臺灣博碩士論文加值系統

　　此研究是運用影像處理的技巧配合統計分析方式，計算棉網板上的棉粒個數及其大小分佈。在預備實驗中，利用照像機隨機取樣200個棉粒樣本，利用一HP型的桌上型掃描器補捉棉粒的光學照片影像，經影像二值化，影像遮罩及棉粒大小計算分析後，計算出棉粒的大小，再根據棉粒大小分佈，定義出棉粒的大小等級。本系統的操作過程則是將梳棉條（或精梳條）6~12條經羅拉牽伸裝置，製成一均勻的棉網板，其中取樣的大小寫l0xl0(in2)。本研究另外應用類神經網路 (Neural Network) 的技術，來辨認棉網中的棉粒、棉殼與含雜。
　　研究結果顯示棉粒的外觀為不規則形；棉粒的大小分佈介於0.1296mm2~3.9872mm2 之間。此研究中吾人將棉粒大小區分為8個等級，分別為0.045mm2~0.08mm2，0.08mm2~0.16mm2、0.18mm2~0.32mm2、0.32mm2~0.50mm2、0.50mm2~0.98mm2、0.98mm2~1.81mm2、1.81mm2~3.61mm2、3.65mm2~7.29MM2，其佔有率分別為：0%，9%，17%，21%，31%，17%，4.5%，0.5%。對棉粒個數的檢出結果而言，本系統與 ASTM 視覺檢驗方法及 USTER AFIS-N 的平均偏差率分別為17%及32%，且單位重量總個數本系統檢測值皆略大於 AFIS-N的檢測值，而略小於 ASTM 的檢測值。對於棉粒與含雜的辨認而言，皆可獲得96%以上的正確辨識率。

　　The nep counts and its distribution in web were studied by applying the image processing technique with the statistical method. Two hundred samples of nep were caught with a HP scanner. After that, the nep's images were processed with digital image processing technique including image thresholding and masking so that the nep counts and its distribution can be calculated. In this study, we also define a new calibration for the level of nep size according to the distribution from those 200 nep samples. A web formation is carried out with a roller drafting device. Six to twelve carded slivers or combed slivers can be available. A size of 10x10(in2) of the web is used as the testing sample. The artificial neural network technique was applied in this study to order to identify the trash and fibers from the neps.
　　The experimental results show that the neps have irregular shapes in their appearances. The nep sizes are ranging from 0.1296~3.9872 mm2. The nep sizes are divided into 8 levels which are 0.045~0.08 mm2, 0.08~0.16 mm2, 0.18~0.32 mm2, 0.32~0.50 mm2, 0.50~0.98 mm2, 0.98~1.81 mm2, 1.81~3.61 mm2, 3.65~7.29 mm2, respectively. Their contents in percentage are 0, 21%, 31%, 17%, 4.5% and 0.5%, respectively. The mean deviation of nep counts between our system and the ASTM test method is 17 % , and that between our system and the USTER AFIS-N is 32% in comparison with the test results.
　　Our system has more total amount of nep counts in unit weight than the USTER AFIS-N, while it has less than amount of nep counts in unit weight than ASTM test method. Generally, the identification extent can reach 96 % for the identification of trash and fibers from the neps.