Research On Super-resolution Imaging Method Of Vehicle-mounted FMCW Radar

With the increasing demand for travel safety,on-board radar has received widespread attention in the detection and application of road traffic safety.Vehicle mounted millimeter wave radar has the advantages of large signal bandwidth,high measurement accuracy,low manufacturing cost,and strong anti-interference ability,which has good commercial value and broad development prospects.However,in practical applications,due to adverse factors such as hardware size,weather factors,and motion blur,the resolution of images generated by onboard FMCW radar has decreased,which cannot meet the requirements of imaging applications.Therefore,studying the super-resolution imaging method of vehicle mounted FMCW radar has important research significance.Based on the actual scientific research project,the thesis studies the radar super-resolution imaging algorithm based on compressed sensing theory and the radar image super-resolution reconstruction algorithm based on depth learning from the perspective of processing radar echoes and radar images.The specific research contents are described as follows:1)Aiming at the problem of low angle resolution of vehicle borne FMCW radar,the theory of compressed sensing is applied to radar super-resolution imaging.Thesis establish a measurement matrix in the array plane dimension,use the OMP algorithm to reconstruct the signal,and analyze the advantages and limitations of the OMP algorithm.2)Research has been conducted on signal sparse reconstruction algorithms,and an improved OMP algorithm has been proposed to address the limitations of selecting the correct atoms and requiring scene sparsity as prior information in each iteration of the OMP algorithm.The main improvement lies in the introduction of a threshold judgment mechanism,which replaces scene sparsity with the maximum and minimum contrast of the signal,and determines whether the current selected atom is correct in each iteration,ensuring the accuracy of the reconstructed signal.This thesis conducted simulation experiments and field testing experiments.The improved OMP algorithm can also accurately image in unknown scene sparsity,verifying the effectiveness of the improved OMP algorithm in improving radar angle resolution and enhancing the applicability and portability of the algorithm.3)In response to the problem of low spatial resolution of images generated by vehicle mounted FMCW radar,the principle and structure of the USRNet super-resolution reconstruction network were analyzed and studied.The noise level and fuzzy kernel were introduced into the degradation model,and the image super-resolution reconstruction was achieved using iterative data and prior terms under the MAP framework.On this basis,the USRNet++super-resolution reconstruction network is proposed to further combine the low-frequency and high-frequency features of radar images for learning.The experimental analysis verifies that the USRNet++network and USRNet network have better super-resolution reconstruction performance than the traditional super-resolution methods for fuzzy degraded radar images,and the reconstructed radar images have more texture features and details,which is more suitable for super-resolution reconstruction processing of actual vehicle radar chart.4)On the basis of USRNet and USRNet++,in order to address the problem of long training time,the number of channels in each layer of the network during the training process is adjusted.USRNet tiny and USRNet++tiny networks are proposed,which sacrifice the reconstruction effect to a limited extent while improving training efficiency.