| Remote sensing Fourier transform infrared (FTIR) techniques are increasingly being applied to the detection of volatile organic compounds (VOCs) present in the atmosphere. This is made possible because of advancements in technology on several fronts including improvement in hardware, development of novel data analysis strategies, automation of the analysis procedure, and compact spectrometer designs which allow for rugged and mobile spectrometers. Our laboratory has pioneered the development of novel data analysis strategies for the detection of VOCs by combining signal processing and pattern recognition techniques. Signal processing, such as filtering with a digital bandpass filter, is used to extract the analyte specific features from the interferogram data collected by the FTIR spectrometer. Short segments of filtered interferograms are classified into analyte active or inactive classes by the pattern recognition technique. The methodology does not require the use of a stable infrared background and can be implemented in an automated manner. Part of this dissertation describes the successful implementation of this interferogram-based methodology for the detection of methanol and trichloroethylene present in the atmosphere. In addition, an improvement in the digital filter design technique is described.;The development of data analysis strategies for the quantitative determination of organic compounds is the other component of the research presented in this dissertation. It has been established that the use of carefully selected wavelengths instead of the full spectrum for the data analysis produces better results. Wavelength selection is performed by use of genetic algorithms (GAs). The selected wavelengths are used in building a multivariate regression model by partial least-squares (PLS) regression. The model is used for the prediction of analyte concentrations of the samples in a prediction set. The GA is guided by a criterion incorporating the calibration, prediction, and regression model size information to find the optimal subset of wavelengths. This strategy is used for the quantitative determination of analytes in three data sets of increasing matrix complexity: (1) methyl isobutyl ketone in water, (2) glucose in triacetin and bovine serum albumin, and (3) glucose in human serum. |
