Research On Sparse Imaging Method Of Through-the-Wall Radar Based On Deep Network

With its ability to penetrate non-metallic building materials,through-the-wall radar can detect,locate and image hidden targets behind walls or closed indoors under non-contact and non-destructive conditions,and is widely used in civil and military fields.However,the strong clutter generated by the wall reflection will cover up the reflected signal of the target behind the wall,affecting the imaging accuracy and detection performance of the target.In addition,the existing through-the-wall radar imaging methods have problems such as high computational cost,excessive manual intervention and low imaging accuracy in the process of model building and optimization solution.Therefore,this paper expands the optimization solution strategy of radar sparse imaging into a deep network structure and adopt an end-to-end learning strategy to efficiently solve the problem of through-the-wall radar imaging.The main research work is summarized as follows:Firstly,starting from the received signal model of through-the-wall radar and the propagation mechanism of electromagnetic waves,introduces the characteristics of different radar signals and the detection system of through-the-wall radar,and analyzes the characteristics of electromagnetic wave propagation in the detection process.Then,several typical radar imaging methods in the field of through-the-wall radar imaging are used to image the target behind the wall,such as back-projection method,convex optimization method and orthogonal matching pursuit algorithm,which providing the imaging theoretical basis for subsequent work.Secondly,a deep unrolling network inspired by an iterative soft-threshold optimization algorithm is proposed to implement a through-the-wall sparse imaging method based on a learnable network.In the problem of through-wall radar imaging based on sparse representation,a two-stage strategy of “clutter suppression followed by image reconstruction” is usually adopted to solve the problems of wall clutter suppression and image reconstruction respectively.This paper first uses the background pair cancellation method to complete the clutter suppression,and then performs a sparse variation of the radar echo signal,constructs a solution model for a linear inverse problem based on radar imaging according to the sparse representation theory,and thus uses the iterative shrinkage thresholding algorithm to optimally solve this physical problem.However,due to the computational cost and tedious iterative process of the traditional iterative optimization algorithm,the optimization solution steps are further mapped into a deep network structure to form a framework for image reconstruction based on a deep learning fusion radar imaging optimization algorithm.Experimental results show that this scheme can efficiently complete target reconstruction while balancing network complexity and reconstruction performance.Finally,a joint low-rank and sparse decomposition-driven learnable deep network for through-the-wall radar imaging is further proposed.Considering the low-rank characteristics of the wall clutter and the sparse characteristics of the target image in the through-wall-radar scene,the problem is modeled as a regularization optimization problem driven by low-rank and jointly sparse decomposition.A variational framework and a rotation strategy are used to transform the optimization problem into two quasilinear optimization sub-problems and derive their update formulas.Finally,the above iterative update formulas are mapped into a network structure and expanded into a deep iterative network model to form a learnable deep iterative network framework that integrates physical models.Experimental results show that the method is efficient in filtering wall clutter as well as reconstructing the target image,and significantly improves the target imaging accuracy and speed compared with the traditional twostage imaging method.