Study On Terahertz Imaging Based On Compressed Sensing

Terahertz imaging is an important branch of terahertz technology and plays an important role in non-destructive testing,security inspection and biomedicine.The terahertz detector is the core device of the terahertz imaging system.Due to the high cost of fabrication and technical difficulties of array detectors,terahertz imaging currently faces serious problems such as poor imaging quality and high noise.Compared with array imaging technology,terahertz imaging technology based on compression sensing uses a single pixel terahertz detector to realize image sampling and high-precision reconstruction under the condition of far lower sampling rate than Nyquist sampling rate,which has outstanding advantages of low cost and high imaging efficiency,and its development potential in the future is difficult to be estimated.In this paper,the noise performance of a terahertz imaging system based on compressed sensing is analyzed and an optimization scheme is proposed.The main research contents and results are as follows:1.This paper utilizes ZEMAX software to establish a compressed sensing model based on terahertz imaging system to collect compressed signals,and further study the influence of main parameters of light source and optical path on the signal-to-noise ratio of reconstructed image.By analyzing simulation,the peak signal-to-noise ratio(PSNR)of the reconstructed image under the point light source is first obtained,and the different light path parameters including the point light source divergence angle and the distance counterweight between the imaging object and the mask are analyzed.Then the PSNR of reconstructed images under the conditions of Gaussian source and mixed source are studied respectively.The simulation results show that: 1)For the point light source,the divergence angle increases,and the distance between the imaging object and the mask decreases,which is conducive to improving the PSNR of the reconstructed image.The PSNR value can reach up to 50 dB;For combined sources of point light,the optimal value of PSNR is 13 dB;2)For Gaussian sources,the beam waist is increased,and the terahertz beam unevenness irradiated on the mask plate is reduced,which is conducive to improving the PSNR of the reconstructed image.The maximum PSNR value can reach 50 dB;3)In order to eliminate the noise introduced by the source non-uniformity,a light source noise optimization method is proposed.After optimization,the image noise introduced by the source is effectively reduced.For the combined source of point light sources,the object image is completely and clearly reconstructed after optimization,and the image PSNR can reach a maximum value of 50 dB.2.Considering the diffraction effect of the system edge,the time-domain finite difference method FDTD software was used to simulate the object reconstruction image.It is found that the PSNR value of the reconstructed image is much lower than the PSNR under the same conditions of ZEMAX software.The comparative study found that for Gaussian terahertz sources,the effect of light source parameters on PSNR is consistent with the results of ZEMAX software,but the optimal PSNR value is about 50% under ZEMAX conditions.Diffraction at the edge of the hollowed out part of the mask causes part of the beam to no longer be collimated.In the process of beam focusing,the focus point changes,which makes the single-point detection data worse,and ultimately leads to a reduction in the peak signal-to-noise ratio of the reconstructed image.3.A terahertz imaging system was built based on 100 GHz and 250 GHz VDI light sources and unit terahertz detectors.The effect of actual light source parameters on the reconstructed image PSNR was measured and the feasibility of the light source noise optimization method was verified.The experimental results show that the signal emitted by the VDI terahertz source is Gaussian terahertz light in the far field,and there is unevenness.The 100 GHz source is more uneven than the 250 GHz source.After using the light source noise optimization method,the PSNR of the 100 GHz reconstructed image can be increased by about one time,from 6.8dB to 13.2dB,and under 250 GHz light source,the PSNR can reach up to 20.5dB.Experimental results show that the proposed optimization method is more effective for high-frequency terahertz imaging.