結合卷積神經網路、條件生成對抗網路與遷移式學習集成網路於薄膜液晶顯示器Mura瑕疵分類

對於薄膜液晶顯示器（TFT-LCD）產業而言，面板缺陷識別是至關重要的問題。 Mura缺陷會導致屏幕顯示不均勻，Mura缺陷識別是所有缺陷識別中最具挑戰性的問題。近年來，人工智能技術已成功應用於眾多領域，但是，這樣的方法需要大量的訓練圖像數據。TFT-LCD產業，產品差異化和客製化的策略已迫使產業從大規模生產轉向少量多樣的生產模式，在這樣的環境下，很難收集大量的訓練數據；此外，在產線檢查站，透過AOI系統收集到的Mura缺陷的圖像，經常被莫爾條紋污染。莫爾條紋是檢查相機傳感器的像素網格和面板屏幕之間發生干擾的結果，嚴重影響了圖像的視覺品質，導致難以識別Mura缺陷。因此，在不損害圖像品質的情況下去除缺陷圖像中的莫爾條紋是非常重要的。本研究探討了這個問題，並提出了一種使用條件生成對抗網絡（Conditional GAN）從缺陷圖像中消除莫爾條紋的方法。此外，本研究也開發了一種遷移式學習集成網路模型，該模型基於有限訓練數據，聚合了多個卷積神經網絡（CNN），進行Mura缺陷分類。透過使用所提出的方法進行了實際案例研究，針對TFT-LCD面板中的Mura缺陷進行分類，結果顯示，本研究提出的方法為Mura缺陷分類提供了較優的準確性。因此，該方法可以成為TFT-LCD產業中手動分類的可行替代方法。

關鍵字

薄膜液晶顯示器；卷積神經網絡； Mura缺陷；條件生成對抗網絡；莫爾條紋；遷移式學習；集成學習

並列摘要

Display panel defect recognition is a critical concern for thin-film transistor liquid crystal display (TFT-LCD) manufacturers. Mura defects cause uneven screen displays and are the most challenging to detect among all visual defects. In recent years, artificial intelligence technologies have been successfully applied in numerous areas. However, such approaches require large amounts of training image data. Simultaneously, product differentiation and customization strategies have forced the TFT-LCD manufacturing industry to shift from mass production to high-mix, low-volume, and short-life-cycle production. In this environment, collecting a large amount of training data is difficult. Moreover, images with Mura defects captured at the inspection station remain challenging because they are often contaminated with moiré patterns. Moiré patterns, a result of interference between the pixel grids of the inspection camera’s sensor and panel screen, severely affect the visual quality of images and cause difficulty in determining Mura defects. Therefore, removing moiré patterns in defect images without impairing image quality is critical. In this study, we explored this problem and proposed an approach to eliminate moiré patterns from defect images by using a conditional generative adversarial network (CGAN). In addition, we developed a transfer learning ensemble model that aggregates multiple convolutional neural networks (CNNs) based on a denoising network for defect classification in a limited training data set. An industrial case study was conducted to classify Mura defects in TFT-LCD panels by using the proposed approach. The results demonstrated that the proposed method provides improved accuracy for Mura defect classification. This method can therefore become a viable alternative to manual classification in the TFT-LCD manufacturing industry.

並列關鍵字

Thin-film transistor liquid crystal display ； convolutional neural network ； Mura defect ； Conditional Generative adversarial network ； Moiré pattern ； Transfer learning ； ensemble learning

參考文獻

[14] M. Kim, M. Lee, M. An, and H. Lee, “Effective automatic defect classification process based on CNN with stacking ensemble model for TFT-LCD panel,” Journal of Intelligent Manufacturing, vol. 31, no. 5, pp. 1165–1174, 2020, doi: 10.1007/s10845-019-01502-y.

[31] F. Mattias P. Heinrich, M. Stille, and T. M. Buzug, “Residual U-Net convolutional neural network architecture for low-dose CT denoising,” Curr. Directions Biomed. Eng., vol. 4, no. 1, pp. 297–300, 2018. Online]. Available: https://doi.org/10.1515/cdbme-2018-0072

[48] J. Margeta, A. Criminisi, R. C. Lozoya, D. C. Lee, and N. Ayache, “Fine-tuned convolutional neural nets for cardiac MRI acquisition plane recognition,” Comput. Methods Biomech. Biomed. Eng. Imag. Visual., vol. 5, no. 5, pp. 339–349, 2016, doi: 10.1080/21681163.2015.1061448.

[59] J. Long, E. Shelhamer, and T. Darrell, “Fully convolutional networks for semantic segmentation,” in Proc. Computer Vision and Pattern Recognition(CVPR), Boston, MA,USA,doi. 10.1109/CVPR.2015.7298965, 2015.

REFERENCES

Google Scholar

國際替代計量

結合卷積神經網路、條件生成對抗網路與遷移式學習集成網路於薄膜液晶顯示器Mura瑕疵分類

全文下載

主題瀏覽