Quantification of H2S and NO2 using gas-sensor arrays and an artificial neural network

Based on a dedicated sensor array and a novel data pre-processing function, we report the successful application of a multilayer perceptron to a sensor array to give quantitative identification of individual gas concentrations (H2S and NO2) in their gas mixtures. The sensors were produced by the rheotaxial growth and thermal oxidation technique. Our raw sensor's responses covered a range of four magnitudes due to the different responses of the sensors. By comparing several pre-processing methods, we demonstrated that pre-processing of input data has crucial influence on the final performance of neural networks. Our adoption of the pre-processing rule might be of general usefulness for the case of a large range of raw data.