具深度學習之數位全像顯微系統於玻璃基板瑕疵檢測
No Thumbnail Available
Date
2018
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
本論文主要探討運用深度學習在玻璃基板瑕疵檢測的技術及利用數位全像顯微系統得到玻璃基板的複數影像進行瑕疵檢測，透過全像術取得玻璃基板的光學繞射資訊，其中複數影像包含振幅資訊及相位資訊，用深度學習進行學習背景、灰塵、刮痕、污漬、棉絮及水痕的振幅資訊及相位資訊之特性，更進一步探討各種瑕疵之間的差異、特性及辨識結果。在全像術和深度學習的運用，本研究會探討調整參數及改進流程達到檢測系統的辨識正確率最大化，以及影像校正方面使得可以得到品質穩定的複數影像，以利未來可以推廣到透過數位全像顯微系統擷取其他材料的光場資訊進行檢測，增加未來應用層面的潛力。
We present an inspection method based on deep learning for defect classification of glass substrate by using complex wavefronts measurement in digital holographic microscopy. The complex image contains the amplitude and phase. The proposed inspection performed on convolutional neural network can achieve high accuracy of classifying defects, such as dust, crack , fiber, stain and watermark, on the glass substrate. We will adjustment parameters and the image correction to obtain a stable image with stable quality, which has more potential in application in the future.
We present an inspection method based on deep learning for defect classification of glass substrate by using complex wavefronts measurement in digital holographic microscopy. The complex image contains the amplitude and phase. The proposed inspection performed on convolutional neural network can achieve high accuracy of classifying defects, such as dust, crack , fiber, stain and watermark, on the glass substrate. We will adjustment parameters and the image correction to obtain a stable image with stable quality, which has more potential in application in the future.
Description
Keywords
深度學習, 全像術, 瑕疵檢測, 玻璃基板, 摺積神經網路, Deep Learning, Holography, Defect Inspection, Glass Substrate, Convolutional Neural Network