCSP module for lightweight standard convolution.Finally,SPD Conv is used to reconstruct the CNN architecture,replacing the convolution step size and convolution layer to improve detection accuracy.The experimental results show that compared to the original YOLOv5s,the algorithm proposed in this paper improves detection speed by 16.61%,reduces floating point computing requirements by 85.44%,and reduces model weight by 84.83%.It has lower computational requirements and better detection speed,meeting the requirements of lightweight local deployment.

关 键 词：YOLOv5s 印制电路板 目标检测 轻量化 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]