一种改进的YOLOv8s模型用于电子线路板瑕疵检测  

An Improved YOLOv8s Model for Defect Detection in Printed Circuit Boards

作　　者：黄文浩 HUANG Wenhao(China Three Gorges University,College of Computer and Information Technology,Yichang 443002)

机构地区：[1]三峡大学计算机与信息学院,湖北宜昌443002

出　　处：《长江信息通信》2025年第1期57-60,66,共5页Changjiang Information & Communications

摘　　要：电子线路板在生产制作过程中存在瑕疵、缺陷、漏孔等问题,因此在生成制造过程中对其进行瑕疵检测是重要且必要的,这能快速定位问题与弥补一定的经济损失。针对电子线路板瑕疵检测任务,该文提出一种改进YOLOv8s模型的细粒度图像目标检测方法EFF-YOLOv8s。首先,为了细粒度地提取图像特征与提高图像的目标检测准确率,使用EfficientFormerV2作为主干网络,替换YOLOv8s模型原始的卷积块结构;其次为了更突出地表征待检测目标的信息,该文在特征融合阶段基于高效的空间注意力ELA提出一种全新的C2f-ELA模块。最后的实验结果表明文章提出的方法在电子线路板瑕疵检测数据集上的准确率、召回率、mAP@0.5以及mAP@0.5:0.95指标分别能达到97.3%、93.9%、97.1%和55.4%,分别较YOLOv8s模型涨点5.0%、3.0%、2.7%和4.6%。Printed Circuit Boards have defects,leakage holes,etc.,in the production and fabrication process,It is therefore important and necessary to detect defects at this stage,which can quickly locate the problem and make up for certain economic losses.For the task of printed circuit board defect detection,this paper proposes a fine-grained object detector called EFF-YOLOv8s based on YOLOv8s.firstly,in order to extract image features at a fine-grained level and to improve the object detection accuracy of the image,EfficientFormerV2 is utilized as the backbone,replacing the original convolution block structure of the YOLOv8s.Secondly,in order to more prominently characterize the information of the object,this paper proposes a new C2f-ELA module based on the efficient spatial attention during the feature fusion stage.The final experimental results show that The accuracy、recall、mAP@0.5 and mAP@0.5:0.95 of the method proposed in this paper on the printed circuit board defect detection dataset can reach 97.3%、93.9%、97.1%和55.4%,respectively,and rise 5.0、3.0、2.7 and 4.6 points,respectively,compared with the YOLOv8s.

关 键 词：深度学习 目标检测 YOLOv8 瑕疵检测 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]