Super-resolution Imaging, Orientation Analysis and Image Processing with Machine Learning in Sum Frequency Generation Spectroscopy

This work encapsulates different aspects of Sum Frequency Generation (SFG) spectroscopy, microscopy and image processing. It is focused on three areas, namely, super-resolution SFG imaging, molecular orientation analysis for rubrene surface, and improvements in chemical assignment of pixels in SFG imaging. For super-resolution SFG microscopy, a specialized ground state depletion scheme was implemented by modifying the SFG raster scan imaging technique. This strategy deployed a donut shaped depletion beam that suppresses the SFG signal from the peripheral regions of the area illuminated by focused beams used in the raster scan. This improved the spatial resolution by threefold. However, spatial resolution beyond diffraction limit was not attained due to lack of specialized focusing optics. The molecular orientation of the rubrene single crystal surface was determined using SFG spectroscopy with spectra acquired under different azimuthal orientations and polarization combinations. The SFG analysis indicated a 27° tilt from the surface normal for the phenyl rings of the molecules on the rubrene surface. Finally, a Neural Networks (NNs) based machine learning approach was developed to improve chemical assignment for the pixels in SFG imaging. In this technique, NNs were trained through supervised learning in which they were shown spectra with known chemical labels. Afterwards, they were deployed to predict chemical labels for new pixel spectra. The tests returned above 90 % accuracy under high spectral noise conditions; hence, removing pixel binning requirements traditionally encountered in spectral fitting approach for chemical analysis.