P‐42: AI‐Based Rapid Defect Detection Method for Display Screen Appearance

The display manufacturing industry faces growing challenges as pixel sizes shrink and product complexity increases. Traditional defect detection algorithms, while widely adopted, suffer from three critical limitations: (1) Scenario‐specific customization requiring labor‐intensive parameter adjustments during product transitions; (2) Compromised efficiency due to manual balancing between over‐detection (false positives) and underdetection (missed defects); (3) Inability to handle emerging defect types in advanced manufacturing processes. This study proposes a universal AI‐driven inspection framework integrating a dual‐stage detection architecture: A high‐speed screening module rapidly filters defect‐free regions (> 99% elimination rate), followed by a deep learning‐based classification module for precise defect identification. Implemented with an optimized YOLO architecture incorporating global attention mechanisms and dimensionality reduction, our solution achieves a 3% to 5% higher accuracy than baseline models while maintaining real‐time processing capabilities. The system demonstrates strong adaptability across diverse defect types (scratches, ITO defects, particle contamination, etc.) and industrial scenarios, with deployment timelines reduced by 68% compared to conventional methods.