The Research On Electromagnetic Leakage Detection And Recognition Method Of Computer Keyboard Based On Deep Learning

The keyboard is a commonly used computer input device,which inevitably generates electromagnetic leakage during the process of key input,resulting in the risk of information leakage.The existing methods for detecting and identifying keyboard electromagnetic leakage usually require analysis of electromagnetic signals based on known electromagnetic leakage information characteristics and specific environmental conditions,with weak noise resistance and insufficient generalization ability.This article is based on the working mechanism of PS/2and USB keyboards,analyzes the principle of electromagnetic leakage,studies the characteristics of leakage signal types and leakage paths under different conditions,and proposes a targeted detection and recognition algorithm based on deep learning:1.A keyboard electromagnetic signal detection algorithm based on GRU-CN is proposed for the detection and recognition of electromagnetic signals during keyboard communication,which intelligently detects and recognizes the original electromagnetic signals of the keyboard.The algorithm first uses endpoint detection,which is based on the time-domain features of keyboard electromagnetic signals.Its purpose is to preliminarily locate electromagnetic signal samples,eliminate redundant information in the samples,and improve the efficiency of subsequent model training;We have designed a one-dimensional convolutional neural network suitable for feature extraction of keyboard electromagnetic signals,and introduced a gated recursive unit to enhance the network’s memory of long sequence information of keyboard electromagnetic signals.This to some extent solves the problem of gradient vanishing during the process of signal feature extraction,and can more efficiently extract useful information features hidden in the signal,thus obtaining an effective classification model for recognizing keyboard electromagnetic information;A keyboard electromagnetic information localization method was designed,which is based on an improved gradient weighted class activation mapping algorithm and utilizes the algorithm’s visualization function to achieve automatic calibration of red information in keyboard electromagnetic leakage signals.This solves the problem of manually calibrating the position of electromagnetic signals in traditional keyboard electromagnetic leakage signal detection methods.2.In response to the problem of low signal-to-noise ratio and high difficulty in detecting and identifying keyboard leakage signals in complex electromagnetic environments,an algorithm called SADAN is proposed for denoising keyboard leakage signals.The algorithm is based on adversarial game theory,utilizing generative adversarial networks and self-attention mechanisms to continuously enhance the discriminative and denoising abilities of the discriminator and denoiser during the learning process.The SADAN algorithm can better learn the dependency relationship of keyboard electromagnetic leakage signal sequences during the denoising process,model the long-distance relationship between sample points in the sequence,reduce the impact of network layers on relationship acquisition,and simplify the optimization of network parameters.This algorithm achieves noise suppression in keyboard leakage signals,improves their signal-to-noise ratio while retaining effective information,and ultimately obtains denoised leakage signals that can be effectively detected and recognized.The experimental method of simulating and verifying the electromagnetic leakage detection and recognition of computer keyboards based on deep learning proposed in this thesis has been proven to have high accuracy in identifying the original electromagnetic signals of keyboards,thereby improving the signal-to-noise ratio of noisy signals and achieving effective recognition of keyboard leakage signals.