Research On Airborne SAR Moving Target Imaging Technology Based On Deep Learning

Synthetic Aperture Radar(SAR)imaging technology has the advantages of all-day,all-weather and can penetrate coverings.It plays a huge role in the military and civilian fields.For SAR moving target imaging,traditional imaging algorithms can't achieve high-resolution and accurate imaging of moving targets,and problems such as blurred imaging and enlarged sidelobes will occur,which will result in degradation of imaging quality.Based on this,this paper proposes a SAR moving target imaging method based on deep learning.The main content includes the following aspects:Firstly,the imaging geometric model of airborne SAR stationary and moving targets is established.Then the article briefly explains the principle and implementation process of the classic Range Doppler(Range Doppler,R-D)algorithm and the influence of target movement on SAR imaging is analyzed.The simulation experiment is carried out.Secondly,in view of the defocus and blur problems in SAR moving target imaging,a SAR moving target imaging algorithm based on the combination of R-D algorithm and deep learning is proposed.The R-D algorithm coarsely compensates the echo data to obtain the preprocessed image.The images pass the cascaded neural network to obtain high-resolution images.In order to increase the generalization ability of the network,random point targets and area targets(MNIST data set)are used as scene targets to generate radar echo when constructing the training set.In addition,the channel data processing module is added to filter out the useless features in the moving target image information and selectively strengthen the important channel information to accelerate the network convergence speed.In order to further reflect the robustness of the cascaded neural network,simulation experiments are carried out on the situation of random noise and missing echo in the imaging scene.The test results show that the designed cascaded neural network can still obtain high-resolution SAR motion target images in a complex environment.Finally,the traditional radar imaging system(echo data domain--image domain--CNN)based on Convolutional Neural Networks(CNN)is optimized.This paper proposes a processing method from the echo data domain to CNN directly.The ?-K algorithm firstly is used to roughly compensate the radar echo data to get the two-dimensional spectrum form,and then the moving target is focused and imaged by CNN to get a clear SAR image.