Related Experiment Video
Updated: Nov 12, 2025
11:54
Microfluidic Platform with Multiplexed Electronic Detection for Spatial Tracking of Particles
Published on: March 13, 2017
9.6K
A PCB Electronic Components Detection Network Design Based on Effective Receptive Field Size and Anchor Size
Jing Li1,2, Weiye Li3, Yingqian Chen3
1School of Mechanical Engineering, Jiangsu University, Zhenjiang 212000, China.
Computational Intelligence and Neuroscience
|March 17, 2021
Summary
This study introduces an improved YOLOv3 network for real-time electronic component detection on printed circuit boards (PCBs). The new method enhances accuracy and reduces model size for efficient manufacturing quality inspection.
Area of Science:
- Computer Vision
- Machine Learning
- Manufacturing Technology
Background:
- Automated quality inspection in electronics manufacturing relies on accurate component recognition.
- Small, similar-looking electronic components on PCBs pose challenges for traditional visual object detection methods.
- Existing deep learning models like Faster-RCNN, SSD, and original YOLOv3 struggle with efficiency and accuracy for PCB component detection.
Purpose of the Study:
- To design a real-time, efficient, and accurate electronic component detection network for PCBs.
- To address the challenges of detecting small and visually similar components.
- To improve the quality inspection efficiency in electronic product manufacturing.
Main Methods:
- Developed a real-time electronic component detection network using YOLOv3 with effective receptive field size and anchor size matching.
- Calculated and visualized effective receptive field sizes of convolutional neural network (CNN) layers via gradient backpropagation.
- Proposed a modular YOLOv3 composition strategy for flexible network design.
- Implemented an algorithm for matching effective receptive field size and anchor size for a lightweight detection network.
Main Results:
- Achieved the highest detection mean average precision (mAP) of 95.03% on the PCB electronic component dataset.
- Reduced the model parameter size to approximately 1/3 of the original YOLOv3, resulting in the smallest memory footprint.
- Demonstrated the second-best performance in terms of floating-point operations (FLOPs) compared to Faster-RCNN, SSD, and original YOLOv3.
Conclusions:
- The proposed YOLOv3-based network significantly enhances the accuracy and efficiency of electronic component detection on PCBs.
- The method offers a lightweight and effective solution for visual object detection in manufacturing quality inspection.
- This approach provides a valuable tool for improving the overall quality control process in the electronics industry.
