| Snapshot imaging spectroscopy obtain the image data for restoring spectral data cube in one or a few shutters,which can effectively improve the speed of information acquisition.Snapshot Fourier transform imaging spectrometer is a kind of computational snapshot imaging spectrometer with compact structure and good seismic resistance.Snapshot imaging spectroscopy obtains spectral data cube by imaging different spectral-related components in one detector at the same time.Therefore,the spatial resolution of a single spectral image is inevitably reduced.Computational snapshot imaging spectrometers generally facing the inefficient computation efficiency because of the complex reconstruction algorithm.In this paper,the key technology of snapshot Fourier transform imaging spectrometer and its clinical application is studied.The specific research contents are as follows: 1.The basic principle of snapshot Fourier transform imaging spectrometer is analyzed,and the structural design of the prototype is realized.The influence mechanism of double Wollaston prism structure parameters on the performance of the Fourier transform spectrometer was analyzed in detail,and the parameters were optimized according to the application requirements.The prototype of the compact snapshot Fourier transform spectrum imager is developed.2.Aiming at solving the problem of the low spatial resolution of the prototype,a super-resolution method is studied.To obtain high spatial resolution image,the traditional schemes introduced additional light path which increases the size.This project utilizes the slight deviation between subgraph arrays to directly reconstruct high-quality panchromatic images by deep learning,thus keeping the system compact.In the training stage,the general panchromatic image and snapshot spectral image were collected simultaneously using beam splitting.After the training,the common panchromatic image optical path is removed,and the trained network can be used to reconstruct high-quality panchromatic images.The specially designed deep convolutional network can enhance the contrast and space resolution simultaneously.This method can obtain a panchoromatic image whose PSNR is generally higher than 33 d B and SSIM is higher than 0.95 with a scale factor of 3.5.Through this network,an additional high-quality panchromatic image can be obtained quickly without changing the structure of the instrument.The spectral cube can be fused with the panchromatic image to improve the data quality.3.Aiming at enhancing the computational efficiency of the spectral reconstruction algorithm of snapshot Fourier transform spectrometer,a parallel spectral reconstruction method was proposed after analyzing the data characteristics of the instrument and the concurrency of the algorithm.To parallelize the sub-image registration process,the relative positions between sub-images of the system were optically designed to be fixed and a template-based parallel registration method is proposed.This method uses a checkerboard to pre-register the prototype and the registration parameters are stored as coordinates.By resample all subgraphs using the hardware interpolation characteristics of GPU texture memory,the sub-images will be aligned efficiently.In the registration process,the transformation model uses block linear transformation to segment the small deformation of the subgraph.The overall effect is better than the affine modle.After optimization,the spectral restoration time of the prototype is reduced to about 90 ms,which is less than the image transmission rate of the detector.The spectral resolution of the prototype is better than 300 cm-1,and it has good spectral imaging ability.4.The application of the snapshot imaging spectrometer in the real-time monitoring of patients with systemic lupus erythematosus was explored.Benefit from the portable and real-time spectral imaging capability of our compact snapshot spectrometer,this paper established the rash activity model based on the spectral characteristics of specific skin manifestations of systemic lupus erythematosus and obtained the corresponding relationship between the rash activity and the disease status of patients.It was found that the rashes activity coefficients extracted from the spectral data of the patients were consistent with the artificial rating of the rashes activity status of the patients,and the obtained rashes activity coefficients showed a normal distribution on the corresponding artificial rating.By comparing the changes of skin rash activity coefficient with the changes of complement C3/C4,ds DNA and SLEDAI rating that indicate the patient's disease status,it was found that the change of skin rash activity coefficient was consistent with the change of patient's disease activity.Therefore,the snapshot Fourier transform imaging spectrometer can be used to assist in monitoring the activity of systemic lupus erythematosus. |
PDF Full Text Request