Research On Snapshot Imaging Spectrum Restoration Based On GPU

Imaging spectroscopy, which is the combination of imaging technology and spectrocopy, has been widely used in various fields for its outstanding capabilities on multi-dimensional information acquisition and processing. Owing to its ability of obtaining the whole 3D data cube by one exposure, snapshot imaging spectroscopy, has enormous potential in the field of real-time detectioni and has become a hot research topic.However, mostly, the innovation or improvement of structure has been overemphasized and the study on spectrum restoration neglected in the field of snapshot imaging spectrocopy. In addition, the classic spectral image software mainly processes off-line spectral data and cannot work with the real-time snapshot imaging equipment.This, to some extent, retricts the development and applications of snapshot imaging spectroscopy.This paper, taking the homemade snapshot imaging spectrometer as research platform, aims to study the spectrum restoration algorithm and design the spectral reconstruction system by integrating the real-time acquisition and data processing together.Firstly, the stucture and principle of snapshot imaging spectrometer are studyed and the error that may exist in the system is minutely discussed. Based on that, the spectrum restoration algorithm research, especially the research on image registration with sub-pixel accuracy and non-uniform data processing algorithm is carried out.Secondly, a GPU-based spectrum restoration algorithm is proposed and realized to meet the need of real-time spectrum processing. The results prove that, the novel GPU-based parallel spectrum restoration algorithm proposed and realized in this paper,can achieve the same reconstruction accuracy but more than 20 times in computational efficiency when comparing with the traditional CPU serial algorithm.Thirdly, the spectral restoration system of snapshot imaging spectrometer is established by integrating real-time acquisition and data processing together.Finally, in order to verify the performance of the spectral restoration system, the spectral resolution, spatial resolution, spectral restoration accuracy, acceleration ratio of GPU-based algorithm and the sensitivity of metamerism effect are tested.