An End-to-End Marking Recognition System for PCB Optical Inspection

In the context of integrating automation and developing high-precision visual inspection systems for hardware assurance, extracting information about the components on printed circuit boards plays a crucial role, as they are susceptible to fraudulent intrusions that can cause unfathomable damage to the electronic supply chain. Recognizing component labels and reference locators on the PCB surface with high precision is a critical element of such systems. While most of the state-of-the-art text detection and recognition methods focus on the challenges regarding the characteristics of the natural scene images, very few works clearly identify such edge case scenarios in PCB assurance domain. In our study, we identified those circumstances with sound justification and used carefully selected data augmentation approaches to create a better-performing end-to-end marking recognition system for PCB optical inspection.