| Summary: | 碩士 === 國立臺北科技大學 === 有機高分子研究所 === 90 === The inspection of fabric defects is the most important process for ensuring the quality of final products. Traditionally, manual inspections are not only time wasted but also labour intensive. To solve these problems, lots of researches apply updated computer visual technologies to the “Automatic Inspection of Fabric Defects System”. However, most of those researches focus on the inspection of woven fabrics or nonwoven fabrics. The researches and applications for the inspections of lace fabrics, jacquard fabrics, print fabrics or others are rare. Therefore, this study concentrates on improving the classification and recognizing accuracy of lace fabric defects by employing an image processing technology with an artificial neural network theory to solve the defect-recognition difficulty of lace fabrics caused by a slightly shift or a rotation of fabrics during a dynamic inspection process.
The approach of this study is to compare the defect-recognition results between theoretical lace fabric images and real lace fabric images. Firstly, Area-Scan CCD camera is applied to acquire theoretical and real images of lace fabrics defects. Secondly, the theoretical image of lace fabric is then shifted and rotated by the computer to simulate the variances of lace fabric images generated by a dynamic acquisition process. Then, the binary technology is employed to distinguish the differences between the images of lace fabrics and background. In order to tell the feature values of lace fabrics defects, this study utilizes the image processing analysis technologies of “moment invariants” and “total black pixel”. All the feature values of defect images are consisted into feature vectors as the input vectors of a neural network to classify the defects. Finally, the real images of lace fabrics can be acquired to verify the system. From the results, it tells that the accuracy of theoretical defect images classification of lace fabrics is 100% but the real defect images classification of lace fabrics is 93.33%. Therefore, it can be suggested that this application of acquiring features values of defect images is suitable for the “Automatic Inspection of Lace Fabric Defects System” to improve product quality and reduce the cost of lace fabric processing and inspection.
|
|---|
