General countermeasure disturbance generation method based on electromagnetic signal modulation type identification of automatic encoder

Abstract

The invention discloses an automatic encoder-based general countermeasure disturbance generation method for electromagnetic signal modulation type identification, and the method comprises the steps: building a signal classification model, and enabling a signal data set to be outputted in a high-precision prediction manner; obtaining the structure and weight parameters of the model, randomly sampling a signal sample from the training set, generating corresponding signal confrontation disturbance by using a Deepfool white box attack algorithm, splicing the signal confrontation disturbance into asignal disturbance matrix, inputting the signal disturbance matrix into an automatic encoder for training, and obtaining output data of an encoding layer after training; the data can reserve global features of original data, so that the classification precision of the classification model is greatly reduced.

Description

technical field [0001] The invention relates to a general anti-disturbance generation method based on the identification of the electromagnetic signal modulation type of an automatic encoder, and the invention belongs to the field of deep learning security. Background technique [0002] With the rapid development of deep learning, it has been widely used in many fields. For example, applications in fields such as face recognition, unmanned driving, and classification of radio signal modulation types are affecting and changing our lives all the time. However, the security issues of deep learning are also gradually emerging. As a complex system, deep learning is also subject to threats from all parties. Such threats mainly include the following three aspects: First: model stealing, hackers through certain The technical means to illegally steal the models deployed on the user's local or central server. Second: data poisoning, by adding abnormal data to the training data, the ...

AI Technical Summary

Problems solved by technology

[0003] For the generation of adversarial samples, the general method used is to use certain adversarial attack algorithms to generate corresponding adversarial samples, that is, a normal sample corresponds to an adversarial sample. Such adversarial sample generation requires a lot of time and expensive calculations. resource

In addition, such an adversarial example generation method needs to obtain all input samples, which is difficult to achieve in many cases

Images