Summary: | 碩士 === 朝陽大學 === 工業工程與管理研究所 === 87 === In textile industry, the manufacturing process of textile fabrics is continuous and mass productive. The printing of textile fabrics is also a continuous process and inspectors need to examine printed textile fabrics for a long time. This inspection process is time-consuming and tedious. The inspectors may be too tired to make correct decisions in the judgements of printing faults of textile fabrics. This research applies computer vision techniques and statistical testing methods in the automated detection of printing faults of textile fabrics for decreasing the probabilities of judgement errors.
The purpose of this research is to develop a computer aided visual inspection system to detect printing faults of textile fabrics. The sources of printing faults of textile fabrics can be classified into black, blue and red colors in this research. The relations between different colors of printing faults and the abilities of printing faults detection of different statistical methods can be obtained. Two kinds of statistical methods, chi-square test and multiple comparison methods, are used in this research. The multiple comparison methods include LSD method, Duncan's test, Newman-Keuls test and Tukey's test. Each of the statistical multiple comparison methods has different critical value in the decision of hypothesis testing, this results in different detection effects.
From the experimental results of applying the statistical methods in printing faults detection of textile fabrics, we find that the detection of black printing faults is very easy and the detection of red printing faults is very hard. Among the effects of printing faults detection of textile fabrics of the statistical methods, the Tukey's test is the best method. The Tukey's test has 0.001 type I error from detecting black and blue printing faults and the type II error is 0. The Tukey's test has 0.004 type I error and 0.052 type II error from detecting red printing faults. Generally, the detection effects of the multiple comparison methods are better than those of other methods based on texture features.
|