Research On Concentration From Fourier Transform Infrared Spectroscopy Inversion Technology Based On DSP

Nowadays, with the human production and living there’s unavoidable discharge ofpollutants to the atmosphere causing air pollution that has brought great harm to humans,animals and plants, destruction of the ecological balance. Therefore, the monitoring ofpollutants has guiding significance on the protection of the environment.Gas concentration detection Based on fourier transform infrared spectroscopy (FTIR) iswith the measurement of the gas spectral information as the basis for quantitative analysis. Inspite of great quantity of algorithms for gas concentration detection, regardless of the choicesalgorithm, it must be carried on computer. Due to the limitations of the platform，the measuredspectral information has to be read to computer before the concentration analysis, thereforeFTIR gas concentration detection system has poor performance in real time aspect. To solvethe problem, it’s required to integrate concentration analysis algorithm into embedded chip inorder to achieve the miniaturization of the gas concentration detection system integration.Taking the development and application of Fourier transform infrared spectroscopytechnology as a starting point, this paper analyzed the research status using FTIRspectroscopy data for qualitative and quantitative at home and abroad. Proposed firstconstructed BP neural network that can be used for concentration inversion on MATLAB,then make the network transplanted into the DSP. Build the infrared data the spectralacquisition systems, with the background of air, the precision control of the mass flowcontroller is in order to build the samples of different concentrations of CO, take the differentconcentration of the sample as the basis for modeling, as the concentration of inversion modeltraining set. The main research content of this paper is to transplant the inversion algorithmsof concentration to DSP platform, the following major tasks:(1) Utilize the quality controller to build different concentrations of CO and air samples as thebasis of the gas concentration inversion model. (2) Reduce spectral data noise with S-G algorithm, and remove the spectral data drift usingpolynomial iterative fitting, providing better spectral data for subsequent processing.(3) The spectral data collected use principal component analysis (PCR) for feature extractionto reduce the neural network input points, in order to simplify the network model.(4) Make use of BP neural network to build a gas concentration inversion model using thevalidation set to test the model and achieve the neural network model with C, to run toachieve a concentration of inversion function on the platform of DSP.