A Defense Method Oriented to Signal Sampling Gradient Attack

Abstract

The invention discloses a defense method for signal sampling gradient attack, comprising: (1) screening the wireless signal to be attacked; (2) randomly taking a number of sampling signal points in any direction in the wireless signal; The signal points are offset to obtain the estimated samples; (4) use the estimated samples to calculate the loss function and gradient information of the sampled signal points; (5) divide the important sample pixels according to the gradient information; (6) within the set pixel threshold range, the Sampling gradient attack on important sample pixels; (7) Repeat steps (2) to (6) to iteratively find the optimal disturbance and adversarial samples; (8) Add the obtained adversarial samples to the training sample set, and use the training samples Set to train the signal classification model to obtain a new signal classification model; (9) use the new signal classification model to classify the signal to be classified to achieve defense against attacks.

Description

technical field [0001] The invention belongs to the research field of deep learning algorithms and data security in the field of artificial intelligence, and specifically relates to a defense method for signal sampling gradient attacks. Background technique [0002] Inspired by neuroscience, deep learning can obtain more accurate classification results than general algorithms by learning and calculating the potential connections of large amounts of data, and has powerful feature learning and feature expression capabilities. The core of deep learning is to use a neural network with huge parameters for feature extraction. Typical neural networks include convolutional neural network (CNN) and recurrent neural network (RNN), which are suitable for processing image data and time series data respectively. [0003] In the ImageNet competition over the years, the convolutional neural network has continuously developed from AlexNet and VGGNet to ResNet and InceptionNet, showing stron...

AI Technical Summary

Problems solved by technology

[0006] For some black box models that can more accurately distinguish the type of wireless signal transfer, although the invisible internal structure has greatly increased the difficulty of attack, there are already a variety of attack methods that show strong attack capabilities against black boxes

Images