AYOLO-BASEDSYSTEMFORREAL-TIME DETECTIONOFDEFECTS INPRINTED CIRCUIT BOARDS

Abstract

Printed Circuit Boards (PCBs) are the backbone of modern electronic devices and quality of PCB is paramount for reliability of the system. Traditional inspection tech niques for PCB defects by human hand are generally slow, laborious and not very pre cise in the case of defects which are very small and intricate.This paper provides a framework for automated inspection system for PCBs based on a YOLO (You Only Look Once) architecture which is coupled with a Single Head Self-Attention (SHSA) mechanism to boost the feature representation and inspection ability. The aim is to detect various type of PCB defects using the combination of real-time detection by YOLO and attention based reasoning provided by SHSA to enhance the importance of feature regions and minimize redundancy from backgrounds. Attention mechanism learns the feature space of defect categories by paying special attention to the signif icant features while disregarding backgrounds in image thereby making the detection of defects, with different sizes, shapes, and intensity variations, robust. The system analyzes PCB image, extracts distinctive features and predicts bounding box for defect regions. This framework is composed of stages such as data preprocessing, feature extraction, multi scale feature fusion, attention enhancement, and detection of defects. These defect classes in PCBs are annotated to train and test the deep learning based object detection system. The experimental validation has been done by reporting vari ous object detection metrics like Precision, Recall, mean Average Precision (mAP) and inference speed. From the empirical results, it is evident that attention mechanisms are beneficial for capturing intricate features and enhancing defect detection accuracy in PCBs. The present work highlights the significance of attention based feature enrich ment along with real time object detection technique for industrial automation. The proposed system is extensible and feasible for automating the industrial inspection.