English  |  正體中文  |  简体中文  |  全文筆數/總筆數 : 64745/64745 (100%)
造訪人次 : 20457494      線上人數 : 318
RC Version 7.0 © Powered By DSPACE, MIT. Enhanced by NTU Library IR team.
搜尋範圍 查詢小技巧:
  • 您可在西文檢索詞彙前後加上"雙引號",以獲取較精準的檢索結果
  • 若欲以作者姓名搜尋,建議至進階搜尋限定作者欄位,可獲得較完整資料
  • 進階搜尋


    請使用永久網址來引用或連結此文件: http://ir.lib.ncu.edu.tw/handle/987654321/1810


    題名: 特徵選擇與決策理論在紗線分級之研究;Combining Feature Selection with Decision Theory for Yarn Grading
    作者: 連信仲;Hsin-Chung Lien
    貢獻者: 機械工程研究所
    關鍵詞: 特徵選擇;決策理論;紗線分級;Feature Selection;Decision Theory;Yarn Grading
    日期: 2003-06-24
    上傳時間: 2009-09-21 11:32:17 (UTC+8)
    出版者: 國立中央大學圖書館
    摘要: 紗線等級評估是市場評估紗線價值的一個重要指標,本文利用紗線特徵數據,經由特徵選擇、圖形識別及倒傳遞神經網路等理論進行紗線等級評估。實驗過程是藉由學習樣本之紗線強度差異(Single yarn strength cv%),100公尺重量變異(100m weight cv%),紗線的均勻度變異(Yarn evenness cv%),黑板棉粒數(black-board neps no/g),紗線斷裂強度(Single yarn break-strength gf/tex)及100公尺重量公差(100 m weight tolerance %)等六項紗線特徵,作為6個輸入之紗線屬性向量(property vectors, PVs),經屬性向量輸入至決策理論以評估紗線之客觀等級。 本文首先利用圖形識別理論尋求簡單有效之紗線等級評估系統。在紗線分級的學習過程中,經由Karhunen-Loeve (K-L) expansion方法,將紗線屬性向量轉換至單位正交主軸向量(principal axis vectors, PAVs)上,達成特徵選取與降階之目的,應用Bayes Classifier方法及最小距離方法,求取評估紗線等級之決策函數,並在不降低辨識率之前提下重複上述學習過程,尋找可辨識最低維度主軸向量之轉換矩陣及決策函數。測試過程中,決策函數藉主軸向量進行紗線等級評估,藉由實驗結果證實,本方法辨識向量數目將減少到只需一維主軸向量,即可進行快速而有效的紗線等級評估,達到簡化辨識系統之目的,並進而獲得公正客觀之紗線等級評估結果。 在倒傳遞神經網路結合圖形識別理論尋求精確之紗線等級評估系統。辨識結果顯示將紗線之六項屬性向量作為輸入向量,以倒傳遞神經網路進行紗線等級評估,可獲得客觀且精確之評估結果,為提升系統性能,本文更以倒傳遞神經網路結合K-L expansion方法進行紗線分級,經由實驗結果證實,K-L expansion方法將有效降低輸入向量之亂度與維度,使倒傳遞神經網路簡化至只需一維主軸向量作為辨識向量,即可達到辨識之效能,而當倒傳遞神經網路以四維主軸向量作為辨識向量時,則紗線等級評估系統將具有較佳之強健性及再學習能力,並可達到精確之評估結果。 最後本文藉由計算紗線族群有效距離(effective distance between clusters EDC)之方法作為特徵選擇( Feature Selection )之依據,以達到提供明顯而有效之紗線分級特徵,並進而提升紗線分級之作業效益。實驗結果顯示EDC方法對主軸向量(principal axis vectors PAVs)作特徵選擇時,其平均特徵誤排數目及平均誤排總距離分別只有K-L expansion方法之33.3%及16.7%,而且EDC方法亦可直接應用於屬性向量之特徵選擇上,在不降低辨識精度下可減少PVs量測項目,達到相對於前文之紗線分級方法中,分別在計算PAVs之時間上及量測PVs之項目上同時提升16.7%之效益。 The present research investigates a simple and effective system for grading textile yarns by pattern recognition theory. During the learning processes of textile yarn grading, yarn property vectors (PVs) can be transferred to the principal axis vector (PAV) of the orthonormal function with Karhunen-Loeve (K-L) expansion, so as to select features and then reduce dimensions. Whereas, through the Bayes Classifier (BC) or the Minimum Distance (MD) method, the decision function for evaluating textile yarns grading can be obtained. Then under the premise of not sacrificing identification rate, the learning process can be repeated to search for the transfer matrix and the decision function with PAVs of the lowest identifiable dimension. The present system is shown to create a quick and effective grading system for textile yarns by using PAVs of only one dimension, thus both simplifying the identification system and providing objective grading results. The grade of textile yarns is an important index in evaluating the yarn’s market value. This paper uses the backpropagation neural network (BNN) and Karhunen-Loeve (K-L) expansion method to construct a new and highly accurate grading system. Outcomes show that a highly accurate and neutral grading system can be obtained if the BNN learning sample is comprehensive or by adopting BNN with relearning technique (Self-Healing). Considering the possibility of reducing the dimension of BNN input vectors without losing the accuracy, this paper preprocesses the BNN grading system using the K-L expansion. Experiments demonstrate that the K-L expansion provides a way to reduce the input dimensions, and that a single principle axis value of BNN with K-L expansion grading system is able to grade textile yarns. In addition, the experiment demonstrates that as the input dimensions are reduced to four in a self-healing neural network with K-L expansion, the grading system provides the high accuracy and robustness. By using the effective distance between clusters (EDC) as the basis for feature selection, this paper achieves a significant and effective feature for textile yarn grading, and further upgrades the operational efficiency of such grading. The results, such as the feature selection processing to principal axis vectors (PAVs) by EDC, show that The features’ average number and average total distance of mistaken ranking by EDC were only 33.3% and 16.7% of those by the Karhunen-Loeve (K-L) expansion, respectively. Furthermore, EDC can be applied directly to the feature selection of property vectors (PVs) and reduce the measured items of PVs without lowering the identification precision. Compared with the textile yarn grading proposed earlier, EDC provides 16.7% greater efficiency both in measuring PVs and in calculating PAV1 time.
    顯示於類別:[機械工程研究所] 博碩士論文

    文件中的檔案:

    檔案 大小格式瀏覽次數
    0KbUnknown535檢視/開啟


    在NCUIR中所有的資料項目都受到原著作權保護.

    社群 sharing

    ::: Copyright National Central University. | 國立中央大學圖書館版權所有 | 收藏本站 | 設為首頁 | 最佳瀏覽畫面: 1024*768 | 建站日期:8-24-2009 :::
    DSpace Software Copyright © 2002-2004  MIT &  Hewlett-Packard  /   Enhanced by   NTU Library IR team Copyright ©   - 回饋  - 隱私權政策聲明