Research On Radar Target Recognition Algorithm Based On High Resolution Range Profile

Since the invention of radar,it has played an extremely important role in many fields.A large number of radar technologies have been developed rapidly.People are also exploring new radar technologies.Existing radar target detection is far from meeting the needs of radar detection technology.Coupled with the increasingly mature broadband radar technology,radar is required to complete target identification.Therefore,radar target recognition technology has become the focus of research in the field of target recognition.High-resolution one-dimensional range profile(HRRP)is the vector sum of target scattering point echoes obtained from wideband radar signals projected on radar rays.It not only provides the geometric shape and structural,features of targets,but also.contains many key features of target recognition.Therefore,target recognition using high resolution range profile has become a hot spot in the field of radar target recognition.In this paper,the high resolution range profile of ship model is studied and tested in data simulation,feature extraction and target recognition based on convolution neural network.Firstly,the basic theory of high-resolution range image is studied,the imaging principle of high-resolution range image is analyzed,and the image characteristics of high-resolution range image are analyzed and discussed.Aiming at the three sensitivity problems of amplitude sensitivity,attitude sensitivity and translation sensitivity in the image characteristics of high-resolution range image,the causes of the three sensitivity problems are analyzed and corresponding solutions are proposed.The results show that the processed high-resolution range image call better solve its own sensitivity problems.Secondly,the acquisition methods of high-resolution range images are discussed.The acquisition methods can be divided into actual measurement method and theoretical simulation method.The actual measurement method includes field measurement method and compact field measurement method.There are many theoretical simulation rules,which can be mainly divided into using professional electromagnetic simulation software to build models,obtaining distance images through simulation and using point scattering models to obtain distance images of targets.In this paper,the theoretical simulation method is mainly adopted,and the electromagnetic simulation software FEKO and its corresponding high-frequency algorithm are used to simulate the high-resolution range images of ship targets under different attitude angles in the sea environment.The simulation efficiency of the high-frequency algorithm in FEKO software and the influence of polarization mode on the simulated high-resolution range images are compared.The experimental results show that the method of simulating high-resolution range profile using FEKO software is still effective and can be used in the following simulation experiments.Thirdly,the feature extraction method of high resolution range profile is studied.The principles of principal component analysis,linear discriminant analysis and kernel principal component analysis are analyzed.On this basis,a two-way two-dimensional principal component analysis method based on kernel function is proposed.Subspace features of high-resolution range images are extracted by various feature extraction methods.Finally,the feature information extracted by different methods is classified by support vector machine classifier.The classification ability of each algorithm and the influence of attitude angle change on recognition rate are analyzed and compared.The experimental results show that the algorithm proposed in this paper has greatly improved the recognition rate method,and the improvement is more obvious when the range of attitude angles is reduced.Finally,the convolution neural,network method of depth learning is introduced to recognize the high-resolution range profile of ship targets.This paper introduces the development process and working principle of convolution neural network and TensorFlow deep learning system.A complete convolution neural network is built by using TensorFlow system.Based on the original convolution neural network structure,the convolution neural network structure is improved,and a dual-channel parallel convolution neural network structure model is proposed.By optimizing the activatioh function and depth learning parameters,the deep-level attribute features of the target contained in the high-resolution range profile are fully extracted,automatic feature extraction is realized,and the optimal network structure is finally constructed.In the simulation experiment,the influence of different network configuration,activation function and learning rate on the classification ability of the convolutional neural network is analyzed,and the recognition accuracy of the convolutional neural network and the improved convolutional neural network are compared.The experimental results show that the improved convolutional neural network has higher recognition rate and robustness.