Global contextual attention augmented YOLO with ConvMixer prediction heads for PCB surface defect detection

Area of Science:

• Computer Vision
• Artificial Intelligence
• Machine Learning

Background:

• Printed circuit board (PCB) inspection faces challenges with detecting numerous small targets and complex background textures, leading to missed and false detections.
• Existing object detection models struggle with the fine-grained details required for accurate PCB defect identification.

Purpose of the Study:

• To develop an advanced object detection model, GCC-YOLO, specifically designed to enhance the detection of small targets in PCB inspection.
• To improve the precision, recall, and overall accuracy of PCB defect detection while maintaining computational efficiency.

Main Methods:

• Proposed a Global Contextual Attention augmented YOLO model with ConvMixer prediction heads (GCC-YOLO).
• Utilized a high-resolution feature layer (P2) for enhanced detail and positional information of small targets.
• Integrated a global contextual attention module (GC) with a C3 module in the backbone for noise suppression and feature enhancement.
• Employed a bi-directional weighted feature pyramid (BiFPN) for effective feature fusion and reduced information loss.
• Introduced a ConvMixer module combined with a C3 module for a novel prediction head to improve small target detection and reduce parameters.

Main Results:

• GCC-YOLO demonstrated significant improvements over YOLOv5s on a PCB dataset, increasing Precision by 0.2%, Recall by 1.8%, mAP@0.5 by 0.5%, and mAP@0.5:0.95 by 8.3%.
• The proposed model achieved a smaller model volume and faster reasoning speed compared to other state-of-the-art algorithms.
• Enhanced capability in detecting small targets and suppressing background noise was confirmed.

Conclusions:

• GCC-YOLO effectively addresses the limitations of detecting small targets and complex backgrounds in PCB inspection.
• The integration of global contextual attention and ConvMixer prediction heads offers a promising approach for improving object detection performance in specialized industrial applications.
• The model provides a balance between detection accuracy and computational efficiency, making it suitable for real-world PCB manufacturing quality control.