Verification of Unwanted Design Components on PCBs: A Comparative Analysis of YOLOv10 Variants

The Automotive and Industry 4.0 sectors face significant challenges in verifying unwanted design components, traditionally reliant on manual PCB inspection methods like X-ray imaging, which are slow and limited in defect classification. This study evaluates the performance of various YOLOv10 variants for detecting PCB defects. Findings reveal that YOLOv10-L achieves the highest mean average precision (mAP50-95) at 53.17% within a confidence interval of 50-95%, while YOLOv10-N offers computational efficiency at 52.95%. The YOLOv10n model, developed through hyperparameter tuning, achieves superior performance with 96.025% mAP50 and 96.7% Precision-Recall, addressing crucial needs in automotive PCB fault detection.