Enhanced YOLOv11 framework for high precision defect detection in printed circuit boards

Area of Science:

• Computer Vision
• Deep Learning
• Electronic Manufacturing

Background:

• Automated defect detection in Printed Circuit Boards (PCBs) is crucial for electronic reliability.
• Conventional inspection methods suffer from subjectivity, high labor costs, and limited adaptability.
• Existing deep learning models may lack efficiency and accuracy for complex PCB defect identification.

Purpose of the Study:

• To develop an enhanced deep learning framework, YOLOv11-PCB, for high-accuracy automated defect detection in PCBs.
• To improve upon existing methods by integrating novel modules for adaptive feature extraction and refined bounding box regression.
• To achieve real-time processing speeds while maintaining superior detection performance.

Main Methods:

• The proposed YOLOv11-PCB framework incorporates an Efficient Multi-Scale Attention (EMA) module for adaptive feature extraction.
• A Content-Aware ReAssembly of Features (CARAFE) mechanism is utilized for dynamic receptive field adjustment.
• A refined Efficient Intersection over Union (EIoU) loss function is implemented to optimize bounding box regression.

Main Results:

• YOLOv11-PCB achieved 99.5% mAP@0.5 and 90.7% mAP@0.5:0.95 on the Peking University PCB dataset, surpassing the baseline by 9.7%.
• On the DeepPCB dataset, it attained 98.9% and 81% mAP, demonstrating significant improvements.
• The system processed at 227.2 frames per second (FPS), outperforming state-of-the-art methods in both accuracy and efficiency.

Conclusions:

• YOLOv11-PCB demonstrates robust and efficient detection of critical PCB defects like solder bridges, missing components, and fractures.
• The framework meets industrial throughput requirements while offering enhanced accuracy.
• The proposed innovations contribute to advancing automated optical inspection in electronics manufacturing.