| Summary: | 碩士 === 元智大學 === 資訊管理學系 === 106 === Defect inspection has become more difficult with the complex process and high product-mix in the manufacturing process of thin film crystal liquid crystal display panel (TFT-LCD). The existing study of defect classification is not easy for the expert through the human eye experience, but also the threshold of many categories is ambiguous. This study constructs an ensemble classification model based on convolutional neural networks, and generates images based on simultaneous processing. The characteristics of the change learning can be detected and classified according to the defect category. In this study, the validity of the deep learning network model was verified by the defect image data in a thin film crystal liquid crystal display panel process, and the experimental comparison was made through the ensemble method to explore the images that the model could not correctly distinguish and the individual images, and also find the pros and cons of each model.
|
|---|
