Improved Several Competition Neural Networks And Their Application For Patten Classification

Neural network learning algorithm consists of two ways, one is supervised that using a sample of a given standard classification or simulate; the other is unsupervised that provides only learn manner or certain rules and specific learning content very with environment where system lie (namely the input signal situation), then system can automatically find environmental characteristics and laws and has more akin to the function of the human brain. Compared to supervised learning, unsupervised learning of a late start has greater space for its research. This paper improves Fuzzy Adaptive Resonance Theory (Fuzzy ART) Neural Network and lastly proposes a new unsupervised classification method. The main contributions of this thesis are given as follows:(1) By analyzing several classical neural network models, we understand the future trend and direction of neural network, and mainly study the models of both SOM Neural Network and Fuzzy ART.(2) We deeply study the model structure and algorithm theory of the SOM Neural Network and validate its performance by experiments of Iris, Wine, Remote sensing data and gray image classification.(3) After deeply studying Fuzzy ART Neural Network, we improve its learning rate and further reduce the complexity of the calculation. Experimental results for Iris data, Remote sensing data and gray image verify the validity of our new method.(4) We propose an unsupervised pattern recognition method of combining SOM with Fuzzy ART and the advantages of new method are shown by several experiments.