| With the rapid development of optical imaging technology,basic hardware such as new light sources,sensors,and light modulators have been continuously improved and innovated,and new imaging models have continued to emerge.Optical imaging systems have already developed from black and white cameras that can only acquire twodimensional spatial information of objects to new imaging systems that can acquire optical information in multiple dimensions(space,polarization,spectrum,etc.).All dimensional optical information of the photons emitted or reflected by the target scene can be represented as(χ,y,z,θ,φ,λ,t,Ψ,χ)and called the plenoptic function.The optical information of each dimension reflects the different physical properties of the target independently of each other,and can be combined together to play a common advantage.Therefore,the more complete recording of optical information,that is,the more dimensions of optical information,and the higher the resolution in each dimension,is of great significance for the detection and analysis of targets by optical means.However,due to the huge gap between the data acquisition capabilities of existing detectors and the massive data required to record the multi-dimensional optical information of the target,it is impossible to achieve high-dimensional imaging by increasing the detector data bandwidth in the short term.The current popular computational imaging methods can break through the bottleneck of detector data acquisition capability by means of compression acquisition and computational reconstruction,which provides a useful idea for realizing multi-dimensional imaging.However,computational imaging methods are still underdeveloped in terms of the number of imaging dimensions and the resolution of each dimension.The core scientific problems in this field can be summarized as:how to use efficient compression acquisition and computational reconstruction methods to record and reconstruct target light information with more dimensions and higher resolution.In view of the above scientific problems,this paper carries out the following research contents:1.Aiming at the goal of optical imaging systems to record more dimensional light information,this paper proposes a snapshot-type polarized spectral light field imaging system.In this paper,the basic theory of focusing light field imaging technology is firstly described,and the scale-depth space conversion algorithm for depth information reconstruction is introduced in detail.Based on the focused light field imaging technology,a snapshot polarized light field imaging scheme is explored,and a multiimage registration algorithm based on depth information is designed,which provides a theoretical basis and technical basis for the realization of multi-dimensional computational imaging.At the same time,this paper combines the channel-type polarization spectroscopy technology and the snapshot-type spectral imaging technology based on the microlens array,and proposes a snapshot-type channel-type polarization spectroscopy imaging scheme based on frequency domain multiplexing.Finally,according to the idea of combining space-frequency domain multiplexing,this paper combines the snapshot channel polarization spectral imaging scheme with the focusing light field imaging technology to establish a channel interferometric multi-dimensional imaging system and theoretical model.By effectively controlling the compression and collection efficiency of information in different optical dimensions,the system can break through the bottleneck effect of detector bandwidth on the acquisition of optical dimensions in multi-dimensional imaging technology,and realize that the light field of target light can be efficiently compressed and collected in real time through a single detector.,polarization and spectral information.2.Aiming at the problems that the resolution of each dimension restricts each other in the snapshot polarized spectral light field imaging system,and the traditional reconstruction methods relying on channel segmentation and Fourier transform have cumbersome steps and low resolution,this paper proposes a method based on channel redundancy information.High-precision reconstruction method of polarization and spectrum.First of all,it is clear that the interference channel crosstalk is the core bottleneck that the frequency domain compression model is difficult to achieve highresolution solution.Based on the self-built polarization and spectral database training network,the polarization and spectral channels are separated and redundant information is extracted to complete the high-precision polarization-spectrum.reconstruction.This method can effectively learn the forward conversion process of multi-dimensional imaging system from interference compression data to polarization and spectral information during information reconstruction,and further improves spectral resolution and polarization reconstruction accuracy while completing phase correction.3.Based on the snapshot-type polarized spectral light field imaging prototype,this paper firstly conducts imaging experiments on indoor and outdoor complex static and dynamic scenes,which intuitively shows that the multi-dimensional imaging system has the ability to record the polarization,spectrum and 3D information of the target in real time.In addition,this paper uses experimental methods to test the performance of the multi-dimensional imaging prototype in various dimensions,including spectral reconstruction accuracy and spectral resolution,polarization reconstruction accuracy,spatial lateral resolution,depth reconstruction accuracy,etc.The experimental results show that the computational reconstruction method represented by deep learning improves the spectral resolution and polarization reconstruction accuracy,so that the multi-dimensional imaging system has excellent indicators in various dimensions such as polarization,spectrum,spatial lateral direction,and depth.At the same time,the imaging experiment results of indoor and outdoor static and dynamic complex scenes further confirm the feasibility and practicability of the snapshot polarized spectral light field imaging system,which can be used in target detection,military de-camouflage and other applications in the future. |