| Radar target recognition is a key technology of modern radar,and plays an important role in the fields of anti-missile air defense,resource exploration and national defense.Through the analysis and processing of radar echoes,the target position,velocity,radar cross-sectional area and jet engine modulation can be obtained for classification and identification of targets.Narrowband low-resolution radars are often limited by resolution to accurately identify targets.In this thesis,the feature information acquired by multiple radars is data fusion,and the redundancy and complementarity between multiple radar echo information are utilized to form a multi-angle description of the target and improve the recognition accuracy.The thesis studies radar feature modeling,radar target recognition and multi-radar data fusion algorithms.The main work is as follows:1.Research on multi-radar data fusion algorithm.The research summarizes the multi-radar data fusion algorithm and the principle of genetic algorithm,and studies the data fusion six-level model.A multi-radar data fusion algorithm based on improved adaptive genetic algorithm is proposed.The SoftSign function is used to improve the genetic operator and the neural network is used to improve the fitness function.According to the high-resolution one-dimensional range image feature of radar target,the improved adaptive genetic algorithm proposed in this thesis is used to further fuse multi-radar serialized data,and the 84-dimensional feature vector is selected to participate in the final modeling.Experiments show that the improved adaptive genetic algorithm proposed in this thesis can effectively fuse multiple radar data.The accuracy of target recognition after multi-radar fusion is significantly higher than that of single radar,and it is improved when the signal-to-noise ratio is low.The effect is more obvious.2.Research on support vector machine classification algorithm.A SAR(Synthetic Aperture Radar,SAR)image target recognition algorithm based on SURF(Speeded Up Robust Features)and support vector machine is proposed.The algorithm extracts the key points of SAR image features by SURF algorithm,and then clusters a certain number of keywords through image feature word bag algorithm.Finally,the support vector machine is used to train the classification model for keywords.The model realizes the automatic recognition function of the target in the SAR image,and uses the MSTAR(Moving and Stationary Target Acquisition and Recognition,MSTAR)radar image dataset training model to test and achieve 96.89%accurate recognition in ten target classification and recognition.rate.Experiments show that the algorithm can effectively classify SAR image targets and has the characteristics of fast training speed.3.Research on neural network classification algorithm.An improved Convolutional Neural Network(CNN)model is proposed for the characteristics of SAR images.The network model consists of basic structures such as multiple convolutional layers,pooled layers and fully connected layers,using error inverse propagation algorithm.Train network parameters.The model implements automatic recognition of targets in SAR images and uses the MSTAR radar image dataset to train models and test them.Only the original image was cropped,and the recognition accuracy was 97.32%in the ten target classification recognition.Experiments show that the algorithm can achieve target recognition function with high accuracy.4.Target JEM(Jet Engine Modulation,JEM)features and high-resolution onedimensional range image studies.The mechanism of target JEM generation is analyzed,and the principle of high-resolution one-dimensional range image and its characteristics of amplitude sensitivity,translation sensitivity and attitude sensitivity are deeply studied.In this thesis,the JEM feature model of four targets and the high-resolution one-dimensional range image model are established by FEKO simulation software,and the simulation data is generated in batches.For the JEM feature,the amplitudes of all radar JEM simulation data are normalized,and the data matrix of each radar is combined.The above data is used to compare the single radar target recognition and multi-radar fusion target recognition.Experiments show that multi-radar data fusion is also effective for JEM features,and has a better improvement effect when the signalto-noise ratio is low. |
