跳到主要內容

臺灣博碩士論文加值系統

(44.201.99.222) 您好!臺灣時間:2022/12/05 22:56
字體大小: 字級放大   字級縮小   預設字形  
回查詢結果 :::

詳目顯示

我願授權國圖
: 
twitterline
研究生:彭銘德
論文名稱:使用影像投影法與類神經網路於工件辨識之研究
論文名稱(外文):Part Recognition Using Image Projection and A Neural Network
指導教授:侯東旭侯東旭引用關係
指導教授(外文):Hou, Tung-Hsu
學位類別:碩士
校院名稱:國立雲林科技大學
系所名稱:工業工程與管理技術研究所
學門:工程學門
學類:工業工程學類
論文種類:學術論文
論文出版年:1997
畢業學年度:85
語文別:中文
論文頁數:74
中文關鍵詞:工件辨識影像投影法類神經網路
外文關鍵詞:part RecognitionImage ProjectionNeural Network
相關次數:
  • 被引用被引用:1
  • 點閱點閱:579
  • 評分評分:
  • 下載下載:0
  • 收藏至我的研究室書目清單書目收藏:1
  電腦現覺已廣泛的應用在工件自動辨識與自動化檢驗系統中。本研究提出一套結合影像投影法與類神經網路的工件辨識系統。在此系統中,先使用影像投影法來獲得工件之影像投影分佈圖,為使本系統能辨識各種方位的工件,本研究中提出一套快速有效的旋轉正規化方法。於完成投影後,再萃取此投影分佈圖之特徵,最後結合倒傳遞類神經網路,而形成一個工件自動辨識系統。此系統不受工件大小、位移及旋轉之影響,且有極佳之辨識能力。
  為驗證本系統之辨識能力,本研究中做了一項可行性評估,並將此系統與文獻上之簽字動差法做比較。可行性評估發現,類神經網路輸出層使用直接編碼方式之辨識率較使用二元碼編碼方式為高,故使用類神經網路於工件辨識時,應使用直接編碼方式。本研究最後透過兩個研究個案,比較投影法與文獻上之簦字動差法的優劣,實驗結果發現,投影法的辨識正確率可高達98.8%,較簽字動差法為佳,而辨識時間僅較簽字動差法慢0.31秒左右。
  Computer visions have been widly applied in parts recognition and automated inspection. In this research, a new part recognition system is developed. the system is built, basically, by combining image projection and back propagation neural network. By applying image projection procedure on a part at the X and Y direction, we can acquire projection histograms which can represent the shape of the test part. By combining the features extracted from the projection histogram and the back-propagation neural network, the new part recognition system is developed. In this research, a new orientation normalization scheme for image projection is also developed. Therefore, the system is invariant to translation, rotation, and scaling variations.
  In order to demonstrate the feasibility of the proposed system, a feasibility study is conducted and the proposed system is compared with a traditional system. The feasibility study reveals that the back-propagation neural network with the direct output coding yields the better recognition accuracy than the network with the binary output coding. Therefore the direct output coding should be used in the back-propagation neural network for the proposed part recognition system. It addition, two cases are used to compared the proposed system with the signature modeling system. The comparison reveals that the proposed system has higher recognition accuracy than the signature modeling system, although its processing speed is 0.31 second slower than that of the signature modeling system. Therefore, the proposed system is feasible for used in part recognition applications.
QRCODE
 
 
 
 
 
                                                                                                                                                                                                                                                                                                                                                                                                               
第一頁 上一頁 下一頁 最後一頁 top