Research On SAR Imaging Algorithm Based On Convolutional Neural Network

Synthetic aperture radar(SAR)is a kind of high-resolution imaging radar system which is not affected by time and climate.It can not only recognize camouflage and penetrate the cover,but also observe the ground all day and all day.It plays an important role in both military and civil fields.The traditional SAR imaging algorithm is easy to be affected by the external environment and the moving state of the target.Its imaging speed is slow,the imaging quality is reduced,and the recognition degree of the target image is reduced.So it is very important for SAR imaging system to improve the resolution of target image and realize fast imaging.Based on this,this paper proposes an imaging method based on depth learning,which can quickly image the target with high resolution.The main contents of this paper are as follows:First of all,the basic principle of airborne SAR is introduced.Through the establishment of SAR still echo model,the echo expression of the target is obtained,and the imaging process of SAR moving target is simply described.The classic Range Doppler algorithm(R-D)and the SAR imaging algorithm based on Oversampling smoothing(OSS)are analyzed.The algorithm flow is given respectively,and the simulation experiment is carried out.Secondly,the network layer of the convolutional neural network structure is briefly described,and one of the classic networks,u-net,is elaborated in detail.According to the characteristics of SAR imaging,the network structure is adjusted,the network model is optimized,and the training data set is obtained through simulation.The trained network can improve the SAR imaging resolution.The simulation of the scene with noise and the absence of echo signal are carried out respectively,and the ideal target images are obtained.Finally,the echo expression of the moving target in the two-dimensional frequency domain is deduced,and the influence of the moving target on the imaging result is explained by analyzing the expression.Aiming at the moving target in the strong background,a new imaging method is proposed.Displaced Phase Center Antenna(DPCA)technology is used to preprocess the imaging result of the algorithm,and then the trained neural network is used as an imaging processor to realize the refocusing of moving objects.