A lightweight detection algorithm of PCB surface defects based on YOLO

Shiwei Yu ¹, Feng Pan ¹, Xiaoqiang Zhang ¹

⁰CGN Digital Technology Co., Ltd., Shanghai, China.

Plos One +

|

April 18, 2025

View abstract on PubMed

Summary

This study introduces a lightweight YOLO-based algorithm for PCB defect detection, significantly reducing computational load and model size while improving accuracy. The enhanced model is ideal for deployment on resource-constrained platforms.

Area Of Science

Electronics
Computer Vision
Artificial Intelligence

Background

Printed Circuit Board (PCB) defect detection faces challenges with low accuracy and high computational demands.
Existing algorithms often struggle with cluttered backgrounds and subtle defect variations.

Purpose Of The Study

To develop a lightweight and accurate PCB defect detection algorithm.
To optimize the model for efficient deployment on low-arithmetic platforms.

Main Methods

Utilized GhostNet in the Backbone for model lightness.
Improved the neck network with depthwise separable convolution to reduce parameters.
Integrated Swin-Transformer with C3 module (C3STR) to handle complex image backgrounds and defect types.
Replaced PANet with Bidirectional Feature Pyramid Network (BIFPN) for enhanced multi-scale feature fusion.

Main Results

Achieved a 47.2% reduction in parameter count.
Reduced GFLOPs by 48.5% and model weight by 42.4%.
Observed a 2.4% increase in mean Average Precision (mAP) with only a 2.0% decrease in Frames Per Second (FPS).

Conclusions

The proposed lightweight YOLO-based algorithm offers a superior balance between accuracy and computational efficiency for PCB defect detection.
The model's reduced complexity makes it highly suitable for real-time applications on embedded systems and low-arithmetic platforms.