Neuro-fuzzy pattern recognition (Q2703392)

This collection of articles contains recent works on methodology and applications of fuzzy Neural Networks (NNs) for pattern recognition. Chapter 1 proposes a five-layer fuzzy NN implementing a fuzzy rule-based inferencing system. The NN structure and the training mechanism allow for simultaneous feature selection and system identification. Feature extraction for unsupervised classification through a NN model is described in Chapter 2. The procedure involves minimization of a fuzzy feature evaluation index, and is compared with a principal component analysis network. NEWLINENEWLINENEWLINEChapter 3 analyzes the current trends in computational intelligence (taken as a synonym of soft computing) contrasted to artificial intelligence, and highlights the need for synergetic models between NNs, fuzzy logic and evolutionary computation. A neuro-fuzzy-genetic classifier is proposed in this line. Chapter 4 presents clustering through variants of fuzzy c-means, fuzzy competitive learning networks, modified Hopfield nets and chaotic neural networks. The performance of the algorithms is demonstrated on synthetic and real images. Chapter 5 discusses the fuzzy min-max classifier, its improved version called adaptive resolution classifier, and variants thereof. A generalized min-max model is advocated, and a training procedure is suggested.NEWLINENEWLINENEWLINEChapter 6 explains granular computing and demonstrates how to characterize information granules in terms of their size and variability. The authors suggest a neural classifier for granular patterns. Fuzzy Adaptive Resonance Theory models, called FasART are presented in Chapter 7, together with their applications to document management and handwritten character recognition. Interpretability of the extracted rule bases is also discussed. The application part of the book commences with a study on speech recognition in a noisy environment using adaptive noise cancellation and evolving fuzzy neural networks.NEWLINENEWLINENEWLINEChapter 9 tackles the problem of land mine detection by applying Choquet fuzzy integral for information fusion. Different fuzzy measures for calculating the fuzzy integral are presented and compared with a heuristically estimated measure. A partly supervised approach to segmentation of multi-spectra MR brain images is proposed in Chapter 10. At the first stage, a fuzzy clustering algorithm is employed, and at the second stage, a feed-forward NN delivers the exact segmentation that can be useful during surgery. NEWLINENEWLINENEWLINEChapter 11 describes a neuro-fuzzy control system for autonomous vehicle lane following. The design is aimed at a low cost implementation without degrading the system's performance.