Design And Implementation Of Signal Recognition Method Based On Prototype Network And Incremental Learning

Radio frequency fingerprint identification has been the subject of in-depth study and interest as a high-security identity authentication technique.The main focus of current research in this area is the use of device RF fingerprint differences to distinguish between trusted and malicious devices in addition to wireless device access authentication.The RF fingerprinting technique in the conventional closed-set scenario is only appropriate for detecting signals of known types of equipment and is unable to identify signals of unknown types of equipment that appear in test samples.Open-set RF fingerprinting is the classification and identification of signal transmitters in an open-set environment using RF fingerprinting technology.The open-set signal identification algorithm’s inadequate resilience,insufficient recognition accuracy,and high cost of manual labeling data are still issues that the present open-set RF fingerprint recognition is still dealing with.The primary contributions of this thesis are as follows:1.The robustness and recognition accuracy of conventional closed-set recognition models are poor when applied to the issue of signal detection of unknown types of equipment in open-set scenarios.This thesis designs a semi-supervised prototype network-based open-set recognition model.In order to lower the cost of manual labeling,a lot of unlabeled data is used in the semi-supervised learning method.It is also used to strengthen the creation of sample prototypes and improve the robustness and accuracy of the model’s identification of known classes in open-set scenarios.The soft threshold approach is also designed in this thesis to increase the validity of the open set recognition model for the discrimination of unknown classes.2.Unknown signals must be used as new knowledge supplementary input to the open-set recognition model for learning in order to classify and recognize unknown types of equipment signals in open-set scenarios.However,the open-set recognition model will have catastrophic forgetting problems,showing this thesis designs an incremental learning algorithm based on fine-tuning to address the phenomenon of declining signal recognition ability,ensuring that the model can learn new knowledge to expand the ability of signal recognition while recognizing the performance of old tasks.3.An intelligent management system for wireless devices is created on the basis of the novel features and useful applications of the algorithm model previously mentioned.The system’s primary duties include displaying the access authentication equipment’s signal,recognizing and detecting input signals,and managing the access authentication equipment intelligently.This thesis designs an open-set recognition model for identifying known device signals and detecting unknown device signals in the open-set scenario.The original scheme has a security flaw,so it’s important to increase the open-set recognition model’s capacity for recognition.As a result,this thesis designs an algorithm based on expanding and recognizing new unidentified device signals through fine-tuning incremental learning.The open-set recognition model put forth in this thesis serves as the foundation for the design of a wireless device intelligent management system for signal recognition.