Research On Ground Radar Target Classification And Recognition Technolog

The acquisition of target information is of great significance on the battlefield,and the technology of radar target classification and recognition is born from this,which has a wide range of application in the military and civil fields in the 21 st century of information globalization.Low resolution radar is still the mainstream at home and abroad due to its advantages of low cost,simple equipment and wide application.Therefore,it is worthwhile to study the target classification and recognition technology of low resolution ground radar.This paper mainly studies the targetclassification and recognition algorithm of low resolution ground radar,then analyzes the echo of personnel,vehicles and UAVs to define 24 kinds of characteristic parameters such as time domain waveform irregularity and power spectrum entropy.Decision tree,naive bayes,k-nearest neighbor,linear discriminant analysis and support vector machine are used for multi-target classifier design and simulation.The super parameters of the classifiers are optimized by grid search,random search and Bayesian optimization methods.Aiming at the support vector machine which is difficult to carry out multi classification directly,a tree structure support vector machine classification algorithm based on bayesian optimization is proposed.Considering the characteristics of multiple classifiers,a multi-level fusion multi-objective classification algorithm based on score decision is proposed to solve the problem of low recognition rate caused by poor classification performance of a single classifier and insufficient utilization of sample features.In order to further improve the classification accuracy,a single hidden layer back propagation neural network with multiple output which uses time-domain echo as the input is used to realize the task of ground radar multi-objective classification and recognition.In view of the limited description ability caused by its simple model,taking the time-domain echo,power spectrum and power transform spectrum as three channel inputs,a feature fusion multitarget classification and recognition algorithm based on convolutional neural network is proposed,and the effects of different network structure,activation function,optimizer and other parameter settings on the recognition performance are compared and analyzed.For the degradation and over fitting problems caused by network deepening,a fusion feature multiobjective classification algorithm based on multi-scale wide residual network is proposed.The identity mapping network with multi-scale branches is used to reduce the over fitting risk,the convolution kernel is widened to reduce the network depth and ensure the network performance.Other super parameters of the network are set by bayesian optimization method.Compared with the algorithm based on convolutional neural network,the accuracy is improved by 3%.In view of the problem that the real-time performance of classification is reduced due to many parameters and large network scale,the auto-encoder is used to reduce the dimension of data and improve the ability of classifier to select parameters independently.Finally,this paper implements the ground radar target classification and recognition algorithm on the domestic signal processing platform.Through the experiments of actual outfield test,the accuracy of the algorithm proposed in this paper can reach more than 94%.