Optimum Pattern Recognition Method for the Electronic Nose System

An electronic nose (EN) is an artificial olfactory system that tries to perform the same task as human olfactory system and widely used in the gas analysis field. EN consists of an array of chemical sensors possessing broad specificity, coupled to electronics and software that allow feature extraction – extraction of salient data for further analysis, together with pattern recognition – identification of sample odour. Here, we present a neural architecture based on several self-organizing maps that counteract the parameter drift problem for an array of chemical gas sensors. The neural architecture named mSom, since m is the number of odours to be recognized and is mainly constituted of m maps, each one of them approximate the statistical distribution of a given odour. Competition occurs both in each single map and between maps for the selection of the minimum map distance in the Euclidean/Mahalanobis distance space. Fuzzy-Cmeans has been used to find three optimum centres for the training data and then initializing SOM weights (system was called mSomFuzzy) with these centres instead of random weights. The system is called mSomFuzzyMaha, while the Mahalanobis distance is being used as a distance metric. The main aim of the research work is to investigate the optimum pattern recognition method for the electronic nose system under normal conditions subjected to drift, no matter how different the concentrations. Although chemical patterns from the sensor array should be the same for a particular sample, the actual responses are affected by many factors such as temperature, humidity, and sensor drift.