The Research Of Radar Signal Classification Based On Machine Learning Algorithm

Radar signal sorting is the key technology in radar countermeasure reconnaissance system,one of the most important measures of radar countermeasure information processing technology is signal sorting level.While trying to improve the measurement accuracy of radar receiver parameters,efficient and accurate sorting method is also the focus of research.In recent years,machine learning algorithm has become a hot topic in various fields,and has achieved many practical and effective results,and it has also greatly promoted the rapid development of artificial intelligence field.How to apply machine learning algorithm to signal sorting,which is based on traditional radar signal sorting technology,has become one of the important research directions for solving radar signal sorting in complex electromagnetic environment at present.For the problem that parameter overlap is serious and the conventional three-parameter(pulse width PW,angle of arrival DOA,carrier frequency RF)sorting method has low sorting accuracy,this paper studies the wavelet packet feature extraction method and extracts a very good class.The intra-pulse characteristics of the radar signal within the characteristics of intra-clustering degree and inter-class separation degree,that is,the wavelet packet characteristic Wpt6,and the specific extraction steps are analyzed.Similar to conventional parameters,the signal modulation mode is the same,with only two radiation sources with different characteristics,this paper,from the angle of radar transmitter,extracts the dual spectral features(also called individual features)of radar signals and optimizes them by using the double spectral diagonal slice which is advantageous to the recognition and processing of machine learning algorithm,Finally,using support vector machine(SVM)algorithm to select multiple diagonal slice data,simulation results show that this feature can effectively distinguish between the two very small radiation source signals.Aiming at the problem that the accuracy rate is not high by using single sorting algorithm,this paper proposes the SOFM-kmeans algorithm which combines the self-organizing feature map neural network(SOFM,Self-Organizing Feature Map)and the k-means algorithm to select the conventional three parameters.This method mainly adopts the idea of first division and then,and the two algorithms are complementary to each other,which can reduce the complexity of time domain and have more robust classification and recognition ability.The simulation results show that the accuracy of the SOFM-kmeans algorithm is improved compared with the two kinds of single algorithms.In order to determine the size of sorting before SOFM network sorting,in this paper,an improved algorithm of SOFM network with scale self-tuning is put forward,and on the basis of conventional three parameters,it is put forward that the parameters of pulse Wpt6 participate in the signal sorting,and the results of simulation of different parameters are compared with the same algorithm,the simulation results show that for conventional three-parameter overlap is particularly serious,and even very close to the problem of the accuracy of sorting is very low,the added signal pulse characteristics effectively improve the sorting accuracy rate.