The use of deep convolutional networks to identify hidden drone control signals in conditions of urban radio frequency noise

Abstract

This article discusses the problem of identifying hidden drone control signals in conditions of intense urban radio frequency noise using deep learning methods, in particular, deep convolutional neural networks. The relevance of the study is determined by the increasing threat of unauthorized control of aircraft in conditions of high density of electromagnetic interference typical of urban areas. The introduction substantiates the need to develop effective algorithms capable of detecting and recognizing subtle signals from the background noise. To achieve this goal, the methodological part of the article proposes a convolutional neural network architecture adapted to the specifics of radio frequency data. The stages of data collection and preprocessing, including signal segmentation, noise filtering, and the use of augmentation methods, are described. Particular attention is paid to the choice of architectural hyperparameters, activation functions, and regularization methods to minimize overfitting of the model. The experimental part is based on modeling the urban radio frequency space with the inclusion of real data, which made it possible to evaluate the effectiveness of the proposed methodology in conditions close to real ones. As a result of the study, high accuracy of identification of control signals has been demonstrated, even in the presence of significant levels of background noise. The results show that the proposed approach is capable of detecting hidden signals with an accuracy level of over 90%, which significantly increases the reliability of detecting unmanned vehicles in urban environments. The difficulties associated with the non-stationarity of the radio frequency environment are also discussed, and ways are proposed to further optimize the network to increase resistance to changing interference. In conclusion, the authors emphasize the importance of using deep convolutional networks for radio frequency monitoring tasks in the context of ensuring the safety of urban infrastructure. This research opens up prospects for the development of automated early warning systems and the detection of potentially dangerous drone control signals in difficult real-world conditions, which can be used to increase the level of protection of critical facilities and the urban environment as a whole.