IDENTIFICATION AND CONCENTRATION MEASUREMENTS OF ATMOSPHERIC POLLUTANTS THROUGH A NEURAL-NETWORK ANALYSIS OF PHOTOTHERMAL SIGNATURES

We propose here to use a neural network as a help in the identificatio n of individual components of a gaz mixture and in the evaluation of t he concentration level of each component as they appear in real photoa coustic spectra in presence of background signals and noises. The meth od is based on the dependance of the activation functions with regard to the input level controled variation and on the desactivation of the first activated outputs.