SAR Image Target Recognition And Classification Based On Deep Learning

Synthetic Aperture Radar(SAR)is a kind of advanced microwave imaging equipment,it can still get high-resolution images,which has the advantages of all day,all-weather,high processing gain and strong anti-jamming ability,especially in the harsh environment where the optical remote sensing equipment can not work normally,such as rainy days,haze and other environments.SAR system can work in harsh conditions and its unique imaging characteristics make it have important application value in both military and civil fields.In recent years,along with the development,of remote sensing technology,satellites can provide high-resolution images,which provides more abundant information for the research of target recognition algorithm based on SAR.Compared with the traditional image information,SAR image does not have a rich spectrum and the data size is often limited.The traditional image processing technology for target detection relies too much on artificial feature extraction,and its robustness and generalization ability are relatively weak.With the continuous development of deep learning in computer and other fields,SAR image recognition based on deep learning has attracted more and more attention from researchers at home and abroad.In this paper,SAR image target recognition and classification based on deep learning are deeply studied.Firstly,the hybrid method of SAR image feature extraction and matching feature is studied,which can be used for SAR image automatic registration more accurately and effectively,Finally,the SAR image target recognition and classification method based on Fast.er R-CNN is studied to further improve the recognition and classification efficiency.The main contributions of this paper are as follows.1.We propose a hybrid method of SAR image feature extraction and matching feature.SAR image is difficult to express intuitively,and it has special imaging mechanism and speckle noises.In this paper,a hybrid method of feature extrac-tion and matching is proposed.This method combines Gaussian Guidance Filter(GGF)with scale invariant feature transform to obtain blob features,and it ob-tains angle features from GGF.At the same time,fast full consensus algorithm and complete graph method arc combined to match features,which effectively re-duces speckle effect.Experimental results demonstrate that the proposed method can effectively improve the accuracy of SAR image recognition.2.We propose a SAR image target recognition method based on improved Convolutional Neural Network(CNN).In the traditional CNN method,there is a problem that the convergence rate is very slow.In this paper,we use the new method of transferring parameters to train CNN network,and connect the clas-sifier in the last layer of CNN to make the output layer.Finally,we do further training fine-tuning on the parameters training to find the global best.Exper-iment al results demonstrate that the proposed method has better recognition effect and success rate than Support Vector Machine(SVM)and feedforward neural network.3.We propose a Faster R-CNN SAR image target recognition and classification method.In order to improve the accuracy and efficiency of deep learning method in SAR image target recognition and classification,this paper proposes a Faster R-CNN method model.In this method,a Regional Proposal Network(RPN)is introduced to predict the object range and score of the location,and it can generate high-quality regional suggestions,so as to improve the overall target recognition and classification accuracy.The experimental results demonstrate that the recognition accuracy and efficiency are improved.