CM-UNetv2: An Enhanced Semantic Segmentation Model for Precise PCB Defect Detection and Boundary Restoration

Area of Science:

• Electronics Manufacturing
• Computer Vision
• Artificial Intelligence

Background:

• Printed circuit boards (PCBs) are crucial in electronics, requiring precise defect detection for quality assurance.
• Traditional methods struggle with small, irregularly shaped PCB defects amidst complex textures.
• Automated defect detection is vital for reliability and manufacturing efficiency.

Purpose of the Study:

• To introduce CM-UNetv2, a novel semantic segmentation network for enhanced PCB defect detection.
• To address the challenges of detecting small, complex defects in PCB manufacturing.
• To improve the accuracy and robustness of automated PCB inspection systems.

Main Methods:

• Developed CM-UNetv2, a semantic segmentation network with three innovative architectural modules.
• Incorporated Parallelized Patch-Aware Attention (PPA) for multi-scale feature representation.
• Introduced Dual-Stream Skip Guidance (DSSG) for refined semantic and spatial information preservation.
• Designed Frequency-domain Guided Context Mamba (FGCMamba) with Spatial Guidance Cross-Attention (SGCA) and Frequency-domain Self-Attention Solver (FSAS) for improved boundary restoration and efficiency.

Main Results:

• CM-UNetv2 achieved state-of-the-art performance on MeiweiPCB and KWSD2 datasets.
• Demonstrated superior accuracy in detecting small PCB defects.
• Showcased enhanced boundary accuracy and overall segmentation robustness compared to existing methods.
• Validated the effectiveness of PPA, DSSG, and FGCMamba modules in improving detection capabilities.

Conclusions:

• CM-UNetv2 offers a significant advancement in automated PCB defect detection.
• The proposed architectural innovations effectively address the limitations of traditional methods.
• The network provides a robust and accurate solution for ensuring electronic product quality and reliability.