Research Of Radar Signal Recognition Method Based On Ensemble Deep Learning

Deep learning has strong ability of data expression as a hot algorithm in current research.It not only can overcome the defect of the poor generality of the traditional neural network architecture,but also can learn the important features by the trained model,which are difficult to extract for some artificial methods.Therefore,the intelligent technology based on deep learning has been developed rapidly and has a wide application space,which has brought great impetus to the progress of various fields of society.Modern radar emitter signals have the characteristics of variable parameters,various forms and complex rules of change.There are often weaknesses of large amount of computation,complicated process and low efficiency in conventional signal sorting recognition methods,which seriously affect the recognition of radar signals in complex systems.Deep learning relies on multi-layer deep network model,which can learn abundant implicit information.As a result,it is of great significance to use deep learning to automatically learn and identify a large amount of complex radar emitter pulse data.The specific research work of this article is as follows:1、The development status,merits and drawbacks of existing radar signal recognition methods are introduced,as well as the development and basic principles of deep learning algorithm.For the existing artificial feature extraction methods of radar signal,it has some problems like the difficulty to quickly extract the core feature,complex calculation and limited expression ability.A signal recognition method based on deep belief network model is proposed,which can automatically learn and analyze signals,so as to get information that can reflect the essential characteristics of signals.It can improve the performance of radar signal sorting and recognition.2、It improves the depth model by stacking generalization learning thought,and builds an integration method of multi-layer deep belief network to avoid parameter optimization,and dispose of existing issues,including the lack of single model learning ability and low degree of accuracy.By studying simulated pulse signals from eight different radars,it can class and identify those signals by integrating the learning results of multiple models stacked.Compared with the original depth model,this method can effectively improve the accuracy of signal recognition.Compared with other signal classification methods,the experimental results show that this algorithm can achieve better classification results,and then reveals the effectiveness and superiority.3、By linear integration of the posterior probability of each layer of depth model and learning from the loss function and regularization parameter,a new data set is obtained.The final classification results are determined by the decision layer to class the data set,so as to further improve the performance of the model.By comparing the results of the linear ensemble with other different classification methods in the simulation experiments,it shows that this measure can not only improve the recognition rate,but also enhance the stability and overcome the overfitting problem to some extent.