Research On Small Sample Radar Emitter Recognition

Radar emitter recognition is to extract features by receiving the electromagnetic signal sent by the emitter,and determine the signal emitter according to the prior information.However,traditional recognition method based on template matching has been difficult to meet the needs of emitter recognition in modern war.Moreover,when the pulse repetition rate is low or the target is an advanced non cooperative target,it is difficult to obtain a large number of samples.At present,most of the research on emitter recognition is based on multiple samples to form a large sample data set.In the small sample environment with less data,the emitter recognition rate is low because a small number of samples can not well reflect the target characteristics.Based on the above background,Aimde at the small sample environment and development status of emitter recognition methods,this paper studies three key technologies of sample expansion,feature extraction and classifier design in emitter recognition in small sample environment.The effectiveness of the proposed method is verified by a real sample set.The main work is summarized as below:(1)Aimed at the problem of insufficient data size of the small samples,SMOTE algorithm is applied to study the imbalance problem into small sample expansion,and uses the weighting strategy to give weight according to the Euclidean distance between a few samples and other samples,synthesize different numbers of new samples,and expand small samples to form large samples from the data level,and compares the application of machine learning algorithm in ten different rules Recognition performance under modular small sample data set.The experimental results show that the improved SMOTE improves the quality of extended samples and the classification effect;(2)Aimed at the problem that it is difficult for the conventional features to reflect the target characteristics in the small sample environment,at the feature extraction level,a radiation source in pulse feature extraction method combining transient and steady-state features is proposed,and the random forest recognizer is used for recognition.According to the experimental results,the feature extraction method can extract the deep essential features of the radiation source signal and identify the same type of radiation source It shows good recognition performance.(3)Aimed at the problem of model over fitting in small sample environment,in the aspect of classifier design,this paper proposes improved random forest classifier through weighting strategy according to the characteristics of small samples,evaluate the recognition effect of decision tree with kappa coefficient,and give weight according to this,so that the decision tree with high recognition performance has greater voting weight in the voting stage,so as to realize the application of random forest algorithm in small samples The recognition performance is improved.Experimental results on real data sets show that the improved weighted random forest algorithm improves the recognition performance of small samples compared with the unmodified algorithm.