Search

Motivated by biological olfactory coding principles, rank-code-based classifiers have recently been proposed to facilitate integration of hardware-friendly gas identification platforms with an array of gas sensors. These classifiers operate on a simple principle of generating rank-codes by ranking the gas sensors’ features instead of treating them as a multi-dimensional vector as in computation-intensive state-of-the-art gas classifiers. However, the performance of the rank-code-based classifiers is limited when distinguishing information about all the target gases is not found in the ranks of the gas sensors’ features, but in their values. In this paper, we introduce a computationally efficient alternative solution to overcome this limitation. In this solution, an original multi-gas classification problem is decomposed into pairwise classifiers, and the gas is then predicted with the weighted binary decision codes in each of these classifiers, where each element of the code is generated by exploiting features individually. The weighted binary decision codes are formed by first using the nearest centroid approach, which exploits the mean value of each gas sensor’s feature to generate binary decision codes, and then, a simple approach is used to assign a weight to each element of the code, depending upon its capability to discriminate the gases in each pairwise classifier. The added advantage of this classification approach is that two computationally efficient metrics are introduced to access the classifiers’ applicability to the given data set and certainty about the prediction of any test sample. A classification performance of 97.87% is achieved with this approach on an extensive data set of ten gases experimentally obtained with Figaro series gas sensors, and this is increased to 100% by rejecting 3.37% of samples for which certainty about their predictions is below a 25% confidence level. [ABSTRACT FROM PUBLISHER]

In this paper, we propose a biologically inspired rank-order-based classifier to facilitate the development of a smart electronic olfaction system, because the state-of-the-art pattern recognition algorithms are not suitable for the said objective due to their computationally intensive nature. In order to mimic biological olfactory rank codes, the features of gas sensors in an electronic olfaction system are ranked to form rank codes in our classifier instead of treating them as multidimensional data points. This classifier relies on probability rank tables, which are built for all target gases, because one-to-one mapping between the rank codes and the target gases is not practically found with the existing gas sensor technology. The table for each target gas records the probability of each sensor ID at each rank, and hence also serves as a visual interpretation tool. Due to the limited diversity in electronic olfaction system rank codes compared with biological olfactory codes, gas pairs are considered in this classifier by transforming the original multi-gas identification problem into pairwise classification problems in order to evaluate the discriminatory power between the rank codes for each pair of target gases. This discriminatory power at each rank not only helps to determine the applicability of the classifier but also serves as a weight for that rank in order to improve the classification performance. In addition, insightful quantitative feedback is integrated into the classifier in order to avoid misclassifications, and as a result, we achieve a 100% classification performance at the cost of a 3.79% rejection of uncertain predictions with a data set of an array of FIS inc. and Figaro inc. gas sensors exposed to five target gases. [ABSTRACT FROM PUBLISHER]

Abstract: This paper presents a monolithic 4×4 tin oxide gas sensor array together with on-chip multiplexing and differential read-out circuitry. A robust fabrication process focusing on the integration of the CMOS circuitry and the microelectromechanical systems (MEMS) structure is first described. The gas sensor uses a surface micro-machined micro-hotplate based structure. Platinum is selected as the material of choice for both the micro heater and the sensor’s electrodes, for its good thermal stability and compatibility with the sensing film. Different post-treatments are used to modify the characteristics of the tin oxide gas sensors hence improving the sensor’s selectivity of the overall sensor array. Furthermore, in contrast to the conventional voltage divider read-out technique, a novel differential read-out circuit (DRC) for tin oxide gas sensors is proposed. The DRC applies a constant current to drive the sensor and uses a unit-gain single stage amplifier (inverter) to generate a fully differential output, directly from the voltage drop across the sensor. The output of the DRC is simply proportional to the difference between the voltage on the two electrodes of the sensor but not to the transistor parameters such as mobility and threshold voltage, neither to the supply voltage. The monolithic sensor array and its pre-processing circuitry have been implemented in our in-house 5μm process. Experimental results showed good linearity at the output of the DRC for a wide range of sensor resistance variation (over 20MΩ). The fabricated micro-hotplate sensor array was tested for four target gases. Results show good sensitivity and interesting thermal characteristic leading to only 15.5mW power consumption for 300°C operating temperature. [Copyright &y& Elsevier]

Gas identification represents a big challenge for pattern recognition systems due to several particular problems such as nonselectivity and drift. The purpose of this paper is twofold: 1) to compare the accuracy of a range of advanced and classical pattern recognition algorithms for gas identification for the in. house sensor array signals and 2) to propose a gas identification ensemble machine (GIEM), which combines various gas identification algorithms, to obtain a unified decision with improved accuracy. An integrated sensor array has been designed with the aim of identifying combustion gases. The classification accuracy of different density models is compared with several neural network architectures. On the gas sensors data used in this paper, Gaussian mixture models achieved the best performance with higher than 94% accuracy. A committee machine is implemented by assembling the outputs of these gas identification algorithms through advanced voting machines using a weighting and classification confidence function. Experiments on real sensors' data proved the effectiveness of the system with an improved accuracy over the individual classifiers. An average performance of 97% was achieved using the proposed committee machine. [ABSTRACT FROM AUTHOR]

In this paper, a novel gas identification approach based on Gaussian process (GP) combined with principal components analysis is proposed. The effectiveness of this approach has been successfully demonstrated on an experimentally obtained dataset. Our aim is the identification of different gases with an array of commercial Taguchi gas sensors (TGS) as well as microelectronic gas sensors. The proposed approach is shown to outperform both K nearest neighbor (KNN) and multilayer perceptron (MLP) classifiers. [ABSTRACT FROM AUTHOR]

Abstract: In this paper we compare the accuracy of a range of advanced density models for gas identification from sensor array signals. Density estimation is applied in the construction of classifiers through the use of Bayes rule. Experiments on real sensors’ data proved the effectiveness of the approach with an excellent classification performance. We compare the classification accuracy of four density models, Gaussian mixture models, Generative topographic mapping, Probabilistic PCA mixture and K nearest neighbors. On our gas sensors data, the best performance was achieved by Gaussian mixture models. [Copyright &y& Elsevier]