| dc.description.abstract | In recent years the thin-film transistor liquid crystal display (TFT LCD) has a high demand in the market, the high necessity of the product quality control and the requirements of the various defect detections are more stringent. The high defect detection rate is the basic requirement of the quality control process. The use of the conventional human visual inspection methods to find the defects of the TFT LCD is not accurate, and will consume a large amount of resources. The automatic defect inspection method is necessary to this industry. In addition to find the defects, the types of the defects should be recognized as well.
Here, we propose a method based on the optical interference patterns sensing method to find the interference fringes, and use the image processing to enhance the contrast of the interference fringes, and the recognition rate for the later process could be increased. The neural network method is used to learn the defects and to identify the types of the defects. The paper is to focus on the mura defect inspection and the classification. In the beginning, before the learning process, about 3% of misjudges were happened, if the threshold of the certainty factor is set to be 90%. After neural network retrain these samples, the result of the rest of the images, got a non misjudge level (100% correct). The neural network system gets a non-misjudge level after retraining. The elapse time for the inspection of one panel is less than 1 sec. The inspection procedure is processed before the injection of the liquid crystal (LC) into the lattice panel. The defect panels could be sorted out, so that the later process and the waste of the materials could be avoided, it is called the pretest process of the TFT LCD product.
| en_US |