Translator Disclaimer
12 July 1993 IR spectrometer using spectral pattern recognition: a feasibility study
Author Affiliations +
The application of a neural network for processing of the output signal of an infrared (IR) emission mode spectrometer is investigated in this paper. A set of spectral patterns representative of 16 different compounds has been simulated using normal distribution line profiles. These data have then been combined with atmospheric transmittance and path radiance. The network, following training, has been presented with a test set consisting of perturbed versions of the spectra. Perturbations analyzed were line width, peak height, and center variations. The last effect is due to slit image curvature caused by a finite length slit in our hypothetical spectrometer. The purpose of the atmospheric and optical analysis was to insure a realistic estimate of phenomena expected in a field application. The network was found to recognize the input patterns correctly over a broad range of perturbation parameters. We propose that once a satisfactory set of connection weights is established, these should be transferred to a parallel processor (electronic or optical). The network considered in this paper proved capable of generalization under all but the most extreme conditions. Such performance allows by passing of intermediate signal processing for spectral analysis. Consequently, this sort of a system would form a fast and accurate spectral recognition instrument capable of operation under unpredictable conditions.
© (1993) COPYRIGHT Society of Photo-Optical Instrumentation Engineers (SPIE). Downloading of the abstract is permitted for personal use only.
Michael R. Descour and Eustace L. Dereniak "IR spectrometer using spectral pattern recognition: a feasibility study", Proc. SPIE 1900, Charge-Coupled Devices and Solid State Optical Sensors III, (12 July 1993);

Back to Top