期刊名称：Lecture Notes in Engineering and Computer Science
印刷版ISSN：2078-0958
电子版ISSN：2078-0966
出版年度：2018
卷号：2235&2236
页码：90-95
出版社：Newswood and International Association of Engineers
摘要：Real-time inspection of glass substrates to detect
defects has been very important because of the rapid growth in
the flat panel industry. Since the wiring pitch of the glass
substrate becomes increasingly narrower, it is difficult to detect
defects from the time-series data obtained by the non-contact
inspection machine because the data involves much noise. This
study proposes machine learning-based methods of detecting
defects in glass substrates with high precision in a short time.
Several feature quantities are constructed not only to
distinguish defects with noise but also to specify waveform types.
In addition, numerical experiments are conducted using actual
data to show the effectiveness of the proposed method.
