Research On 2.52 Terahertz In-Line Digital Holographic 3D Scene Reconsrtuction Algorithm

The strong absorption or strong scattering characteristics of teraher tz light waves make it widely used in coaxial digital holographic imaging.Holographic imaging forms a hologram through light wave interference,which records all the information of the object light wave,so the original object light wave can be reconstructed by processing the hologram.The main work of this paper is to obtain target amplitude information through holograms,including two-dimensional planes and three-dimensional continuous objects.Firstly,the influencing factors of the reconstruction amplitude image quality of the double-exposure algorithm are simulated and studied;on this basis,the continuous terahertz coaxial digital holographic three-dimensional imaging model and the deep learning tool are used to realize the reconstruction of the terahertz image.In order to make the simulation model of the original amplitude image obtained by the double-exposure algorithms closer to the reality,the influencing factors of the reconstructed image quality are first studied under the ideal light source.Taking into account the non-uniformity of the background plate material,only Gaussian noise with different variances is added to the background plate to simulate and study this effect;secondly,considering the interference factors of the holographic reco rding process,noise is added to the sample and holographic plane to study the influence of the optical system on the reconstruction result,respectively gives the change curve of the objective evaluation value with the noise variance.Taking into account the processing errors in the production process of the background plate,the background plate transmittance is set to be cyclically inclined to simulate this effect.After the hologram is acquired under the ideal light source,the constraints for obtaining a better reconstruction result are inconsistent with the experimental hologram.Therefore,the actual illumination light source is added to the simulation model to make the acquired hologram closer to reality.Due to the uneven energy distribution of the background light source,the influence on the reconstruction quality was simulated by changing the position of the target on the background plate,and the results were qualitatively analyzed by using a logarithmic histogram.In order to further improve the quality of the reconstructed amplitude image obtained by using the HRNet structure,according to the characteristics that the test sample is only composed of horizontal and vertical lines,the shape of the diagonal line and the curve is added to the training sample to improve the robustness of the network training,realize the reconstruction of simulation images by expanding the training sample set,and further analysis of the adaptability of the network to the axial offset distance,on this basis,the reconstruction of the experimental terahertz image is realized.The U-Net structure is used to realize the digital holographic reconstruction of continuous three-dimensional objects,and the residual structure can be added to the initial network to achieve higher quality reconstruction.The influence of the network hyperparameters and the Dropout layer on the reconstruction results is analyzed;the adaptability of the network to system noise and axial offset is continued to be analyzed.The difference of the reconstruction effect of the simulated and experimental terahertz images with different network structures is compared and analyzed.