Deep Learning Driven Real-Time PCB Inspection Using an Optimized YOLO v9 Architecture

深度学习 人工智能 计算机科学 建筑 计算机视觉 工程类 系统体系结构 特征(语言学) 人工神经网络 自动目标识别 可视化 钥匙(锁) 工程制图
作者
Jigar Subodhchandra Sarda,Rohan Vaghela,Akash Kumar Bhoi,Chang-Won Yoon,Mangal Sain
出处
期刊:Computers, materials & continua 卷期号:: 1-10 被引量:2
标识
DOI:10.32604/cmc.2026.080112
摘要

Printed circuit boards (PCBs) are essential components that strongly influence the performance and reliability of modern electronic systems. However, minor and visually subtle manufacturing defects can degrade product quality and pose serious challenges for automated inspection systems. Existing deep learning–based methods often struggle to simultaneously achieve high detection accuracy, real-time processing speed, and compact model size. This study proposes an enhanced approach for real-time PCB defect detection using advanced object detection models. A dedicated dataset of bare PCBs was developed and carefully annotated with six defect categories: open circuits, missing holes, spurs, mouse bites, short circuits, and spurious copper. Multiple YOLO-based models were trained and evaluated on this dataset, among which the YOLOv9-small model demonstrated superior performance, achieving a mAP@50 of 99.2% and a mAP@50–95 of 63.1%. To further enhance computational efficiency, optimization techniques such as pruning and quantization were applied to the YOLOv9-small model. These optimizations reduced model size by 36.4% and increased processing speed by 24.7%, while maintaining high detection accuracy with a mAP@50 of 98.6%. Overall, the experimental results demonstrate that the optimized YOLOv9-based model provides a highly accurate, efficient, and practical solution for automated PCB defect detection in real-world electronics manufacturing environments.
最长约 10秒,即可获得该文献文件

科研通智能强力驱动
Strongly Powered by AbleSci AI
科研通是完全免费的文献互助平台,具备全网最快的应助速度,最高的求助完成率。 对每一个文献求助,科研通都将尽心尽力,给求助人一个满意的交代。
实时播报
1秒前
1秒前
1秒前
mayli完成签到,获得积分10
1秒前
1秒前
2秒前
2秒前
Bibabo完成签到,获得积分10
2秒前
hhhhhha完成签到,获得积分10
3秒前
3秒前
充电宝的应助被sunny采纳,获得10
3秒前
酷酷迎彤完成签到 ,获得积分10
4秒前
隐形曼青的应助被meehan采纳,获得30
4秒前
4秒前
枳酒发布了新的文献求助10
4秒前
4秒前
biggerxiaoshuaui完成签到,获得积分10
5秒前
6秒前
xiaohaitao完成签到,获得积分10
6秒前
独自受罪发布了新的文献求助10
6秒前
6秒前
6秒前
欢喜可乐完成签到 ,获得积分10
6秒前
zzzzw发布了新的文献求助10
6秒前
cheng完成签到,获得积分10
7秒前
hzz发布了新的文献求助10
7秒前
向连虎发布了新的文献求助10
8秒前
WWWUBING完成签到,获得积分10
8秒前
希望天下0贩的0的应助被初景采纳,获得50
8秒前
8秒前
saturn发布了新的文献求助10
8秒前
Akim的应助被yanyh采纳,获得10
9秒前
百年孤独完成签到,获得积分10
9秒前
9秒前
柒芋完成签到,获得积分10
10秒前
10秒前
luo发布了新的文献求助10
10秒前
超级冰夏完成签到,获得积分20
11秒前
不安的契发布了新的文献求助10
11秒前
11秒前
高分求助中
(应助此贴封号)【重要!!请各用户(尤其是新用户)详细阅读】【科研通的精品贴汇总】 10000
拉伸高分子熔体中缠结行为的模拟和小角中子散射研究 1500
John McDowells Bildungstheorie 800
Analyses de reconnaissances de dettes passées devant les échevins d'Ypres (1249-1291) 600
Handbook of Pharmaceutical Excipients. 9th ed 400
Synthesis, Application and Future Perspectives of Smart Nano-materials - Part 1 400
Breaking the World 300
热门求助领域 (近24小时)
化学 材料科学 医学 生物 纳米技术 计算机科学 工程类 有机化学 化学工程 生物化学 物理 复合材料 内科学 催化作用 光电子学 人工智能 细胞生物学 无机化学 物理化学 数学
热门帖子
关注 科研通微信公众号,转发送积分 8155816
求助须知:如何正确求助?哪些是违规求助? 9552109
关于积分的说明 21708107
捐赠科研通 7744978
什么是DOI,文献DOI怎么找? 3426225
关于科研通互助平台的介绍 2551237
邀请新用户注册赠送积分活动 2469958