Specific Emitter Identification of small sample comvmmunication radiation source based on IQ graph features

 In a real battlefield environment, it is impossible for us to collect enough local radiation source data. Small sample learning becomes more and more important. Through continuous development, CNN neural networks have a strong ability to process image classification. Under the condition of small samples, in order to use the most mature CNN neural network, this paper proposes a method of converting one-dimensional IQ data into two-dimensional IQ graph features to perform classification tasks for small samples. Due to the repeatability of the IQ map of the data and the difference of the individual, this method has 99.5% accuracy in identifying different individuals on ultrashort wave radio stations. Compared with the bispectrum feature, the IQ map feature has a strong generalization ability. This method has a simple feature transformation, and the CNN network has a mature technology for processing picture classification, which has strong practicality.