Research On 3D Tomography Algorithm Based On 0.11 Terahertz Continuous Wave

Terahertz waves are electromagnetic waves with frequencies between 0.1 THz and10 THz,which can penetrate some opaque objects such as plastic,fabric,paper,etc.They are widely used in the fields of imaging and non-destructive testing.Terahertz imaging as a new generation of imaging technology is developing rapidly,terahertz nondestructive testing is the main application area of terahertz imaging,in which terahertz three-dimensional chromatography imaging technology is widely used in security inspection,nondestructive testing and biomedical imaging and other fields.In the field of terahertz imaging,the image reconstruction algorithm is an extremely important part,which directly affects the image reconstruction quality and image reconstruction time.High-quality reconstructed images require sufficient projection data acquisition,which is time-consuming for terahertz tomography systems and also makes the algorithm computation time longer.In this thesis,to solve the problems of classical image reconstruction algorithms in terahertz tomography,two imaging methods based on compressive sensing and deep learning are proposed to achieve high-quality image reconstruction for terahertz continuous wave tomography in the electronics band,and successfully reconstruct the internal structural defects of the imaged samples in three dimensions.The main research contents and innovations of this thesis are as follows:1.The principle of terahertz continuous wave laminar imaging and the terahertz continuous wave laminar imaging system are studied,and a 0.11 THz laminar imaging experimental system based on a vector network analyzer with the ratio of received wave to transmitted wave amplitude as the projection value is built.The algorithm is optimized based on the terahertz Gaussian beam according to the difference between terahertz waves and X-rays,and the algorithm is validated by simulation using Matlab.The signal-to-noise ratio of the images reconstructed by the algorithm after optimization using the Gaussian model was improved by 17.32%.The algorithm was applied to an imaging experimental system and successfully reconstructed the internal structures of foam,metal and paper cups with 3D images.2.The laminar imaging algorithm based on compression perception is studied,and the ordered subset maximum expectation imaging algorithm based on full variation is proposed,and the image reconstruction results of this algorithm and the filtered backprojection and other algorithms are analyzed by simulation.Analyzing the signal-tonoise ratio and structural similarity indexes of the simulation reconstruction results,the algorithm proposed in this thesis can reconstruct high quality images with 2/5 less projection data,and the images are smoother.The algorithm proposed in this thesis can effectively improve the image reconstruction quality,reduce the acquired data and shorten the time of data acquisition.And combined with the imaging experimental system for 3D image reconstruction of pig bones,it can effectively reconstruct the internal pore structure of pig bones.3.A deep learning-based imaging algorithm is proposed,an AUTOMAP network is built,a terahertz simulation data set is constructed based on the terahertz projection data acquired by the experimental system,and the network is trained with this training set.The image reconstruction is performed using filtered inverse projection,full variational ordered subset maximum expectation,and deep learning algorithms.The results are analyzed using indicators such as signal-to-noise ratio and structural similarity,and the reconstruction results of the deep learning method improve 35.34% in signal-to-noise ratio and 20.42% in structural similarity compared to the compressed perception algorithm.The algorithm was applied to the imaging experimental system and reconstructed three randomly placed iron bars,and the positions and contours of the iron bars were effectively reconstructed.The results show that the algorithm proposed in this thesis,based on compressed sensing and deep learning,has a large improvement compared with the classical imaging algorithm,and can effectively solve the problems of the existing classical algorithm which requires a large amount of projection data and long computation time,and can reconstruct better quality images faster,which has great potential in the field of terahertz 3D image reconstruction.