Radar Image Processing Based On Convolutional Neural Network

Synthetic Aperture Radar(SAR)uses the relative motion of the radar and the target to synthesize a number of smaller real antenna aperture radars into an equivalent antenna aperture radar through data processing.SAR images have the characteristics of all-day and all-weather,and are widely used in various fields such as economy and military.SAR target recognition uses advanced radar image information to determine and identify other attributes such as target type and model.Because of its excellent characteristics,it is widely used in military battlefield reconnaissance and precision strike.With the development of the times,SAR target recognition technology has been developed more and more,and gradually matures.For the purpose of SAR image target recognition,the following work is performed:Firstly,aiming at the classification optimization method of SAR image,a new method of SAR image classification based on is proposed: Canny-WTD-CNN.The edge features extracted by Canny operator are fused with the wavelet features extracted by wavelet threshold denoising method as the input of the convolutional neural network.Softmax is used as a classifier to classify and detect SAR images.This method can well preserve the edge features of the image while removing the SAR image noise,and effectively improve the accuracy of the algorithm.Secondly,based on the original algorithm,Propose a new SAR target recognition algorithm MOPSO-WTD-Canny-CNN,the MOPSO was used to optimize the threshold in the wavelet threshold denoising algorithm and the double threshold in the Canny edge feature extraction algorithm to construct a new optimized network structure—MOPSO-WTD-Canny.While denoising,the edge information of the image is better preserved,and the optimalthreshold value found is used in the WTD-Canny-CNN algorithm for classification and recognition.Finally,the MSTAR public data set is used for simulation experiments to compare the recognition accuracy of various denoising algorithms and edge extraction algorithms after feature fusion,and the experimental comparison with other proposed algorithms shows that the algorithm has higher accuracy.