Application of Image Analysis and Machine Learning on the Fabric Pilling Evaluation

• Main Authors: Hau-Chiun Yang, 楊浩群
• Other Authors: Mei-Ling Huang
• Format: Others
• Language: zh-TW
• Published: 2015
• Online Access: http://ndltd.ncl.edu.tw/handle/52352386308954469306

碩士 === 國立勤益科技大學 === 工業工程與管理系 === 103 === Due to the lack of raw materials in early Taiwan, raw materials of textile industry were mainly depended on import. During the mid-term, petrochemical industries have refined the raw materials of textile to produce nylon and other fabrics. However, causing by unventilated and airtight, industries nowadays keep on developing fabrics with moisture absorbent, sweat relieving, and other innovative functions. Under the variety of requirements in functional fabrics of the international market, the industries in Taiwan have been actively and continuously developing new products and renewing manufacturing equipment to extend their international market. With the rise of the environmental awareness, industries have combined environmental friendly concepts to form artificial functional fabrics with coffee grounds or splintered recycling bottles. After that, they developed a complete production system coupled with imported cotton, wool and other natural fiber supplements. Textile industry in Taiwan, therefore, has become the top international functional textiles. Before shipment, innovative functional fabrics and traditional fabrics must pass through ’SGS international standard testing’ for carried wear testing and followed by the visual classification. Although Taiwan has the innovative technology and new equipment to enhance the added value of functional fabrics, unfortunately, the testing and classification was based on a traditional manual method. Not only the classification standards stayed unimproved, but also the cases misclassification temporarily occurred. Therefore, this study used image processing for image reconstruction. The difference of the choice of the image was first judged by the method of objective image quality evaluation. Next, a database of wear wool fabric ball parameters was created according to the wear resistant number of cases. Then, applying objective machine learning on fabric classification. In addition, the usage-rate differences of wear wool fabric ball parameters database was established to calculate the value of parameter boundaries among different fabric grades. At last, explore the possibility of the classification criteria established by this method.