Research On Individual Radiator Recognition Based On Deep Learning

As an important part of modern electronic warfare and intelligence systems,individual identification technology of radiation sources is an important guarantee for improving national security and improving electronic warfare capabilities.In order to realize the rapid and accurate identification of radiation source individuals,and to solve the problems that the traditional radiation source individual identification features are not obvious,cannot be well adapted to streaming data,and the traditional individual identification generalization ability is insufficient,this paper develops a radiation source based on deep learning.Research on source individual identification technology.Aiming at the lack of feature extraction ability of shallow network and the uncertainty of the influence of feature extraction method on signal,a research on individual identification method of radiation source based on deep learning network is proposed.Bi-spectral transform,IQ two-way analysis,short-time Influence of features such as Lie transform and Hilbert-Huang transform on individual identification of radiation sources.The simulation results show that the individual identification technology of radiation sources based on deep learning has the advantages of strong generalization ability and no need for human intervention,and the identification accuracy is better than that of traditional methods.Aiming at the problem that the individual data of the radiation source has a strong dependence on features,this paper adopts a method of feature fusion of four features,such as Hilbert-Huang transform,short-time Fourier transform,fuzzy function and Bi-spectral transform.It also has a good inhibitory effect on noise.Aiming at the problems that the traditional radiation source individual identification methods cannot process streaming data well,the training cost is high,or the storage data consumes resources,this paper proposes a radiation source individual identification method based on incremental learning.Based on the convolutional neural network,the original model is obtained by training the existing data.When new data arrives,the incremental learning method based on regularization or playback is used to perform incremental training on the data.The simulation results show that the method proposed in this paper effectively utilizes the training results of the old model,and also has better recognition performance for new types of data,greatly reducing the required storage space and training time,and improving the anti-noise ability.In view of the insufficient generalization ability of individual radiation source identification,it is difficult to transfer to other radiation source individual identification fields and the "cold start" and "slow convergence" of the model,this paper proposes a radiation source individual identification method based on transfer learning.The model parameter freezing and MMD error are combined with the migration learning method of classification error.In this paper,the original irrelevant or weakly related source domain and target domain are brought closer to reduce the distance between the target and the source domain,and the previously learned knowledge of the source domain is transferred to the target domain,without the need for the target domain to be re-solved for optimization.Process.Finally,in order to improve the overall network performance,a hyperbolic space module is added before the network output layer to improve the network’s expressive ability and the ability to mine information between data.After iteration,the model after transfer learning is finally obtained.The simulation results show that this method improves the generalization ability of the model and solves the problem of slow convergence,and the training accuracy is also improved to a certain extent.