Research On SAR Image Target Recognition Based On Deep Learning

Synthetic aperture radar(SAR)is widely used in military and civil Synthetic aperture radar because of its ability to monitor targets in real-time,all-time and all-weather.The traditional methods of SAR image target recognition are complicated,and the precision of target recognition is not very high.In recent years,with the development of deep learning technology,the research of SAR image target recognition has made great progress,compared with the traditional SAR image target recognition methods,the recognition accuracy and recognition rate are improved,but there are still some problems to be solved.In this paper,the method of depth learning is applied to target recognition in SAR images.A depth convolutional neural network model and a target recognition method are established and designed based on attention mechanism and other techniques,and its performance has carried on the experiment and the analysis,the main research work is as follows:(1)A new method of multi-azimuth SAR image target recognition based on joint loss function is proposed.In this method,two SAR images of the same target at continuous azimuth are added linearly to form a new fusion image,which is input into the network to train the network model,at the same time,the Softmax loss function is combined with the improved center loss function to construct a new joint loss function supervisory training process,in addition,a batch normalization layer behind the convolution layer is designed to speed up the convergence of the model and prevent over-fitting.The experimental results show that the proposed method can achieve good recognition performance under the condition of large pitch angle difference and standard operation.(2)A new method of SAR image target recognition based on multi-attention mechanism is proposed,which improves the channel attention mechanism and the spatial attention mechanism,and enhances its generalization,furthermore,the Dropout and residual shrinkage network is introduced into the residual module,which can avoid over-fitting and reduce the effect of noise in SAR image to some extent.The experiment verifies the effectiveness of the algorithm by adding different degrees of random noise to the MSTAR data set and reducing the number of training samples,and under the condition of extended operation,good recognition results are obtained,which shows that the proposed method has good stability.(3)Combining the theoretical results with the practical application,a SAR image target recognition system is designed and implemented,which includes three functional modules:login interface,visual image selection and image recognition.By uploading the SAR image to the system and selecting the corresponding recognition algorithm module,the image classification can be recognized quickly and effectively.Compared with manual recognition,the recognition efficiency is greatly improved.Finally,the research work of this paper is summarized and prospected.