Electromagnetic Signal Target Recognition Method And Application In Open Set Scenario

Electromagnetic signal target recognition is a common technology in the information technology era,which identifies the belonging electronic equipment by extracting the electromagnetic signal characteristics.It plays a very important role for communication and signal detection,adopted in military and other scenarios.For example on the battlefield,the technology can determine whether the opposite side is using special electronic equipment by listening to the special electromagnetic signal.Moreover,the action intention can be analyzed from the detected information,including time,location and type of the target.The electromagnetic signal target recognition in the open set scenario refers to the accurate identification of specific kinds of electromagnetic signal targets(also called known class electromagnetic signal targets)in an environment where the electromagnetic signal target types are uncertain,and the rest of electromagnetic signal targets are uniformly identified as "unknown class electromagnetic signal targets".Unlike the traditional electromagnetic signal target recognition method,this scenario can only train the classifier under the condition of sample data of a given target,and unpredictable electromagnetic signal target data will appear in the testing phase.If the traditional electromagnetic signal target recognition method is directly used to train the obtained classifier,the unknown class of electromagnetic signal targets will be recognized as the electromagnetic signal targets with the most similar extracted features to them,leading to the false recognition of the unknown class of electromagnetic signal targets and reducing the recognition rate of the whole method.Although the data comprehensiveness can be ensured by collecting all electromagnetic signal target data to improve the probability of misclassification of unknown class electromagnetic signal targets,it is unrealistic in the environment of rapid development of information technology and wide variety and popularity of electronic devices.The main work of this paper is divided into three parts to address the above issues as follows:1.To address the problem that the unknown class electromagnetic signal target cannot be accurately recognized in the open set scenario,this paper uses mutual inverse points to separate the known class electromagnetic signal target and the unknown class electromagnetic signal target feature space,and achieves the recognition of the unknown class electromagnetic signal target by the threshold method.To improve the feature separation effect,this paper transforms the electromagnetic signal target recognition problem into an image-like classification problem and extracts electromagnetic signal target features using neural networks in the field of deep learning.2.For the problem of feature separation and compression of electromagnetic signal targets in open-set scenarios,this paper uses the temporal characteristics of electromagnetic signal time-frequency matrix samples,extracts the global features of electromagnetic signal targets using Tranformer series models,uses Wasserstein distance as a distance method to measure data distribution,and integrates central loss to compress the electromagnetic signal target feature space,so as to achieve optimization of feature separation and constraints.3.In order to explore the engineering value of the electromagnetic signal target recognition algorithm in the open set scenario proposed in this paper,this paper designs and implements an electromagnetic signal target recognition system integrating various open set recognition algorithms with the algorithm of this paper as the core.The system includes data set generation,sample presentation,model training,target recognition and other functions.In this paper,the above method and other advanced methods are evaluated experimentally on two datasets of cell phone electromagnetic signal and radio station electromagnetic signal,and the experimental results show that the method in this paper has achieved steady improvement in each index.This paper’s electromagnetic signal target recognition system incorporates a variety of open-set recognition algorithms,which is of high value and significance for studying algorithms and solving engineering problems.