Research On Intelligent Sorting And Recognition Of Radar Emitters

Radar emitter sorting and recognition is an important part of radar reconnaissance system,which plays an important role in the collection of intelligence and estimation of threats in radar countermeasures.As the electromagnetic environment becomes more and more complex,the density of radar pulse stream increases,and the modulation mode of signal increases,the data processed by radar emitter sorting and recognition becomes more and more complex and the data volume becomes larger and larger,which leads to the degradation of the effectiveness and accuracy of the traditional sorting and recognition methods.In addition,the traditional sorting methods are based on the measured Pulse Descriptive Word and intra-pulse modulation characteristics designed and calculated by hand.The sorting results are affected by artificial design features and parameter measurement errors,and the robustness of the method is poor.How to improve the performance of radar emitter sorting and recognition system,while reducing the reliance on a priori knowledge and manual experience and improving the intelligence,is the focus of current research in this field.In order to improve the performance and intelligence of radar emitter sorting and recognition,this thesis analyzes the defects of the sorting and recognition methods based on machine learning,and on this basis,focuses on the application of artificial intelligence methods in radar emitter sorting and recognition.Based on the deep clustering and knowledge graph theory in the field of artificial intelligence,this thesis realizes the sorting and model identification of radar emitter signals.The main research contents of this thesis are as follows.Aiming at the problems that the traditional sorting algorithm requires manual extraction of intra-pulse features,the extracted features contain less signal information,and the mismatch between feature distribution and clustering algorithm,a radar emitter sorting method based on deep clustering is proposed.The method is based on the theory of deep clustering,using two auto-encoder networks to extract parameter features and intra-pulse features of radar signals,and contact the two types of features to form joint features,and then using k-means clustering algorithm to cluster the joint features to achieve signal sorting.The method can automatically extract signal features using neural networks,reducing manual intervention.At the same time,combining clustering and feature extraction,alternating network finetuning and sample clustering enables the network to learn features that are easy to be distinguished by the clustering algorithm,improving the accuracy of sorting.Experiments show that the method has higher correct sorting rate and better performance of anti-noise than the traditional method under the same conditions.In addition,to address the problem that k-means clustering requires a preset number of clusters,this thesis proposes to use the density peak clustering algorithm to estimate the number of clusters,and experiments show that the number of clusters can be correctly estimated using the Density Peaks Clustering algorithm under the condition of high signal-to-noise ratio.Aiming at the problems of poor interpretability and low recognition rate of traditional recognition algorithms based on machine learning,a radar emitter model recognition method based on knowledge graph is proposed.It mainly includes two aspects:(1)the construction of knowledge graph in the field of radar emitters.Starting from the data composition of the signal,the radar emitter knowledge system is firstly constructed,and then entities and relations are extracted from the structured radar emitter data,and finally the knowledge is represented in the form of a triad to obtain the radar emitter domain knowledge graph,and the Neo4 j graphical database is used to store and represent it;(2)the model identification of radar emitter.The Trans H model is used to learn the entities,relationships and relationship planes in the knowledge graph,mapping them into vector form,and by calculating the score function of the model,the "radar model" attribute of each sample is inferred,so as to complete the identification of the radar emitter model.This method visualizes the radar emitter database and improves the interpretability and intelligence of the identification process.The experiments show that the recognition performance of the method is better than that of the traditional method,and the analysis of the score function value calculated by Trans H can further realize the recognition of unknown radar emitter models and automatic expansion of the knowledge graph.