Intelligent Systems for Machine Olfaction: Tools and Methodologies

Intelligent Systems for Machine Olfaction: Tools and Methodologies

Evor L. Hines (University of Warwick, UK) and Mark S. Leeson (University of Warwick, UK)
Indexed In: SCOPUS
Release Date: March, 2011|Copyright: © 2011 |Pages: 354
ISBN13: 9781615209156|ISBN10: 1615209158|EISBN13: 9781615209163|DOI: 10.4018/978-1-61520-915-6


Intelligent systems are systems that, given some data, are able to learn from that data. This makes it possible for complex systems to be modeled and/or for performance to be predicted. In turn, intelligent systems’ functionality can be controlled through learning/training, without the need for a priori knowledge of their structure.

Intelligent Systems for Machine Olfaction: Tools and Methodologies introduces new, state-of-the art applications of intelligent systems to researchers and developers in the area of machine olfaction. Readers will benefit from in-depth analyses of fundamental theories, potential trends, and key literature in the field, making this work both a source of application examples that can be readily implemented and a practical guide for the implementation of solutions in other scenarios.

Topics Covered

The many academic areas covered in this publication include, but are not limited to:

  • Computational Intelligence
  • Evolutionary algorithms
  • Gas dispersal models
  • Gas distribution modeling
  • Gas source localization
  • Image content description techniques
  • Kernel methods
  • Mobile robot
  • Sensor selection
  • Teleolfaction

Reviews and Testimonials

Researchers in the Machine Olfaction community, as well as users wishing to know more about the underpinning technologies, will find this book to provide a useful update in the latest state of the art.

– Santiago Marco, Universitat de Barcelona and Institute for Bioengineering of Catalonia, Spain

Engineers discuss some of the fundamental and generic issues that underpin the application of intelligent systems to machine smelling, then present a series of specific applications to a range of olfaction tasks across industrial, medical, and horticultural topics.

– Book News, Reference - Research Book News - August 2011

Table of Contents and List of Contributors

Search this Book:


The subject of Intelligent Systems (IS) has expanded considerably since its beginnings in the 1940s, and this volume considers the application of IS to machine olfaction, an area that has itself undergone substantial growth in recent years. IS includes a broad range of complementary techniques that provide attractive solutions to many hard and nonlinear problems. Modern work in IS includes methods such as Artificial Neural Networks, fuzzy systems, evolutionary algorithms, support vector machines, particle swarm optimization, memetic algorithms, and ant colony optimization. In addition, hybrid combinations also play a significant role with inter alia, neuro-fuzzy, neuro-genetic, and fuzzy-genetic systems firmly established in the literature. IS methods offer particular advantages in the robust handling of the large datasets containing uncertainties which have become more common with the growth of modern data storage capacity. The ability of IS solutions to learn from the data, extracting patterns, and to explore very large, multi-modal solution spaces taking into account multiple, often conflicting objectives, has further consolidated their place in the modern optimization toolbox.

The first section of the book presents some of the fundamental and more generic issues that underpin the application if IS to machine olfaction. The second section presents a series of specific applications to a range of olfaction tasks across industrial, medical, and horticultural topics. 

Chapter one provides the reader with a thorough review of feature or sensor selection for machine olfaction. It covers the need for variable selection followed by a critical review of the different techniques employed for reducing dimensionality. Further, examples from the literature are used to illustrate the application of the various techniques machine olfaction followed by coverage of sensor selection and array optimization. In addition to conclusions, the chapter ends with a visionary look toward the future in terms of how the field may evolve.

In the second chapter, the combination of data from a number of sensors, known as Multisensor Data Fusion (MDF), is considered. The requirement is to provide comprehensive and reliable information, and MDF has expanded rapidly in tandem with improvements in computing power and the emergence of new sensors over the last 30 years. One of the key aims in MDF is the elimination of redundant, noisy, or irrelevant information to discover an optimal subset from an array of high dimensionality. Since the signals in MDF are constantly evolving, an opportunity is provided for Evolutionary Computation (EC) algorithms to assist in this aim. Here, the application of three EC algorithms to widely used datasets is described, demonstrating that ECs are of great utility in the MDF task. This leads nicely into the role of incremental learning considered in the next chapter. 

Machine olfaction presents a significant challenge for pattern recognition systems because it is relatively difficult to obtain quality training data, and the samples arrive in batches widely spaced in time. Furthermore, industrial users are generally reluctant to supply samples but would like instant results. These characteristics mean that suitable pattern recognition algorithms need to be flexible in that they must update an existing, stable, and plastic classifier without affecting the classification performance on old data. Incremental learning algorithms offer the required features, and in chapter three, a range of such algorithms for machine olfaction are reviewed.

The final chapter of this part (chapter four) is concerned with the causes of the errors and inconsistencies that contribute to the problems of MDF. Given the wide application of electronic noses (e-noses) in environmental monitoring, food production, and medicine, understanding the imperfections they display is crucial so that they may be compensated for. Thus, in this chapter, the impact of sensor interference and noise on measurement repeatability is considered. Probability density functions and power spectra of noise from real sensors are presented to deliver a pragmatic view of sensor imperfection effects on repeatability.

The use of electronic noses for odor recording in the atmosphere is challenging because of the turbulent airflows present. Chapter 5 addresses this challenge by utilizing visual information to enhance the sensitivity of an electronic nose by combining senses. The system also offers the prospect of teleolfaction, where users can sniff distant objects whilst watching video images of the objects in real time via the Internet. Thus, users can perceive much of the sensation of reality of the object even at the remote site. Although there have been several reports on the fusion of vision and olfaction in virtual environments, the system described is unique because it is able to record and reproduce olfactory information synchronized with visual information.

In the sixth chapter, the determination of the distribution gas is considered and an alternative sought to current computationally expensive physical models. In the chapter, kernel models are introduced that treat gas measurements as random variables, enabling the gas distribution predictions, but making no strong assumptions about the gas distribution’s functional form. In addition to two-dimensional models, the kernel density estimation algorithm is extended to three dimensions, with additional incorporation of wind information and time-dependent changes of the random process. The methods covered are discussed based on experimental validation using real sensor data.

Chapter 7 tackles the problem of a machine olfaction colorimetric sensor array description method for the visual enhancement of volatile organic components (VOCs). The solution presented utilizes structured light patterns, demonstrating the similarity between the colorimetric and the structured pattern interpretation. Once the calorimetric data has been transformed into a structured form, existing structured pattern description methods may be employed. Different methodologies for retrieving, combining, and selecting the most appropriate structured feature sets are presented, demonstrating increased classification rates.

The important topic of plant disease diagnosis, using the specific example of tomato plants, forms the basis of chapter 8. Since plants that are subject to infection typically release exclusive volatile organic compounds (VOCs), these may be detected by appropriate sensors. In this contribution, an electronic nose (EN) is used to sample the VOCs emitted by control plants and plants artificially infected with powdery mildew and spider mites. The EN data were analyzed using Fuzzy C-Mean Clustering and Self-Organizing Maps with results that indicate that this is a promising automated crop pests and disease detection method.

The ninth chapter continues the theme of the previous one by again examining powdery mildew and spider mites on tomatoes using and electronic nose (e-nose), but this time in a greenhouse setting. A commercial e-nose was used to collect data from tomato plants grown in an isolated controlled greenhouse environment. Principal Component Analysis and Grey System Theory are utilized to analyze the data, producing noticeable groupings in the sensor responses between healthy and infected plants. The results show that the approach is potentially a highly effective in greenhouse tomato plant health monitoring.

Chapter 10 returns to the area of gas sources, in this case, the location of a single gas source from a set of localized gas sensor measurements. Nonlinear least squares fitting is applied to a two dimensional gas distribution grid map and the parameters learned using Evolution Strategies (ES), a special type of Evolutionary Algorithm (EA). By considering the best fit to the statistical gas distribution map, an improved estimate of the gas source position is derived. A comparison of methods to ascertain the true source position in made based on gas distribution mapping experiments with a mobile robot. 

The final chapter addresses a topic from medicine, namely the classification of eye bacteria. Human eyes are constantly exposed to airborne organisms, with those that cause eye infection able to proliferate rapidly. There is thus a requirement for fast and accurate classification of eye bacteria that may be achieved by smelling the distinctive and specific characteristic odors arising in different diseases using an e-nose. The focus of the work in this contribution is the enhancement of classification by introducing the bagging technique to both multi-layer perceptron (MLP) and decision tree (DT) classification. In the case of the former, the accuracy significantly improved, and in the latter, the misclassification rate is reduced. Using bagging, the dimension of the data is reduced to enhance training and testing.
These chapters give an indication of wide ranging areas of machine olfaction, in which IS methods have found application, and they present a representative selection of the available approaches. The IS field continues to evolve rapidly with more data constantly available, coupled with new hardware and software. This collection will be particularly useful as a reference for graduate students and researchers in engineering, computer science, system sciences, and Information Technology, as well as or practitioners in a range of industries. It collects together some of the major techniques that are available to make sense of the wealth of data that may be obtained with modern machine olfaction technology.

Evor L. Hines
University of Warwick, UK

Mark S. Leeson
University of Warwick, UK

Author(s)/Editor(s) Biography

Evor L. Hines joined the School of Engineering at Warwick in 1984. He was promoted to Reader in 2005 and to a personal chair in 2009. He obtained his DSc (Warwick) in 2007 and is a Fellow of both the Institute of Engineering and Technology and the Higher Education Academy, in addition to being a Chartered Engineer. His main research interest is concerned with intelligent systems and their applications. Most of the work has focused on artificial neural networks, genetic algorithms, fuzzy logic, neurofuzzy systems and genetic programming. Typical application areas include, inter alia, intelligent sensors such as the electronic nose, medicine, non-destructive testing, computer vision, and telecommunications. He has co-authored in excess of 230 articles and supervised over 30 research students in addition to currently leading the Information and Communication Technologies Research Group in the School of Engineering.
Mark S. Leeson received the degrees of BSc and BEng with First Class Honors in Electrical and Electronic Engineering from the University of Nottingham, UK, in 1986. He then obtained a PhD in Engineering from the University of Cambridge, UK, in 1990. From 1990 to 1992 he worked as a Network Analyst for National Westminster Bank in London. After holding academic posts in London and Manchester, in 2000 he joined the School of Engineering at Warwick, where he is now an Associate Professor. His major research interests are coding and modulation, ad hoc networking, optical communication systems and evolutionary optimization. To date, Dr. Leeson has over 180 publications and has supervised nine successful research students. He is a Senior Member of the IEEE, a Chartered Member of the UK Institute of Physics, and a Fellow of the UK Higher Education Academy.