Anti-side-channel-attack electrocardiogram data transmission method and system based on machine learning

Abstract

The invention discloses an electrocardiogram data transmission method and system for resisting side channel attack based on machine learning, and the method comprises the steps: carrying out the Fourier transform and frequency domain screening of an electrocardiogram signal data set through a data preprocessing step, and achieving the preliminary data compression; performing feature selection and model training by using a random forest model, identifying and selecting most valuable features from the data through a SelectKBest or RFECV method, and training the random forest model by using the features; after the model training is completed, a randomized redundancy filling step: introducing random redundancy information into the compressed electrocardiosignal data so as to enhance the side channel attack resistance of the data; and carrying out performance evaluation on electrocardiogram data transmission after randomized redundancy filling through a side channel attack resistance evaluation step. According to the method, by introducing the random forest model and the randomized redundancy filling technology, the side channel attack resistance of electrocardiogram data transmission is improved, and meanwhile the high data compression rate and transmission efficiency are kept.

Description

Technical Field

[0001] The present invention relates to the field of biomedical signal data transmission, and in particular to a machine learning side-channel attack resistant data transmission method and system applicable to electrocardiogram (ECG) signals. Background Art

[0002] Electrocardiogram (ECG) signals are a key tool for monitoring cardiac electrical activity. They can provide detailed information about cardiac rhythm, heart rate, and heart health, and are essential for the diagnosis of cardiovascular diseases such as arrhythmias and myocardial ischemia. In each cardiac cycle, the waveform characteristics of ECG signals, such as P waves, QRS complexes, and T waves, are specific reflections of cardiac electrophysiological activity. However, continuous monitoring of ECG signals generates a huge amount of data, which poses a challenge to data storage and transmission.

[0003] In order to reduce the amount of data that needs to be transmitted, compression of ECG signal data becomes a necessary means. However, compressed data may lead to uneven distribution of 0s and 1s in data transmission, which increases the risk of side channel attacks and greatly affects the security of data. Side channel attacks (SCA) use the behavioral information of the system during execution to recover secret data. Even under the provable security model, cryptographic systems are vulnerable to side channel attacks. When there is a data dependency between the data processed by the device and its side channel leakage (such as running time, energy curve, electromagnetic radiation, etc.), the attacker can use this leaked information to recover secret data such as keys. This attack provides the attacker with additional information and reduces the difficulty of recovering unknown keys. Unlike traditional cryptanalysis, the cracking technology of side channel attacks is independent of or linearly related to the key length. When the leaked information is large enough, the side channel attack can recover the key at only a small cost.

[0004] In order to enhance the security of data transmission, researchers have proposed a data transmission strategy based on machine learning to resist side-channel attacks. These methods include masking and hiding techniques, randomized cache design, etc. Masking technology introduces random values ​​to conceal the actual operation, reducing the risk of information leakage, but may increase computational complexity. Randomized cache design reduces the information available to attackers by changing the cache mapping method, but this method may lead to a decrease in cache performance. In order to further enhance security, the implementation of randomized redundant information is proposed. This method introduces random redundant information in data transmission, making it more difficult for attackers to analyze data, thereby improving the system's ability to resist attacks. The advantage of this method is that it can effectively resist side-channel attacks without significantly reducing system performance.

[0005] Although the randomized redundant information method based on machine learning has made some progress, there are still some problems. For example, the problem of high computational complexity leads to reduced algorithm efficiency. Secondly, the randomization process may lead to uneven distribution of information, which affects the stability and accuracy of the model. In addition, the filling strategy of redundant information is not flexible enough and may not be able to adapt to different types of data sets. In order to solve these problems, the random forest method can be considered for redundant filling. Random Forest (RF) fills in missing values ​​by constructing multiple decision trees, making the filled data random and uncertain, and better reflecting the true distribution of these unknown data. This method can not only effectively process ECG data, but also improve the accuracy and reliability of the filled values.

[0006] Therefore, it is necessary to propose a machine learning data transmission method suitable for ECG signals that is resistant to side channel attacks. While meeting reasonable data density and system throughput, it combines random forest and self-adaptation feature extraction algorithms and uses randomized redundant data filling to enable ECG signal transmission to achieve higher anti-side channel attack performance. Summary of the invention

[0007] In order to overcome the deficiencies of the above-mentioned prior art, the present invention provides a machine learning anti-side channel attack data transmission method and system applicable to ECG signals. By introducing a random forest model and randomized redundant filling technology, the anti-side channel attack performance of ECG data transmission is improved, while maintaining a high data compression rate and transmission efficiency.

[0008] The technical solution of the present invention is as follows:

[0009] A method for transmitting electrocardiogram data based on machine learning to resist side channel attacks is characterized in that it includes the following steps:

[0010] S1. Data preprocessing: Load the ECG signal dataset from the data source, perform Fourier transform to obtain spectrum information, and divide the dataset into training set and test set;

[0011] S2. Random forest model construction: Use the RandomForestClassifier or RandomForestRegressor class in Python's Scikit-learn library to build a random forest model, set the number of decision trees and random seed parameters to handle missing data in the dataset and extract adaptive features;

[0012] S3. Feature selection: Identify and select the features that contribute most to model performance from the preprocessed data using the SelectKBest or RFECV method;

[0013] S4. Model training: Use the feature-selected dataset to train the random forest model to improve the model's ability to recognize and process ECG signals;

[0014] S5. Randomized redundant filling: random redundant information is introduced into the compressed ECG signal data, and the filling position and amount of redundant data are determined by using the random forest algorithm;

[0015] S6. Anti-side channel attack evaluation: Evaluate the anti-side channel attack performance of ECG data transmission after randomized redundant padding.

[0016] Furthermore, in step S1), the method further includes performing frequency domain screening on the ECG signal data to achieve preliminary data compression.

[0017] Furthermore, in step S1), it also includes defining a loss function, which combines the signal distribution evaluation of the ECG signal spectrum sequence with the chi-square coefficient and the adaptive feature, as well as an auxiliary loss function introduced according to different targets to be measured, and optimizes the loss function successively through multiple iterations and gradient descent methods.

[0018] Furthermore, the target to be measured includes the compression amount of single-beat ECG signal preprocessing data, the number of random redundant fillings, the transmission data throughput and the uniformity of data distribution.

[0019] Furthermore, in step S3, feature selection is performed by calculating the importance of features and selecting features according to their importance to optimize the input feature set of the model.

[0020] Furthermore, in step S5, the randomized redundant filling is performed by defining the Gini coefficient and the chi-square coefficient to evaluate the importance of data distribution and features, and using a gradient descent algorithm to solve the optimal value of the target to be measured, thereby achieving accurate reconstruction and redundant filling of data.

[0021] Further, the method further comprises the following steps:

[0022] Step S7. Iterative optimization: Through multiple iterations, the model parameters are updated according to the current feature importance, and the corresponding feature weights are calculated to further optimize the feature selection and model performance.

[0023] This method is implemented based on the TensorFlow deep learning framework and Python's Scikit-learn library, and supports automatic derivation of the target measurement and its gradient function.

[0024] Second, the present invention also provides a system for implementing the above method, which is characterized in that it includes:

[0025] Data preprocessing module, used to load and prepare ECG signal datasets;

[0026] Random forest model building module, used to build random forest models;

[0027] Feature selection module, used to identify and select the most valuable features from the preprocessed data;

[0028] Model training module, used to train random forest models and evaluate their performance;

[0029] A randomized redundant filling module is used to introduce random redundant information into the compressed ECG signal data;

[0030] The anti-side channel attack evaluation module is used to evaluate the anti-side channel attack performance of ECG data transmission.

[0031] Furthermore, an iterative optimization module is included for further optimizing feature selection and model performance through multiple iterations.

[0032] Compared with the prior art, the present invention has the following beneficial effects:

[0033] 1. By combining the random forest model construction and adaptive feature selection algorithm, when processing ECG signals with complex spectra, it is possible to analyze areas with densely distributed data and quickly locate the target point, which not only improves the processing efficiency of the algorithm, but also ensures the accuracy and reliability of ECG signal analysis.

[0034] 2. The introduction of randomized redundant padding data randomization strategy makes this method have advantages in data transmission throughput and algorithm robustness, and can provide transmission data streams that are resistant to side channel attacks. This method can not only effectively resist side channel attacks and protect the security of transmitted data, but also significantly improve the data transmission throughput and ensure efficient data transmission. At the same time, this strategy also enhances the robustness of the algorithm, enabling it to maintain stable performance in the face of various complex environments and attack methods.

[0035] 3. Particularly suitable for application scenarios with high requirements for data security and transmission throughput. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art are briefly introduced below.

[0037] Figure 1 The machine learning-based ECG data transmission system architecture that resists side-channel attacks proposed by the present invention;

[0038] Figure 2 The architecture flow of the randomized data filling algorithm based on machine learning proposed by the present invention;

[0039] Figure 3 The principle of the incomplete data positioning algorithm based on random forest proposed by the present invention;

[0040] Figure 4 The adaptive feature selection algorithm architecture proposed by the present invention;

[0041] Figure 5 The anti-side channel data transmission performance evaluation model proposed by the present invention; DETAILED DESCRIPTION

[0042] In order to make the purpose, advantages, technical solutions and implementation methods of the present invention clearer, the present invention is further described in detail below in conjunction with the embodiments and drawings. The exemplary implementation methods of the present invention and their descriptions are only used to explain the present invention and are not intended to limit the present invention. Based on the embodiments of the present invention, other embodiments obtained by ordinary technicians in this field without making creative work are all within the scope of protection of the present invention.

[0043] The present invention proposes a data transmission method and system based on machine learning to resist side channel attacks, which utilizes the processing ability of random forest for missing data and adaptive feature extraction ability, combines the TensorFlow deep learning framework and the Scikit-learn library in Python, such as Figure 1 As shown in the figure, it mainly includes the following modules: data preprocessing, random forest model construction, feature selection, model training, randomized redundant filling and anti-side channel attack evaluation.

[0044] The data preprocessing module is responsible for loading and preparing the dataset to ensure effective training of the random forest model. First, the dataset is loaded from the data source and divided into training and test sets to evaluate the generalization ability of the model. The random forest model building module uses the RandomForestClassifier or RandomForestRegressor class in the Scikit-learn library to build the model. During the model building process, key parameters such as the number of decision trees (n_estimators) and random seeds (random_state) are set to ensure the stability and accuracy of the model.

[0045] The feature selection module uses methods such as SelectKBest or RFECV to identify and select the most valuable features. The input feature set of the model is optimized by calculating the importance of the features and selecting them according to their importance. The model training module uses the selected features to train the random forest model and evaluates the performance of the model through accuracy or other evaluation indicators. The present invention combines the algorithm flow of data preprocessing, random forest model construction, adaptive feature selection, and randomized redundant filling to provide a transmission data stream that is resistant to side channel attacks. Figure 2 The design requirements and architectural process shown are introduced.

[0046] In this example, the original ECG signal and heartbeat tag are used as the data input of the system, and combined with Fourier transform, the data compression and preprocessing of the signal are realized in the processing of complex spectrum signals. While meeting the transmission throughput requirements of the ECG channel, the compressed data is randomly padded to achieve the characteristics of anti-side channel attack.

[0047] In this example, the ECG signal is first Fourier transformed to obtain the signal's spectral information, and preliminary compression is achieved through frequency domain screening. Then, the random forest model is used to build a computational graph on the machine learning platform to extract features and classify the preprocessed data. The random forest model constructs multiple decision trees to iteratively process the compressed sensing sampled data. In each iteration, the model parameters are updated according to the current feature importance, and the corresponding feature weights are calculated to further optimize the feature selection and model performance. During the iterative processing, the random forest can effectively capture the nonlinear relationship in the data and improve the classification effect of complex data sets. Finally, the algorithm flow of adaptive feature selection and randomized redundant filling is combined to ensure the anti-side channel attack characteristics of the data stream, and accurate data reconstruction is achieved through multiple rounds of updates. After each round of update, the model performance is evaluated and the parameters are adjusted to optimize the model performance. This algorithm not only improves the data compression efficiency, but also takes into account the accuracy and computational cost during the model training process, which is suitable for real-time data processing and storage optimization.

[0048] like Figure 1 Before neural network feature extraction, data compression preprocessing is required according to the periodic characteristics of the ECG signal. Combined with the random forest model construction, adaptive feature selection, and random redundant filling algorithms, the input ECG signal and heartbeat label are preprocessed and randomized to fill in redundancy, so that the transmitted data can obtain anti-side channel attack characteristics.

[0049] The present invention transforms the transmission problem of ECG signals against side channel attacks into an optimization problem of the target to be measured. The target band measurement includes the compression amount of the single-beat ECG signal preprocessing data, the number of random redundant padding, the transmission data throughput and the uniformity of data distribution. The present invention uses a gradient descent algorithm to solve the optimal value of the target band measurement, and uses a joint framework based on a deep learning platform-random forest algorithm to implement it. Multiple iterations are performed to optimize successively. The loss function and its gradient are calculated in each iteration, and the loss function is updated according to the gradient descent method.

[0050] According to the data sequence of ECG signal, R peak position sequence and heartbeat label sequence, the design related data is read in the deep learning platform. When the ECG data sequence of the i-th cycle is x, the frequency domain sequence is X, the frequency domain sequence is divided into K sample sets, and the distribution probability of data "1" in the k-th non-sample is p k , the Gini coefficient Gini(k) can be defined as:

[0051]

[0052] K sample sets are used to construct K classification and regression trees to form a random forest to analyze the data. The voting mechanism is used, and the evaluation results are:

[0053]

[0054] Among them, H(X) represents the combined classification algorithm, h i (X) represents a single decision tree classification algorithm, and Y is the distribution probability of the target data "1" (in this description, Y = 50%) + 3%). I represents the indicative function.

[0055] In this description, according to the spectral distribution characteristics of ECG signals, we divide the data into M frequency bands, and the "1" distribution of the M frequency band data is used as the selected feature in this architecture. Through the random redundant data filling of the jth round, and the redundant data filling position is determined by the random forest algorithm, the evaluation coefficient H(X) of data filling is calculated, and the chi-square coefficient of each feature is obtained as follows:

[0056]

[0057] Then the sum of the chi-square coefficients of the total feature set is:

[0058]

[0059] The present invention uses a chi-square coefficient sum that combines the signal distribution evaluation of the ECG signal spectrum sequence with the adaptive features as the loss function of the algorithm:

[0060] Loss=α*H(X,Y)+β*W(X,M)

[0061] Among them, α and β are weight parameters used to balance the impact of different losses. When the target to be measured is the system transmission ECG throughput after redundant filling, the amount of single-beat ECG spectrum sequence data Q after filling is used as the auxiliary loss function, and the loss function of this algorithm can be converted to:

[0062] Loss=θ*Q+(1θ)[α*H(X,Y)+β*W(X,M)]

[0063] Among them, θ, α and β are weight parameters, and the optimal value of the above formula is when the Loss function takes the minimum value.

[0064] The present invention is implemented using the TensorFlow deep learning framework and the Scikit-learn library in Python, which supports the automatic derivation of the target measurement and its gradient function, so that the system architecture can adopt a unified implementation process for different optimization goals, while meeting the data throughput of ECG signal transmission, making the data transmission resistant to side channel attacks.

Claims

1. A method for transmitting electrocardiogram data based on machine learning to resist side channel attacks, characterized in that: The following steps are involved: S1. Data preprocessing: Load the ECG signal dataset from the data source, perform Fourier transform to obtain spectrum information, and divide the dataset into training set and test set; S2. Random forest model construction: Use the RandomForestClassifier or RandomForestRegressor class in Python's Scikit-learn library to build a random forest model, set the number of decision trees and random seed parameters to handle missing data in the dataset and extract adaptive features; S3. Feature selection: Identify and select the features that contribute most to model performance from the preprocessed data using the SelectKBest or RFECV method; S4. Model training: Use the feature-selected dataset to train the random forest model to improve the model's ability to recognize and process ECG signals; S5. Randomized redundant filling: random redundant information is introduced into the compressed ECG signal data, and the filling position and amount of redundant data are determined by using the random forest algorithm; S6. Anti-side channel attack evaluation: Evaluate the anti-side channel attack performance of ECG data transmission after randomized redundant padding.

2. The ECG data transmission method based on machine learning to resist side channel attacks according to claim 1 is characterized in that: In step S1), the method also includes frequency domain screening of the ECG signal data to achieve preliminary data compression.

3. The method for transmitting ECG data based on machine learning to resist side channel attacks according to claim 1 or 2, characterized in that: In step S1), it also includes defining a loss function, which combines the signal distribution evaluation of the ECG signal spectrum sequence with the chi-square coefficient and the adaptive feature, as well as an auxiliary loss function introduced according to different targets to be measured, and optimizes the loss function successively through multiple iterations and gradient descent methods.

4. The ECG data transmission method based on machine learning to resist side channel attacks according to claim 3 is characterized in that: The target measurements include the compression amount of single-beat ECG signal preprocessing data, the number of random redundant fillings, the transmission data throughput and the uniformity of data distribution.

5. The ECG data transmission method based on machine learning to resist side channel attacks according to claim 1 is characterized in that: In step S3, feature selection is performed by calculating the importance of features and selecting features according to their importance to optimize the input feature set of the model.

6. The ECG data transmission method based on machine learning to resist side channel attacks according to claim 1 is characterized in that: In step S5, the randomized redundant filling is to evaluate the importance of data distribution and features by defining the Gini coefficient and the chi-square coefficient, and to solve the optimal value of the target to be measured by using the gradient descent algorithm to achieve accurate reconstruction and redundant filling of data.

7. The method for transmitting ECG data based on machine learning to resist side channel attacks according to any one of claims 1 to 6, characterized in that: The following steps are also included: Step S7. Iterative optimization: Through multiple iterations, the model parameters are updated according to the current feature importance, and the corresponding feature weights are calculated to further optimize the feature selection and model performance.

8. The method for transmitting ECG data based on machine learning to resist side channel attacks according to any one of claims 1 to 6, characterized in that: This method is implemented based on the TensorFlow deep learning framework and Python's Scikit-learn library, and supports automatic derivation of the target measurement and its gradient function.

9. A system for implementing the method according to any one of claims 1 to 6, characterized in that: include: Data preprocessing module, used to load and prepare ECG signal datasets; Random forest model building module, used to build random forest models; Feature selection module, used to identify and select the most valuable features from the preprocessed data; Model training module, used to train random forest models and evaluate their performance; A randomized redundant filling module is used to introduce random redundant information into the compressed ECG signal data; The anti-side channel attack evaluation module is used to evaluate the anti-side channel attack performance of ECG data transmission.

10. The system according to claim 9, characterized in that It also includes an iterative optimization module for further optimizing feature selection and model performance through multiple iterations.

Images