Design Of Snapshot Spectral Imaging System And Research On High-Resolution Spectral Reconstruction Algorithm

Surface array snapshot spectral imaging system is a research hotspot in the field of spectral imaging.It solves the problem of low temporal resolution in traditional spectral imaging systems.However,although the spectral imaging systems of regional array snapshot has rich spectral information,the spatial resolution of spectral images is generally low.In view of this,this thesis has carried out a deep research on the snapshot spectral imaging system,proposed a snapshot spectral imaging system design and high-resolution spectral reconstruction algorithm,to achieve the target high-spatial resolution hyperspectral image under a single exposure condition.Firstly,this paper designs a snapshot spectral imaging system based on microlens array beam splitting.Low-spatial resolution hyperspectral and high-spatial resolution RGB images of the same target were obtained by means of spectral imaging and RGB imaging duplex design,respectively.The spectral imaging optical path is mainly composed of an image telecentric lens,a microlens group,and a surface dispersive spectral imaging unit.The optical information of the target is collected and imaged by the image side telecentric lens on a microlens group,which is segmented and collimated.The collimated beam is focused by dispersion through a surface dispersive spectral imaging unit,and finally a low spatial resolution hyperspectral image is obtained by a grayscale camera;Another way to get a high spatial resolution RGB image of the target directly from the RGB camera.The optical model is simulated by ZEMAX software and the imaging performance is analyzed and evaluated.In addition,an experimental platform was set up to verify the effectiveness of the spectral imaging system.Experimental results show that the spectral imaging system has good imaging performance and high imaging quality.Secondly,based on the imaging characteristics and fusion targets of high spatial resolution RGB images and low spatial resolution hyperspectral images,a dual optical path high-resolution spectral reconstruction algorithm is proposed.This algorithm adopts a multi-scale feature extraction mechanism and a multi-level scale fusion mechanism to fully fuse the spectral information of the low spatial resolution hyperspectral image of the same target with the structural features of the high spatial resolution RGB image,obtaining a high spatial resolution hyperspectral image of the target.The experimental results show that the proposed method achieves peak signal-to-noise ratios of 39.8504 and 42.9646 respectively on the CAVE and Harvard datasets,and spectral angle mapping reaches 0.0685 and 0.1585 respectively,which is a certain improvement compared to other methods.It can reduce spectral distortion and improve the spatial resolution of spectral images.