A method, device and equipment for preprocessing magnetoencephalogram signals

Abstract

The present application relates to the technical field of brain-computer interfaces, and discloses a magnetoencephalogram signal preprocessing method, apparatus, and device. The method includes: performing multi-strategy filtering on an original magnetoencephalogram signal and fusing the filtering results to obtain a fused filtering signal; the multi-strategy filtering includes at least one filtering method applicable to a linear system and at least one filtering method applicable to a non-linear system; performing frequency decomposition and feature extraction on the fused filtering signal to obtain signal modal features; and performing signal reconstruction based on the signal modal features that meet the correlation condition to obtain a reconstructed magnetoencephalogram signal. The beneficial effects of the present application are as follows: Adaptive noise cancellation is performed on the magnetoencephalogram signal through multi-strategy filtering, thereby combining the advantages of different filtering methods, reducing the influence of the dynamic attributes of the magnetoencephalogram signal on signal classification, effectively improving the quality of the magnetoencephalogram signal, and further improving the classification effect and classification accuracy of the magnetoencephalogram signal, providing more accurate magnetoencephalogram signal features for subsequent research.

Description

Technical Field

[0001] This application relates to the technical field of brain-computer interfaces, and in particular, to a method, device, and equipment for preprocessing magnetoencephalogram signals. Background Art

[0002] A brain-computer interface (BCI) refers to a technology that deciphers specific intentions from brain activity signals and converts them into machine instructions to control external devices. Based on brain-computer interface technology, users can use event-related potentials (ERPs) triggered by motor imagery (MI) as an active neural activity feature as a control signal in an active-brain-computer interface (Active-BCI) system to control external devices. However, due to the high impedance of the brain skull, it is not conducive to the passage of high-frequency signals, resulting in significant signal attenuation in the actually collected electroencephalogram (EEG) signals. Moreover, as a scalar field, inverting or tracing the EEG signals will cause the loss of spatial features in the EEG signals, affecting the study of brain nerve activities. In contrast, magnetoencephalogram (MEG) signals are not affected by attenuation when penetrating the brain skull, and as a vector field, MEG signals are superior to EEG signals in spatio-temporal resolution. Therefore, MEG signals can replace EEG signals in brain-computer interface technology, and brain-computer interface systems using MEG signals emerge in an endless stream.

[0003] In related technologies, the signal quality of MEG signals still needs to be improved. Summary of the Invention

[0004] This application provides a method, device, and equipment for preprocessing MEG signals, which perform adaptive noise cancellation on MEG signals through multi-strategy filtering, thereby combining the advantages of different filtering methods, reducing the influence of the dynamic attributes of MEG signals on signal classification, and effectively improving the quality of MEG signals.

[0005] To achieve the above object, the main technical solutions adopted in this application include:

[0006] In a first aspect, an embodiment of this application provides a method for preprocessing MEG signals, where the method includes:

[0007] Perform multi-strategy filtering on the original MEG signal and fuse the filtering results to obtain a fused filtered signal; where the multi-strategy filtering includes at least one filtering method applicable to a linear system and at least one filtering method applicable to a nonlinear system;

[0008] Perform frequency decomposition and feature extraction based on the fused filtered signal to obtain signal modal features;

[0009] Reconstruct the original magnetoencephalogram (MEG) signal according to the signal modal characteristics that meet the relevance condition to obtain a reconstructed MEG signal.

[0010] The MEG signal preprocessing method proposed in the embodiments of the present application combines a filtering method applicable to a linear system and a filtering method applicable to a nonlinear system, uses a multi-strategy filtering method to adaptively eliminate noise from the original MEG signal, and performs signal reconstruction based on the obtained fused filtering signal, thereby obtaining a preprocessed reconstructed MEG signal. The reconstructed MEG signal can serve as the signal basis for MEG signal classification and provide accurate MEG signal characteristics for the research of brain-computer interface technology. Compared with the related technology, the present application combines multiple filtering methods, processes the stationary noise in the original MEG signal through a linear filtering method, and processes the complex nonlinear signals in the original MEG signal through a nonlinear filtering method, reduces the influence brought by the nonlinear interference in the original MEG signal, effectively improves the filtering effect, and improves the robustness of the MEG signal. Fuse the obtained filtering results to obtain a fused filtering signal, so as to combine the advantages of different filtering methods to improve the quality of the MEG signal, and further improve the classification effect and classification accuracy of the MEG signal.

[0011] Optionally, the multi-strategy filtering of the original MEG signal and the fusion of the filtering results to obtain a fused filtering signal include

[0012] Perform linear system filtering on the original MEG signal to obtain a linear filtering signal;

[0013] Perform nonlinear system filtering on the original MEG signal to obtain a nonlinear filtering signal;

[0014] Fuse the linear filtering signal and the nonlinear filtering signal to obtain the fused filtering signal.

[0015] Optionally, the performing linear system filtering on the original MEG signal to obtain a linear filtering signal includes:

[0016] Update the estimation weight of the sample reference value corresponding to the original MEG signal according to the fused filtering signal at the previous moment;

[0017] Use the sample reference value and the updated estimation weight of the sample reference value to obtain the motion artifact noise matrix of the original MEG signal;

[0018] Perform linear system filtering on the original MEG signal according to the motion artifact noise matrix to obtain the linear filtering signal.

[0019] Optionally, the original MEG signal includes a radial signal and a tangential signal; the performing nonlinear system filtering on the original MEG signal to obtain a nonlinear filtering signal includes:

[0020] Based on the fused filtering signal of the radial signal at the first moment and the external stimulus input signal at the second moment, obtain the predicted state of the radial signal at the current moment as the radial prediction state; wherein, both the first moment and the second moment are before the current moment;

[0021] Based on the fused filtering signal of the tangential signal at the first moment and the external stimulus input signal at the second moment, obtain the predicted state of the tangential signal at the current moment as the tangential prediction state;

[0022] Perform component coupling on the radial prediction state and the tangential prediction state to obtain the coupled prediction state of the original magnetoencephalogram signal; and perform state observation on the coupled prediction state to obtain the non-linear filtering signal.

[0023] Optionally, the step of fusing the linear filtering signal and the non-linear filtering signal to obtain the fused filtering signal includes:

[0024] Adjust the preset weight of the linear filtering signal according to the signal-to-noise ratios of the linear filtering signal and the non-linear filtering signal respectively to obtain the fusion weight of the linear filtering signal;

[0025] Perform weighted calculation on the linear filtering signal and the non-linear filtering signal according to the fusion weight to obtain the fused filtering signal.

[0026] Optionally, the step of performing signal reconstruction on the original magnetoencephalogram signal according to the signal modal features that meet the relevance condition to obtain the reconstructed magnetoencephalogram signal includes:

[0027] Perform information entropy detection on the signal modal features to obtain the relevance of the signal modal features;

[0028] Screen the signal modal features according to the relevance condition to obtain the signal reconstruction features that meet the relevance condition;

[0029] Perform signal reconstruction according to the signal reconstruction features to obtain the reconstructed magnetoencephalogram signal.

[0030] Optionally, the method further includes:

[0031] Classify the reconstructed magnetoencephalogram signal to obtain the classification result of the magnetoencephalogram signal.

[0032] Optionally, the step of classifying the reconstructed magnetoencephalogram signal to obtain the classification result of the magnetoencephalogram signal includes:

[0033] Input the reconstructed magnetoencephalogram (MEG) signals and the corresponding label set into a deep learning neural network model to classify the reconstructed MEG signals and output the classification result of the MEG signals; wherein, the deep learning neural network model is trained by the reconstructed MEG signals before the current moment.

[0034] In a second aspect, an embodiment of the present application provides a MEG signal preprocessing device, the device includes:

[0035] A fusion filtering module, configured to perform multi-strategy filtering on the original MEG signals and fuse the filtering results to obtain a fusion filtering signal; wherein, the multi-strategy filtering includes at least one filtering method applicable to a linear system and at least one filtering method applicable to a non-linear system;

[0036] A feature extraction module, configured to perform frequency decomposition and feature extraction based on the fusion filtering signal to obtain signal modal features;

[0037] A signal reconstruction module, configured to reconstruct the original MEG signals according to the signal modal features that meet the relevance condition to obtain reconstructed MEG signals.

[0038] In a third aspect, an embodiment of the present application provides a computer device, including: a memory and a processor, the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the computer instructions to execute the method according to any one of the above embodiments.

[0039] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, on which computer instructions are stored, and the computer instructions are used to cause a computer to execute the method according to any one of the above embodiments.

[0040] In a fifth aspect, an embodiment of the present application provides a computer program product, including computer instructions, and the computer instructions are used to cause a computer to execute the method according to any one of the above embodiments. Description of the Drawings

[0041] In order to more clearly illustrate the specific embodiments of the present application or the technical solutions in the prior art, the following will briefly introduce the drawings required for the description of the specific embodiments or the prior art. Obviously, the following drawings are some embodiments of the present application. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0042] Figure 1 It is a step diagram of the MEG signal preprocessing method provided by the embodiment of the present application;

[0043] Figure 2a In the embodiment of the present application, it is at the position a schematic diagram of the signal modal feature obtained from the component;

[0044] Figure 2b In the embodiment of the present application, it is at the position a schematic diagram of the signal modal feature obtained from the component;

[0045] Figure 3 a step diagram for obtaining the fused filtering signal in the embodiment of the present application;

[0046] Figure 4 a step diagram for linear system filtering in the embodiment of the present application;

[0047] Figure 5 a step diagram for non - linear system filtering in the embodiment of the present application;

[0048] Figure 6 a step diagram for signal fusion in the embodiment of the present application;

[0049] Figure 7 a step diagram for obtaining the reconstructed magnetoencephalogram signal in the embodiment of the present application;

[0050] Figure 8a In the embodiment of the present application, it is at the position a schematic diagram of the reconstructed magnetoencephalogram signal obtained from the component;

[0051] Figure 8b In the embodiment of the present application, it is at the position a schematic diagram of the reconstructed magnetoencephalogram signal obtained from the component;

[0052] Figure 9 a module diagram of the magnetoencephalogram signal pre - processing device provided in the embodiment of the present application;

[0053] Figure 10 a schematic structural diagram of a computer device provided in the embodiment of the present application. Specific embodiments

[0054] To make the objectives, technical solutions, and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are some, but not all, of the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative efforts belong to the scope of protection of the present application.

[0055] Brain-Computer Interface (BCI) refers to the technology that deciphers specific intentions from brain activity signals and converts them into machine instructions to control external devices. Based on BCI technology, users can use the event-related potential (ERP) triggered by motor imagery (MI) as an active neural activity feature to serve as a control signal in an active-BCI system, thereby controlling external devices.

[0056] In current related technologies, the brain activity signals utilized by BCI technology are usually the electroencephalogram (EEG) signals of users. EEG signals are non-invasive signals with high temporal resolution, capable of reflecting changes in brain activity in real time, which is conducive to real-time control and interaction. Compared with other brain activity signals, EEG signal devices have lower costs and are easy to use, suitable for long-term wearing and real-time monitoring.

[0057] However, in actual situations, due to the high impedance of the brain skull, it is not conducive to the passage of high-frequency signals. During the signal acquisition process, when EEG signals penetrate the skull and reach the electrodes, there will be significant signal attenuation. In addition, EEG signals are a scalar field. When conducting signal source tracing or inversion of EEG signals in in-depth studies of brain nerve activities, the spatial characteristics in EEG signals will be lost, affecting the study of brain nerve activities.

[0058] With the development of atomic magnetometers, magnetoencephalogram (MEG) signals are widely used as a type of brain activity signal. In contrast to EEG signals, MEG signals are not affected by attenuation when penetrating the brain skull, and as a vector field, MEG signals are superior to EEG signals in terms of spatio-temporal resolution. Therefore, MEG signals can replace EEG signals and be applied in BCI technology, and BCI systems applying MEG signals emerge in an endless stream.

[0059] Due to the characteristic that atomic magnetometers are vulnerable to low-frequency disturbances, when measuring MEG signals through atomic magnetometers, MEG signals will inevitably be contaminated by physiological and non-physiological artifacts. Especially in the application of real-time closed-loop brain-machine interface (BMI), effective means are needed to process the artifact noise that appears in MEG signals.

[0060] In the related art, the means for removing artifacts from magnetoencephalogram (MEG) signals include Independent Component Analysis (ICA) and Principal Components Analysis (PCA). However, the calculation processes of ICA and PCA are highly complex, and the obtained filtering effects are not ideal enough. In the actual working environment of a brain-computer interface, the related art is difficult to provide a comprehensive and sufficient method to classify all the characteristic components in MEG signals, which is also a defect of the current brain-computer interface system based on atomic magnetometers.

[0061] Based on the above problems, the present application provides a method, device and equipment for preprocessing MEG signals. The method includes: performing multi-strategy filtering on the original MEG signal and fusing the filtering results to obtain a fused filtering signal; the multi-strategy filtering includes at least one filtering method applicable to a linear system and at least one filtering method applicable to a nonlinear system; performing frequency decomposition and feature extraction on the fused filtering signal to obtain signal modal features; and performing signal reconstruction according to the signal modal features that meet the correlation condition to obtain a reconstructed MEG signal.

[0062] The method for preprocessing MEG signals provided by the present application combines the filtering methods applicable to linear systems and the filtering methods applicable to nonlinear systems, uses the multi-strategy filtering method to perform adaptive noise cancellation on the original MEG signal, and performs signal reconstruction according to the obtained fused filtering signal, so as to obtain a preprocessed reconstructed MEG signal. The reconstructed MEG signal can be used as the signal basis for MEG signal classification and provide accurate MEG signal features for the research of brain-computer interface technology.

[0063] Compared with the related art, the present application combines multiple filtering methods, processes the stationary noise in the original MEG signal through a linear filtering method, and processes the complex nonlinear signals in the original MEG signal through a nonlinear filtering method, reduces the influence brought by the nonlinear interference in the original MEG signal, effectively improves the filtering effect, and improves the robustness of the MEG signal. The obtained filtering results are fused to obtain a fused filtering signal, so as to be able to combine the advantages of different filtering methods to improve the quality of the MEG signal, and further improve the classification effect and classification accuracy of the MEG signal.

[0064] The method for preprocessing magnetoencephalogram signals provided in this specification can be applied to classifying magnetoencephalogram signals collected by an atomic magnetometer. The type of the atomic magnetometer can be any one of magnetometers such as a superconducting quantum interference device (SQUID) magnetometer, an optically pumped atomic magnetometer (OPM), a coherent population trapping (CPT) atomic magnetometer, a spin-exchange relaxation-free (SERF) magnetometer, or a non-linear magneto-optical rotation (NMOR) atomic magnetometer. When this application is applied to different types of atomic magnetometers, certain adaptive modifications are required. It can be understood that after being adaptively modified, this application can also be used to filter and classify magnetoencephalogram signals collected by other devices, or other signals other than magnetoencephalogram signals.

[0065] According to an embodiment of the present application, an embodiment of a method for preprocessing magnetoencephalogram signals is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. And although the logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in a different order than here.

[0066] In this embodiment, a method for preprocessing magnetoencephalogram signals is provided, which can be used for the magnetoencephalogram signals collected by the above-mentioned atomic magnetometer. Referring to Figure 1 as shown, the method includes:

[0067] S100. Perform multi-strategy filtering on the original magnetoencephalogram signal, and fuse the filtering results to obtain a fused filtering signal; wherein, the multi-strategy filtering includes at least one filtering method applicable to a linear system and at least one filtering method applicable to a non-linear system.

[0068] S200. Perform frequency decomposition and feature extraction based on the fused filtering signal to obtain signal modal features.

[0069] S300. Reconstruct the original magnetoencephalogram signal according to the signal modal features that meet the correlation condition to obtain a reconstructed magnetoencephalogram signal.

[0070] Among them, the original magnetoencephalogram signal can be a signal obtained by measuring a user who is performing motor imagery using an atomic magnetometer. The number of atomic magnetometers can be multiple, which are respectively arranged at different positions on the user's head. Each atomic magnetometer at each position can include multiple data channels. Exemplarily, the number of atomic magnetometers can be five, and the positions where the atomic magnetometers are arranged can follow the 10-20 international standard lead system, and are respectively arranged at , , , and the positions of, where is the reference position, is and the area between, is and the area between, each position containing two data channels respectively, and the data channels represent the radial magnetic field component and the tangential magnetic field component of that position.

[0071] Multi-strategy filtering can be a filtering process including two or more different filtering methods. The filtering methods include at least one filtering method applicable to a linear system and at least one filtering method applicable to a nonlinear system. Among them, the filtering methods applicable to a linear system can be one or more of Kalman filtering, H∞ filtering, least mean square filtering, recursive least squares filtering, and minimum mean square error filtering, etc. The filtering methods applicable to a nonlinear system can be one or more of particle filtering, adaptive filtering, Gaussian process regression filtering, neural network filtering, and fuzzy filtering, etc.

[0072] It should be noted that this application can be repeated during the real-time process of collecting magnetoencephalogram signals, preprocessing the real-time collected magnetoencephalogram signals, so as to be used for real-time classification of magnetoencephalogram signals, and quickly and accurately obtaining the characteristic component signals corresponding to the current motor imagery task of the user.

[0073] Specifically, multi-strategy filtering is performed on the original magnetoencephalogram signals collected by an atomic magnetometer, and the filtering method used can be selected based on the actual scenario requirements according to the characteristics of the filtering method itself. After multi-strategy filtering, the filtering results of various filtering methods can be obtained, and the filtering results are fused to obtain a fused filtering signal, so as to combine the advantages of different filtering methods and further improve the filtering effect on the original magnetoencephalogram signals.

[0074] It should be noted that in this application, the original magnetoencephalogram signals are filtered by the filtering methods applicable to a linear system, so as to remove the stationary noise in the original magnetoencephalogram signals and effectively reduce the interference of noise on the original magnetoencephalogram signals. In this application, the original magnetoencephalogram signals are also filtered by the filtering methods applicable to a nonlinear system, so as to effectively process the nonlinear behaviors and nonlinear characteristics in the original magnetoencephalogram signals, reduce the influence brought by nonlinear interference, and improve the quality and robustness of the original magnetoencephalogram signals. Compared with the technical solutions in the related art that only use a single filtering method, this application combines the filtering results obtained by different filtering methods, and while utilizing the noise reduction ability of the linear filtering method, also utilizes the ability of the nonlinear filtering method to handle the nonlinear interference in the signals, comprehensively improving the signal quality of the original magnetoencephalogram signals from different angles and providing a high-quality signal data basis for the subsequent signal reconstruction and signal classification processes.

[0075] Exemplarily, the filtering method applicable to a linear system can be H∞ filtering, and the filtering method applicable to a nonlinear system can be particle filtering. H∞ filtering has strong robustness and adaptability, and has less dependence on the system model, with a wide range of applications. Particle filtering is applicable to nonlinear and non-Gaussian systems, can handle complex problems in high-dimensional state spaces, has high adaptability, and does not require an accurate system model. In the case of only using H∞ filtering, the filtering process cannot handle the nonlinear behavior and characteristics in the original magnetoencephalogram signals, resulting in limited filtering effects. In the case of only using particle filtering, due to the high computational complexity of particle filtering, the computational amount of the filtering process will be too large. By combining the filtering results of H∞ filtering and particle filtering respectively, the linear and nonlinear parts in the original magnetoencephalogram signals can be processed separately, improving the filtering effect and at the same time enhancing the robustness and signal quality of the original magnetoencephalogram signals.

[0076] Furthermore, taking the fused filtering signal as the signal basis, frequency decomposition is performed on the fused filtering signal, and the part related to the brain activity signal is extracted from the fused filtering signal. Exemplarily, the extracted signal frequency range can be between 8 Hz and 30 Hz, where the α-wave signal between 8 Hz and 12 Hz represents the brain activity signal of the user in a waking and quiet state, and the β-wave signal between 12 Hz and 30 Hz represents the brain activity signal of the user in a busy state with active thinking. By collecting the signals in these two ranges, the user's intention can be accurately identified, thus obtaining effective signal support.

[0077] Specifically describing the process of frequency decomposition, the method of frequency decomposition can be the Discrete Wavelet Transform (DWT) algorithm. Using the discrete wavelet transform algorithm to perform frequency decomposition on the fused filtering signal, the fused filtering signal is decomposed into multiple narrowband signals as the initial frequency band signals. In this embodiment, according to the characteristics of the magnetoencephalogram signals and the power frequency interference characteristics, the discrete wavelet transform algorithm uses the db4 wavelet as the wavelet basis function, the downsampling multiple is 2, and the number of wavelet decomposition levels is 2.

[0078] Specifically describe the process of feature extraction. Further, the method of modal feature extraction can be the Empirical Mode Decomposition (EMD) algorithm. The process of extracting modal features from the initial frequency band signal includes: According to the maximum points of the initial frequency band signal, connect the maximum points with a cubic spline curve to obtain the upper envelope. Correspondingly, according to the minimum points of the initial frequency band signal, connect the minimum points with a cubic spline curve to obtain the lower envelope. Calculate the modal features of the initial frequency band signal based on the average value of the upper envelope and the lower envelope, and its form can be expressed according to the following formula:

[0079]

[0080] Wherein, is the average value of the upper envelope and the lower envelope; represents the initial modal features of the initial frequency band signal. After obtaining the initial modal features, judge whether they meet the conditions of the Intrinsic Mode Function (IMF). The conditions of the intrinsic mode function can include: 1. Within the entire data segment, the number of extreme points and the number of zero-crossing points must be equal or differ by at most one; 2. At any moment, the average value of the upper envelope formed by the local maximum points and the lower envelope formed by the local minimum points is zero, that is, the upper envelope and the lower envelope are locally symmetric with respect to the time axis. In addition to the above conditions, corresponding intrinsic mode function conditions can also be selected according to actual scenario needs.

[0081] If the initial modal features of the initial frequency band signal meet the conditions of the intrinsic mode function, then take this initial modal feature as the first modal feature . If the initial modal features of the initial frequency band signal do not meet the conditions of the intrinsic mode function, then take this initial modal feature as the basic data, and repeat the above steps of obtaining the upper envelope and the lower envelope and obtaining the modal features until the initial modal features that meet the conditions of the intrinsic mode function are obtained as the first modal feature .

[0082] To calculate other modal features, remove the first modal feature from the initial frequency band signal to obtain the intermediate frequency band signal, and its form is expressed according to the following formula:

[0083]

[0084] Wherein, is the intermediate frequency band signal. For the intermediate frequency band signal Perform the above steps of obtaining the upper envelope and the lower envelope to obtain the modal features, and removing the modal features from the intermediate frequency band signal for updating until the updated intermediate frequency band signal becomes a monotonic function. At this time, the signal modal features can be obtained. And the residual, the form of the residual is expressed according to the following formula:

[0085]

[0086] Where is the residual; is the intermediate frequency band signal after repeating the above steps times.

[0087] It should be noted that in this application, the discrete wavelet transform algorithm and the empirical mode decomposition algorithm are combined. The discrete wavelet transform algorithm provides multi-scale signal decomposition ability to extract signals with different frequency components, and the empirical mode decomposition algorithm adaptively decomposes and extracts features from the extracted signals to obtain the feature components in the signals. By combining the two methods, the performance of this application in blind source separation is improved, so that the key signal modal features in the magnetoencephalogram signal can still be separated under the interference of large power frequency noise, providing more accurate signal features for magnetoencephalogram signal classification.

[0088] Taking as an example, referring to Figures 2a to 2b shown, Figure 2a shows six signal modal features and the residual obtained from the component at the position, Figure 2b shows six signal modal features and the residual obtained from the component at the position. It can be seen from Figure 2a that for the signal modal features obtained from the component, the correlation between the signal modal features , and is relatively high compared to other signal modal features. Similarly, it can be seen from Figure 2b that for the signal modal features obtained from the component, the correlation between the signal modal features , and is relatively high compared to other signal modal features.

[0089] Further, based on the extracted signals, the original magnetoencephalogram (MEG) signals are reconstructed to obtain reconstructed MEG signals. Through signal reconstruction, the interference of irrelevant components in the original MEG signals is removed, and only the effective information is extracted, thereby improving the quality of the MEG signals and enhancing the expression effect of key feature components in the MEG signals for use in subsequent MEG signal classification processes and improving the accuracy of MEG signal classification.

[0090] The MEG signal preprocessing method provided in this embodiment combines a filtering method applicable to a linear system and a filtering method applicable to a nonlinear system, uses a multi-strategy filtering method to adaptively eliminate noise from the original MEG signals, and performs signal reconstruction based on the obtained fused filtering signals, thereby obtaining preprocessed reconstructed MEG signals. The reconstructed MEG signals can serve as the signal basis for MEG signal classification and provide accurate MEG signal features for the research of brain-computer interface technology.

[0091] Compared with the related art, this application combines multiple filtering methods, processes the stationary noise in the original MEG signals through a linear filtering method, and processes the complex nonlinear signals in the original MEG signals through a nonlinear filtering method, reducing the influence of nonlinear interference in the original MEG signals, effectively improving the filtering effect, and enhancing the robustness of the MEG signals. The obtained filtering results are fused to obtain fused filtering signals, thereby being able to combine the advantages of different filtering methods to improve the quality of the MEG signals and further improving the classification effect and classification accuracy of the MEG signals.

[0092] Refer to Figure 3 As shown, as an embodiment of this application, multi-strategy filtering is performed on the original MEG signals, and the filtering results are fused to obtain fused filtering signals, including

[0093] S110. Perform linear system filtering on the original MEG signals to obtain linear filtering signals.

[0094] S120. Perform nonlinear system filtering on the original MEG signals to obtain nonlinear filtering signals.

[0095] S130. Fuse the linear filtering signals and the nonlinear filtering signals to obtain fused filtering signals.

[0096] Specifically, among the filtering methods used in multi-strategy filtering, it includes at least one filtering method applicable to linear systems and at least one filtering method applicable to nonlinear systems. In this application, the filtering method applicable to linear systems is used to perform linear system filtering on the original magnetoencephalogram (MEG) signal to remove the stationary noise in the original MEG signal and obtain a linearly filtered signal; and the filtering method applicable to nonlinear systems is used to perform nonlinear system filtering on the original MEG signal to reduce the influence brought by the nonlinear interference in the original MEG signal and obtain a nonlinearly filtered signal. It can be understood that multiple different filtering methods can perform noise cancellation on the original MEG signal from different angles. After fusing multiple filtered signals, it is possible to combine the respective advantages of different filtering methods, thereby improving the filtering effect, effectively enhancing the signal quality of the MEG signal, and further providing an accurate signal basis for MEG signal classification.

[0097] Furthermore, the method of fusing the linearly filtered signal and the nonlinearly filtered signal can be weighted calculation. At this time, the fused filtered signal can be expressed according to the following formula:

[0098]

[0099] Wherein, is the fused filtered signal; is the linearly filtered signal; is the nonlinearly filtered signal; is the preset weight.

[0100] Referring to Figure 4 As shown, as an embodiment of this application, performing linear system filtering on the original MEG signal to obtain a linearly filtered signal includes:

[0101] S112. Update the estimated weight of the sample reference value corresponding to the original MEG signal according to the fused filtered signal at the previous moment.

[0102] S114. Obtain the motion artifact noise matrix of the original MEG signal by using the sample reference value and the estimated weight of the updated sample reference value.

[0103] S116. Perform linear system filtering on the original MEG signal according to the motion artifact noise matrix to obtain a linearly filtered signal.

[0104] Specifically, the linear system filtering method can be H∞ filtering. During the process of performing linear system filtering on the original MEG signal, the H∞ filter is used to perform channel-by-channel analysis on the original MEG signal, and the state is updated and noise is filtered according to the corresponding H∞ adaptive rule.

[0105] Furthermore, there is a corresponding H∞ adaptive rule for H∞ filtering with a time-varying weight assumption, and its form is expressed according to the following formula:

[0106]

[0107]

[0108]

[0109] wherein, is the estimated weight vector of the sample reference value; is the measurement sample point of the reference value vector; is the original magnetoencephalogram signal; is the noise matrix of the motion artifact; is the magnetoencephalogram signal after removing the motion artifact through H∞ filtering; is the noise covariance matrix,

[0110]

[0111] wherein, is the bound of the energy-energy gain from the interference to the output estimation error; is the prior information of the noise covariance matrix at time, and its update formula is expressed according to the following formula:

[0112]

[0113] wherein, is the prior information of the speed of change of the weight over time, is the identity matrix; The closer it is to 1, the closer the behavior of the H∞ filter is to the optimal filter. When is satisfied, it can be ensured that the system maintains stable performance under various degrees of interference.

[0114] It should be noted that during the H∞ filtering process, the estimated weight vector of the sample reference value is updated according to the fusion filtering signal at the previous moment of the current moment. By using the fusion filtering signal, the advantages of the nonlinear system filtering method are combined, the effect of the H∞ filtering is improved, and the signal quality is enhanced. In this application, the original magnetoencephalogram signal is filtered through a linear system, so as to filter the possible environmental noise, instrument noise and possible stationary noise in the original magnetoencephalogram signal, and reduce the influence of noise interference on the original magnetoencephalogram signal and subsequent research and analysis.

[0115] Referring to Figure 5 shown, as an embodiment of this application, the original magnetoencephalogram signal includes a radial signal and a tangential signal; the original magnetoencephalogram signal is subjected to nonlinear system filtering to obtain a nonlinear filtering signal, including:

[0116] S122. Obtain the predicted state of the radial signal at the current moment, as the radial prediction state, based on the fused filtered signal of the radial signal at the first moment and the external stimulus input signal at the second moment; wherein, both the first moment and the second moment are before the current moment.

[0117] S124. Obtain the predicted state of the tangential signal at the current moment, as the tangential prediction state, based on the fused filtered signal of the tangential signal at the first moment and the external stimulus input signal at the second moment.

[0118] S126. Perform component coupling on the radial prediction state and the tangential prediction state to obtain the coupled prediction state of the original magnetoencephalogram signal; and perform state observation on the coupled prediction state to obtain the non - linear filtered signal.

[0119] Specifically, the non - linear system filtering method can be particle filtering. In the process of performing non - linear system filtering on the original magnetoencephalogram signal, use the particle filter to perform channel - by - channel analysis on the original magnetoencephalogram signal to perform state estimation on the original magnetoencephalogram signal, and perform step - by - step state update on the original magnetoencephalogram signal according to the state transition model and the observation model.

[0120] Furthermore, the state transition model and the observation model of the particle filtering are represented by the following formulas:

[0121]

[0122]

[0123] Wherein, is the state estimation value of any measurement point at time is the radial magnetic field component of any measurement point at time the state estimation result after multi - strategy filtering and fusion of filtering results, is the tangential magnetic field component of any measurement point at time the state estimation result after multi - strategy filtering and fusion of filtering results; is the radial magnetic field component of any measurement point at time the external stimulus input received for motor imagery, is the tangential magnetic field component of any measurement point at time the external stimulus input received for motor imagery, if there is no external stimulus input, neither of them is represented; is the coupling coefficient; is the process noise in the radial direction of any measurement point at time is the process noise in the tangential direction of any measurement point at time is the measured value of any measurement point at time is the observation noise of any measurement point at time and is the weight coefficient.

[0124] It should be noted that due to the actual brain response mechanism, when the user performs a motor imagery task, the characteristic signal of the brain activity signal generally appears within 600 ms or 1000 ms after the external stimulus input. Therefore, there is a delayed feedback in the motor imagery task. In the state transition model of the particle filter, by designing the first time and the second time, a non-linear delay is introduced into the state transition model to describe the brain response mechanism in the actual situation. It can be understood that the first time, that is, the time is related to the sampling rate of the original magnetoencephalogram signal; the second time, that is, the time is related to the occurrence time of the external stimulus input.

[0125] Furthermore, there are magnetic field signals in different directions in the cerebral cortex. By measuring with an atomic magnetometer, the magnetic field component in the radial direction and the magnetic field component in the tangential direction of the cerebral cortex can be obtained, and there is a signal coupling behavior between these two magnetic field components, which affects the signal intensity, clarity, and spatial distribution characteristics of the two magnetic field components. In the state transition model of the particle filter, by designing the coupling coefficient , the signal coupling behavior between the radial magnetic field component signal and the tangential magnetic field component signal is characterized, so that the state transition model is more in line with the state transition situation in the actual situation.

[0126] In addition, it should be noted that, similar to the H∞ filter, in the process of particle filter, both the state transition model and the observation model are updated according to the fusion filter signal of the previous time. By using the fusion filter signal, the advantages of the H∞ filter and the particle filter are combined, the state prediction effect of the particle filter is improved, and the signal quality is further improved.

[0127] In this application, by filtering the original magnetoencephalogram signal with a non-linear system, the non-linear components in the original magnetoencephalogram signal are processed, the influence caused by the non-linear interference in the original magnetoencephalogram signal is reduced, the filtering effect is effectively improved, and the robustness of the magnetoencephalogram signal is improved.

[0128] Specifically describe the process of particle filter. First, initialize the particles and the corresponding weight values, and randomly sample to generate particles , and the initial weight value of each particle is , where the number of particles is one percent of the total number of samples. Update the weight value of each particle, and its update formula is expressed according to the following formula:

[0129]

[0130] where, is the weight value of the particle at the previous moment; is the observation probability density with the given particle state being .

[0131] Secondly, normalize the updated particle weight values, and the normalized particle weight value form is expressed according to the following formula:

[0132]

[0133] Update the state estimation result at the moment which is the current moment according to the normalized particle weight value, and its update formula is expressed according to the following formula:

[0134] .

[0135] Referring to Figure 6 shown, as an embodiment of the present application, fuse the linear filtering signal and the non - linear filtering signal to obtain a fused filtering signal, including:

[0136] S132. Adjust the preset weight of the linear filtering signal according to the signal - to - noise ratios of the linear filtering signal and the non - linear filtering signal respectively to obtain the fusion weight of the linear filtering signal.

[0137] S134. Perform weighted calculation on the linear filtering signal and the non - linear filtering signal according to the fusion weight to obtain the fused filtering signal.

[0138] Specifically, the adjustment formula of the preset weight is expressed according to the following formula:

[0139]

[0140] where, represents the signal - to - noise ratio of the linear filtering signal, represents the signal - to - noise ratio of the non - linear filtering signal. The higher the signal - to - noise ratio, the higher the quality of the signal. By increasing the weight corresponding to the signal with a higher signal - to - noise ratio, the fusion effect is improved to adapt to the noise changes in the actual scenario, and further improve the signal quality of the fused filtering signal.

[0141] Referring to Figure 7As shown, as an embodiment of the present application, signal reconstruction is performed on the original magnetoencephalogram (MEG) signal according to the signal modal features that meet the relevance conditions to obtain a reconstructed MEG signal, including:

[0142] S232. Perform information entropy detection on the signal modal features to obtain the relevance of the signal modal features.

[0143] S234. Screen the signal modal features according to the relevance conditions to obtain signal reconstruction features that meet the relevance conditions.

[0144] S236. Perform signal reconstruction according to the signal reconstruction features to obtain a reconstructed MEG signal.

[0145] Specifically, calculate the information entropy between the signal modal features obtained from the same magnetic field components, and determine the relevance between any two signal modal features according to the obtained results. It can be understood that in other embodiments, the relevance calculation can also be performed by other methods to select signal modal features that meet the conditions.

[0146] Furthermore, the form of the reconstructed MEG signal is represented by the following formula:

[0147]

[0148] where, is the reconstructed MEG signal; is the number of signal modal features that meet the relevance conditions. Exemplarily, can be 3.

[0149] Taking the position as an example, referring to Figures 8a to 8b shown, Figure 8a shows the signal waveforms of the component at the position after different steps. Figure 8b shows the signal waveforms of the component at the position after different steps. It can be seen that the component and the original MEG signals of the component are greatly affected by noise, and the characteristic components cannot be distinguished. After multi-strategy filtering processing, it can be seen that even when the amplitude of the power frequency interference is significantly higher than the amplitude of the original MEG signal, the motion artifacts in the original MEG signal can still be effectively removed, achieving a good noise removal effect. After signal reconstruction processing, the obtained reconstructed MEG signal only contains characteristic components related to the motor imagery task, and the signal quality is high, providing a high-quality signal basis for subsequent MEG signal classification. In this embodiment, the accuracy of MEG signal classification by the reconstructed MEG signal is increased by about five percentage points compared with the accuracy of a single algorithm.

[0150] As an embodiment of the present application, the method further includes:

[0151] S400. Classify the reconstructed magnetoencephalogram (MEG) signals to obtain the MEG signal classification result.

[0152] Specifically, perform signal classification according to the reconstructed MEG signals to distinguish the signal components in the MEG signals that belong to different intentions and actions, and obtain the characteristic component signals of the reconstructed MEG signals as the MEG signal classification result. It can be understood that the characteristic component signals of the reconstructed MEG signals respectively represent different action intentions of the user. By classifying the MEG signals, the usability of the MEG signals is enhanced, the control effect of the brain-computer interface technology on external devices is improved, and at the same time, accurate MEG signal characteristics can be provided for subsequent research on brain-computer interface technology.

[0153] As an embodiment of the present application, classifying the reconstructed MEG signals to obtain the MEG signal classification result includes:

[0154] S410. Input the reconstructed MEG signals and the corresponding tag set into a deep learning neural network model to classify the reconstructed MEG signals and output the MEG signal classification result; wherein, the deep learning neural network model is trained by the reconstructed MEG signals before the current moment.

[0155] Specifically, the MEG signal acquisition process can be divided into three stages. In the first stage and the second stage, the deep learning neural network model is trained according to the motor imagery data of the user's left hand and right hand respectively, and the MEG signals of the user are not classified at this time. During the training process of the first stage and the second stage, the user receives an instruction signal and performs corresponding motor imagery according to the instruction signal. The MEG signals of the user at this time are collected by an atomic magnetometer, and the MEG signals and the corresponding tag set are used as training data to train the deep learning neural network model. In the third stage, the trained deep learning neural network model is used to perform real-time classification on the reconstructed MEG signals to obtain the characteristic component signals in the user's MEG signals as the MEG signal classification result for output.

[0156] Correspondingly, please refer to Figure 9 , an embodiment of the present application provides a device for preprocessing MEG signals, and the device includes:

[0157] A fusion filtering module 910, configured to perform multi-strategy filtering on the original MEG signals and fuse the filtering results to obtain a fusion filtering signal; wherein, the multi-strategy filtering includes at least one filtering method applicable to a linear system and at least one filtering method applicable to a nonlinear system.

[0158] The feature extraction module 920 is configured to perform frequency decomposition and feature extraction based on the fused filtered signal to obtain signal modal features.

[0159] The signal reconstruction module 930 is configured to perform signal reconstruction on the original magnetoencephalogram signal according to the signal modal features that meet the correlation condition to obtain a reconstructed magnetoencephalogram signal.

[0160] In some alternative embodiments, the fused filtering module 910 includes:

[0161] A linear filtering unit configured to perform linear system filtering on the original magnetoencephalogram signal to obtain a linearly filtered signal.

[0162] A non-linear filtering unit configured to perform non-linear system filtering on the original magnetoencephalogram signal to obtain a non-linearly filtered signal.

[0163] A signal fusion unit configured to fuse the linearly filtered signal and the non-linearly filtered signal to obtain a fused filtered signal.

[0164] In some alternative embodiments, the linear filtering unit includes:

[0165] A weight update sub-unit configured to update the estimated weight of the sample reference value corresponding to the original magnetoencephalogram signal according to the fused filtered signal at the previous moment.

[0166] A noise determination sub-unit configured to obtain the motion artifact noise matrix of the original magnetoencephalogram signal by using the sample reference value and the estimated weight of the updated sample reference value.

[0167] A noise filtering sub-unit configured to perform linear system filtering on the original magnetoencephalogram signal according to the motion artifact noise matrix to obtain a linearly filtered signal.

[0168] In some alternative embodiments, the original magnetoencephalogram signal includes a radial signal and a tangential signal; the non-linear filtering unit includes:

[0169] A radial prediction sub-unit configured to obtain the prediction state of the radial signal at the current moment as the radial prediction state according to the fused filtered signal of the radial signal at the first moment and the external stimulus input signal at the second moment; wherein both the first moment and the second moment are before the current moment.

[0170] A tangential prediction sub-unit configured to obtain the prediction state of the tangential signal at the current moment as the tangential prediction state according to the fused filtered signal of the tangential signal at the first moment and the external stimulus input signal at the second moment.

[0171] A component coupling sub-unit configured to perform component coupling on the radial prediction state and the tangential prediction state to obtain the coupled prediction state of the original magnetoencephalogram signal; and perform state observation on the coupled prediction state to obtain a non-linearly filtered signal.

[0172] In some alternative embodiments, the signal fusion unit includes:

[0173] A weight adjustment subunit, configured to adjust a preset weight of the linear filtering signal according to the signal-to-noise ratios of the linear filtering signal and the non-linear filtering signal respectively, so as to obtain a fusion weight of the linear filtering signal.

[0174] A weighted calculation subunit, configured to perform weighted calculation on the linear filtering signal and the non-linear filtering signal according to the fusion weight, so as to obtain a fusion filtering signal.

[0175] In some alternative embodiments, the signal reconstruction unit includes:

[0176] A correlation degree calculation subunit, configured to perform information entropy detection on the signal modal features to obtain the correlation degree of the signal modal features.

[0177] A correlation degree screening subunit, configured to screen the signal modal features according to the correlation degree condition, so as to obtain signal reconstruction features that meet the correlation degree condition.

[0178] A signal reconstruction subunit, configured to perform signal reconstruction according to the signal reconstruction features to obtain a reconstructed magnetoencephalogram signal.

[0179] In some alternative embodiments, the device further includes:

[0180] A signal classification module, configured to classify the reconstructed magnetoencephalogram signal to obtain a magnetoencephalogram signal classification result.

[0181] In some alternative embodiments, the signal classification module includes:

[0182] A feature component signal classification unit, configured to input the reconstructed magnetoencephalogram signal and the corresponding label set into a deep learning neural network model to classify the reconstructed magnetoencephalogram signal and output a magnetoencephalogram signal classification result; wherein, the deep learning neural network model is trained by the reconstructed magnetoencephalogram signals before the current moment.

[0183] The further function descriptions of the above-mentioned various modules and units are the same as those in the corresponding embodiments above, and will not be elaborated here.

[0184] The magnetoencephalogram signal preprocessing device in this embodiment is presented in the form of functional units. Here, the unit refers to an ASIC (Application Specific Integrated Circuit) circuit, a processor and a memory that execute one or more software or fixed programs, and / or other devices that can provide the above functions.

[0185] Please refer to Figure 10 ,Figure 10 This is a schematic structural diagram of a computer device provided by an embodiment of the present application. As shown in the figure, the computer device includes: one or more processors 10, a memory 20, and interfaces for connecting various components, including a high-speed interface and a low-speed interface. Each component communicates with each other using different buses and can be installed on a common motherboard or installed in other ways as needed. The processor can process instructions executed within the computer device, including instructions stored in the memory or on the memory to display graphical information of the GUI on an external input / output device (such as a display device coupled to the interface). In some alternative embodiments, if necessary, multiple processors and / or multiple buses can be used together with multiple memories and multiple memories. Similarly, multiple computer devices can be connected, and each device provides some necessary operations (such as a server array, a set of blade servers, or a multi-processor system). Figure 10 Here, a processor 10 is taken as an example.

[0186] The processor 10 can be a central processing unit, a network processor, or a combination thereof. Among them, the processor 10 can further include a hardware chip. The above hardware chip can be an application-specific integrated circuit, a programmable logic device, or a combination thereof. The above programmable logic device can be a complex programmable logic device, a field programmable gate array, a general array logic, or any combination thereof.

[0187] Among them, the memory 20 stores instructions executable by at least one processor 10, so that the at least one processor 10 executes the method shown in the above embodiment.

[0188] The memory 20 can include a program storage area and a data storage area. Among them, the program storage area can store an operating system and application programs required for at least one function; the data storage area can store data created according to the use of the computer device, etc. In addition, the memory 20 can include a high-speed random access memory and can also include a non-transitory memory, such as at least one disk storage device, a flash memory device, or other non-transitory solid-state storage devices. In some alternative embodiments, the memory 20 can optionally include a memory remotely set relative to the processor 10, and these remote memories can be connected to the computer device through a network. Examples of the above network include but are not limited to the Internet, an enterprise intranet, a local area network, a mobile communication network, and combinations thereof.

[0189] The memory 20 can include a volatile memory, such as a random access memory; the memory can also include a non-volatile memory, such as a flash memory, a hard disk, or a solid-state drive; the memory 20 can also include a combination of the above types of memories.

[0190] The computer device further includes a communication interface 30 for the computer device to communicate with other devices or a communication network.

[0191] The embodiments of the present application also provide a computer-readable storage medium. The methods according to the embodiments of the present application can be implemented in hardware, firmware, or be implemented as computer code that can be recorded on a storage medium, or be implemented as computer code that is originally stored in a remote storage medium or a non-transitory machine-readable storage medium and downloaded through a network and will be stored in a local storage medium, so that the methods described herein can be stored in such software processes on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. Among them, the storage medium can be a magnetic disk, an optical disk, a read-only memory, a random access memory, a flash memory, a hard disk, or a solid-state drive, etc.; further, the storage medium can also include a combination of the above types of memories. It can be understood that a computer, a processor, a microprocessor controller, or programmable hardware includes a storage component that can store or receive software or computer code. When the software or computer code is accessed and executed by the computer, the processor, or the hardware, the methods shown in the above embodiments are implemented.

[0192] The embodiments of the present application provide a computer program product. The computer program product includes computer instructions, and the computer instructions are stored in a computer-readable storage medium. The processor of the computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device executes the methods of any embodiment of the present application.

[0193] Although the embodiments of the present application are described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of the present application, and such modifications and variations all fall within the scope defined by the appended claims.

[0194] The systems, devices, modules, or units illustrated in the above embodiments can be specifically implemented by a computer chip or an entity, or by a product with certain functions. A typical implementation device is a computer. Specifically, the computer can be, for example, a personal computer, a laptop computer, a cellular phone, a camera phone, a smart phone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or a combination of any of these devices.

[0195] For the convenience of description, when describing the above devices, they are described separately as various units according to functions. Of course, when implementing the present application, the functions of each unit can be implemented in the same or multiple software and / or hardware.

[0196] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code.

[0197] The present application is described with reference to the flowcharts and / or block diagrams of methods, apparatuses (systems), and computer program products according to the embodiments of the present application. It should be understood that each flow and / or block in the flowchart and / or block diagram can be implemented by computer program instructions, and the combination of the flows and / or blocks in the flowchart and / or block diagram can also be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing devices generate means for implementing the functions specified in Figure 1 one or more of the flows Figure 1 or blocks or the combination of the flows and / or blocks.

[0198] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer-readable memory generate a manufactured article including instruction means, and the instruction means implements the functions specified in Figure 1 one or more of the flows Figure 1 or blocks or the combination of the flows and / or blocks.

[0199] These computer program instructions can also be loaded onto a computer or other programmable data processing device, so that a series of operation steps are executed on the computer or other programmable device to generate a computer-implemented process, and thus the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in Figure 1 one or more of the flows Figure 1 or blocks or the combination of the flows and / or blocks.

[0200] It should also be noted that the term "comprise", "include" or any other variant thereof is intended to cover non-exclusive inclusion, such that a process, method, commodity or device comprising a series of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, commodity or device. Without further limitation, an element defined by the statement "comprising an..." does not exclude the presence of additional identical elements in the process, method, commodity or device comprising said element.

[0201] Each embodiment in this specification is described in a progressive manner. For the identical or similar parts among the embodiments, reference can be made to each other, and the differences between each embodiment and other embodiments are emphasized. In particular, for the system embodiments, since they are basically similar to the method embodiments, they are described relatively simply, and reference can be made to the relevant parts of the method embodiments for the relevant content.

[0202] The above description is only for the embodiments of the present application and is not intended to limit the present application. For those skilled in the art, various changes and modifications can be made to the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included within the scope of the claims of the present application.

[0203] Although the embodiments of the present application are described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of the present application, and such modifications and variations fall within the scope defined by the appended claims.

Claims

1. A method for preprocessing magnetoencephalogram signals, characterized in that It is used for preprocessing the raw magnetoencephalogram (MEG) signals collected in real time, and the raw MEG signals include radial signals and tangential signals; the method includes: Performing multi-strategy filtering on the raw MEG signals at the current moment, and fusing the filtering results to obtain the fused filtering signal at the current moment; wherein, the multi-strategy filtering is a filtering process including two or more different filtering methods, including at least one linear filtering method applicable to a linear system and at least one non-linear filtering method applicable to a non-linear system; specifically, performing linear system filtering on the raw MEG signals at the current moment according to the fused filtering signal at the previous moment of the current moment to obtain a linear filtering signal; performing non-linear system filtering on the raw MEG signals at the current moment according to the fused filtering signals at the previous moments of the current moment to obtain a non-linear filtering signal; fusing the linear filtering signal and the non-linear filtering signal to obtain the fused filtering signal at the current moment; wherein, the non-linear filtering signal is obtained by performing state observation on the coupled prediction state, and the coupled prediction state is obtained by component coupling of the tangential prediction state of the tangential signal at the current moment and the radial prediction state of the radial signal at the current moment; Performing frequency decomposition and feature extraction based on the fused filtering signal at the current moment to obtain signal modal features; Performing signal reconstruction on the raw MEG signals according to the signal modal features satisfying the relevance condition to obtain reconstructed MEG signals.

2. The method according to claim 1, characterized in that, The step of performing linear system filtering on the raw MEG signals at the current moment according to the fused filtering signal at the previous moment of the current moment to obtain a linear filtering signal includes: Updating the estimated weight of the sample reference value corresponding to the raw MEG signals at the current moment according to the fused filtering signal at the previous moment; Obtaining the motion artifact noise matrix of the raw MEG signals by using the sample reference value and the updated estimated weight of the sample reference value; Performing linear system filtering on the raw MEG signals according to the motion artifact noise matrix to obtain the linear filtering signal.

3. The method according to claim 1, wherein The step of performing non-linear system filtering on the raw MEG signals at the current moment according to the fused filtering signals at the previous moments of the current moment to obtain a non-linear filtering signal includes: Obtaining the prediction state of the radial signal at the current moment, as the radial prediction state, according to the fused filtering signal of the radial signal at the first moment and the external stimulus input signal at the second moment; wherein, both the first moment and the second moment are before the current moment; Obtaining the prediction state of the tangential signal at the current moment, as the tangential prediction state, according to the fused filtering signal of the tangential signal at the first moment and the external stimulus input signal at the second moment; Performing component coupling on the radial prediction state and the tangential prediction state to obtain the coupled prediction state of the raw MEG signals at the current moment; and performing state observation on the coupled prediction state to obtain the non-linear filtering signal.

4. The method according to claim 1, characterized in that, Fusing the linear filtering signal and the non - linear filtering signal to obtain the fused filtering signal includes: Adjusting a preset weight of the linear filtering signal according to the signal - to - noise ratios of the linear filtering signal and the non - linear filtering signal respectively to obtain a fusion weight of the linear filtering signal; Performing weighted calculation on the linear filtering signal and the non - linear filtering signal according to the fusion weight to obtain the fused filtering signal.

5. The method according to claim 1, wherein Reconstructing the original magnetoencephalogram signal according to the signal modal features satisfying the correlation condition to obtain a reconstructed magnetoencephalogram signal, including: Performing information entropy detection on the signal modal features to obtain the correlation degree of the signal modal features; Screening the signal modal features according to the correlation condition to obtain signal reconstruction features satisfying the correlation condition; Performing signal reconstruction according to the signal reconstruction features to obtain the reconstructed magnetoencephalogram signal.

6. The method according to any one of claims 1 to 5, characterized in that, The method further includes: Classifying the reconstructed magnetoencephalogram signal to obtain a classification result of the magnetoencephalogram signal.

7. The method according to claim 6, characterized in that Classifying the reconstructed magnetoencephalogram signal to obtain a classification result of the magnetoencephalogram signal, including: Inputting the reconstructed magnetoencephalogram signal and the corresponding label set into a deep - learning neural network model to classify the reconstructed magnetoencephalogram signal and output the classification result of the magnetoencephalogram signal; wherein, the deep - learning neural network model is trained by the reconstructed magnetoencephalogram signals before the current moment.

8. A magnetoencephalogram signal preprocessing device, characterized in that, For pre - processing a real - time acquired original magnetoencephalogram signal, the original magnetoencephalogram signal includes a radial signal and a tangential signal; the device includes: A fusion filtering module, configured to perform multi - strategy filtering on the original magnetoencephalogram signal at the current moment and fuse the filtering results to obtain the fused filtering signal at the current moment; wherein, the multi - strategy filtering is a filtering process including two or more different filtering methods, including at least one linear filtering method applicable to a linear system and at least one non - linear filtering method applicable to a non - linear system; specifically, performing linear - system filtering on the original magnetoencephalogram signal at the current moment according to the fused filtering signal at the previous moment of the current moment to obtain a linear filtering signal; performing non - linear - system filtering on the original magnetoencephalogram signal at the current moment according to the fused filtering signal at a moment before the previous moment of the current moment to obtain a non - linear filtering signal; fusing the linear filtering signal and the non - linear filtering signal to obtain the fused filtering signal at the current moment; wherein, the non - linear filtering signal is obtained by observing the coupled prediction state, and the coupled prediction state is obtained by component - coupling the tangential prediction state of the tangential signal at the current moment and the radial prediction state of the radial signal at the current moment; A feature extraction module, configured to perform frequency decomposition and feature extraction based on the fused filtering signal at the current moment to obtain signal modal features; A signal reconstruction module, configured to reconstruct the original magnetoencephalogram signal according to the signal modal features satisfying the correlation condition to obtain a reconstructed magnetoencephalogram signal.

9. A computer device, characterized in that, Including: A memory and a processor, which are communicatively connected to each other. Computer instructions are stored in the memory, and the processor executes the computer instructions to perform the method according to any one of claims 1 to 7.

Images