Summary: | 碩士 === 國立臺灣科技大學 === 自動化及控制研究所 === 99 === This study identified and classified micro defects in LED grain. The LED grain inspection was divided into three parts in this study, which are inspection of LED grain shape, inspection of electrode zone of LED grain and inspection of luminous zone of LED grain. The inspection of LED grain shape contains two parts, which are grain shape integrity and finger zone integrity. The inspection of electrode zone of LED grain is divided into two stages, 1) the ratio of defects in the electrode zone is calculated as the judgment base of LED grain quality, 2) three common defects in the electrode zone, probe, scrape and contamination are classified and identified. Finally, the luminous zone of LED grain will be inspected for four defect types, such as remaining of metal, dust, breakdown and contamination.
As for the image processing procedure of this study, first, in sample selection, the Normalized Cross Correlation is used for template matching, so as to extract the image of single LED grain, and then the image subtraction, image binarization, morphology and image mask are used for defect detection and location. Afterwards, the image features are captured for the detected defects, the extracted features are classified into geometric features (e.g., area, perimeter, centroid, etc.) and texture features (e.g., gray level co-occurrence matrix, histogram texture feature, etc.). Finally, the decision tree J48 algorithm is used for classification and identification. The overall recognition rate of three-stage inspection flow is higher than 90%. Therefore, it is proved that the detection method proposed in this study can be used in detection and classification of micro defects in LED grain. The defective LED grains can be detected effectively by using the detection system proposed in this study, and the product yield rate can be increased and the production cost can be reduced.
|