Defect detection method of printed circuit boards based on EDF-YOLOv10

Area of Science:

• Computer Vision
• Artificial Intelligence
• Electrical Engineering

Background:

• Printed circuit board (PCB) defect detection faces challenges with small object representation and slow model convergence.
• Existing methods struggle with complex industrial environments and robust feature extraction.

Purpose of the Study:

• To propose an improved YOLOv10 algorithm (EDF-YOLOv10) for enhanced PCB defect detection.
• To develop a real-time, co-optimized hardware and software detection system.
• To improve feature representation for small defects and accelerate model convergence.

Main Methods:

• Incorporated Efficient Channel Attention (ECA) for key channel feature extraction.
• Utilized Dynamic Snake Convolution (DSConv) for improved geometric structure recognition of small targets.
• Implemented Focaler-CIoU loss to enhance hard sample learning and convergence.
• Applied extensive data augmentation to the PKU-Market-PCB dataset for robustness.
• Developed a real-time detection system with industrial cameras, controllable lighting, and a PyQt5 GUI.

Main Results:

• The EDF-YOLOv10 model achieved mAP@0.50 of 90.6% and mAP@0.50:0.95 of 48.4%.
• Demonstrated improvements of 3.0% and 1.6% over the baseline model, respectively.
• The real-time detection system successfully integrated hardware and software for practical application.

Conclusions:

• The proposed EDF-YOLOv10 algorithm effectively addresses limitations in PCB defect detection, particularly for small objects.
• The developed real-time system offers a robust solution for industrial applications.
• This work provides a valuable methodological reference for complex industrial defect detection scenarios.