Intelligence And Anti-intelligent Processing Techniques Of Non-cooperative Signal

Non-cooperative signal processing technology has a wide range of applications in both civil and military fields.Classical signal processing methods based on statistical signal theory are stretched in the face of rapidly changing electromagnetic target signals in complex environments.With the rapid development of artificial intelligence technology represented by deep learning,many scholars have turned their attention to the research in this field to solve the current "non-" and "multiple" radiation sources facing non-cooperative signal processing with deep learning tools.The problem of“change”,but little attention to the vulnerability of deep learning,especially when it is affected by anti-intelligence techniques such as anti-samples,the deep learning model will be fatally attacked.This paper will focus on the current problems of non-cooperative signal processing.For the adaptability of complex environments,this paper focuses on improving the low SNR adaptability of non-cooperative signal detection algorithms.The idea is to fully combine the physical characteristics of the signal and the advantages of the deep learning model to adaptively extract features.Deep learning structure and data annotation method;for the problem of deep learning vulnerability,this paper analyzes the impact of the confrontation sample on the deep learning model,and on this basis,adopts the regularization method of confrontation training,which improves the robustness of the deep learning model.Based on the above ideas,there are three main innovations in this paper:1.From the perspective of gradient descent,the deep learning parameter optimization and anti-sample processing methods are unified.An electromagnetic data intelligent and anti-intelligence processing architecture is proposed,which is detailed from data acquisition and data annotation to network design and activation functions.Description2.A blind detection model based on CLDNN improved non-cooperative signal is proposed,which solves the problem that the signal blind detection algorithm can not use the signal prior information and the weak adaptability under low SNR environment.The signal detection problem is completely transformed into a classification recognition problem suitable for deep learning processing;3.A modified learning deep learning model based on confrontation training is proposed,which fully considers the vulnerability of the deep learning model.After the confrontation training,the model is more robust and can resist the attack against the sample to a certain extent.Improved recognition performance of the original sample.