A Fast Object Detection-Based Framework for Via Modeling on PCB X-Ray CT Images

For successful printed circuit board (PCB) reverse engineering (RE), the resulting device must retain physical characteristics and functionality of original. Although applications RE are within discretion executing party, establishing a viable, non-destructive framework for analysis is vital any stakeholder in PCB industry. A widely-regarded approach uses x-ray computed tomography (CT) to produce three-dimensional volumes with several slices data corresponding multi-layered PCBs. However, noise sources specific CT variability from designers hampers thorough acquisition features necessary RE. This article investigates deep learning as successor current state-of-the-art detecting vias on images; key building block designs. During RE, offer an understanding PCB’s electrical connections across multiple layers. Our method improvement earlier iteration which demonstrates significantly faster runtime quality results comparable or better than state-of-the-art, unsupervised iterative Hough-based method. Compared method, 4.5 times discrete image scenario 24.1 volumetric scenario. The upgrades prior version include feature-based detection real-world usability adaptive post-processing methods improve detections.