YOLOv12-Based PCB Defect Classification Post-Etching via SMOTE Data Balancing

Abstract

Printed Circuit Boards (PCBs) are critical components in electronic devices, requiring high precision during manufacturing to ensure optimal circuit functionality. The etching process in PCB fabrication is prone to copper track defects, particularly in liquid transfer techniques, which are susceptible to poor ink adhesion that leads to wrinkled, widened, or broken tracks. To support fast and accurate quality inspection, this study implements YOLOv12 to classify the condition of PCBs post-etching. The dataset consists of 650 PCB images with a resolution of 640 × 640 pixels (200 non-defective and 450 defective PCB images). This research evaluates the performance of four optimizers (Adam, AdamW, SGD, and RMSProp) and analyzes the impact of applying the Synthetic Minority Over-sampling Technique (SMOTE) to address class imbalance. Experimental results demonstrate that without SMOTE, RMSProp achieved the highest overall performance with an accuracy of 93.846% and an F1-Score of 93.635%. However, applying SMOTE significantly boosted the performance of the Adam optimizer, attaining the highest overall accuracy of 95.385% and an F1-Score of 95.415%.