Deep U-NET Based Heating Film Defect Inspection System

This study introduces a real-time, high-resolution image inspection system that utilizes multiple cameras and deep learning algorithms for the real-time detection of pinholes and scratches on large-area heating films. To accommodate the repetitive inspection processes inherent in products with consistent patterns, the system operates at the region level rather than the frame level. By modifying the U-Net architecture, the system achieved precise segmentation of the inspection area, enabling real-time detection of microscale pinholes and scratches. Additionally, a sticker marker was developed to label the defective regions detected on the film. The proposed system was experimentally validated in an actual production environment, where it demonstrated an impressive 96.6% accuracy in area inspection and a 97.5% defect detection rate at a transportation speed of 12 m/min. These results serve as clear evidence of the effectiveness and practicality of the automatic detection capability facilitated by deep learning in production processes.