Research On Terahertz Super-Resolution Stress Measurement Method Based On Convolutional Neural Network

Terahertz imaging technology has great application prospect in non-destructive testing,biomedical imaging,safety inspection,astronomical science and so on.Terahertz waves are highly penetrable to optically opaque materials,so they can be used to obtain information inside the material.However,due to the limitations of the size of the terahertz spot and the diffraction limit of the terahertz wave,the low resolution of the terahertz imaging becomes an important reason and problem that restricts the development of the terahertz measurement technology.With the rapid development of deep learning,the image super resolution algorithm based on convolutional neural network(CNN)has achieved exciting image super-resolution effect.In this paper,CNN is applied to super resolution reconstruction of terahertz scanning imaging and super-resolution stress field measurement for terahertz.The details are as follows:Firstly,this paper develops an image super-resolution composite algorithm suitable for terahertz time domain spectral(THz-TDS)scanning imaging,which can improve the resolution of system scanning imaging without changing the hardware conditions.The Gaussian blurred kernel of THz-TDS system is evaluated through calibration experiment and analysis of the experimental system and actual situation.The down-sampling model of the terahertz imaging is obtained by analysis.The super-resolution composite algorithm is constructed by combining deconvolution algorithm,interpolation algorithm and CNN algorithm to reconstruct low resolution terahertz images,and the results are better than those of each algorithm alone.The results of scanning imaging experiments of hole structures and imaging experiments of amplitude data of the diametrical loaded disk show that the proposed algorithm can improve the resolution of terahertz scanning imaging with good effect.Secondly,low spatial resolution is not only one of the disadvantages of terahertz scanning imaging,but also one of the reasons limiting the development of terahertz stress measurement.In this paper,the numerical simulation of THz-TDS image under plane stress state is realized by considering Fresnel formula,stress optical effect and Gaussian blur effect of terahertz spot.By applying the down-sampling model of terahertz three-dimensional signal directly to terahertz signal,the modulation model from plane stress state to THz-TDS signal is established.So the simulation training sets which can be used for training network model is obtained.Finally,this paper combines the principle of full-field stress measurement based on THz-TDS technology with the super-resolution convolutional neural network(SRCNN)algorithm to establish an end-to-end mapping algorithm for low resolution stress field to high resolution stress field.By applying the trained SRCNN model to numerical simulation stress field and real experimental stress field,the spatial resolution of stress field calculated from the captured THz-TDS signal is improved.In summary,this paper studies the super-resolution composite algorithm for terahertz scanning imaging and carries out experimental verification.The CNN algorithm of super resolution stress measurement based on THz-TDS is studied,and the network model from low resolution stress field to high resolution stress field is obtained,and the experimental verification of stress measurement is carried out.