Individual Recognition Based On Transfer Learning For Wireless Network Device

EasyChair Preprint 1659

Abstract

Due to the complexity of the electromagnetic environment and the non-cooperative characteristics of the target in the process of individual recognition of wireless network devices, the amount of data acquired is small, and the data required for training the model cannot be enough. For this small samples situation, a new method of individual recognition of wireless network devices based on transfer learning is proposed. We combine a convenient signal processing method with transfer learning in this paper. The acquired signal is divided into source domain and target domain. We extract the feature of transient signal and get the weight of feature.Then we transfer the knowledge of the source domain to the target domain. The training model is reconstructed by adjusting the weight of the source domain samples, which solves the problem of insufficient training samples in the small samples situation. The results show that in the case where the number of target domain samples is less than 30, the recognition rate has improved obviously.It enriches the application of transfer learning in the field of individual recognition.

Keyphrases: Transfer Learning, individual recognition, small samples

@booklet{EasyChair:1659,
  author    = {Zhibing Yu and Ruilun Zong and Shuangqiu Li},
  title     = {Individual Recognition Based On Transfer Learning For Wireless Network Device},
  howpublished = {EasyChair Preprint 1659},
  year      = {EasyChair, 2019}}