top

知識分享

結合數位全像術與機器學習之透明基板瑕疵檢測

建立日期：2018/10/22
  • 作者： 台師大光電研究所/林昱志; 林佑勳、蔡昕澔、鄭超仁；文化大學電機系/張簡光哲、杜翰艷
  • 出處： 2018 AOI論壇與展覽論文集
  • 內容： th the development of manufacturing technologies, the demand for industrial defect inspection is increased to guarantee product quality. Unfortunately, the traditional manual inspection methods by human inspectors are associated with low accuracy in quality control procedures due to human vision limit, fatigue, and oversight. In addition, learning curve variability is dependent on the inspector experience and may result in intra- and inter-inspector reliability. Therefore, machine vision systems are proposed to efficiently and accurately inspect the undesired defects on products with different materials products. Recently, the transparent substrate is becoming widely used as materials for optics and electronics product manufacturing. Therefore, development of machine vision inspection for the transparent substrate has increased its importance for the quality control. In the past decades, numbers of machine vision methods were proposed for defect detection in various transparent products. For example, the previous study proposed an automated inspection method with specific backlight modules and filters to increase the performance of the quality control process for glass products. On the other hand, by using computer-assisted approaches, several studies were proposed to automatically and efficiently detect different defect patterns on the surface of glass products, such as Otsu thresholding, edge detection method, wavelet transform, Fuzzy logic analysis, and Markov random field analysis. Moreover, to accurately discriminate the detected defect patterns, the characteristics of the defect patterns were acquired via the specific feature extraction, and classifiers, such as artificial neural network and the AdaBoost method were trained by the features for classification. However, the inspection accuracy of machine vision systems using camera-based photography may be limited by two-dimensional (2D) information of intensity images, especially for defect classification. Recently, several studies proposed defect inspection based on analysis of 3D information obtained by optical coherence tomography (OCT). In these methods, sectional examination of OCT was applied at different depths of a glass product to extract 3D information of the refractive index to improve inspection accuracy. However, the examination procedure was accomplished via a point-to-point/line-to-line scanning, which was associated with high running time. Hence, a new medium, called digital holography (DH), which is associated with the fast capturing procedure and 3D diffraction characteristics are needed to assess wavefront information of glass substrate and to provide complex (amplitude and phase) images for improving defect detection and classification.
  • 檔案下載：全文下載
  • 下載次數：90
回上頁