基于YOLO-G的PCB缺陷检测算法
PCB defect detection algorithm based on YOLO-G
苏佳 1罗都 1贾欣雨 1梁奔 1冯康康1
作者信息
- 1. 河北科技大学 信息科学与工程学院,河北 石家庄 050018;视航实验室,河北 石家庄 050018
- 折叠
摘要
PCB存在线路设计多样、缺陷面积小且各缺陷之间特征相似等问题,导致缺陷检测精度较低,本文提出使用YOLO-G缺陷检测算法提高检测精度.首先,基于YOLOv7 基础模型,使用SPD-Conv模块解决池化过程中像素信息丢失的问题.其次,引入SimAM模块从三维角度区分重要特征信息,提高模型对不同程度特征表示的感知能力.最后,使用BiFPN融合结构提高低层细节和高级语义信息的多路径交互融合,提高小目标位置的高效定位与分类.实验数据表明:该算法平均精度均值和召回率较原算法分别提升 8.2%和 5.8%,提高了PCB小目标的检测性能和检出率,证明了算法的有效性.
Abstract
PCB defects have problems such as diversified line designs,small defect areas,and similar characteristics between each defect,lead to defect detection precision is low.In this paper,the purpose of using YOLO-G algorithm for PCB defect detection is to improve defect detection precision.Firstly,based on the YOLOv7 basic model,the SPD Conv module is used to solve the problem of pixel information loss during the pooling process.Secondly,introducing the SimAM module to distinguish important feature information from a three-dimensional perspective,improving the model's perception of different degrees of feature representation.Finally,using BiFPN fusion structure to improve multi-path interactive fusion of low-level details and high-level semantic information,and improve efficient localization and classification of small target positions.The experimental data show that the mean Average Precision(mAP)and recall rates of this algorithm have increased by 8.2% and 5.8% respectively compared to the original algorithm,greatly improving the detection performance and rate of PCB small targets,proving the effectiveness of this algorithm.
关键词
目标检测/YOLOv7/缺陷检测/平均精度均值/召回率Key words
object detection/YOLOv7/defect detection/mAP/recall引用本文复制引用
基金项目
国家自然科学基金青年基金(62105093)
出版年
2024
NETL
NSTL
万方数据