Communication Specific Emitter Identification Based on Transfer Learning

Abstract

In the application of specific communication emitter identification(CSEI), the emitter slight features will change along with different times, places and conditions, which makes training samples and test samples obey different distribution and makes most machine learning algorithms not work well. To solve this problem, this paper introduced transfer learning to CSEI. Compared with traditional machine learning, transfer learning does not relax the assumption that the data is independent and distributed. It can explore useful knowledge from the data which is different from the target domain but similar to the source domain. It can also learn the classifier when the labeled samples in the same domain are insufficient. Extensive experiments show the effectiveness of the new method.