Publication

Abstract :

Ensuring the quality of printed circuit boards is crucial for the reliability of electronic devices. Traditional manual inspection methods are labor-intensive and prone to errors, underscoring the need for automated solutions. This study presents an automated deep learning-based approach for printed circuit board inspection and defect detection by leveraging segmentation and object detection models. The investigation focuses on the performance of the SAM and an integrated YOLO-SAM pipeline. A high-resolution dataset comprising PCB images annotated with six common defect types—missing hole, mouse bite, open circuit, short, spur, and spurious copper—is collected and pre-processed for training and evaluation. Initial experiments with SAM reveal limitations in segmenting defects without guided prompts, particularly board with missing holes. To address this, YOLO is integrated with SAM to provide localized bounding box prompts, enhancing segmentation accuracy. The YOLO-SAM model is trained for 100 epochs and evaluated using precision, recall, F1-score, mean average precision, Intersection over Union, and Dice score. Results demonstrate that the YOLO-SAM pipeline significantly outperforms standalone SAM, achieving more accurate and robust detection across multiple folds. This two-stage approach offers a reliable solution for automated, high-precision printed circuit boards’ defect analysis.