Pattern Recognition Methods For Electronic-nose Based On Feedforward Neural Networks

Electronic-nose is an electronic olfaction system which can recognize the odour. Pattern recognition plays an important role in the electronic-nose technology. Artificial neural network (ANN) has the advantage than other traditional methods for pattern recognition, because of its strong fault-tolerance, good nonlinear mapping ability and self-adaptive learning ability. Recently, the artificial neural network has been one of the main methods for electronic-nose pattern recognition.This paper contains two primary parts:Firstly, a new method of eliminating abnormal samples by using neural network is proposed. In order to show the proposed method's effectiveness and feasibility, BP neural network and double parallel feedforward neural networks are set up to find the disturbed samples existed in the samples set. Experiment results show this method having high identification rate.The second part mainly study on the pattern recognition methods for electronic nose based on feedforward neural network. Besides the usual feedforward neural networks such as BP neural network, higher-order neural network and double parallel feedforward neural network, a modified locally connected higher-order neural network based on Sigma-Sigma-Pi neural network is presented. They are used to forecast the gaseous density according to sensors array's output signal. The batch gradient algorithm is used to train the four neural networks. By comparing network cost and generalization ability of these neural networks, the best neural network model for pattern recognition is found out.