The possibility of increasing the efficiency of Savitzky–Golay filter for THz absorption spectra noise reduction as a preprocessing step was considered. The studies were carried out on the model atmosphere gas mixtures typical for swampy area. The noisy absorption spectra of the gas mixtures in 0.1-1.0 THz spectral range were calculated using data from HITRAN database. A sliding window variant of Savitzky–Golay filter was proposed and analyzed. The analysis of noise reduction quality was evaluated by the criterion of proximity of two curves. Using Savitzky-Golay filter together with the sliding window method was shown to provide a significant improvement in the noise reduction compared to standard variant of this filter.
This article describes the methods and approaches used by us to solve the problem of a high error in the determination of a component with a low concentration in a gas mixture. The approaches based on the modification of the machine learning model were considered, the approach to the generation of the training sample was changed, an iterative method for increasing the accuracy of the model results was proposed.
The regression model was applied to solve the problem of restoration of the concentration of components in a gas mixture using infrared absorption spectra. A solution to the problem of determining the concentrations of individual components of a multicomponent gas mixture is suggested and tested.
An important role in component analysis with spectral methods has a spectral resolution of used tools. The most useful and perspective methods to improve spectral resolution is decreasing of impulse response function (IRF) and improving resolution using superresolution (SR) reconstruction methods. We have analyzed different types of neural networks (convolution neural network, multilayered perceptron) for improving the spectral resolution of initial absorption spectra. The used approach is based on an association of a high-resolution and a low-resolution spectrum. The latter was constructed from high-resolution spectra to which IRF and some random noise were added. Highresolution spectra were generated using the HITRAN database. Most optimal architectures of neural networks to improve spectral resolution were defined.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.