Research On Target Recognition And Classification Of SAR Image In Suppressive Jamming Environment Based On Deep Learning

Compared with optical imaging,SAR imaging has the advantages of all-weather and all-time,and plays an irreplaceable role in military and civil fields.In recent years,with the rapid development of deep learning technology,the method of SAR target recognition using convolutional neural network has been widely studied,and many achievements have been produced.In radar countermeasure,SAR system is in the circumstance of suppressing jamming,the image quality is greatly reduced,and the recognition rate of SAR target recognition system will also be reduced.Aiming at this problem,this thesis studies a method to effectively improve the accuracy of target recognition and classification based on deep learning technology in suppressing interference environment.The specific research contents are as follows:Firstly,this thesis introduces the imaging model and algorithm of SAR,describes the basic characteristics of SAR image,discusses the basic process of SAR target recognition,and analyses the basic principles of several important jamming modes of active suppression jamming.When the SAR system is in the environment of suppressed jamming,the target signal is submerged by the suppressed jamming signal,and a lot of useful information of the target will be lost in the SAR image,which has a great impact on the recognition and classification of the target.Based on MSTAR data set,two image denoising models based on depth convolution neural network,which have achieved good performance in optical image denoising,are analyzed and improved to apply in SAR image processing.The performance of deep learning networks depends not only on the network structure,but also on the data set and parameter tuning.The experimental results show that the two denoising networks based on deep learning have better performance than the traditional denoising methods.The improved conv-deconv symmetric neural network can achieve higher peak signal-to-noise ratio(PSNR),but the image is blurred and the target edge details are lost more.DnCNN with residual learning and batch normalization strategy can retain more target features while denoising,and has faster learning speed.It is suitable for deep CNN model of SAR image denoising under jamming conditions.At present,there are fewer kinds of SAR image data sets,and there is no SARimage data set under suppressed jamming conditions.In this thesis,based on MSTAR data set,ten kinds of target data sets under various noise jamming conditions are obtained.Experiments show that the recognition accuracy of SAR target recognition system based on in-depth learning decreases significantly when the signal to jamming Ratio reaches or exceeds 10 dB,and the classification ability is completely lost when the signal to jamming Ratio ratio exceeds 20 dB.In this paper,we use DnCNN network to denoise SAR images disturbed by noise,and use data without noise,added noise and denoised respectively to train network models and carry out comparative experiments.According to the experimental results,this thesis proposes a method that can significantly improve the SAR target recognition effect under the interference conditions.This method can significantly improve the performance of SAR target recognition when the signal to jamming Ratio is in the range of 30 dB,and achieve effective classification and recognition of SAR target under certain intensity of noise interference.