Research On Deep Belief Network Model-based Radar Signal Classification

Deep learning is a hot research direction in the field of machine learning.Deep learning has powerful data representation ability.Its essence is to extract the features contained in the data by constructing a multi-hidden layer structure of neural network and a large number of data,so as to improve the accuracy of classification or prediction.Radar signal classification is one of the key technologies of radar signal processing.With the emergence of new radar system,the traditional signal classification methods have achieved worse and worse results.Deep learning can automatically learn data features,which is of great significance in the field of signal classification.The specific research work of this paper is as follows:Firstly,aiming at the problem of low data dimension and low classification accuracy when using Pulse Description Word(PDW)parameters to classify radar signals,this paper proposes a method to increase the data characteristics of the parameters to be classified by using existing PDW parameters and processing certain data without adding other measurement parameters,so as to improve the classification accuracy.The simulation results show that this method can significantly reduce the classification error rate.Secondly,aiming at the problem of how to determine the optimal parameter setting of Deep Belief Network(DBN)model under PDW parameters,a lot of simulation experiments have been carried out,and the effects of network layers and nodes,learning rate,training sample size and training times on the classification results have been analyzed.A set of definite parameter settings can be obtained,which can ensure that the model can basically play a good role in classification.The RMSProp algorithm in the adaptive learning rate is added to the DBN model,which not only accelerates the training speed of the model,but also improves the classification accuracy.Finally,in view of the good generalization ability of support vector machine,DBN model and support vector machine are combined to achieve better classification results.In addition,according to the principle of ensemble learning,two kinds of DBN model frameworks based on ensemble learning are proposed.One is to integrate multiple DBN models with different classifiers,the other is to add classifiers to each restricted Boltzmann Machine(RBM)output position under the same DBN model,and then integrate them.Thesimulation results show that the accuracy of the two methods is higher than that of the original DBN model,and the prediction results are more stable.