Application of sensor dynamic response analysis to improve the accuracy of odour-measuring systems

A system consisting of a matrix of three semiconductor gas sensors was applied to the classification of different orange juices. The sensor matrix responses were sampled in short time intervals. Such responses were processed by discrete wavelet transform (DWT) together with the k-nearest neighbour (kNN) classification algorithm or by the probabilistic neural network (PNN). The obtained results show that both types of signal processing (DWT + kNN and PNN) applied provide very good class separation for time response analysis, while in the case of the static response analysis the correct classification coefficients are much lower. It is shown that the analysis of the sensor's time response can be an efficient way of increasing both the accuracy level and the immunity to external noise in e-nose systems. The possibility of reducing the number of sensors without decreasing the system performance is also demonstrated. Additional experiments have shown that for both processing methods, the results obtained with the dynamic response of a single sensor were better than those reached with the three-sensor array measured in static conditions.