Research On Key Techniques Of Synthetic Aperture Radar Target Recognition

In the fields of daily production,economic development and national security,SAR image target recognition has been widely and profoundly applied to these fields.Therefore,it is of great significance to study the target recognition technology of SAR images.This dissertation is based on the background of realistic requirements,and studies the method of SAR image target recognition.This paper will start research from the following aspects:1.We analyzed the traditional SAR image target recognition technology and its development in detail,and learned the advantages and disadvantages of the entire target recognition process.At the same time,we are clear that the feature extraction part can be improved in the entire process.In addition,we made a detailed analysis of deep learning techniques and their development,and it is clear that deep learning techniques can be applied to target recognition tasks of SAR images.Based on this,we introduced the concept of end-to-end.In other words,the model based on neural network algorithm is used for target recognition of SAR images.2.Based on the common full-connection neural network,we used the neural network model based on convolution unit to identify the target of SAR image,and we used the data under the condition of SOC and EOC-1 on the MSTAR data set to validate the model.In addition to the different models,the improvement also includes the connection between the final feature output layer of the model and the classifier,from the traditional full connection to the convolutional connection,and we made a comparison of the two on the data set.The recognition results found that the accuracy of the model identification of the convolutional layer is higher than the other one,which verifies the effectiveness of the improved model in this chapter.3.We made optimization on the basis of convolutional neural network.Neural network model based on residual unit is used to identify target of SAR image,and we used the data under the condition of SOC and EOC-1 on the MSTAR data set to validate the model.We made comparisons with the method described in the previous chapters.At the same time,to increase the robustness and generalization ability of the model,the improvement is made to fuse the model described in the previous chapter with the one in this chapter.It is found that the accuracy of the fused model is higher,and the method used in this chapter can be verified to increase the recognition accuracy of the model.4.Based on the residual unit-based network,we tried another way of optimization.The neural network model based on the dense convolution unit was used to identify the target of the SAR image,and the data under the SOC condition was used to verify the effect of the model on the MSTAR data set.But the recognition result obtained is not very ideal.5.The algorithm used in the full text is summarized,and the future research on SAR image target recognition is discussed.This dissertation has studied and improved the SAR image target recognition algorithm,compared the experimental results under different models,and has certain significance for the development of SAR image target recognition technology.