Research On ISAR Imaging Algorithm Based On Deep Learning

Inverse Synthetic Aperture Radar(ISAR)can achieve high-resolution imaging of moving targets,which plays an important role in the identification,discrimination and classification of targets.It has all-day,all-weather,high processing gain,strong penetrating power,etc.It plays an important role in both military and civilian fields.Because inverse synthetic aperture radar is the imaging of non-cooperative targets,imaging resolution and imaging speed are the key issues.Based on this,this paper proposes a deep learning-based imaging method,which can quickly perform high-resolution imaging result of maneuvering targets.The main contents of this paper are as follows:Firstly,the difference between ISAR and Synthetic Aperture Radar(SAR)imaging is compared.The generation of one-dimensional range image in ISAR imaging is introduced.From the data acquisition,the basic principle of ISAR imaging is studied,and the classic ISAR imaging is analyzed.The Range Doppler(RD)Algorithm is verified correctly by the simulation experiments,and the principle of convolutional neural network is briefly introduced.Secondly,an ISAR imaging method based on deep learning is proposed.The U-shaped full convolutional network structure commonly used in medical images is applied to radar imaging,and the network structure is improved for the characteristics of ISAR imaging.The improved network runs faster.A new ISAR imaging training method is proposed.It analyzes how to obtain training data and trains the simulated training set.After training,the network can effectively improve the quality of ISAR imaging.The characteristics of ISAR imaging under the condition of radar echo absence and random noise in the scene are simulated and analyzed.The training methods for these two cases are proposed.In this experiment,the algorithm can obtain a clear ISAR image.Finally,geometric modeling of the maneuvering target is carried out to analyze the influence of the motion of the object on the echo and the principle of motion compensation of the minimum entropy algorithm.Aiming at the maneuvering target,a new imaging algorithm is proposed,keep the training data and the network model unchanged.Firstly,the minimum entropy algorithm is used for motion compensation and then the neural network u-net is used for prediction.The experimental results show that the algorithm can effectively perform high-resolution imaging of maneuvering targets.