標題: 混合式類神經網路之研究
A Study of Hybrid Neural Network
作者: 劉榮智
Jung Chih Liu
許根玉;謝太炯
Ken Y. Hsu;Tai Chiung Hsieh
光電工程學系
關鍵字: 類神經網路;楔形-還形取樣;neural network;wedge-ring sampling
公開日期: 1993
摘要: 在本論文中我們探討一種混合式類神經網路。它是由一個光學前置處理器 和一種類神經網路模型所組合而成的圖像辨識系統。類神經網路的基本架 構是由兩層的感知器網路所構成,而其學習法則係選用反向錯誤傳播法則 來作為網路的訓練。光學前置處理器主要包含一組富氏轉換透鏡和一個取 樣CCD偵測器。CCD的取樣方式係模擬楔形-環形取樣原理,而富氏轉換透 鏡的作用則是用來取得輸入圖像的功率頻譜。由於這種光學前置處理器具 有可程式規劃的能力,因而能適合於各種圖像的取樣。在實驗中輸入圖像 的特徵經由光學前置處理器的取樣,並透過類神網路的訓練可得到一組適 當的網路權值,以作為網路的分類依據。實驗的結果顯示出此系統具有位 移不變、大小不變、或旋轉不變的圖像分類特性。最後,我們將此系統應 用於指紋的分類,其結果顯示本系統能對數個不同的拇指指紋做出正確的 分類。 The thesis investigates an opto-electronic hybrid neural network for image classification. The system consists of an optical pre-processor and a neural network model. A two-layer perceptron is used to construct the neural network which uses the back-error propagation algorithm for training. The optical pre-processor combines a Fourier transform lens and a sampling CCD detector. The Fourier transform lens is used for obtaining the power spectrum of the input patterns. The CCD detector is configured to simulate the wedge-ring detector for sampling the power spectrum. In experiment, the sampled data are used as the input to the neural network for training. After training the system is used for pattern classification. Thumbprints are used for the classification experiment. The experimental results how that the system is invariant when the input pattern isifted, scaled, or rotated.
URI: http://140.113.39.130/cdrfb3/record/nctu/#NT820123008
http://hdl.handle.net/11536/57638
Appears in Collections:Thesis