整合模糊類神經方法於分類器及影像分割處理之應用

Abstract

本論文分別對於分類器的設計及影像分割處理兩種應用提出整合型模糊類神經網路解決方法。首先，我們提出一模糊認知學習法則及其應用於可處理數值資料及由模糊若則法則來表示的專家知識的分類器。我們擴展傳統的線性認知網路成二階的認知網路以提供更大彈性的判別函數。為了使類神經網路能夠處理語文變數，模糊集合的水準集合被引進至認知學習法中，在輸出模糊函數的不同水準下，根據輸入模糊函數及其對應之非模糊目標直，可以導出模糊認知學習演算法則。此外，模糊認知學習法則可利用修正的口袋演算法，即模糊口袋演算法，來解決如集合重疊等無法完全分開的分類問題。模擬結果證明模糊認知學習演算法的效果。 針對物體擷取，本文提出一個以模糊邏輯為基礎的改良式單層認知影像分割網路。我們選取一個S型灰階轉換函數，並配合輸入影像的灰階統計圖，將該影像圖素的灰階值映設到[0,1]區間；接著我們採用輸出神經元之輸出值的線性模糊指標，作為影像分割系統的誤差函數，以發揮神經網路的學習能力。所提出的系統能成功地將物體擷取出來。為了更進一步提昇分割網路的能力，本系統再加入模糊若則法則，以適應性調整輸出神經元活化函數的門檻值，期能達到對影像局部特性做最佳的配合；我們依據圖素的邊界強度與垂直位置的模糊若則法則，適應地推論此門檻值。經由比較紅外線影像的模擬結果，所提出之改良式單層認知分割網路，在加入模糊若則法則後，可得到更佳的分割結果。 此篇研究結果顯示，將人類知識以模糊法則的方式融入所設計之數值演算法中，對結果有進一步的提昇，且應可利用於分類或影像分割以外的問題。

This thesis presents two new integrated fuzzy neural approaches for classifier design and image segmentation processing. First, we propose a novel learning algorithm of Fuzzy Perceptron Neural Networks (FPNN) for classifiers that utilize expert knowledge represented by fuzzy if-then rules as well as numerical data as inputs. The conventional linear perceptron network is extended to a second-order one, which is much more flexible for defining a discriminant function. In order to handle fuzzy numbers in neural networks, level sets of fuzzy input vectors are incorporated into perceptron neural learning. At different levels of the input fuzzy numbers, updating the weight vector depends on the minimum of the output of the fuzzy perceptron neural network and the corresponding nonfuzzy target output that indicates the correct class of the fuzzy input vector. This minimum is computed efficiently by employing the modified vertex method to lessen the computational load and the training time required. Moreover, the pocket algorithm is introduced into our fuzzy perceptron learning scheme, called fuzzy pocket algorithm, to solve the nonseparable problems. Simulation results demonstrate the effectiveness of the proposed FPNN model. Second, in this thesis, we propose a fuzzy-logic-based modified single-layer perceptron (MSLP) image segmentation network for object extraction. We select a sigmoid gray level transfer function with the aid of the input image histogram and map the input gray levels into the interval [0,1]. Then we adopt the linear index of fuzziness of the output nodes as the error function of the image segmentation system to incorporate the learning capability of a neural network. Our scheme can successfully extract objects from the background. To further enhance the capability of the segmentation system, the proposed network is incorporated with fuzzy if-then rules to adaptively adjust the threshold of the activation function of the MSLP output neuron for best matching the local characteristics of the image. Fuzzy if-then rules involving the edge intensities and vertical positions of pixels are reasoned to determine the threshold adaptively. From the results of segmenting forward looking infrared (FLIR) images, better segmentation images have been obtained by incorporating fuzzy if-then rules with the MSLP segmentation technique. As demonstrated in this study, it is promising and worthwhile to incorporate human knowledge in terms of fuzzy logic into a designed numerical algorithm, which can further improve the performance, not just for the classification or segmentation problem we have presented. 1.1 Overview 1.2 The Neuro-Fuzzy Approaches for Solving Problems 1.2.1 Fuzzy Perceptron Learning and Its application to Classifier 1.2.2 A Fuzzy Logic-Based Neural Network for Image Segmentation Problem 1.3 Thesis Outline 2. LITERATURE SURVEY 3. THE FUZZY PERCEPTRON NEURAL NETWORK 3.1 Fuzzy Function and The Extension Principle 3.2 Structure of the Fuzzy Perceptron Neural Network 3.3 Fuzzy Perceptron Learning by the Modified Vertex Method 3.4 Fuzzy Pocket Algorithm 3.5 Multiclass Classification 3.6 Simulation 3.6.1 Simulation 1 3.6.2 Simulation 2 4. APPLYING FUZZY LOGIC IN THE MODIFIED SINGLE-LAYER PERCEPTRON IMAGE SEGMENTATION NETWORK 4.1 Single-layer Perceptron Net and Delta Learning Rule 4.2 The Modified Single-layer Perceptron Segmentation Network for Object Extraction 4.2.1 The MSLP Network Architecture 4.2.2 Sigmoid Gray Level Transfer Function of the Input Image Pixels 4.2.3 The Learning Procedures of the MSLP Segmentation Network 4.2.4 Simulation Results 4.3 Incorporating Fuzzy If-then Rules With The Modified Single-layer Perceptron Network 4.3.1 Fuzzy Reasoning in Adjusting the Threshold of Output Neurons 4.3.2 Simulation Results 5. CONCLUSION