Ground Radar Automatic Target Recognition Algorithm And Implementation

Under battlefield conditions,obtaining radar target attributes means that the enemy's target can be determined and the degree of threat of the target can be determined,thus giving birth to target recognition technology.In view of the large size and high manufacturing cost of highresolution broadband radar,it is not suitable for the application environment of portable ground reconnaissance radar,so portable ground reconnaissance radar is usually low-resolution radar.Therefore,it is of great significance to carry out research on automatic target recognition algorithms for low-resolution ground reconnaissance radars.This paper introduces in detail the feature extraction,classifier design and hyperparameter optimization methods of radar automatic target recognition technology,as well as the signal preprocessing and data acquisition process of low-resolution ground reconnaissance radar,which provides a basis for subsequent algorithm design.For the feature extraction of the timedomain waveform and spectrum structure of the radar echo,five classifiers,namely Naive Bayes,decision tree,linear discriminant analysis,K-nearest neighbor,and support vector machine,are used to classify the samples and analyze their recognition performance.Lattice search,random search and Bayesian optimization are three hyperparameter optimization methods to find the best hyperparameter settings of the classifier,and a weighted iterative multiclassifier fusion algorithm is proposed to solve the defect that a single classifier cannot make full use of all sample features.In order to reduce the manual intervention in the process of target recognition,shallow neural network is applied to the field of ground radar automatic target recognition.Regarding the problem of its small network scale and weak model description ability,the radar echo is converted into three-channel data.Convolutional neural network feature fusion target recognition algorithm,and analyze the network recognition performance under different convolutional layer structure,activation function,optimizer and learning rate.Aiming at the problem of neural network degradation,a multi-scale residual neural network target recognition algorithm is proposed.The multi-scale residual neural network is used to achieve identity mapping to solve the problem of difficult training of deep networks,and it is combined with the feature fusion target recognition algorithm based on convolutional neural network.In comparison,the correct recognition rate of human and vehicle target samples is increased by 2%.The Bayesian hyperparameter optimization method is used to find the best hyperparameter settings of the network,which overcomes the influence of human factors in the network design process based on experience setting hyperparameters.On this basis,the selfencoder is used to reduce the data dimension to reduce the parameter calculation scale,and solve the real-time impact of target recognition after the network scale becomes larger.Finally,the ground radar automatic target recognition algorithm is realized on the signal processing platform based on DSP.According to the data in the sample library and the actual field test,the algorithm designed in this paper can achieve a correct recognition rate of more than 95%.