An Optimization Method for PCB Surface Defect Detection Model Based on Measurement of Defect Characteristics and...

Area of Science:

• Computer Vision
• Artificial Intelligence
• Manufacturing Technology

Background:

• Printed Circuit Boards (PCBs) are critical electronic components.
• Detecting diverse, complex PCB surface defects is challenging due to low resolution and background similarity.
• Existing methods struggle with the intricate nature of PCB surface anomalies.

Purpose of the Study:

• To develop an optimized algorithm for accurate and efficient PCB surface defect detection.
• To address the challenges posed by defect complexity, low feature resolution, and background resemblance.
• To enhance the performance of deep learning models for automated PCB quality inspection.

Main Methods:

• Proposed YOLOv8_DSM algorithm incorporating CSPLayer_2DCNv3 with deformable convolution for adaptive feature extraction.
• Introduced Shallow-layer Low-semantic Feature Fusion Module (SLFFM) with bi-level routing attention (BRA) for enhanced feature fusion and defect-background discrimination.
• Utilized feature map separation-based SPDConv for downsampling and MPDIoU as the bounding box loss function.

Main Results:

• YOLOv8_DSM achieved a mean Average Precision (mAP) of 63.4% (0.5:0.9 IoU), a 5.14% improvement over the baseline YOLOv8.
• The model demonstrated a high processing speed of 144.6 Frames Per Second (FPS).
• The algorithm was successfully deployed in a practical PCB quality inspection system.

Conclusions:

• The YOLOv8_DSM algorithm offers a significant advancement in PCB surface defect detection.
• The integration of deformable convolution, shallow-layer feature fusion, and attention mechanisms effectively tackles detection challenges.
• The model's high accuracy and speed make it suitable for real-world industrial PCB quality control applications.