An introductory concept to 'Human like' odor descriptions by grounding linguistic symbols

Abstract

When humans fail to recognize a new odor we rely on our previous knowledge of known odors to describe linguistically the attributes, quality or even the experience surrounding the detected smell. Meanwhile, for an electronic nose a new odor is encoded in the form of numerical patterns and is recognized by signal processing and pattern recognition techniques. In this paper we present a system that is capable of interpreting the results from an electronic nose by using linguistic terms to generate a description of the sample. We have shown that unknown odors can be described by using combinations of known odor terms. Furthermore, we show that the odor of a substance can be used to establish the quality of that substance by generating an appropriate description. To obtain the sensor results for linguistic processing, two biologically inspired designs of an electronic nose were tested. The specifications of both these electronic noses are presented. A large part of the system involves a combination of different data processing techniques such Principal Component Analysis, Artificial Neural Networks and Fuzzy classification. It was found that the success of the data processing improved when several techniques were used in combination rather than using one technique alone. The novelty of this work is in the final phase of the system that generates a multi-part linguistic description associating sensor results to a more human like representation.