Research On Automatic Target Recognition Algorithm Based On SAR Image

The use of Synthetic Aperture Radar(SAR)Automatic Target Recognition(ATR)technology in the field of typical military target recognition and civil satellite remote sensing monitoring has made it a hot trend in related research at home and abroad.The SAR's own unique imaging mechanism causes its generated image to be affected by concomitant speckle noise,perspective shrinkage,and top-bottom inversion.The existence of these problems makes the processing of SAR images more challenging than general optical images,and the related research progress is relatively slow.This paper adopts deep learning technology and proposes some new solutions around some key problems in SAR image target recognition.First of all,this paper does not use the public network architecture to design a classification network.By processing the SAR image of the open library,normalizing the original image and other processing operations,the operation of the subsequent neural network model is simplified.Digesting and absorbing the advantages and disadvantages of the original neural network model,redesigning a network model suitable for SAR image recognition,absorbing the structure of the original deep convolution model,optimizing the parameters of the original model,and introducing hyperparameters set up.Image processing requirements for automatic identification and classification can be met.Through experimental verification,it has a good recognition function for the common coherent noise of synthetic aperture radar.The image recognition rate of the generated library reached 98.39%,and a good recognition accuracy was obtained.Secondly,for the target detection problem in SAR images,the mainstream target detection framework is adopted.Considering that the vehicle target is a small target and the background of the clutter is more complicated,the SAR target detection framework of the Faster R-CNN framework is selected,combined with the VGG16 network.Mark,train,and test the MSTAR data set.Finally,a set of SAR image ATR system with certain practical significance is formed,and the effectiveness of the system is verified by the MSTAR database image.Since the entire process does not require manual participation of feature selection,the efficiency of target recognition is improved.