Signal Processing using Wavelets for Enhancing Electronic Nose Performance
No Thumbnail Available
Files
Date
2007-07-19
Authors
Journal Title
Series/Report No.
Journal ISSN
Volume Title
Publisher
Abstract
In recent years, many new technologies of electronic devices that mimic the mammalian olfactory system, electronic noses (e-noses), have been developed in many research institutions and commercial organizations around the world. These devices have been used in a wide range of applications such as food and beverage quality, environmental monitoring, medical diagnosis. Over the past decade, many researchers have spent a great deal of effort improving e-nose performance and also extended the use of the e-nose devices, not only for discriminating or classifying different odor samples, but also for quantifying an ingredient of a given odor sample.
This dissertation focuses on two technical areas. First, an implementation of an e-nose signal processing system is developed to improve classification performance for small portable devices with fast response times and reduced cost. Second, the signal processing system is extended to odor mixture analysis. The advances made this research are based on a modern signal processing technique, specifically wavelet analysis. Ultimately, the performance of e-nose devices is highly dependent on the quality of features from the sensors' response. Therefore, a new transient feature extraction method using wavelet decomposition to capture the transient sensor's response has been developed. The evaluation of these transient features shows promising results in terms of classification performance, number of sensors employed in the e-nose device, and simplification of the classifier. For handling different types of odor samples, a simplified multiple classifier system is developed based on an "odor type signature." Analyzing mixtures of odors is a challenge for e-nose systems. Herein a new method is developed for predicting a sensor's response to mixtures of odors. The combination of wavelet decomposition and reconstruction is adopted to implement the mixed odor sensor-response predictor.
Description
Keywords
wavelet reconstruction, wavelet decomposition, transient-feature extraction, sensor selection, machine olfaction, odor-type signature, Odor classification
Citation
Degree
PhD
Discipline
Electrical Engineering
