Semi-Supervised Defect Detection Method with Data-Expanding Strategy for PCB Quality Inspection.

Area of Science:

• Computer Vision
• Machine Learning
• Manufacturing Quality Control

Background:

• Printed circuit board (PCB) defect detection is vital for manufacturing quality.
• Current deep learning methods require extensive, costly labeled datasets.
• Semi-supervised learning (SSL) offers a solution by utilizing unlabeled data.

Purpose of the Study:

• To develop an effective semi-supervised learning (SSL) method for PCB defect detection with reduced labeling costs.
• To improve detection accuracy when using a limited number of labeled samples.
• To mitigate the disturbance caused by unlabeled samples during training.

Main Methods:

• Proposed a novel semi-supervised defect detection method with a data-expanding strategy (DE-SSD).
• Introduced a batch-adding strategy (BA-SSL) to effectively leverage unlabeled data with minimal disturbance.
• Implemented a data-expanding (DE) strategy using labeled samples from external datasets to augment the target dataset.

Main Results:

• The DE-SSD method achieved competitive results in PCB defect detection using fewer labeled samples.
• Demonstrated state-of-the-art performance on the DeepPCB dataset.
• Achieved at least a 4.7 mAP improvement compared to existing methods.

Conclusions:

• The proposed DE-SSD effectively reduces data labeling costs for PCB defect detection.
• The method enhances detection accuracy, particularly with limited labeled data.
• DE-SSD offers a robust and efficient solution for automated PCB quality control.