Intelligent Signal Sorting And Radar Working Mode Recognition Based On Deep Learning

The recognition of the radar work mode mainly includes the sorting processing of multiple mixed radar signals received by the receiver and the judgment of the work mode of a single radiator signal after sorting.With the increasing complexity of the electromagnetic environment,conventional reconnaissance methods are facing great challenges in radar signal sorting and work mode recognition.In recent years,the rapid development of deep learning,the combination of deep learning and radar reconnaissance technology will help to improve the intelligence level of radar and improve the sorting and recognition performance.In this thesis,the method of intelligent sorting and work mode recognition of radar signals based on deep learning models has been studied.The main work of thesis is as follows:1.It systematically introduces the whole process of radar signal sorting and work mode recognition,and analyzes the types of common radar signals,the characteristic parameters and traditional sorting algorithms commonly used in radar signal sorting.The typical methods for radar work mode recognition are studied,and the shortcomings of traditional methods are analyzed through experiments.2.The intelligent sorting technology of radar signals based on subtle features is studied.First,a pre-sorting method based on the amplitude envelope of different radiation sources is proposed,which can reduce the workload of main sorting.In view of the problem of poor sorting effect caused by traditional parameters when there are a large number of radiation sources,starting from the signal level,we propose subtle features that can be used for signal sorting,and use deep convolutional neural networks to take the combination of signals and features as input,experiments have verified the effect of the extracted subtle features on the sorting.The accuracy of signal sorting using subtle features is increased by an average of 7.5% than that of not using subtle features.3.The work mode recognition of multifunctional radar in the presence of missing pulses and false pulses is studied.By adding 4 feature parameters,an increased method for radar "word-phrase" modeling is proposed,which accurately describes radar words and radar phrases It realizes the correction of error pulses,and applies the correlation algorithm of data mining to the research of radar work mode recognition.Simulation experiments show that when the missing pulse rate is 40%,the model accuracy of the increased method is increased by 8%,and when the false pulse rate is 40%,the model accuracy of the increased method is increased by 6%.In the case of false pulses,the average confidence in the recognition of radar operating modes is increased by 6%.Finally,the CNN-RNN network model that uses the combination of convolutional neural network and recurrent neural network realizes the recognition of radar phrase sequences.The results show that the recognition accuracy of the CNN-RNN model is increased by 2.2% on average compared with the use of a separate network model.Compared with the traditional multi-parameter joint method and the syntactic method,the recognition accuracy is increased by 7% and 5% respectively.