Improved Gas Source Localization with a Mobile Robot by Learning Analytical Gas Dispersal Models from Statistical Gas Distribution Maps Using Evolutionary Algorithms

The method presented in this chapter computes an estimate of the location of a single gas source from a set of localized gas sensor measurements. The estimation process consists of three steps. First, a statistical model of the time-averaged gas distribution is estimated in the form of a two-dimensional grid map. In order to compute the gas distribution grid map the Kernel DM algorithm is applied, which carries out spatial integration by convolving localized sensor readings and modeling the information content of the point measurements with a Gaussian kernel. The statistical gas distribution grid map averages out the transitory effects of turbulence and converges to a representation of the time-averaged spatial distribution of a target gas. The second step is to learn the parameters of an analytical model of average gas distribution. Learning is achieved by nonlinear least squares fitting of the analytical model to the statistical gas distribution map using Evolution Strategies (ES), which are a special type of Evolutionary Algorithm (EA). This step provides an analysis of the statistical gas distribution map regarding the airflow conditions and an alternative estimate of the gas source location, i.e. the location predicted by the analytical model in addition to the location of the maximum in the statistical gas distribution map. In the third step, an improved estimate of the gas source position can then be derived by considering the maximum in the statistical gas distribution map, the best fit, as well as the corresponding fitness value. Different methods to select the most truthful estimate are introduced, and a comparison regarding their accuracy is presented, based on a total of 34 hours of gas distribution mapping experiments with a mobile robot. This chapter is an extended version of the conference paper (Lilienthal et al., 2005).

Odor Reproduction with Movie and Its Application to Teleolfaction

The authors of this chapter study the odor reproduction system synchronously with a movie. The system is made up of an odor sensing system and an olfactory display. The fruit flavors were recorded with movies using a digital video camera and the odor sensing system. The results of the sensory tests showed that the odor information recorded using the proposed method is appropriate for the smell regeneration associated with the movie. Next, the authors propose a tele-olfaction system synchronous with visual information. The olfactory display system was placed remotely from the odor sensing system, and both of them were connected via Internet. In addition to the olfactory system, a Web camera captures image around the sniffing point and that image appears at the computer display connected to the olfactory display at remote site. Moreover, the mobile stage with its sniffing point and the Web camera remotely controlled by a user was introduced so that he/she could interactively approach a smelling object. The questionnaire survey at the exhibition revealed that a user can enjoy smell synchronous with movie in real time even if he/she stays at the remote site.

Making an Electronic Nose Versatile

Because of these factors, it is necessary to make the system flexible in such a way that the system is able to update an existing classifier without affecting the classification performance on old data, and such classifiers should have the property as being both stable and plastic. Conventional pattern classification algorithms require the entire dataset during training, and thereby fail to meet the criteria of being plastic and stable at the same time. The incremental learning algorithms possess these features, and thus, the electronic nose systems become extremely versatile when equipped with these classifiers. In this chapter, the authors describe different incremental learning algorithms for machine olfaction.

Tomato Plant Health Monitoring

Electric noses (e-noses), taking their inspiration from the human olfactory system, have been extensively used in food quality control and human disease monitoring. This chapter presents the e-nose as a potential candidate for health monitoring and disease and pest detection on tomato plants. Two common problems in greenhouse tomatoes, namely powdery mildew and spider mites, are considered. An experimental arrangement is described based on a commercial 13-sensor e-nose where tomato plants are grown in an isolated, controlled environment inside a greenhouse. Attention is paid to the preliminary results of data post-processing using two different techniques. First, Principal Component Analysis is employed and demonstrates clear evolution of the components as the plants develop disease or infestation. Subsequently, Grey System Theory enables the identification of clear groupings in the sensor responses and thus the reduction of the model, producing stronger trend differences in the Principal Component between healthy and unhealthy plants. The results, although preliminary, show that the e-nose with appropriate data post-processing is a promising approach to monitoring the development of tomato plant diseases and infestations.

Detection of Diseases and Volatile Discrimination of Plants

The diagnosis of plant diseases is an important part of commercial greenhouse crop production and can enable continued disease and pest control. A plant subject to infection typically releases exclusive volatile organic compounds (VOCs) which may be detected by appropriate sensors. In this work, an Electronic Nose (EN) is employed as an alternative to Gas Chromatography - Mass Spectrometry (GC-MS) to sample the VOCs emitted by control and artificially infected tomato plants. A case study in which powdery mildew and spider mites may be present on tomato plants is considered. The data from the EN was analyzed and visualized using Fuzzy C-Mean Clustering (FCM) and Self-Organizing Maps (SOM). The VOC samples from healthy plants were successfully distinguished from the infected ones using the clustering techniques. This study suggests that the proposed methodology is promising for enhancing the automated detection of crop pests and diseases and may be an attractive tool to be deployed in horticultural settings.

Statistical Gas Distribution Modeling Using Kernel Methods

Gas distribution models can provide comprehensive information about a large number of gas concentration measurements, highlighting, for example, areas of unusual gas accumulation. They can also help to locate gas sources and to plan where future measurements should be carried out. Current physical modeling methods, however, are computationally expensive and not applicable for real world scenarios with real-time and high resolution demands. This chapter reviews kernel methods that statistically model gas distribution. Gas measurements are treated as random variables, and the gas distribution is predicted at unseen locations either using a kernel density estimation or a kernel regression approach. The resulting statistical models do not make strong assumptions about the functional form of the gas distribution, such as the number or locations of gas sources, for example. The major focus of this chapter is on two-dimensional models that provide estimates for the means and predictive variances of the distribution. Furthermore, three extensions to the presented kernel density estimation algorithm are described, which allow to include wind information, to extend the model to three dimensions, and to reflect time-dependent changes of the random process that generates the gas distribution measurements. All methods are discussed based on experimental validation using real sensor data.

Evolutionary Algorithms for Multisensor Data Fusion

The aim of Multisensor Data Fusion (MDF) is to eliminate redundant, noisy or irrelevant information and thus find an optimal subset from an array of high dimensionality. An important feature of MDF is that the signals are constantly evolving instead of being static. This provides an opportunity for Evolutionary Computation (EC) algorithms to be developed to solve MDF tasks. This chapter describes the application of three EC algorithms to widely used datasets. Comparative studies were performed so that relative advantage and disadvantages of the different approaches could be investigated. From this study, authros found that ECs performed in the feature selection stage can greatly reduce the dataset dimensionality and hence enhance the MDF system performance; when being used in a way to represent knowledge, ECs can dramatically increase rules when input data are not clustered.

Feature Selection and Sensor Array Optimization in Machine Olfaction

In this context, the main objective of this chapter is to provide the reader with a thorough review of feature or sensor selection for machine olfaction. The organization of the chapter is as follows. First the ‘curse of dimensionality’ and the need for variable selection in gas sensor and direct mass spectrometry based artificial olfaction is discussed. A critical review of the different techniques employed for reducing dimensionality follows. Then, examples taken from the literature showing how these techniques have actually been employed in machine olfaction applications are reviewed and discussed. This is followed by a section devoted to sensor selection and array optimization. The chapter ends with some conclusions drawn from the results presented and a visionary look toward the future in terms of how the field may evolve.

Noise and Repeatability of Odorant Gas Sensors in an E-Nose

The impact of the characteristics of the sensors used for electronic nose (e-nose) systems on the repeatability of the measurements is considered. The noise performance of the different types of sensors available for e-nose utilization is first examined. Following the theoretical background, the probability density functions and power spectra of noise from real sensors are presented. The impact of sensor imperfections including noise on repeatability forms the basis of the remainder of the chapter. The impact of the sensors themselves, the effect of data pre-processing methods, and the feature extraction algorithm on the repeatability are considered.

Enhancing the Classification of Eye Bacteria Using Bagging to Multilayer Perceptron and Decision Tree

Eye bacteria are vital to the diagnosis of eye disease, which makes the classification of such bacteria necessary and important. This chapter aims to classify different kinds of eye bacteria after the data were collected by an Electronic Nose. First the Multi-layer perceptron (MLP) and decision tree (DT) were introduced as the algorithm and the base classifiers. After that, the bagging technique was introduced to both algorithms and showed that the accuracy of the MLP had been significantly improved. Moreover, bagging to the DT not only reduced the misclassification rate, but enabled DT to select the most important features, and thus, decreased the dimension of the data facilitating an enhanced training and testing process.