Radio frequency fine feature information extraction method and system based on semi-supervised meta learning

Abstract

The invention discloses a radio frequency fine feature information extraction method and system based on semi-supervised meta learning. The method comprises the following steps: acquiring corresponding non-label data and label data; acquiring corresponding pseudo label data according to the non-label data; updating a student network weight according to the pseudo tag data; acquiring a feedback coefficient of the teacher network; according to the feedback of the student network, obtaining the gradient change of the teacher network weight; calculating the gradient update of the teacher network according to the label data; according to the teacher network weight gradient of the non-label data under the condition of automatic enhancement processing, updating the teacher network weight, and returning the student network weight; and re-executing the steps until the execution is finished, and outputting the student network weight. According to the method, communication radiation source subtle feature information extraction and individual identification are taken as targets, and on the basis of a semi-supervised meta learning model, individual identification of large-scale equipment based on radio frequency subtle features is completed through simultaneous training and mutual promotion of a student network and a teacher network.

Description

technical field [0001] The invention belongs to the technical field of communication feature extraction, in particular to a method and system for extracting radio frequency subtle feature information based on semi-supervised meta-learning. Background technique [0002] Some progress has been made in the research on the feature perception and extraction of radio frequency radiation sources, but the current radio frequency subtle feature sensing and extraction technology cannot meet the urgent needs of the current large-scale network for network security and spectrum sensing. The specific reasons are as follows: First, the existing radio frequency subtle feature recognition methods have poor generalization ability and low accuracy; second, the current supervised learning scheme requires large-scale labeled data, which is time-consuming; third, the existing deep network model cannot Fully mine the distribution feature information behind the signal samples. To sum up, how to de...

AI Technical Summary

Problems solved by technology

[0002] Some progress has been made in the research on the feature perception and extraction of radio frequency radiation sources, but the current radio frequency subtle feature sensing and extraction technology cannot meet the urgent needs of the current large-scale network for network security and spectrum sensing. The specific reasons are as follows: First, the existing radio frequency subtle feature recognition methods have poor generalization ability and low accuracy; second, the current supervised learning scheme requires large-scale labeled data, which is time-consuming; third, the existing deep network model cannot Fully mine the distribution feature information behind the signal samples

Images