Research On Cluster Analysis Of Communication Reconnaissance Data

As the electromagnetic battlefield environment becomes more and more complex,a vast amount of reconnaissance data of the parameter-level is intercepted and reported by communication reconnaissance equipment continuously.To mine the potential information embedded in the reconnaissance data of the parameter-level,the techniques for analyzing and processing reconnaissance data play a significant role in modern information war,and become more and more important in controlling battlefield and capturing information advantage.To satisfy the requirement of wireless network countermeasure in the battlefield,clustering analysis methods of reconnaissance data,which are used to mine the hidden information,are studied for the cases of different battlefield environments.The main work of this thesis is summarized as follows.1.A two-stage clustering algorithm based on the SOFM network is proposed in this thesis.The algorithm can achieve the communication reconnaissance data clustering under the condition that the clustering number is unknown.This algorithm consists of two stages: rough clustering based on SOFM network and fine clustering based on k-means algorithm.In the rough clustering stage of the proposed algorithm,the initial clustering results are obtained based on the weight distribution of the winning neurons using the self-organizing learning process of the SOFM network.In the second stage,the weight centers of the winning neurons and clustering number are determined by the method of density peak,and the winning neurons are clustered by the k-means algorithm.Then the clustering results are mapped to the original communication reconnaissance data.Simulation experiments show that the two-stage clustering algorithm based on the SOFM network can achieve the clustering effectively under the condition that the clustering number is unknown,and the clustering accuracy is higher.2.A method for the clustering of irregular data with noise based on spectral clustering is proposed to achieve the communication reconnaissance data clustering in the case that the distribution of communication stations is irregular geographically.Due to the task,geographical environment and other factors,irregular distribution of communication stations may make traditional clustering algorithm ineffective in clustering.Moreover,the existing outlier detection algorithms to eliminate noise in the communication reconnaissance data under this condition have poor performance,so it needs a kind of outlier detection algorithm to detect irregular shape data sets and complex distributed multidimensional data sets.Therefore,this thesis proposes an outlier detection algorithm based on similarity pruning to eliminate noise in the communication reconnaissance data.Then the data without noise is clustered through the spectral clustering algorithm.The simulation experiments show that the method can effectively eliminate the effect of irregular location distribution of communication stations and communication reconnaissance data with noise and has better clustering results.3.A method for communication reconnaissance data analysis based on semi-supervised spectral clustering is studied.This method can be used in the case that the non-corporative side has some prior knowledge in the reconnaissance data analysis.It has significance to make full use of this prior knowledge to guide the clustering process of communication reconnaissance data.Firstly,the format of prior knowledge known by the non-corporative side is analyzed.Then,the prior knowledge is used to determine the constraints of the communication relationship between the communication stations,and the pairwise constraints between the communication stations are expanded.Finally,by semi-supervised spectral clustering method using the constraints of the relationship between the communication stations,the clustering process of communication reconnaissance data is completed.Simulation experiments show that the method can take advantage of the prior knowledge known by the non-corporative side effectively to assist communication reconnaissance data clustering process,and improve the accuracy of communication reconnaissance data clustering.