| Summary: | 碩士 === 國立東華大學 === 電機工程學系 === 97 === The purpose of this thesis is to develop a defect inspection system for anti-reflection (AR) glass. Using a high resolution line scan charge-coupled device (CCD) with camera link card and a linear motor drive system loaded with glass, the image of AR glass is grabbed under high speed scan. A linear stage with 600mm stroke, 3m/s maximum speed and 1μm encoder resolution is used in this automatic optical inspection (AOI) bench. A 314mm × 476mm AR glass with 1.6mm thickness is inspected in this study.
The study is focused on the classification of the defects of the AR glass which are consisted of scratch, fingerprint, particle, print scratch,and pinhole with two lighting sources. In the image processing, first,image enhancement and denoising are carried out. Then, the scanned image is segmented using binary method with adaptive threshold values.Moreover, rapid density-based clustering applications with noise (RDBSCAN) is adopted for defect clustering. Furthermore, four features are defined and calculated as the training patterns for the classification of defects using back-propagation neural network (BPNN). The defined features are the average grey level in the defect image, the contrast, the correlation, and the compactness. Finally, the experimental results show that the investigated method can detect and classify various defects effectively for AR glass.
|
|---|
