Recognition of odor mixture using fuzzy-LVQ neural networks with matrix similarity analysis

An artificial odor recognition system has been developed recently. However, recognizing the odor mixture is rather difficult by the use of a limited number of sensors. We have constructed an artificial odor recognition system based on 16 sensors of 20 MHz quartz resonators. Various neural systems, i.e. backpropagation neural system, probabilistic neural system and fuzzy-LVQ, are then studied and applied as the neural classifier of the developed system. Results of experiments confirmed that the F-LVQ shows higher recognition rate compared with that of two other neural systems. Improving the F-LVQ is then conducted by incorporating the matrix similarity analysis to form FLVQ-MSA, showing the highest average recognition rate of 80% on determining three-mixture odors.