PCB Defect Identification and Localisation Using YOLOv10m

The proposed work aims to implement a YOLOv10m-based neural network model for automated PCB defect detection in order to overcome the constraints of conventional, manual and semi-automated techniques. The increasing complexity of PCB designs in modern electronics requires highly accurate and efficient defect detection mechanisms to ensure manufacturing reliability. Thus, there is a need for a robust model to detect various PCB defects efficiently. To build the robust model, the proposed work was trained on a diverse dataset containing all the six types of defects. Data augmentation techniques were used and the images increased from 1386 to 10,668 images and used Yolo based algorithm to train the model. The experimental results demonstrate a mean detection precision (mAP) of 99.1%, surpassing existing methods such as YOLOv7 and Faster R-CNN in both precision (98%) and recall (98.4%). The findings demonstrate the potential of YOLOv10m for high-throughput PCB inspection, offering a reliable solution to improve manufacturing efficiency and product quality.