Space-borne SAR Imaging Algorithm Of P-band Based On Deep Learning

P-band spaceborne Synthetic Aperture Radar(SAR)is highly penetrating and sensitive to biological matter,so it is an effective means of detecting hidden targets such as surfaces and vegetation.However,the P-band SAR signal is affected by the background ionospheric dispersion characteristics and the ionospheric irregularity and random fluctuation characteristics,which leads to the serious defocus of radar image.In radar imaging,if the resolution of radar image is improved in hardware,the production cost will be increased.Therefore,improving the image quality from the perspective of imaging algorithm becomes the most reasonable method.This paper aims to improve the imaging quality of radar affected by ionospheric effect by deep learning method.The specific content is as follows:Firstly,according to the characteristics of the ionosphere,the stratification of the ionosphere,the parameters of the ionosphere and the reference model of the ionosphere are introduced.Combined with the traditional Range Doppler Algorithm(RDA),the dispersion and scintillation effects of the ionosphere are introduced,and the phase errors of the two effects on the time-domain echo signals are analyzed.Secondly,aiming at the spaceborne SAR system crossing the ionosphere,the geometric model of spaceborne imaging is established,and an imaging algorithm based on the combination of RD algorithm and deep learning is proposed.In order to prevent the over-fitting of the network caused by the small number of radar image data samples and the gradient disappearing caused by the deepening of the network,this paper adds residual network at the connection of up-sampling and down-sampling of U-net.The effectiveness of the algorithm is proved by the test of the point target with different ionospheric parameters.Finally,for the lack of semantic information in resampling process of Res-unet,which will contain a lot of useless background information in feature connection,a neural network based on attention mechanism is proposed.Based on the characteristics of P-band imaging as an area target,the MNIST data set is taken as the data sample,and the echo of the data sample is simulated,and then the RD algorithm is used to simulate the ionospheric SAR image,which is used as the input of the network.In addition,when the two networks separately process the MNIST data set when the ionospheric effect exists at the same time,the error and peak signal-to-noise ratio graphs during the training process show that RA-unet is better than Res-unet in processing SAR images,and the training time required is less than Res-unet,which proves that RA-unet is more suitable for radar real-time image processing.