Mask Attention-Based Vision Transformer (MA-ViT) for PCB Defect Detection

PCBs (Printed Circuit Boards) are valuable components in a wide variety of electronic devices, which are used for industry purposes. PCB defects can have detrimental effects on system performance, resulting in malfunctions, failures, compromised connectivity, and signal distortion. It is essential to detect and classify these defects promptly and effectively. In light of the advancements made in artificial intelligence, this topic has garnered significant attention in the research community, resulting in numerous dedicated research and publications. A recently published automated model utilizing CNN-based autoencoders has shown promise in detecting and classifying various problems. However, some challenges such as overfitting, transitive invariance, and model performance need to be addressed explicitly. To address these challenges, we propose a Mask Attention-based Vision Transformer (MA-ViT) model, which offers a promising solution. The MA-ViT model has demonstrated exceptional performance, showcasing its effectiveness in overcoming the aforementioned issues.