Skip to Main Content
An electronic odor discrimination system had been developed by using four quartz resonator-sensitive membranes basic-resonance frequencies at 10 MHz as a sensor and analyzed the measurement data through a back propagation (BP) as the pattern recognition system. The developed system showed high recognition probability to discriminate various single odors to its high generality properties; however, the system had a limitation in recognizing the fragrances mixture. This system also had other disadvantages, such as classifying the unknown category of odor as the known category of odor. In order to improve the performance of the proposed system, development of the sensor and other neural networks (NNs) are being sought. This paper explains the improvement of the capability of that system. In this experiment, the improvement is conducted not only by replacing the last hardware system from four quartz resonator-basic resonance frequencies at 10 MHz with new 16 quartz resonator-basic resonance frequencies at 20 MHz, but also by replacing the pattern classifier from BP NNs with the variance of BP, probabilistic NNs, and fuzzy-neuro learning vector quantization (FNLVQ). Matrix similarity analysis (MSA) is then proposed to increase the accuracy of the FNLVQ, to become FNLVQ-MSA neural systems in determining the best exemplar vector, for speeding up its convergence. The purpose of the recent study is to construct a new artificial odor discrimination system for recognizing the fragrance mixtures, in addition to recognizing the unknown fragrance mixtures. The use of new sensing systems and FNLVQ-MSA has produced higher capability, compared to the previously mentioned system.