Research On SAR Ship Target Recognition Method Based On Deep Learning In Complex Number Domain

The imaging and recognition technology of sporting ships has an important role in many fields.In maritime management,SAR can monitor and manage ships entering and leaving the port throughout the day;in marine environmental protection,it can detect marine pollution caused by oil pollution in time.It can detect and warn enemy ships in military.Traditional SAR moving target recognition mainly uses the amplitude information of SAR images to manually extract features and combine with classifiers to perform moving target recognition.For a ship target moving at sea,in addition to its own sailing motion component,there is also a three-dimensional rotation component.Especially in high sea conditions,the motion component causes the target to defocus and submerge it in the clutter background,based on the amplitude image domain.SAR moving target recognition method is difficult to detect.In this paper,we study the multi-dimensional information perception method of SAR ship target based on CV-CNN.Without the motion compensation of the target,using the different expressions of the ship's moving target and the static target in the complex domain,CV-EstNet and CV-Motion Net,CV-Refocus Net,CV-RotNet four complex domain networks respectively complete SAR motion ship target speed estimation,SAR motion ship target identification,SAR three-dimensional rotating ship target refocusing,SAR three-dimensional rotating ship target identification.First of all,this article first studies the basic imaging processing methods of SAR,and performs simulation analysis of the algorithm.Aiming at the echo simulation of the ship target,the moving ship target echo model is analyzed,the Doppler center change and Doppler modulation frequency change of the moving target echo are studied,and a SAR ship based on the combination of ray tracing and 3D model is proposed Target imaging model,and simulation and simulation to build a SAR ship target simulation sample library.Based on the high score No.3 SAR image and the corresponding AIS data to form a SAR ship target measured sample library.Secondly,study the complex domain CNN network architecture and basic theory,elaborate the basic principles of deep learning,real domain convolutional neural network and complex domain convolutional neural network,based on CNN and CV-CNN respectively carry out SAR stationary ship target recognition experiment,give Comparative analysis of the performance of th e two networks.Then,this paper analyzes the problem of moving ship target offset and defocus.EstNet and CV-EstNet are designed from the perspective of real domain network and complex domain network,respectively,to achieve the speed estimation of SAR moving ship target,and complete Corresponding simulation data experiments give comparative analysis of the performance of the two networks.At the same time,on the basis of EstNet and CV-EstNet,two networks,Motion Net and CV-Motion Net,are designed to realize the target recognition of SAR moving ships,and the corresponding simulation and measured data experiments are completed,and a comparative analysis of the performance of the two networks is given.Finally,this paper analyzes the three-dimensional rotation characteristics of the ship target at sea,studies the effect of the three-dimensional rotation component on the Doppler frequency of the SAR ship target,and then designs Refocus Net and CV-Refocus Net from two perspectives of the real domain networ k and the complex domain network.In order to realize the refocusing of the SAR three-dimensional rotating ship target,and complete the corresponding simulation and measured data experiments,the comparative analysis of the performance of the two networks is given.At the same time,on the basis of Refocus Net and CV-Refocus Net,two networks,RotNet and CV-RotNet,are designed to realize SAR three-dimensional rotating ship target recognition,and analyze and compare the performance of the two networks.