Scattering Correction Methods of Infrared Spectra Using Graphics Processing Units

Abstract

Fourier transform infrared (FTIR) microspectroscopy has been used for many years as a technique that provides distinctive structure-specific infrared spectra for a wide range of materials (e.g., biological (tissues, cells, bacteria, viruses), polymers, energy related, composites, minerals). The mid-infrared radiation can strongly scatter from distinct particles, with diameters ranging between 2-20 micrometer. Transmission measurements of samples (approximately 100 micrometers x 100 micrometers x 10 micrometers) with distinct particles. will be dominated by this scattering (Mie scattering). The scattering distorts the measured spectra, and the absorption spectra appear different from pure absorbance spectra. This thesis presents development and implementation of two algorithms for processing of FTIR spectra and evaluating the resulting mid-FTIR images. The first procedure removes Mie scattering spectral features, and shows resulting spectra and images to confirm that scattering intensity has been minimized, and the second procedure is a spatial deconvolution algorithm which is used to improve the contrast and fidelity of the imaging data. Both the algorithms discussed in this thesis were implemented using Graphics Processing Units (GPUs) for fast hyperspectral processing by exploiting the parallelism in distributed computational environment. 30x speedup was achieved in spatial deconvolution algorithm implementation as compared with MATLAB implementation of the same problem specifications. Scattering correction implementation on GPU achieved 10x speedup for single iteration as compared with previous MATLAB implementation. Next, some tests were run on real datasets and its' GPU implementation time is compared with previous implementation on CPUs. In the end some future directions and prospects are mentioned.