Research On The Application Of Compressed Sensing Algorithm In Millimeter Wave Imaging For Safety Inspection

In modern times,social security problems occur frequently,the shortages of some traditional security inspection methods have become more and more obvious.millimeter wave(MMW)is widely used in body contraband objects detection imaging for its penetrability and non-ionization property,it has become another important method in the field of detection imaging after optical imaging,microwave imaging,and infrared imaging.However,the resolution of a MMW image that can meet the recognition requirements is at the expense of a huge amount of data.For example,an omnidirectional human body MMW image needs hundreds of millions data,there is no doubt a great burden on data acquisition,transportation and high hardware costs.In addition,most of the data collected according to the Nyquist theorem is redundant,there is only a small part of the information really useful for imaging,which resulting in a waste of data resources.Compressed sensing,an emerging theory provides a theoretical support for solving the problem of sparse imaging.Based on this theory,signal can be accurately reconstructed even a small amount of sampling data.This paper is about the study of millimeter-wave security imaging based on compression sensing to achieve the goal of sparse sampling and ensuring high quality reconstructed image,The main contents are as follows:1.This paper introduced the characteristics of millimeter wave and the basis of frequency selecting for imaging,the MMW imaging system and some knowledges of image reconstruction.In addition,the theory of compressed sensing that related to the characteristic of millimeter wave is also introduced,including signal sparseness,measurement matrix and optimization algorithm.2.We researched the methods of applying compression sensing in 2D and 3D MMW imaging.There are two cases about 2D MMW imaging as follow: the 2D sparse imaging by using single frequency of plane scanning method and wide-band frequency of linear array scanning method.From 2D to 3D,a method of establishing wavenumber domain phase function observation matrix and Fourier operator is applied,which can construct a linear measurement model to avoid the complexity of large matrix inversion.In the aspect of using sparse priori information,the 3D total variation methodis applied to improve the reconstruction quality by adding the prior information in the gradient domain of frequency dimension.3.To solving the problem of fuzzy edge for the reason of the excessive smoothing caused by the total variation regularization in image restoration,we used a weighted method to self-adapt restrain the gradient smoothing degree to improve the resolution of the edge.In addition,we used the reconstructed image by self-iteration as the reference image to obtain section priori to guide the subsequent image reconstruction,it can enhance quality of image again.The regularization constraint model established by two methods can improve the quality of reconstructed image obviously.