| Imaging through scattering such as atmosphere,water and biological tissue is a longterm concern in the field of optics.In recent years,computational imaging technology has attracted the attention of researchers,especially the application of deep learning method in solving the inverse problem of optical imaging,and has achieved breakthrough results.Light scattering tends to diminish as the wavelength increases,as is typically the case for human tissue and the atmosphere.Thus there may be advantages to use the longest wavelength possible for which the absorption is not excessive.Many applications in security and health care would benefit from infrared detection and thermal imaging technologies.However,detection devices working in infrared region,especially far-and mid-infrared region,suffer from low sensitivity,high readout noise,complex apparatus and also high cost.One promising approach is frequency upconversion into the visible(VIS)or nearinfrared domain.Instead of directly detecting,the IR signal is first transferred to the higher frequency range through nonlinear optical process,which preserves(mostly)the spatial,temporal,and spectral information of the signal.The convertion signal is then detected by a standard sensitive camera,thus expanding the working spectral range of cameras.In this thesis,infrared image up-conversion through second harmonic generation is combined with the deep-learning-based speckle imaging.In our experiment,a spatial light modulator illuminated by 1064 nm laser beam is used to load the image from EMNIST database as the imaging target objects.Let the beam reflected by SLM pass through a ground glass and focused into the nonlinear optical crystal KTP to generate second harmonic light.The recovering of up-conversion speckle imaging of objects is realized through deep learning.Two kinds of optical paths of infrared image upconversion are discussed,and then the optical path used in the experiment,the process of experimental data acquisition and the methods used in neural network training are introduced.The convolution neural network structure and loss function used in the experiment are studied.Finally,the imaging results of the experiment are evaluated quantitatively by means of mean square error and structural similarity.This technology combines the robustness and generalization of deep learning method compared with other speckle imaging methods,and the advantages of infrared light in penetrating scattering medium and biological tissue imaging,and avoids the disadvantages of infrared detector,such as poor sensitivity,low efficiency and high cost.This technology has potential applications in the fields of biological imaging,pattern recognition and infrared remote sensing.Spiral phase contrast imaging has been employed in optical imaging for edge enhancement.Nonlinear spiral phase contrast technology has been also used for edge enhancement in infrared imaging.In the thesis,we proposed to combine the nonlinear spiral phase contrast technology and deep learning speckle imaging to realize the edge enhancement in speckle imaging.We have finished the experiment data acquisition.The deep learning training and optimizing of the experimental results is in process.In recent years,due to its unique orbital angular momentum,vortex beam has attracted the attention of researchers,and has been widely used in optical communication,imaging,rotational speed detection,particle manipulation and other fields.However,the waist spot size and far-field divergence angle of vortex beam will increase with the increase of its orbital angular momentum,which has an adverse effect on the coupling propagation and micro-particle manipulation of high-order vortex beams.In order to overcome this limitation,researchers have proposed the perfect vortex beam,whose beam size is independent of its orbital angular momentum.Further,grafted perfect vortex beam has been proposed,which has different orbital angular momentum in different transverse region of the beam.In this thesis,perfect vortex beams and grafted perfect vortex beams are experimentally generated.The frequency doubling experiments of perfect vortex and grafted perfect vortex are carried out.The orbital angular momentum of the grafted perfect vortex beams after frequency doubling is verified with a Mach-Zehnder interferometer.This work is helpful to further expand the application of vortex beam in the field of particle manipulation. |
