臺灣博碩士論文加值系統

織物之外表特徵為織物的重要性質之一，在傳統上，一般織物之檢測方式仍使用人工方式來檢測，然而在長時間之工作狀態下以及人為主觀的檢測方式，往往造成其檢驗之結果不客觀，且檢驗者本身也有疲倦及傷害視力等問題產生。針對以上之缺點，本研究所利用之數位影像系統可用來代替人工作為檢驗織物之工具，且其效率也非一般人工方式所能比擬。 在本研究中，我們分別以動態和靜態之織布瑕疵辨識作比較，其流程大致可分為如下。首先，選擇欲檢測之胚布及瑕疵型式，本研究所選擇之胚布為單層平紋織布，而欲檢測之瑕疵為一般織布較容易發生之破洞、油污、缺經及缺緯等四種。而在影像處理部份，我們使用1 4096之高解析度Line-Scan camera作為動態織物影像之擷取，另外再以解析度512 512之Area-Scan camera作為靜態織物影像之擷取，而其中Area-Scan 數位照相機的應用主要是要與Line-Scan數位照相機作比較，之後再將兩者所擷取之影像傳送到電腦作處理分析，其分析過程內容包括有濾波、二值化等

Fabric appearance is one of the most important properties of fabrics. Traditionally, fabric detects are usually examined by observers, and the test methods via observers are so subjective that the accuracy can not be satisfactory enough. At the same time, the process is so tedious that observers usually cause an eyestrain and easily tired. According to the aforementioned considerations, a image recognition system instead of using human inspectors is applied and the detection efficiency is proved from the exIn the scheme, a static and a dynamic test of the fabric fault are represented, respectively. In the first phase, the fabric samples and the fault types are chosen. The samples are plain fabrics and the fault types are holes, oil stains, weft lacking and warp lacking. Second, the 1’4096 line-scan camera with a high resolution is used for grabbing the dynamic fabric image and the 512’512 area-scan camera is employed for grabbing the static fabric image. After the image is grabbed, it is passed to computer In the neural network, three nodes are adopted in the input layer. The maximum length, maximum width, and the gray-level are represented as the nodes of the input layer, respectively. The neural network system can be applied to an on-line fault classification after an off-line learning. Because BP can solve the nonlinear equation, a powerful tool for on-line fault classification can be shown.