Electromagnetic Big Data Retrieval Algorithm Based On Deep Learning

In both military and civilian fields,thousands of electromagnetic signals are usually received in spectrum monitoring,signal reconnaissance,electromagnetic information security detection and other tasks.Searching for some specific electromagnetic signals among the numerous and complicated signals is like looking for a needle in a haystack for the detection personnel.In the current era of big data,isolated analysis of a signal is no longer in line with the demand.It is often necessary to accurately and quickly find similar signals that users are interested in from a large number of signals,so as to extend the isolated study of a signal to the overall study of the change trend of a class of signals.However,traditional signal reconnaissance usually uses hand-made features to characterize electromagnetic signals,but these descriptors are still limited and cannot meet the requirements of automatic reconnaissance in the current complex and changeable electromagnetic environment.Compared with manually extracting features,deep neural networks can better obtain the inherent features of electromagnetic signals,and the extracted features are more abundant and accurate.In this paper,the retrieval algorithm of electromagnetic signal is studied based on deep learning method.The specific research content is as follows:Firstly,the structure of deep convolutional neural network is studied and analyzed in detail,the structure of typical convolutional neural network is discussed.This paper describes the concept and application scenarios of metric learning,focuses on the excellent performance of deep metric learning in various fields,and the definition and application scenarios of various typical distance metric functions.Secondly,the image state electromagnetic signal retrieval method based on siamese convolutional neural network is studied.This paper first constructs the electromagnetic signal model,uses the signal’s constellation diagram as the medium and performs feature enhancement operations on it to visualize the electromagnetic signal into a two-dimensional image.Then,the image state electromagnetic signal retrieval model architecture was built,and the siamese convolutional neural network composed of the improved Alex Net network was used to extract the features of paired signal images.The loss was measured by calculating the similarity of feature vectors and the distance of labels in the embedding space.At the same time,the influence of SNR and distance measure function on the retrieval results is studied,and the experiment proves that the proposed model has excellent and stable retrieval performance.Finally,the sequential electromagnetic signal retrieval method based on complex network is studied.First,the signal is preprocessed,the modulating signal dataset constructed in this paper is stored in the form of I\Q two channels,and the signals are assembled into appropriate tensor data.By studying the key components of complex neural network complex convolution kernel,the complex fully connected layer,the complex batch normalization layer and the complex weight initialization,referring to the VGG network model,this paper designs a complex VGG network structure for electromagnetic signal sequence retrieval.The influence of complex network method and real network with the same structure on signal retrieval results is studied.The experimental results show that with the increase of SNR,the signal retrieval model composed of complex VGG is optimized more quickly in the signal retrieval task of this paper,and the retrieval performance is more stable.