Research And Application Of Recognition Technology For Small Sample Radar Signals

As the forerunner of electronic warfare,radar signal recognition plays a vital role in the war.With the advancement of electromagnetic technology and the continuous emergence of new radar systems,the electromagnetic environment of the battlefield has become more and more complex.Therefore,it is difficult to meet the needs of current wars by solely relying on the pulse descriptor word for identification.In addition,since the amount of signal samples that can be collected in a real battlefield environment is quite small,it is difficult to accurately reflect the true distribution of radar signals in the battlefield,and the model training is prone to overfit,the current researches mainly rely on multiple acquisitions of real signal samples to form a large sample data set to complete the experiment.However,this approach often requires a long period,and it is difficult to meet the real-time requirement of wars.Based on the above background,in this paper,the following researches are carried out on the intelligent recognition technology of small sample radar signals:1.Aiming at the problem of poor recognition performance due to the small number of samples in the radar data set,this paper optimizes the recognition effect through the sample expansion methods.This experiment establishes 10 different scale sample sets and builds small sample signal expansion models based on the SMOTE and Borderline SMOTE algorithms,which commonly used in the study of imbalance problems.Besides,the experiment achieves recognition of the expanded signals through three classifier models:decision tree,random forest and KNN.The results show that both the SMOTE algorithm and the Borderline-SMOTE algorithm can improve the recognition accuracy of small sample data sets,and the recognition accuracy of three algorithms gradually increases with the increase of the number of samples both before and after the expansion.2.Secondly,in view of the fact that the characteristics of pulse description word characters are difficult to meet the needs of small sample radar signal identification in complex environments,the paper proposes a feature extraction method.In the recognition part,the paper adopts a support vector machine classifier.The experimental result shows that this feature extraction method can extract the subtle features inside the radar signal,not only to identify different type of radar signals,but also to deeply identify the signals emitted by different radiation sources of the same type.3.From the perspective of improving the performance of the small sample classification model,this paper focuses on the improvement of traditional support vector machine classifier,to make it suitable for the multi-classification problem of radar signals.At the same time,automatic parameter adjustment is realized by combining the Grid Search method to the classifier.Compared with the traditional manual method,this method can find optimal parameters of the radar signal set more quickly and conveniently.The result shows that the classification accuracy has basically been improved.Among them,the classifier with the highest accuracy is the Gaussian kernel function classifier,and the accuracy rate after optimization reaches 99.92%.