Deep learning based defect inspetion in TFT-LCD rib depth detection

• Main Authors: Chao-Ching Ho, Hao-Ping Wang, Yuan-Cheng Chiao
• Format: Article
• Language: English
• Published: Elsevier 2021-12-01
• Series: Measurement: Sensors
• Subjects: Flaw detection; Feature detection; Automatic optical detection; Digital image processing; Convolutional neural network; Model acceleration
• Online Access: http://www.sciencedirect.com/science/article/pii/S2665917421001616

In this research, a set of TFT-LCD rib mark depth detection system was proposed. The system is mainly divided into three parts: hardware, system control and software. For the hardware part, a line scan camera coupled with a telecentric coaxial lense were adopted for shooting. As for the light source, an internal coaxial white light source combined with a white line light source were employed to strengthen characteristic information. For the system control part, Nivdia Xavier AGX was applied. The model weighted data format was changed to INT8 to accelerate the model image prediction speed and shorten the time to 0.18 seconds. For the software part, in order to detect rib mark features, the Unet network was mainly used to carry out feature segmentation, with splitting accuracy reaching 100%.