基于DBC-YOLOv8s改进模型的PCB 缺陷检测方法研究

陈松; 周婷; 陈立爱; 张谦

文章摘要

基于DBC-YOLOv8s改进模型的PCB 缺陷检测方法研究

Research on PCB Defect Detection Method Based on DBC-YOLOv8s Improved Modeling

投稿时间：2025-05-21 修订日期：2025-06-26

DOI：

中文关键词: PCB 表面缺陷识别 YOLOv8s算法小目标

英文关键词: PCB Surface Defect Recognition YOLOv8s algorithm Small Targets

基金项目:安徽省高等学校科学研究重大项目（2022AH040044）

作者	单位	邮编
陈松^*	安徽建筑大学	230601
周婷	安徽建筑大学	230601
陈立爱	安徽建筑大学	230601
张谦	安徽建筑大学	230601

摘要点击次数: 142

全文下载次数: 0

中文摘要:

在印刷电路板（Printed Circuit Board，PCB）的质量检测过程中，微小缺陷目标的精准识别面临诸多挑战，主要表现为待检对象尺寸极小、缺陷特征与背景区域对比度低等问题。本研究提出了一种改进型目标检测算法DBC-YOLOv8s。该算法在YOLOv8主干网络特征融合层采用DCNv2结合C2f生成C2f_DCNv2模块，更精准捕捉不规则和复杂背景下的缺陷特征；其次，在模型的head部分集成了BiFPN，以提升网络对多尺度特征的融合效能；最后，在backbone中嵌入CBAM注意力机制聚焦目标区域，提高模型检测精度和鲁棒性。实验结果表明，该算法检测精度和mAP0.5分别较YOLOv8s提高4.1%和1.5%，参数量和计算量减少18.7%和57.8%，表明本模型在提升检测精度和速度的同时兼顾模型轻量化。

英文摘要:

In the quality inspection process of Printed Circuit Board (PCB), the accurate identification of tiny defective targets faces many challenges, which are mainly characterized by the extremely small size of the object to be inspected and the low contrast between the defective features and the background region. In this study, an improved target detection algorithm, DBC-YOLOv8s, is proposed, which employs DCNv2 in the feature fusion layer of the YOLOv8 backbone network in combination with C2f to generate the C2f_DCNv2 module, which can more accurately capture defective features in irregular and complex backgrounds; secondly, a BiFPN is integrated into the head portion of the model to enhance the fusion efficiency of the network for multi-scale features; finally, a BiFPN is integrated in the backbone network to enhance the fusion efficiency of the network for multi-scale features; lastly, a BiFPN is integrated in the head portion of the backbone network to improve the fusion efficiency of the network for multi-scale features. Finally, the CBAM attention mechanism is embedded in the backbone to focus on the target region to improve the detection accuracy and robustness of the model. The experimental results show that the detection accuracy and mAP0.5 of this algorithm are improved by 4.1% and 1.5%, and the number of parameters and computation are reduced by 18.7% and 57.8%, respectively, compared with YOLOv8s, which suggests that the proposed model takes into account the lightweighting of the model while improving the detection accuracy and speed.

View Fulltext 查看/发表评论下载PDF阅读器

关闭