Sofiane Brahim-Belhouari

Gaussian process for nonstationary time series prediction

In this paper, the problem of time series prediction is studied. A Bayesian procedure based on Gaussian process models using a nonstationary covariance function is proposed. Experiments proved the approach effectiveness with an excellent prediction and a good tracking. The conceptual simplicity, and good performance of Gaussian process models should make them very attractive for a wide range of problems.

Fast and robust gas identification system using an integrated gas sensor technology and Gaussian mixture models

Among the most serious limitations facing the success of future consumer gas identification systems are the drift problem and the real-time detection due to the slow response of most of todays gas sensors. This paper shows that the combination of an integrated sensor array and a Gaussian mixture model permits success in gas identification problems. An integrated sensor array has been designed with the aim of combustion gases identification. Our identification system is able to quickly recognize gases with more than 96% accuracy. Robust detection is introduced through a drift counteraction approach based on extending the training data set using a simulated drift.

Gas identification with microelectronic gas sensor in presence of drift using robust GMM

The pattern recognition problem for real life applications of gas identification is particularly challenging due to the small amount of data available and the temporal variability of the instrument mainly caused by drift. We present a gas identification approach based on class-conditional density estimation using Gaussian mixture models (GMM). A drift counteraction approach based on extracting robust features using a simulated drift is proposed. The performance of the retrained GMM shows the effectiveness of the new approach in improving the classification performance in the presence of artificial drift.

Quantization errors in committee machine for gas sensor applications

In a digital implementation of a gas identification system, the mapping of continuous real parameter values into a finite set of discrete values introduces an error into the system. This paper presents the results of an investigation into the effects of parameter quantization on different classifiers (KNN, MLP and GMM). We propose a committee machine to decrease the classification performance degradation due to the quantization errors. The simulation results show that the committee machine always outperforms a single classifier and the gain in classification performance is greater for a reduced number of bits.

On the use of the transient information for gas identification using microelectronic gas sensor

Gas identification using microelectronic gas sensor represents a big challenge for pattern recognition systems due to a number of problems mainly related to non-selectivity and the drift. The use of preprocessing can often greatly improve the recognition performance. The aim of this paper is to evaluate the performance of different preprocessing techniques using transient information for gas identification from sensor array signals. We compare the classification accuracy of three feature extraction techniques based on steady state value, transient intergrals and dynamic slope. It was found that the transient information plays a critical role in improving the classification performance. In addition the proposed dynamic slope feature extraction technique is hardware friendly which makes the prospect of building a smart gas sensor very promising.

A real-time architecture of SOC selective gas sensor array using KNN based on the dynamic slope and the steady state response

This paper demonstrates that using the dynamic response together with the steady state response greatly improves the classification performance of gas sensors. We propose a SOC VLSI architecture based on the KNN algorithm and operating on both the steady state and dynamic slope response of the data from the gas sensor array. The architecture is based on a current model analog pipelining strategy which allows to share hardware resources between different sensors within the sensor array. This results in significant area savings making the prospect of building low cost and real-time electronic nose microsystem reasonably cheap.