PCB plug-in solder joint defect detection method based on coordinated attention-guided information fusion.

Area of Science:

• Electronics Engineering
• Computer Vision
• Quality Control

Background:

• Printed Circuit Boards (PCBs) are fundamental to electronic devices.
• Effective PCB defect detection is vital for product quality assurance.
• Existing algorithms struggle to balance precision, low false alarm rates, and speed.

Purpose of the Study:

• To develop a high-precision, low-false-alarm, and high-speed PCB plug-in solder defect detection method.
• To enhance the capabilities of existing defect detection algorithms.

Main Methods:

• Utilized an attention-guided pyramid structure for context information fusion.
• Incorporated a coordinated attention network to calibrate features and reduce redundancy.
• Integrated Atrous Spatial Pyramid Pooling (ASPP) into the Darknet53 backbone for multi-scale feature extraction.

Main Results:

• The enhanced network achieved an average detection accuracy of 96.43%, an improvement from 94.45%.
• The proposed method demonstrates superior performance in PCB plug-in solder defect detection.
• The improvements address the limitations of existing algorithms in precision, false alarm rate, and speed.

Conclusions:

• The improved YOLOv3-based method is highly suitable for PCB plug-in solder defect detection.
• The integration of spatial convolution pooling and information fusion significantly enhances detection performance.
• This approach contributes to more robust quality control in electronic manufacturing.