| Summary: | 碩士 === 國立臺灣科技大學 === 高分子系 === 97 === At present, the OLED (Organic Light Emitting Diode) manufacturing industry has not a way of quite accurate OLED surface detection yet. The way of inspect is still by the man-power or by the chromometer primarily. This kind of detection consumes lots of man-power, and makes mistakes easily. Moreover, the tool of detection is quite expensive. In this thesis, an automatic inspection system for the OLED surface defects was developed based on image processing. The OLED surface defects can be divided into three kinds: scratch, luminescent spot and puncture. In image processing, the median filter is first used to reduce the impulse noise of images. Secondly, the high pass filter is used to sharpening the edge pixel. Because the gray values of all defects are close to those of the background, it is difficult to separate the image by a fixed threshold value. First of all, the statistical threshold value decision method is used to choose optimal threshold values with the difference of gray values in image segmentation. Secondly, the opening operator in morphology is used to smooth the contour of defects and erode slight noise. Finally, the compactness, entropy and the specific gravity of the circumscribed circle and defection area are selected as defect features.
In this thesis, eighty one defect samples are collected, by using the SVM (Support Vector Machine) to do the classification. First, the fifteen, twenty, twenty five, thirty, and forty samples are selected randomly. Secondly, the same steps are repeated to take out other two sets just like each of the defect sample set. Finally those sets are use to test the entire samples and we record the recognition rate. When the fifteen samples are used as training data, we can get the recognition rate above 95%. When the twenty are used as training data, we can get the recognition rate above 96%; But when the training samples more than twenty five, we can get the recognition rate of 100%. This experiment proves suitability of the defect features and the reliability of SVM classifier. So, this inspection system can be applied in the OLED surface defect successfully.
|
|---|
