Research On Image Super-resolution Reconstruction Algorithm Based On Wall-penetrating Radar

The radar imaging method is a kind of detection imaging technology with good penetrability and high stability.It has a wide range of applications in many military and civilian fields such as criminal detection,house overhaul,etc.However,the current radar imaging system still has the following limitations:the scene of target detection in obstacles is not fully considered;the body is bulky and the portability is poor;the existing imaging method is greatly interfered by the electromagnetic wave reflection of the wall,and the imaging of the weak target is blurred.Therefore,this paper proposes a portable wall-to-wall radar imaging system,and optimizes the radar imaging model based on deep learning,and finally proves the effectiveness of the model through experiments.First,for the scene where the detection target is inside the wall,the compact ultra-wideband frequency modulated continuous wave(FMCW)radar is selected as the detection front end.At the same time,in order to make up for the problem of the FMCW radar's single scanning range being too small,this paper designs an automated scanning imaging system based on the FMCW radar.The system consists of three parts:detection front-end equipment,data acquisition preprocessing,and image feature reconstruction.It realizes the detection and imaging of the target in the wall by the portable radar system in flexible scenarios,and the size of the scanning area is controllable.In addition,in order to improve the imaging accuracy of the FMCW radar wall detection system and solve the problem of image details and texture blur,this paper studies the super-resolution algorithm based on deep learning and proposes an image super-resolution reconstruction scheme based on the radar imaging system.The scheme deeply analyzes the principles of residual learning and dense connection algorithm,and designs a new dual-channel coupling network BPNCSR based on the dual-channel network architecture,which greatly improves the texture reconstruction and detail restoration of images.At the same time,on the basis of BPNCSR,this paper designed a radar image-based super-resolution reconstruction network RadarNet,using convolutional neural network to extract features from regions of radar multiple scans,and combined with a dense and residual algorithm to suppress wall's echo and noise,finally,the multi-output features are fused through the feature fusion module to realize the feature reconstruction of the entire target image.This paper has conducted sufficient experiments on the methods proposed above,trained the network by combining simulation and real data,and completed the experiments in real scenarios.The results show that the BPNCSR network architecture proposed in this paper is superior to similar mainstream methods.Wall-penetrating radar imaging system and RadarNet algorithm can well reconstruct the internal characteristics of the wall,and can obviously suppress the echo and noise information of the wall.In addition,through the use of command control and multi-state calculation,this paper deploys the improved network model in a dedicated acceleration module based on FPGA design to complete the verification and implementation of RadarNet on mobile platforms.