Deep Learning and Image Processing Techniques for Recognizing Liquid-Crystal Display Array Residue and the Automatic Planning of Laser-Cutting Segments

Abstract

Defect detection in thin-film-transistor liquid-crystal displays (TFT-LCDs) is crucial for ensuring the quality of the display. However, because of the diversity of TFT-LCD panel defects, accurate localization and detection become difficult. To overcome these problems, this study used deep learning and image processing algorithms that automatically detect TFT array defects and presents a laser-cutting path that removes only array defect regions. First, the YOLOv4 model was used to locate the defect and glass region of interest (ROI) in panel images, and a semantic segmentation model (FCN-VGG16) was used to identify defect and glass pixel positions in the ROI. Finally, the effective cutting range of faults was determined using overlap and nonoverlap cutting (ON-cutting) judgment methods. In this investigation, three types of defects were used: D1, D2, and D3. According to the proposed ON-cutting methodology, the error repair rates of D1-, D2-, and D3-type defects were 4.79%, 0%, and 0%, respectively. Therefore, these strategies can help manufacturers enhance the quality of TFT-LCD panels by identifying and locating the optimal glass cutting line and defective pixels.