| Terahertz radiation can penetrate nonmetallic and nonpolar materials and has low photon energy,so it has broad application prospects in biomedical detection and space remote sensing.In recent years,Terahertz imaging technology has been developing towards the direction of continuous improvement of imaging quality and dimension.Therefore,this paper adopts the method of combining subjective and objective.Firstly,the influencing factors of the two-dimensional imaging quality of terahertz in-line digital holography are studied.Then the dimension was extended to three dimensions,and the studies were carried out on the compressed sensing terahertz three-dimensional reconstruction algorithm and the Matlab-based holographic reconstruction network(MHRNet).In order to study the factors affecting the quality of two-dimensional imaging of terahertz in-line digital holography,the effects of noise changes on the reconstructed image were studied when the transmittance of the support plate was 0.93.In the case of Gaussian noise with a mean value of 0 and a standard deviation of 0.04,the influence of the change of the transmittance of the support plate on the reconstructed image was studied,and the variation trend of the optimum transmittance of the support plate with the noise was finally given.In addition,for the four support domain determination methods,simulation studies were carried out on their applicability under the conditions of noise and no noise on the support plate respectively,and finally the application scopes of different support domain determination methods were obtained.In order to study the influence factors of terahertz three-dimensional reconstruction method based on compressed sensing,a series of in-depth studies were carried out on continuous scenes.Firstly,the reconstruction image quality of continuous scene and discrete scene is compared by simulation,which proves that there is mutual influence between all levels of continuous scene.Then,for the continuous scene,the influence of the change of relevant parameters on the reconstruction effect was simulated,and the influence of the number of experimental iterations,sparse limiting parameters,the number of reconstruction planes and the change of TV parameters on the reconstruction image quality was studied under the ideal state.When the number of reconstruction planes is 4,a certain amount of Gaussian noise is added at each sample level to study the selection of corresponding simulation parameters when the noise exists.Change the contrast between the target and the background,and further study the selection of simulation parameters when there is noise at the sample level and the contrast is changed.Considering the efficiency of the compressed sensing terahertz three-dimensional reconstruction algorithm is not high,this paper introduces deep learning into the compressed sensing terahertz three-dimensional reconstruction algorithm,on the basis of the holographic reconstruction network uses a holographic reconstruction based on Matlab(MHRNet).The influence of training data set on reconstruction is discussed and the effect of training parameters and network structure is studied by simulation. |