CN114372493B - Computer cable electromagnetic leakage characteristic analysis method - Google Patents
Computer cable electromagnetic leakage characteristic analysis method Download PDFInfo
- Publication number
- CN114372493B CN114372493B CN202210030092.2A CN202210030092A CN114372493B CN 114372493 B CN114372493 B CN 114372493B CN 202210030092 A CN202210030092 A CN 202210030092A CN 114372493 B CN114372493 B CN 114372493B
- Authority
- CN
- China
- Prior art keywords
- leakage
- electromagnetic
- layer
- signal
- neural network
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000004458 analytical method Methods 0.000 title claims abstract description 23
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 26
- 238000000034 method Methods 0.000 claims abstract description 22
- 238000001514 detection method Methods 0.000 claims abstract description 19
- 238000013528 artificial neural network Methods 0.000 claims description 37
- 238000012549 training Methods 0.000 claims description 35
- 238000004364 calculation method Methods 0.000 claims description 30
- 238000013145 classification model Methods 0.000 claims description 22
- 238000013527 convolutional neural network Methods 0.000 claims description 22
- 230000006870 function Effects 0.000 claims description 19
- 238000011176 pooling Methods 0.000 claims description 15
- 238000012545 processing Methods 0.000 claims description 12
- 239000013598 vector Substances 0.000 claims description 12
- 238000010606 normalization Methods 0.000 claims description 11
- 108010076504 Protein Sorting Signals Proteins 0.000 claims description 9
- 238000007781 pre-processing Methods 0.000 claims description 9
- 230000009466 transformation Effects 0.000 claims description 7
- 238000000605 extraction Methods 0.000 claims description 6
- 238000001228 spectrum Methods 0.000 claims description 6
- 230000004913 activation Effects 0.000 claims description 5
- 230000008569 process Effects 0.000 claims description 4
- 238000006243 chemical reaction Methods 0.000 claims description 3
- 238000013507 mapping Methods 0.000 claims description 2
- 230000006872 improvement Effects 0.000 description 8
- 238000002474 experimental method Methods 0.000 description 7
- 238000012360 testing method Methods 0.000 description 6
- 238000002790 cross-validation Methods 0.000 description 5
- 238000011156 evaluation Methods 0.000 description 4
- 238000007637 random forest analysis Methods 0.000 description 4
- 238000005070 sampling Methods 0.000 description 4
- 230000008859 change Effects 0.000 description 3
- 238000010586 diagram Methods 0.000 description 3
- 230000000694 effects Effects 0.000 description 3
- 238000011478 gradient descent method Methods 0.000 description 3
- 241000282326 Felis catus Species 0.000 description 2
- 238000013135 deep learning Methods 0.000 description 2
- 238000010801 machine learning Methods 0.000 description 2
- 101150116295 CAT2 gene Proteins 0.000 description 1
- 101100326920 Caenorhabditis elegans ctl-1 gene Proteins 0.000 description 1
- 101100126846 Neurospora crassa (strain ATCC 24698 / 74-OR23-1A / CBS 708.71 / DSM 1257 / FGSC 987) katG gene Proteins 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 238000007635 classification algorithm Methods 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 238000013480 data collection Methods 0.000 description 1
- 238000013400 design of experiment Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000005670 electromagnetic radiation Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000005284 excitation Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 210000005036 nerve Anatomy 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 238000011084 recovery Methods 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/02—Preprocessing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/12—Classification; Matching
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- General Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Biophysics (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Computational Linguistics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Signal Processing (AREA)
- Complex Calculations (AREA)
Abstract
The invention discloses a computer cable electromagnetic leakage characteristic analysis method, which comprises the following steps: identifying leakage sources, judging whether information leakage exists or not, and analyzing leakage information content; according to the invention, under the condition of unknown electromagnetic signal characteristics, electromagnetic signals can be analyzed, leakage sources of the electromagnetic signals and hidden leakage information therein can be identified step by step, a layered detection strategy of electromagnetic information leakage is realized under a unified algorithm and technical framework, compared with the traditional method for analyzing the electromagnetic leakage information through operation experience and priori knowledge, the electromagnetic leakage information analysis efficiency is improved, under the condition of low signal-to-noise ratio, the characteristic judgment leakage sources in the signals can be accurately extracted, the optimal value solving speed of gradient descent in a model is higher by adding a batch standardization layer, meanwhile, the fitting problem is effectively avoided, and the electromagnetic leakage information detection efficiency and accuracy can be higher under the condition of small samples by adopting an integrated learning method.
Description
Technical Field
The invention relates to the technical field of electromagnetic leakage analysis, in particular to a computer cable electromagnetic leakage characteristic analysis method.
Background
Along with the rapid development and wide application of electronic equipment, the information security problem of the equipment is increasingly remarkable, and researches find that when the electronic equipment such as a computer, a numerical control system and the like processes information, internal current changes can generate electromagnetic radiation, electromagnetic signals are emitted to the outside, the non-communication electromagnetic signals which are emitted unintentionally carry useful information in the equipment and can be detected and reproduced through a reduction technology, so that sensitive information leakage is caused, the information security is endangered, various cables connected with the computer are main ways for generating electromagnetic conduction leakage, and the electromagnetic signal characteristics of the computer cable leakage are analyzed, so that a judgment basis can be provided for electromagnetic information leakage detection;
At present, because the coding modes used by various cables in information transmission are different, theoretically, each leaked electromagnetic signal should also have different characteristics, but noise interference exists in the actual electromagnetic environment, the leakage signal characteristics of the computer cable are often difficult to distinguish, so that hidden information in the leakage signal is more difficult to identify, and the analysis of the electromagnetic leakage characteristics of the cable is inconvenient.
Disclosure of Invention
Aiming at the problems, the invention aims to provide a computer cable electromagnetic leakage characteristic analysis method, which adopts a deep learning convolutional neural network algorithm to establish a convolutional neural network structure model for electromagnetic information leakage analysis, can analyze electromagnetic signals under the condition of unknown electromagnetic signal leakage sources, accurately predicts the electromagnetic signal leakage sources, improves electromagnetic leakage information analysis efficiency compared with the traditional method for analyzing electromagnetic leakage information through operation experience and priori knowledge, and can accurately extract characteristics in signals to judge the leakage sources under the condition of low signal-to-noise ratio.
In order to achieve the purpose of the invention, the invention is realized by the following technical scheme: a computer cable electromagnetic leakage characteristic analysis method comprises the following steps:
Step one: identifying a source of leakage
Firstly, acquiring a leaked electromagnetic signal through a signal acquisition device, preprocessing a time signal sample by using fast Fourier transform, designing a convolution neural network structure comprising two one-dimensional convolution layers, two pooling layers and a full-connection layer, respectively carrying out model training on signal data on a time domain and a frequency domain and a continuous observation domain, then inputting the electromagnetic signal to be tested into a trained convolution neural network structure model, extracting layer by layer characteristics of the neural network, outputting prediction scores of classifications at the full-connection layer, and finally judging an electromagnetic information leakage source through calculation;
Step two: judging whether information leakage exists or not
Firstly, electromagnetic leakage signals of a leakage source with electromagnetic leakage are collected through a signal collection device, then a fast Fourier transform is used for preprocessing time signal samples, then a convolution neural network structure containing batch standardization layers is designed, model training is respectively carried out on signal data on two observation domains of a time domain and a frequency domain, electromagnetic signals to be tested are input into a trained convolution neural network structure model, layer-by-layer characteristic extraction of the neural network is carried out, various prediction scores are output at a full connection layer, and finally whether information leakage exists in the leaked electromagnetic signals is determined through calculation;
step three: analyzing leakage information content
Firstly, collecting electromagnetic signals leaked from a leakage source through a signal collecting device, and then designing an integrated learning method based on a convolutional neural network to analyze the content of leakage information: firstly, preprocessing a time domain signal sample by using fast Fourier transform to obtain a frequency domain signal representation of the sample, and training frequency domain signal data to obtain a corresponding base classification model; then further calculating the frequency domain signal by using cepstrum analysis to obtain a cepstrum signal of the sample, and training the cepstrum signal to obtain a corresponding base classification model; then, combining the base classification models in a weighted mode to construct a new model, then inputting an electromagnetic signal to be detected into the new model, extracting layer-by-layer characteristics of a neural network in the model, and then performing electromagnetic information leakage detection by singly deciding and outputting the weighted combined base classification models, wherein the training 2 base classification models all adopt the same convolutional neural network structure and comprise a one-dimensional convolutional layer, a batch standardization layer, a pooling layer and a full-connection layer, and the integrated learning weighting operation calculation formula outputted by the full-connection layer is as follows:
wherein: y c represents the final score of the c-th classification, k represents the corresponding base classification model, S is the output of the fully connected layer, ω k is the weight of the base classification model.
The further improvement is that: the fast fourier transform formula is:
Wherein: x (N) is a time domain signal sequence, X (k) is a frequency spectrum signal sequence after calculation, and N is the length of the signal sequence;
The cepstrum analysis formula is:
Wherein: the data after Fourier transformation is divided into a real part and an imaginary part, X (e jw) is a real part of X (n) after Fourier transformation, and Ceps (n) is a calculation result of real cepstrum.
The further improvement is that: the input of the convolutional neural network structure is a frequency spectrum signal, the length is 16384, the convolutional network transmits the data of the input layer to the convolutional layer and the pooling layer, after feature extraction and conversion, the feature vector with 64 channels and the length of 1024 is input into the full connection layer, and finally the full connection layer gathers the features and outputs the result.
The further improvement is that: the convolution function calculation formula of the convolution layer is as follows:
Wherein: And N is the index number of each layer in the neural network, X (N) and X (N+1) are the characteristic vectors input and output by the Nth layer in the convolution calculation process, W (N) is the weight vector of the Nth layer neural network, and B (N) is the bias vector of the Nth layer.
The further improvement is that: the convolution neural network structure adopts a ReLU as an activation function, and the calculation formula of the ReLU function is as follows:
ReLU(x)＝max(0，x)
after the ReLU function, a Dropout function is used, the probability value is set to be 0.5, so that the overfitting of the neural network is restrained, then the maximum pooling is used for downsampling, the characteristics are screened, the network parameters are reduced, and the maximum characteristic value in the pooling domain is output.
The further improvement is that: the full connection layer uses SoftMax as an activation function, and the calculation formula of the SoftMax function is as follows:
Wherein: i is an element number for performing an operation, and S i represents a mapping of the i-th element in the (0, 1) section.
The further improvement is that: the convolutional neural network structure normalizes data by using a batch normalization method after convolutional operation, and the batch normalization algorithm is as follows:
S1, calculating the average value of data output by the previous layer
S2, calculating standard deviation of output data of the previous layer
S3, normalization calculation
S4, reconstructing the normalized data
Wherein: x is the output result of the upper layer, m is the size of the training sample batch, epsilon is a small value close to 0, the situation that the denominator is 0 is avoided, and beta and gamma are learning parameters.
The further improvement is that: the signal acquisition device is composed of calipers and signal processing equipment, wherein the calipers are used for collecting electromagnetic leakage information, and the acquisition precision of the signal processing equipment is 10 bits.
The beneficial effects of the invention are as follows: the invention establishes a convolutional neural network structure model for electromagnetic information leakage analysis by adopting a convolutional neural network algorithm of deep learning, can analyze electromagnetic signals under the condition of unknown electromagnetic signal leakage sources, accurately predicts the electromagnetic signal leakage sources, improves the electromagnetic leakage information analysis efficiency compared with the traditional method for analyzing the electromagnetic leakage information by operating experience and priori knowledge, can accurately extract characteristics in signals to judge the leakage sources under the condition of low signal-to-noise ratio, has the accuracy reaching more than 95 percent, ensures the speed of gradient descent solving the optimal value in the model to be faster by adding a batch of standardized layers, simultaneously avoids the problem of overfitting, and can also have higher electromagnetic leakage information detection efficiency and accuracy under the condition of small samples by adding more leakage sources, thereby optimizing the network structure and parameters of the convolutional nerves to obtain higher accuracy, increasing the generalization capability of the model, ensuring that the model is more efficient and stable, and bringing convenience for identifying hidden information in the leakage signals.
Drawings
In order to more clearly illustrate the embodiments of the invention or the technical solutions of the prior art, the drawings which are used in the description of the embodiments or the prior art will be briefly described, it being obvious that the drawings in the description below are only some embodiments of the invention, and that other drawings can be obtained according to these drawings without inventive faculty for a person skilled in the art.
FIG. 1 is a flow chart of a method according to a first embodiment of the invention;
Fig. 2 is a flowchart of an integrated learning method based on a convolutional neural network according to a first embodiment of the present invention;
Fig. 3 is a diagram showing a collection mode of electromagnetic leakage information according to a second embodiment of the present invention;
FIG. 4 is a graph showing the loss rate of two network structure training according to the second embodiment of the present invention;
FIG. 5 is a graph showing the accuracy of two network structure training according to the second embodiment of the present invention;
FIG. 6 is a schematic diagram of a D-CNN structure according to a second embodiment of the invention;
FIG. 7 is a schematic diagram of the structure of D-CNN-BN according to the second embodiment of the invention;
FIG. 8 is a graph showing the comparison of test accuracy in the second embodiment of the present invention.
Detailed Description
The following description of the embodiments of the present invention will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present invention, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
Example 1
Referring to fig. 1 and 2, the embodiment provides a computer cable electromagnetic leakage characteristic analysis method, which includes the following steps:
Step one: identifying a source of leakage
Firstly, acquiring a leaked electromagnetic signal through a signal acquisition device, preprocessing a time signal sample by using fast Fourier transform, designing a convolution neural network structure comprising two one-dimensional convolution layers, two pooling layers and a full-connection layer, respectively carrying out model training on signal data on a time domain and a frequency domain and a continuous observation domain, then inputting the electromagnetic signal to be tested into a trained convolution neural network structure model, extracting layer by layer characteristics of the neural network, outputting prediction scores of classifications at the full-connection layer, and finally judging an electromagnetic information leakage source through calculation;
The signal acquisition device consists of a caliper and signal processing equipment, wherein the caliper is used for collecting electromagnetic leakage information, and the acquisition precision of the signal processing equipment is 10 bits;
the fast fourier transform formula is:
Wherein: x (N) is a time domain signal sequence, X (k) is a frequency spectrum signal sequence after calculation, and N is the length of the signal sequence;
Step two: judging whether information leakage exists or not
Firstly, electromagnetic leakage signals of a leakage source with electromagnetic leakage are collected through a signal collection device, then a fast Fourier transform is used for preprocessing time signal samples, then a convolution neural network structure containing batch standardization layers is designed, model training is respectively carried out on signal data on two observation domains of a time domain and a frequency domain, electromagnetic signals to be tested are input into a trained convolution neural network structure model, layer-by-layer characteristic extraction of the neural network is carried out, various prediction scores are output at a full connection layer, and finally whether information leakage exists in the leaked electromagnetic signals is determined through calculation;
step three: analyzing leakage information content
Firstly, collecting electromagnetic signals leaked from a leakage source through a signal collecting device, and then designing an integrated learning method based on a convolutional neural network to analyze the content of leakage information: firstly, preprocessing a time domain signal sample by using fast Fourier transform to obtain a frequency domain signal representation of the sample, training frequency domain signal data to obtain a corresponding base classification model, and then further calculating the frequency domain signal by using cepstrum analysis to obtain a cepstrum signal of the sample, and training the cepstrum signal to obtain a corresponding base classification model; then, combining the base classification models in a weighted mode to construct a new model, then inputting an electromagnetic signal to be detected into the new model, extracting layer-by-layer characteristics of a neural network in the model, and then performing electromagnetic information leakage detection by singly deciding and outputting the weighted combined base classification models, wherein the training 2 base classification models all adopt the same convolutional neural network structure and comprise a one-dimensional convolutional layer, a batch standardization layer, a pooling layer and a full-connection layer, and the integrated learning weighting operation calculation formula outputted by the full-connection layer is as follows:
Wherein: y c represents the final score of the c-th classification, k represents the corresponding base classification model, S is the output of the full connection layer, ω k is the weight of the base classification model;
The cepstrum analysis formula is:
Wherein: the data after Fourier transformation is divided into a real part and an imaginary part, X (e jw) is a real part of X (n) after Fourier transformation, and Ceps (n) is a calculation result of real cepstrum.
The input of the convolutional neural network structure is a frequency spectrum signal, the length is 16384, the convolutional network transmits the data of the input layer to the convolutional layer and the pooling layer, after feature extraction and conversion, the feature vector with 64 channels and 1024 length is input into the full-connection layer, and finally the full-connection layer gathers the features and outputs the result.
The convolution function calculation formula of the convolution layer is as follows:
Wherein: And N is the index number of each layer in the neural network, X (N) and X (N+1) are the characteristic vectors input and output by the Nth layer in the convolution calculation process, W (N) is the weight vector of the Nth layer neural network, and B (N) is the bias vector of the Nth layer.
The convolution neural network structure adopts a ReLU as an activation function, and the calculation formula of the ReLU function is as follows:
ReLU(x)＝max(0，x)
after the ReLU function, a Dropout function is used, the probability value is set to be 0.5, so that the overfitting of the neural network is restrained, then the maximum pooling is used for downsampling, the characteristics are screened, the network parameters are reduced, and the maximum characteristic value in the pooling domain is output.
The convolutional neural network structure normalizes data by using a batch normalization method after convolutional operation, and the batch normalization algorithm is as follows:
S1, calculating the average value of data output by the previous layer
S2, calculating standard deviation of output data of the previous layer
S3, normalization calculation
S4, reconstructing the normalized data
Wherein: x is the output result of the upper layer, m is the size of the training sample batch, epsilon is a small value close to 0, the situation that the denominator is 0 is avoided, and beta and gamma are learning parameters.
Example two
The system is characterized in that VGA cables, computer power lines, HDMI cables and unshielded twisted pair cables are used as leakage sources to be detected, data acquisition and testing are carried out in a normal indoor working environment, electromagnetic interference of other electronic equipment is not shielded, a signal acquisition device consists of A.H.systems BCP-620 calipers and signal processing equipment with the model of NI PXIe-5162, the calipers are used for collecting electromagnetic leakage information, the acquisition precision of the signal processing equipment is 10 bits, the acquisition device is mainly used for processing and storing signals, the acquired equipment comprises an LCD display with the model of DELL OptiPLex-3240, an AOC I2490PXC display with an HDMI interface, a host computer and an HP notebook computer, and the electromagnetic leakage information acquisition mode is shown in figure 2;
Collecting electromagnetic leakage samples of four different cables, and under the premise of not considering the recovery leakage information, only identifying leakage source characteristics, wherein the sampling rate is 1M/s, the collected electromagnetic leakage signals are time domain sequences, the signal samples are preprocessed by fast Fourier transform, wavelet transform and cepstrum respectively, and model training is carried out on data on four observation domains of a time domain, a frequency domain, a wavelet domain and a cepstrum domain respectively, and the signal sample length is 16384;
through data collection and processing, each data set on each observation domain contains 5472 electromagnetic information leakage samples, as shown in table 1 below:
Table 1 dataset
| Cable category | Number of samples |
| HDMI cable | 1368 |
| VGA cable | 1368 |
| Computer power line | 1368 |
| Unshielded twisted pair | 1368 |
| Totalizing | 5472 |
Dividing the electromagnetic signal samples in the table 1 into four types according to different cables, marking class labels, wherein the number of samples in each type is 1368, and dividing the electromagnetic signal samples into a training set and a testing set according to the ratio of 4:1;
The leakage source detection adopts the convolutional neural network structure (D-CNN) designed by the invention, as shown in fig. 6, the model is trained by using a supervised learning mode, a gradient descent method is used for iterating the model, the prediction result gradually approaches to a real label, the iteration number (Epochs) of model training is set to 200, 5-fold cross validation is adopted, the average value of 5 rounds is taken as the basis of performance evaluation, the evaluation index of algorithm performance is the accuracy and recall rate, and the algorithm is compared with the classical machine learning classification algorithm limit gradient lifting (Extreme Gradient Boosting, XGBOOST), random Forest (RF) and naive Bayes Bayes, NB) are shown in table 2 below.
Table 2 comparison of leak source detection algorithm performance
As can be seen from table 2, the accuracy and recall of the D-CNN model is significantly better than XGBOOST, RF, and NB models, especially in the frequency domain, where the D-CNN model can obtain optimal performance;
The method provided by the invention can extract the electromagnetic signal characteristics and judge the electromagnetic leakage source on the premise of unknown electromagnetic signal leakage source.
Adopting the same data acquisition mode as in FIG. 3, using a sampling rate of 250M/s to acquire electromagnetic information leakage data of VGA cables and power lines, selecting black-white text images, and using 10 physical pictures of an airplane and a cat in a public dataset CIFAR-10 as excitation of information leakage, and defining electromagnetic signals in a black screen state and a white screen state as samples without information leakage;
Because the electromagnetic signal data volume of a frame of image is overlarge under the sampling rate of 250M/s, the neural network is overlarge, the calculation cost is greatly improved, the redundancy of image information is considered, in order to balance the algorithm execution speed and the detection precision, the signals are downsampled, the downsampled data is subjected to fast Fourier, wavelet and cepstrum transformation, the data of four different observation domains are obtained, and the data distribution is shown in the following table 3:
Table 3 dataset
Classifying black-and-white text images and physical images into the types with leakage information, classifying black-and-white screen images into the types without information, and using 4:1 is divided into a training set and a testing set, and 10% of the training set is taken as a checking set in training;
The present invention improves on the existing MGCNN convolution structure by adding a batch normalization layer (Batch Normalization, BN) between the one-dimensional convolution layer and the ReLU layer, and naming D-CNN-BN as shown in fig. 7.
Comparing XGBOOST, RF, NB, MGCNN with D-CNN-BN five algorithms, training the data in a supervised learning mode, performing iterative training on the model by using a gradient descent method, and using the accuracy as an evaluation index of algorithm performance, wherein the result is shown in the following table 4:
Table 4 comparison of leakage information detection algorithm performance
As can be seen from the results of Table 4, the convolutional neural network algorithm is superior to other machine learning algorithms, wherein the D-CNN-BN method provided by the invention has the best performance;
By combining with leakage source detection, the detection performance on each observation domain is compared, the detection effect of the frequency domain is best, because the electromagnetic signal has stronger cycle and frequency characteristics on the frequency domain, the two-dimensional video information is mapped on the frequency domain, and the characteristics can be extracted and learned through a neural network, so that effective classification and identification are realized;
The two algorithms of MGCNN and D-CNN-BN are compared separately, and as batch standardization operation is added, the performance of the D-CNN-BN in the time domain is greatly improved, the optimization effect of batch standardization on the neural network is further analyzed, and the training conditions of the two network structures are compared, as shown in fig. 4 and 5;
The change conditions of the loss rate and the accuracy rate in the training of the two algorithms in the graph show that after batch standardization is added in the neural network structure, the training loss rate is reduced more rapidly, the convergence speed of the algorithm is accelerated, and therefore the robustness of the model is enhanced;
The algorithm is further verified by using the physical image data set in the table 4, the performance of the leakage image information is compared by D-CNN-BN and MGCNN, the leakage signals of the images of the airplane and the cat 2 are selected for two kinds of recognition detection, 5-fold cross verification is adopted, leakage signal samples generated by 8 images are respectively extracted from the signals of the 2 kinds of signals in each round to serve as a training set, the rest are used as test samples, the detection capability of the algorithm on the unknown image information is verified, the detection effect of the frequency domain signals is proved to be optimal in the early stage, only the frequency domain is selected to serve as an observation domain, and the comparison result is shown in the following table 5:
Table 5 dataset
| Algorithm | Accuracy rate of | Accuracy rate of | Recall rate of recall | F1 |
| MGCNN | 0.8474 | 0.9916 | 0.6995 | 0.7336 |
| D-CNN-BN | 0.8600 | 0.9566 | 0.7669 | 0.8158 |
Table 5 shows that the performance of the D-CNN-BN algorithm is obviously better than that of the MGCNN algorithm, the execution efficiency of the algorithm is also higher than that of the MGCNN algorithm, the detection performance of a neural network model can be improved by batch standardization in an actual electromagnetic environment, the characteristics of a video image hidden in an electromagnetic signal are extracted, and the type of leakage image information is classified and predicted.
Design of experiment
The electromagnetic signal leaked on the VGA cable is collected through the signal collection device, the signal collection device is composed of a caliper and signal processing equipment, the caliper is used for collecting electromagnetic leakage information, the collection precision of the signal processing equipment is 10 bits, and the signal sample is preprocessed through fast Fourier transform and cepstrum analysis.
The experiment selects 20 physical images in the public dataset CIFAR-10 as leakage information, wherein the physical images are divided into two types of airplanes and cats, and 10 physical images are respectively selected.
In the experiment, electromagnetic leakage signals generated by each image are collected at a sampling rate of 1M/s, the collected electromagnetic leakage signals are time domain sequences, the signals are preprocessed by fast Fourier transform and cepstrum analysis respectively, model training is carried out on data on a frequency domain and a cepstrum domain respectively, and an experimental sample data set is shown in the following table 6:
Table 6 leak content detection experiment sample dataset
Samples in the above table were taken as 4:1 is divided into a training set and a test set, the data are trained by using a supervised learning mode, the model is iterated by using a gradient descent method, the iteration times are 50 times, for comparing experimental results, the electromagnetic signal data on a frequency domain are independently subjected to experiments without adopting an integrated learning mode, the experiments are marked as FFT-CNN, the above experiments are all subjected to 5-fold cross validation, the average value of the 5-round experiments is taken as a performance evaluation basis, and the experimental results are shown in figure 8.
As can be seen from fig. 8, the classification accuracy of FFT-CNN in the five-fold cross validation is not stable, the model generated by the algorithm has a relatively obvious change when the training sample set is transformed, the accuracy in the last round of cross validation is only 60%, while the accuracy of EL-CNN (ensemble learning) has no great change and is stabilized to be more than 96% in the five-fold cross validation, because the ensemble learning builds the difference degree between the base classification models by using the data of different observation domains, improves the generalization performance of the model, and improves the classification accuracy and stability by using the combination model, so that the model can accurately detect the leakage information of the electromagnetic signal under the condition of unknown leakage information content, and judges the leakage risk thereof.
The foregoing description of the preferred embodiments of the invention is not intended to be limiting, but rather is intended to cover all modifications, equivalents, alternatives, and improvements that fall within the spirit and scope of the invention.
Claims (8)
1. The computer cable electromagnetic leakage characteristic analysis method is characterized by comprising the following steps of:
Step one: identifying a source of leakage
Firstly, acquiring a leaked electromagnetic signal through a signal acquisition device, preprocessing a time signal sample by using fast Fourier transform, designing a convolution neural network structure comprising two one-dimensional convolution layers, two pooling layers and a full-connection layer, respectively carrying out model training on signal data on a time domain and a frequency domain and a continuous observation domain, then inputting the electromagnetic signal to be tested into a trained convolution neural network structure model, extracting layer by layer characteristics of the neural network, outputting prediction scores of classifications at the full-connection layer, and finally judging an electromagnetic information leakage source through calculation;
Step two: judging whether information leakage exists or not
Firstly, electromagnetic leakage signals of a leakage source with electromagnetic leakage are collected through a signal collection device, then a fast Fourier transform is used for preprocessing time signal samples, then a convolution neural network structure containing batch standardization layers is designed, model training is respectively carried out on signal data on two observation domains of a time domain and a frequency domain, electromagnetic signals to be tested are input into a trained convolution neural network structure model, layer-by-layer characteristic extraction of the neural network is carried out, various prediction scores are output at a full connection layer, and finally whether information leakage exists in the leaked electromagnetic signals is determined through calculation;
step three: analyzing leakage information content
Firstly, collecting electromagnetic signals leaked from a leakage source through a signal collecting device, and then designing an integrated learning method based on a convolutional neural network to analyze the content of leakage information: firstly, preprocessing a time domain signal sample by using fast Fourier transform to obtain a frequency domain signal representation of the sample, training frequency domain signal data to obtain a corresponding base classification model, then further calculating the frequency domain signal by using cepstrum analysis to obtain a cepstrum signal of the sample, training the cepstrum signal to obtain a corresponding base classification model, then combining the base classification model by using a weighting mode to construct a new model, inputting an electromagnetic signal to be detected into the new model, extracting layer by layer characteristics of a neural network in the model, and then carrying out independent decision and output of the weighted combined base classification model to carry out electromagnetic information leakage detection to judge the leakage risk, wherein the training 2 base classification models all adopt the same convolutional neural network structure and comprise a one-dimensional convolutional layer, a batch standardization layer, a pooling layer and a full-connection layer, and an integrated learning weighting operation calculation formula of the full-connection layer output is as follows:
wherein: y c represents the final score of the c-th classification, k represents the corresponding base classification model, S is the output of the fully connected layer, ω k is the weight of the base classification model.
2. The method for analyzing electromagnetic leakage characteristics of a computer cable according to claim 1, wherein: the fast fourier transform formula is:
Wherein: x (N) is a time domain signal sequence, X (k) is a frequency spectrum signal sequence after calculation, and N is the length of the signal sequence;
The cepstrum analysis formula is:
Wherein: the data after Fourier transformation is divided into a real part and an imaginary part, X (e jw) is a real part of X (n) after Fourier transformation, and Ceps (n) is a calculation result of real cepstrum.
3. The method for analyzing electromagnetic leakage characteristics of a computer cable according to claim 1, wherein: the input of the convolutional neural network structure is a frequency spectrum signal, the length is 16384, the convolutional network transmits the data of the input layer to the convolutional layer and the pooling layer, after feature extraction and conversion, the feature vector with 64 channels and the length of 1024 is input into the full connection layer, and finally the full connection layer gathers the features and outputs the result.
4. The method for analyzing electromagnetic leakage characteristics of a computer cable according to claim 1, wherein: the convolution function calculation formula of the convolution layer is as follows:
Wherein: And N is the index number of each layer in the neural network, X (N) and X (N+1) are the characteristic vectors input and output by the Nth layer in the convolution calculation process, W (N) is the weight vector of the Nth layer neural network, and B (N) is the bias vector of the Nth layer.
5. The method for analyzing electromagnetic leakage characteristics of a computer cable according to claim 1, wherein: the convolution neural network structure adopts a ReLU as an activation function, and the calculation formula of the ReLU function is as follows:
ReLU(x)＝max(0，x)
after the ReLU function, a Dropout function is used, the probability value is set to be 0.5, so that the overfitting of the neural network is restrained, then the maximum pooling is used for downsampling, the characteristics are screened, the network parameters are reduced, and the maximum characteristic value in the pooling domain is output.
6. The method for analyzing electromagnetic leakage characteristics of a computer cable according to claim 1, wherein: the full connection layer uses SoftMax as an activation function, and the calculation formula of the SoftMax function is as follows:
Wherein: i is an element number for performing an operation, and S i represents a mapping of the i-th element in the (0, 1) section.
7. The method for analyzing electromagnetic leakage characteristics of a computer cable according to claim 1, wherein: the convolutional neural network structure normalizes data by using a batch normalization method after convolutional operation, and the batch normalization algorithm is as follows:
S1, calculating the average value of data output by the previous layer
S2, calculating standard deviation of output data of the previous layer
S3, normalization calculation
S4, reconstructing the normalized data
Wherein: x is the output result of the upper layer, m is the size of the training sample batch, epsilon is a small value close to 0, the situation that the denominator is 0 is avoided, and beta and gamma are learning parameters.
8. The method for analyzing electromagnetic leakage characteristics of a computer cable according to claim 1, wherein: the signal acquisition device is composed of calipers and signal processing equipment, wherein the calipers are used for collecting electromagnetic leakage information, and the acquisition precision of the signal processing equipment is 10 bits.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210030092.2A CN114372493B (en) | 2022-01-12 | 2022-01-12 | Computer cable electromagnetic leakage characteristic analysis method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210030092.2A CN114372493B (en) | 2022-01-12 | 2022-01-12 | Computer cable electromagnetic leakage characteristic analysis method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114372493A CN114372493A (en) | 2022-04-19 |
| CN114372493B true CN114372493B (en) | 2024-06-18 |
Family
ID=81143992
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210030092.2A Active CN114372493B (en) | 2022-01-12 | 2022-01-12 | Computer cable electromagnetic leakage characteristic analysis method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114372493B (en) |
Families Citing this family (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108680796B (en) * | 2018-05-17 | 2023-09-19 | 集美大学 | Electromagnetic information leakage detection system and method for computer display |
| CN115087341B (en) * | 2022-07-21 | 2022-11-15 | 北京中超伟业信息安全技术股份有限公司 | Electromagnetic signal scrambling method and system based on waveguide window |
| CN117332410B (en) * | 2023-10-25 | 2024-04-12 | 北京航空航天大学 | Electromagnetic leakage red and black signal identification method based on information entropy characteristics |
Family Cites Families (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP6958803B2 (en) * | 2017-06-19 | 2021-11-02 | 日本電信電話株式会社 | Electromagnetic information leakage evaluation device, electromagnetic information leakage evaluation method, and electromagnetic information leakage evaluation program |
| CN108680796B (en) * | 2018-05-17 | 2023-09-19 | 集美大学 | Electromagnetic information leakage detection system and method for computer display |
| CN110490154B (en) * | 2019-08-23 | 2020-09-15 | 集美大学 | Multidimensional leakage information detection method, terminal equipment and storage medium |
-
2022
- 2022-01-12 CN CN202210030092.2A patent/CN114372493B/en active Active
Non-Patent Citations (2)
| Title |
|---|
| 一种基于深度学习的电磁信息泄漏检测方法;茅剑等;《北京航空航天大学学报》;20210203;第2200-2207页 * |
| 基于主动检测的电子设备电磁信息泄漏新型威胁分析;丁建锋;刘文斌;丁磊;张健;严承涛;宋滔;;通信技术;20180410(第04期);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114372493A (en) | 2022-04-19 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN114372493B (en) | Computer cable electromagnetic leakage characteristic analysis method | |
| CN111915437B (en) | Training method, device, equipment and medium of money backwashing model based on RNN | |
| Che et al. | Hybrid multimodal fusion with deep learning for rolling bearing fault diagnosis | |
| CN111027069B (en) | Malicious software family detection method, storage medium and computing device | |
| CN112101426A (en) | Unsupervised learning image anomaly detection method based on self-encoder | |
| CN108680796B (en) | Electromagnetic information leakage detection system and method for computer display | |
| CN109190698B (en) | Classification and identification system and method for network digital virtual assets | |
| CN103310235B (en) | A kind of steganalysis method based on parameter identification and estimation | |
| US11100428B2 (en) | Distributable event prediction and machine learning recognition system | |
| Li et al. | An improved adaboost algorithm for imbalanced data based on weighted KNN | |
| CN117493759A (en) | Gas methane distinguishing method and device based on principal component analysis and vector machine | |
| Wang et al. | A transparent and nonlinear method for variable selection | |
| CN110879821A (en) | Method, device, equipment and storage medium for generating rating card model derivative label | |
| CN113191134B (en) | Document quality verification method, device, equipment and medium based on attention mechanism | |
| KR20220125741A (en) | Method for analyzing lesion based on medical image | |
| US20230334662A1 (en) | Methods and apparatus for analyzing pathology patterns of whole-slide images based on graph deep learning | |
| US20230145002A1 (en) | Connecting adversarial attacks to neural network topography | |
| CN110728615B (en) | Steganalysis method based on sequential hypothesis testing, terminal device and storage medium | |
| CN116720183A (en) | Internal threat behavior detection method and system integrating user multidimensional features | |
| CN117831055A (en) | Document image type identification and classification method | |
| Tian et al. | Performance Evaluation of Machine Learning in Wireless Connected Robotics Swarms | |
| Zhang et al. | A novel sparse enhancement neural network for rolling bearing fault diagnosis | |
| Li et al. | Power quality disturbances classification with imbalanced/insufficient samples based on WGAN-GP-SA and DCNN | |
| Tleis et al. | Machine Learning approach to discriminate Saccharomyces cerevisiae yeast cells using sophisticated image features | |
| Gharaibeh et al. | Automated Lung Cancer Diagnosis Applying Butterworth Filtering, Bi-Level Feature Extraction, and Sparce Convolutional Neural Network to Luna 16 CT Images |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |