Graph-Aided Teacher-Student Framework for Semi-Supervised Object Detection in Transformer Bench Inspection

This paper proposes a semi-supervised object detection method applied to transformer bench inspection, using a teacher-student framework enhanced by a Graph Neural Network (GNN)-based Teacher Assistant. By leveraging spatial dependencies between objects commonly found in transformer bench settings, the GNN refines pseudo-labels to improve detection performance with limited labeled data. Experiments demonstrate significant improvements in mean Average Precision (mAP) and F1-score compared to traditional semi-supervised methods, highlighting the effectiveness of the GNN-based Teacher Assistant in refining pseudo-labels and enhancing model robustness in this specific industrial application.