An electroencephalogram motion artifact identification and hierarchical removal method, device, computing equipment and storage medium

Abstract

This invention discloses a method, apparatus, computing device, and storage medium for identifying and classifying motion artifacts in electroencephalograms (EEGs), belonging to the technical field of EEG signal processing. The method includes: acquiring EEG window data and IMU window data; extracting abnormal EEG features and motion prior features respectively; constructing an artifact confidence map containing channel-level confidence, time-level confidence, and frequency band-level confidence; determining the contamination level of the target EEG region and performing classified removal processing to obtain candidate EEG signals; and fusing the original EEG signal and candidate EEG signals using confidence reconstruction weights to obtain the target EEG output signal. This invention obtains the artifact confidence map through multi-dimensional coupling calculation of IMU motion priors and EEG data, and classifies the processing of EEGs with different levels of contamination. This accurately preserves the effective main information of the original EEG, effectively weakens signal distortion caused by motion coupling artifacts, and improves the overall fidelity and data continuity of the output EEG signal.

Description

Technical Field

[0001] This invention belongs to the technical field of electroencephalogram (EEG) signal processing, specifically relating to a method, apparatus, computing device, and storage medium for identifying and classifying EEG motion artifacts. Background Technology

[0002] Electroencephalogram (EEG) signals have high temporal resolution, making them suitable for applications such as brain-computer interfaces, neurorehabilitation, motor intention recognition, long-term wearable monitoring, and mobile health assessment. Under traditional laboratory conditions, subjects typically remain seated or perform tasks with controlled amplitude, leading to EEG contamination primarily from sources such as electrooculography (EOG), electromyography (EMG), power line interference, and localized interference. However, in scenarios involving walking, running, lower limb rehabilitation training, exoskeleton control, or natural interaction, continuous coupling occurs between the EEG acquisition device and human movement, making motion artifacts a major factor affecting signal usability.

[0003] The sources of motion artifacts are not singular. Head rotation, heel strike impact, changes in body posture, micro-slippage of electrodes relative to the scalp, lead traction, changes in contact impedance, and neck and shoulder muscle activity can all contribute to increased amplitude, low-frequency drift, enhanced spectral peaks near the main motion frequency, or multi-channel synchronous perturbations in EEG recordings. Compared to contamination in EEG and ECG, which are more likely to produce specific patterns, the spectral distribution and spatial diffusion of motion artifacts are more dependent on the acquisition posture, movement intensity, and device wearing conditions. In walking scenarios, gait frequency and its harmonics may overlap with low-frequency EEG rhythms; in head movement or rehabilitation training scenarios, sudden changes in angular velocity and electrode contact perturbations can create short-term high-amplitude impacts.

[0004] Existing methods for processing EEG artifacts mainly include filtering, adaptive filtering, independent component analysis, canonical correlation analysis, artifact subspace reconstruction, empirical mode decomposition, variational mode decomposition, and deep learning denoising. These methods are applicable to static or low-to-medium intensity contamination conditions, but several technical gaps remain when applied to dynamic EEG acquisition. First, many methods directly output denoising results, lacking explicit descriptions of the location and degree of contamination of motion artifacts. Second, when relying solely on EEG statistical features, real neural activity and motion-evoked components are easily aliased in both frequency band and temporal domain structures. Third, removal strategies using fixed thresholds, fixed filtering bands, or uniform intensity are difficult to adapt to different movement intensities and different hardware wearing states. Fourth, holistic reconstruction-based denoising may rewrite usable EEG components while suppressing artifacts.

[0005] IMUs can simultaneously record acceleration, angular velocity, and some attitude change information, which are directly related to the physical source of motion artifacts. Some existing solutions use accelerometers or IMUs as reference signals for adaptive cancellation or to assist in independent component screening. These methods demonstrate the value of motion reference signals in artifact detection; however, existing processing methods mostly remain at the level of reference compensation or overall mapping, and have not fully utilized IMU information to construct a contamination distribution description that can be transferred to subsequent removal and reconstruction stages. For practical dynamic acquisition systems, simply knowing that a certain segment of signal "needs denoising" is insufficient; it is also necessary to know which channels, time windows, and frequency bands the contamination mainly occurs in, and how the processing intensity should change with the degree of contamination. Summary of the Invention

[0006] The purpose of this invention is to address the aforementioned shortcomings of the prior art by providing a method, apparatus, computing device, and storage medium for EEG motion artifact recognition and hierarchical removal, thereby solving at least one of the following problems: (1) Existing methods are not explicit enough in identifying motion artifacts and are difficult to describe the contamination distribution at the channel, time and frequency band levels simultaneously; (2) When relying solely on EEG single-modal features, it is not easy to distinguish between motion-induced components and task-related neural activities; (3) Fixed thresholds or uniform removal intensity cannot be adapted to different motion intensities; (4) Overall denoising or overall restoration can easily cause unnecessary rewriting of effective EEG information in mildly contaminated areas.

[0007] To achieve the above objectives, the technical solution adopted by the present invention is as follows: In a first aspect, a method for identifying and classifying brainwave motion artifacts includes the following steps: S1. Simultaneously acquire multi-channel EEG signals and IMU signals, and preprocess the multi-channel EEG signals and IMU signals to obtain EEG window data and IMU window data; S2. Based on EEG window data and IMU window data, extract EEG abnormality features and motor prior features respectively; S3. Based on the cross-modal coupling features between motor prior features and EEG abnormal features, as well as between EEG window data and IMU window data, construct a artifact confidence map that includes channel-level confidence, time-level confidence, and frequency band-level confidence. S4. Determine the contamination level of the target EEG region based on the artifact confidence map, and perform graded removal processing according to the contamination level to obtain candidate EEG signals; S5. Generate confidence reconstruction weights based on the artifact confidence map, and use the confidence reconstruction weights to fuse the original EEG signal and candidate EEG signal to obtain the target EEG output signal.

[0008] Furthermore, in step S1, preprocessing of the multi-channel EEG signals and IMU signals includes: Time synchronization aligns multi-channel EEG signals and IMU signals to the same time axis; Resampling involves resampling multi-channel EEG and IMU signals to a common sampling rate. Basic filtering involves performing at least one of the following processing methods on multi-channel EEG signals: power frequency notch filtering, bandpass filtering, rereference, and abnormal channel detection; and performing at least one of the following processing methods on IMU signals: bias removal, low-pass filtering, bandpass filtering, and coordinate system correction. The window is segmented synchronously according to the preset window length and preset step size, so that the EEG window data and IMU window data in the same window correspond to the same acquisition time period.

[0009] Furthermore, in S2, the abnormal EEG features include at least one of the following: Abnormal EEG amplitude features are used to characterize baseline drift, spike peaks, or sustained high amplitude changes. Abnormal features of the electroencephalogram (EEG) spectrum are used to characterize abnormal energy enhancement in the low-frequency band, near the main motor frequency, or near harmonics. Abnormal features of EEG channel correlation are used to characterize the deviation of related structures between the target channel and adjacent channels; Local outlier features in EEG are used to characterize the degree of abnormality of a target channel relative to other channels within the same window.

[0010] Furthermore, in S2, the motion prior features include at least one of the following: The time-domain motion energy is calculated based on triaxial acceleration and / or triaxial angular velocity; Acceleration rate of change, angular velocity rate of change, and jerk feature calculated from adjacent sampling points or adjacent windows; The main motion frequency obtained based on IMU spectrum estimation; The harmonic distribution characteristics are calculated based on the harmonic position of the main motion frequency.

[0011] Furthermore, in S3, the cross-modal coupling features between motor prior features and EEG abnormality features, and between EEG window data and IMU window data, include: Temporal correlation between EEG window data and IMU window data within a preset time lag range; Frequency domain coherence between EEG window data and IMU window data within a preset frequency range; The degree of matching between the spectral peaks of EEG window data and the main motion frequencies and harmonics of IMU window data.

[0012] Furthermore, in step S3, constructing a pseudo-trace confidence map that includes channel-level confidence, time-level confidence, and frequency band-level confidence includes: Based on abnormal EEG amplitude, local outlier characteristics, temporal correlation, and motion dominant frequency matching characteristics, calculate the channel-level confidence of the target channel within the target window; The time-level confidence of the target window is calculated based on the channel-level confidence of multiple channels, the IMU time-domain motion energy, motion mutation characteristics, and harmonic distribution characteristics. Based on frequency domain coherence, degree of overlap within the main motion frequency band, and frequency band anomalous energy, calculate the frequency band-level confidence of the target channel in the target window and target frequency band. The channel-level confidence, time-level confidence, and frequency band-level confidence are weighted and fused to obtain a comprehensive artifact confidence map.

[0013] Furthermore, in step S4, the contamination level of the target EEG region is determined based on the artifact confidence map, and a graded removal process is performed based on the contamination level, including: When the confidence level of artifacts in the target EEG region is less than the first threshold, the target EEG region is identified as a slightly contaminated region, and reference signal constraint suppression processing is performed. When the artifact confidence level of the target EEG region is greater than or equal to the first threshold and less than the second threshold, the target EEG region is identified as a moderately contaminated region and decomposition and screening are performed. When the confidence level of artifacts in the target EEG region is greater than or equal to the second threshold, the target EEG region is identified as a heavily contaminated region, and local restoration processing is performed.

[0014] Furthermore, in step S4, the reference signal constraint suppression process is performed, including: Select an IMU dimension or combination thereof from the IMU window data that meets the preset conditions for coupling with the target EEG region, and use it as a reference input; Motion coupling components in the target EEG region are estimated using dynamic subspace projection. Low-damage suppression is applied to the motor coupling components, preserving the original EEG core structure of the mildly contaminated region in the output signal.

[0015] Furthermore, in step S4, the decomposition and filtering process includes: Wavelet decomposition was performed on the moderately polluted region in the target EEG to obtain multiple candidate components; Based on the correlation, coherence, and main motion frequency matching degree between the candidate components and the IMU window data, motion coupling scores are performed on the candidate components. Candidate components that meet the preset conditions for motor coupling scores are suppressed or eliminated, and the EEG signals corresponding to moderately polluted areas are reconstructed using the retained candidate components.

[0016] Furthermore, in step S4, the local recovery process includes: The heavily polluted EEG local blocks, adjacent low-pollution EEG local blocks, corresponding IMU motion prior features and artifact confidence maps are input into the local recovery model; The local recovery model generates the recovered EEG signal corresponding to the heavily polluted EEG local block; The recovered EEG signal is subjected to time-domain boundary continuity constraints and frequency-domain smoothing constraints, and the processed recovered EEG signal is then backfilled into the original EEG sequence.

[0017] Secondly, a device for identifying and classifying brainwave motion artifacts includes: The data acquisition and preprocessing module is used to simultaneously acquire multi-channel EEG signals and IMU signals, and preprocess the multi-channel EEG signals and IMU signals to obtain EEG window data and IMU window data. The EEG abnormality feature and motor prior module is used to extract EEG abnormality features and motor prior features based on EEG window data and IMU window data, respectively. The artifact confidence map generation module is used to construct artifact confidence maps containing channel-level confidence, time-level confidence, and frequency band-level confidence based on the cross-modal coupling features between motion prior features and EEG abnormal features, as well as between EEG window data and IMU window data. The graded removal module is used to determine the contamination level of the target EEG region based on the artifact confidence map, and to perform graded removal processing based on the contamination level to obtain candidate EEG signals; The confidence reconstruction module is used to generate confidence reconstruction weights based on the artifact confidence map, and to fuse the original EEG signal and candidate EEG signal using the confidence reconstruction weights to obtain the target EEG output signal.

[0018] Thirdly, a computing device includes a memory and a processor, wherein the memory stores computer-executable instructions, and the processor executes the computer-executable instructions to implement the steps of the above-described method.

[0019] Fourthly, a computer-readable storage medium having stored thereon computer-executable instructions that, when executed by a processor, implement the steps of the above-described method.

[0020] The method, apparatus, computing device, and storage medium for EEG motion artifact recognition and hierarchical removal provided by this invention have the following beneficial effects: 1. In this invention, IMU signals are used to generate motion priors and participate in artifact confidence map calculations, providing an external physical reference for motion artifact identification. For walking, running, and head movement scenarios, abrupt changes in principal motion frequency, harmonic structure, and angular velocity can serve as a basis for determining the source of EEG abnormalities.

[0021] 2. The artifact confidence map of this invention covers three dimensions: channel, time, and frequency band. Compared with the existing technology that only outputs contamination labels or overall denoising results, this structure can pass the contamination location and degree to subsequent processing steps, so that the removal strategy is no longer determined by a single fixed threshold.

[0022] 3. The graded removal strategy of this invention corresponds different treatment intensities to different levels of contamination. Mildly contaminated areas are mainly treated with low-damage suppression, moderately contaminated areas are treated with decomposition and screening, and severely contaminated areas are treated with triggering local recovery, thus reducing the uniform rewriting of still usable EEG structures.

[0023] 4. The confidence reconstruction mechanism of this invention correlates the proportion of original EEG retention with the confidence level of artifacts; more original signals are retained in areas with lighter contamination, while more artifact removal or recovery results are used in areas with heavier contamination, thereby forming a continuous regulation between motor contamination inhibition and neural information preservation. Attached Figure Description

[0024] Figure 1 This is a flowchart of the EEG motion artifact recognition and hierarchical removal method in this embodiment.

[0025] Figure 2 This is a schematic diagram illustrating the principle of the EEG motion artifact recognition and hierarchical removal method in this embodiment.

[0026] Figure 3 This is a schematic diagram of the synchronization window segmentation between EEG signals and IMU signals in this embodiment.

[0027] Figure 4 This is a schematic diagram illustrating the construction process of channel-level, time-level, and frequency-band-level artifact confidence maps in this embodiment.

[0028] Figure 5 This is a schematic diagram of the hierarchical removal strategy based on artifact confidence in this embodiment.

[0029] Figure 6 This is a schematic diagram of the confidence reconstruction process in this embodiment. Detailed Implementation

[0030] The specific embodiments of the present invention are described below to enable those skilled in the art to understand the present invention. However, it should be understood that the present invention is not limited to the scope of the specific embodiments. For those skilled in the art, various changes are obvious as long as they are within the spirit and scope of the present invention as defined and determined by the appended claims. All inventions utilizing the concept of the present invention are protected.

[0031] Example 1 This embodiment provides a method for identifying and classifying motion artifacts in electroencephalograms (EEGs). In this embodiment, the IMU motion prior not only serves as an additional input but also participates in motion artifact localization, contamination level determination, and output reconstruction weight control, thereby reducing the risk of overprocessing caused by uniform denoising. This embodiment is applicable to, but not limited to, scenarios involving walking, running, head movements, rehabilitation training, and wearable EEG monitoring. (See reference...) Figure 1 and Figure 2 Specifically, it includes the following: S1. Data acquisition and preprocessing; In some embodiments, reference Figure 3 The system acquires multi-channel EEG signals X and IMU signals M from an inertial measurement unit (IMU) that are acquired synchronously with the multi-channel EEG signals. It also preprocesses the multi-channel EEG signals and IMU signals to obtain EEG window data and IMU window data. Preprocessing of multichannel EEG and IMU signals includes: Time synchronization aligns multi-channel EEG signals and IMU signals to the same time axis; Resampling involves resampling multi-channel EEG and IMU signals to a common sampling rate. Basic filtering involves performing at least one of the following processing methods on multi-channel EEG signals: power frequency notch filtering, bandpass filtering, rereference, and abnormal channel detection; and performing at least one of the following processing methods on IMU signals: bias removal, low-pass filtering, bandpass filtering, and coordinate system correction. The window is segmented synchronously according to the preset window length and preset step size, so that the EEG window data and IMU window data in the same window correspond to the same acquisition time period.

[0032] In one specific embodiment, the EEG signals can be acquired by scalp electrodes, flexible electrodes, or wearable EEG devices; the IMU can be positioned on a head cap, forehead, acquisition host housing, or at a location rigidly connected to the electrode system. The IMU signals include at least one of triaxial acceleration, triaxial angular velocity, and attitude estimation.

[0033] In one specific embodiment, if the wearable EEG device and the IMU device share the same acquisition clock, they are directly aligned according to a unified timestamp; if they come from heterogeneous devices, alignment is achieved through trigger pulses, synchronization markers, or interpolation correction. Subsequently, the EEG signal and the IMU signal are resampled to a common sampling rate. The EEG signal undergoes power frequency notch filtering and bandpass filtering, while the IMU signal undergoes debiasing and low-pass or bandpass filtering.

[0034] The two are synchronously divided into windows according to the window length L and the step size H. The data in the k-th window is represented as follows: in, The EEG window data is in the k-th window; The data represents the IMU window data in the k-th window; C represents the number of EEG channels, Q represents the number of IMU dimensions, and T represents the number of sampling points in a single window. The window length can be adjusted according to the task; for walking tasks, it can cover at least one major gait cycle, while for head movement tasks, the window can be shortened to preserve the temporal resolution near pose abrupt changes.

[0035] S2. Based on EEG window data and IMU window data, extract EEG abnormality features and motor prior features respectively; In some embodiments, IMU window data Extracting prior motion features; First, based on the time-domain motion energy calculated from the triaxial acceleration and / or triaxial angular velocity, the time-domain motion energy is calculated. : in, This represents the value of the q-th IMU dimension at the t-th sampling point within the k-th window; It reflects the overall motion intensity within the current window, but it is not directly equivalent to the degree of brain electrical pollution. Therefore, it is used in conjunction with abnormal brain electrical features in subsequent steps. T This represents the total length of the time window. Q The total number of IMU dimensions; Calculate motion abrupt change characteristics; calculate the rate of change of acceleration, rate of change of angular velocity, and jerk features based on adjacent sampling points or adjacent windows; this embodiment takes jerk as an example, and the jerk features can be expressed as follows: : in, This represents the triaxial acceleration vector of the t-th sampling point within the k-th window; This represents the triaxial acceleration vector of the (t-1)th sampling point within the k-th window; It is a norm 2; It is highly sensitive to impacts from heel strikes, rapid head turns, and momentary slippage of equipment.

[0036] Spectral analysis of the IMU signal yields the main motion frequency. The harmonic energy ratio (harmonic distribution characteristics) is calculated based on the position of the harmonics of the main motion frequency. : in, For IMU signals at frequency The power spectrum at that location; The harmonic order; This is an index for the harmonic order; For frequency; Preset frequency range; To prevent constants with a denominator of zero; Used to characterize whether a movement rhythm exhibits a stable periodic structure. It is a minimal positive constant.

[0037] Extracting abnormal brainwave features based on EEG window data; EEG window data Extract abnormal EEG features; define robust amplitude abnormality for the c-th channel: in, and Let represent the robust mean and robust scale of the c-th channel within the k-th window, respectively. Using robust statistics can reduce the influence of extreme shock points on anomaly estimation. The abnormal amplitude characteristics of the electroencephalogram (EEG) Let be the EEG characteristic value of the c-th channel at time t in the k-th window.

[0038] Simultaneously, spectral anomalies, channel correlation anomalies, and local outlier features are extracted. Spectral anomalies mainly focus on energy enhancement in the low-frequency band, near the main motion frequency and harmonics; channel correlation anomalies are used to characterize the deviation of the target channel from the correlation structure of neighboring channels; local outlier features are used to characterize the degree of anomaly of a single channel relative to other channels within the same window.

[0039] S3. Generation of artifact confidence maps; refer to Figure 4 Based on the cross-modal coupling features between motor prior features and EEG abnormal features, as well as between EEG window data and IMU window data, a artifact confidence map containing channel-level confidence, time-level confidence, and frequency band-level confidence is constructed. In some embodiments, cross-modal coupling features between motor prior features and EEG abnormality features, and between EEG window data and IMU window data, include: Temporal correlation between EEG window data and IMU window data within a preset time lag range; Frequency domain coherence between EEG window data and IMU window data within a preset frequency range; The degree of matching between the spectral peaks of EEG window data and the main motion frequencies and harmonics of IMU window data.

[0040] In one specific embodiment, the temporal correlation between EEG window data and IMU window data within a preset time lag range is calculated; To accommodate situations where there is a time delay in the transmission of mechanical disturbances, a preset time delay set can be used. Take the maximum relevant response from within: in, Indicates temporal correlation; This represents the temporal data of the c-th EEG channel within the k-th window. This represents the time series data of the q-th IMU dimension within the same window; higher This indicates that the target channel changes synchronously with a certain motion dimension, but this quantity still needs to be determined in conjunction with abnormal EEG characteristics.

[0041] In one specific embodiment, the frequency domain coherence of EEG window data and IMU window data within a preset frequency range is calculated: in, Indicates frequency domain coherence; This represents the cross-spectrum of EEG and IMU. Indicates the autospectral density of brainwaves. This represents the IMU autospectrum. The frequency domain coherence... Used to describe whether a target channel is coupled with a motion process near a specific frequency.

[0042] In one specific embodiment, the degree of matching between the spectral peaks of the EEG window data and the main motion frequencies and harmonics of the IMU window data; Define the motion frequency enhancement ratio : in, This represents the power spectrum of the c-th EEG channel within the k-th window; the motor dominance frequency enhancement ratio. Linking abnormal EEG spectra with IMU main motor frequencies can reduce the possibility of misinterpretation of single-modal spectral peaks.

[0043] In some embodiments, constructing a pseudo-trace confidence map that includes channel-level confidence, time-level confidence, and frequency band-level confidence includes: Based on abnormal EEG amplitude, local outlier characteristics, temporal correlation, and motion dominant frequency matching characteristics, calculate the channel-level confidence of the target channel within the target window; First, construct the channel score: in, , , and These represent the standardized EEG amplitude abnormality features, temporal correlation, motor dominant frequency matching features, and local outlier features, respectively. Non-negative weights and .

[0044] Further, through Sigmoid mapping, we obtain: in, This represents the channel-level confidence level.

[0045] The time-level confidence of the target window is calculated based on the channel-level confidence of multiple channels, the IMU time-domain motion energy, motion mutation characteristics, and harmonic distribution characteristics. in, Non-negative weights and . The time-level confidence level is used to describe whether the entire window is in a state of high risk of motion contamination. , , These are the standardized time-domain motion energy, jerk characteristics, and harmonic energy ratio, respectively.

[0046] Based on frequency domain coherence, degree of overlap within the main motion frequency band, and frequency band anomalous energy, calculate the frequency band-level confidence of the target channel in the target window and target frequency band. Let the b-th frequency band be The frequency band level score is: in, Indicates the average coherence within the band. This indicates the degree of overlap between the main frequency and harmonics within that frequency band. This indicates abnormal energy within the band.

[0047] The artifact confidence map is obtained by weighted fusion of channel-level confidence, time-level confidence, and frequency band-level confidence. in, , and The weights are non-negative fusion weights and satisfy the following conditions: After generating the comprehensive artifact confidence map, smoothing constraints can be applied between adjacent channels, adjacent windows, and adjacent frequency bands to suppress fragmented judgments caused by isolated high-value points; G is the set of artifact confidence maps.

[0048] S4. Determine the contamination level of the target EEG region based on the artifact confidence map, and perform graded removal processing according to the contamination level to obtain candidate EEG signals; In some embodiments, reference Figure 5 Based on the confidence level of the artifact The target EEG region was graded; and a first threshold was set as... The second threshold is ,and In this embodiment, the threshold can be adaptively adjusted according to the motion intensity of the current window: in, Indicates the initial threshold. This represents the adjustment coefficient. This represents the normalization function. This design makes it easier for high-motion-intensity windows to trigger corresponding contamination level treatments.

[0049] when When the target region is identified as a mildly contaminated area, reference signal constraint suppression is applied. Specifically, reference signal constraint suppression involves selecting a reference input with a high degree of coupling to the target EEG region from the IMU dimension, estimating the motion coupling component using normalized least mean square, recursive least square, or dynamic subspace projection, and then performing low-damage suppression on the target EEG region. This layer does not actively reconstruct the entire waveform.

[0050] when When the target area is identified as a moderately polluted area, a decomposition and screening process is adopted. Specifically, the decomposition and screening process means: first, variational mode decomposition, wavelet decomposition, or empirical mode decomposition is performed on the target area to obtain multiple candidate components; then, the correlation, coherence, and main motion frequency matching degree of each candidate component with the IMU signal are calculated; finally, the components with high motion coupling scores are suppressed, and the target area is reconstructed using the retained components.

[0051] when When the target region is identified as a heavily contaminated area, local restoration processing is employed. Specifically, the input to the restoration model includes heavily contaminated EEG local blocks, adjacent low-contaminated EEG local blocks, corresponding IMU motion prior features, and artifact confidence maps. The restoration model can employ a temporal convolutional autoencoder, a local Transformer encoder, or other models capable of local temporal restoration. The restored EEG signal undergoes boundary continuity constraints before being backfilled into the original EEG sequence.

[0052] S5. Generate confidence reconstruction weights based on the artifact confidence map, and use the confidence reconstruction weights to fuse the original EEG signal and candidate EEG signal to obtain the target EEG output signal. refer to Figure 6 Candidate EEG signals were obtained after graded removal. To avoid completely replacing lightly contaminated areas, this embodiment generates original signals with retained weights based on the comprehensive artifact confidence level. : in, For weight decay parameters; when When smaller, A value close to 1 indicates that the raw EEG signal accounts for a large proportion of the output; when... When it is large, A decrease indicates that the candidate signal after hierarchical removal accounts for a larger proportion.

[0053] The target EEG output signal is: in, Represents raw brain electrical signals. This indicates the target EEG output signal.

[0054] In one implementation, the output phase may also check the residual coherence of the EEG-IMU near the main motion frequency, the energy collapse of the task-related frequency band, and the statistical continuity between adjacent windows. These checks are used to constrain output quality and do not constitute a limitation on specific experimental results.

[0055] Example 2 This embodiment provides a device for identifying and classifying EEG motion artifacts, including: The data acquisition and preprocessing module is used to simultaneously acquire multi-channel EEG signals and IMU signals, and preprocess the multi-channel EEG signals and IMU signals to obtain EEG window data and IMU window data. The EEG abnormality feature and motor prior module is used to extract EEG abnormality features and motor prior features based on EEG window data and IMU window data, respectively. The artifact confidence map generation module is used to construct artifact confidence maps containing channel-level confidence, time-level confidence, and frequency band-level confidence based on the cross-modal coupling features between motion prior features and EEG abnormal features, as well as between EEG window data and IMU window data. The graded removal module is used to determine the contamination level of the target EEG region based on the artifact confidence map, and to perform graded removal processing based on the contamination level to obtain candidate EEG signals; The confidence reconstruction module is used to generate confidence reconstruction weights based on the artifact confidence map, and to fuse the original EEG signal and candidate EEG signal using the confidence reconstruction weights to obtain the target EEG output signal.

[0056] Example 3 This embodiment provides a computing device, including a memory and a processor. The memory stores computer-executable instructions, and the processor executes the computer-executable instructions to implement all or part of the steps of the method in Embodiment 1.

[0057] Example 4 A computer-readable storage medium having stored thereon computer-executable instructions that, when executed by a processor, implement all or part of the steps of the method in Embodiment 1.

[0058] Although specific embodiments of the invention have been described in detail with reference to the accompanying drawings, this should not be construed as limiting the scope of protection of this patent. Various modifications and variations that can be made by a person skilled in the art without inventive effort within the scope described in the claims still fall within the scope of protection of this patent.

Claims

1. A method for identifying and classifying brainwave motion artifacts, characterized in that, Includes the following steps: S1. Simultaneously acquire multi-channel EEG signals and IMU signals, and preprocess the multi-channel EEG signals and IMU signals to obtain EEG window data and IMU window data; S2. Based on EEG window data and IMU window data, extract EEG abnormality features and motor prior features respectively; S3. Based on the cross-modal coupling features between motor prior features and EEG abnormal features, as well as between EEG window data and IMU window data, construct a artifact confidence map that includes channel-level confidence, time-level confidence, and frequency band-level confidence. S4. Determine the contamination level of the target EEG region based on the artifact confidence map, and perform graded removal processing according to the contamination level to obtain candidate EEG signals; S5. Generate confidence reconstruction weights based on the artifact confidence map, and use the confidence reconstruction weights to fuse the original EEG signal and candidate EEG signal to obtain the target EEG output signal.

2. The method for identifying and classifying EEG motion artifacts according to claim 1, characterized in that, In step S1, preprocessing of the multi-channel EEG signals and IMU signals includes: Time synchronization aligns multi-channel EEG signals and IMU signals to the same time axis; Resampling involves resampling multi-channel EEG and IMU signals to a common sampling rate. Basic filtering involves performing at least one of the following processing methods on multi-channel EEG signals: power frequency notch filtering, bandpass filtering, rereference, and abnormal channel detection; and performing at least one of the following processing methods on IMU signals: bias removal, low-pass filtering, bandpass filtering, and coordinate system correction. The window is segmented synchronously according to the preset window length and preset step size, so that the EEG window data and IMU window data in the same window correspond to the same acquisition time period.

3. The method for identifying and classifying EEG motion artifacts according to claim 1, characterized in that, In S2, the abnormal EEG features include at least one of the following: Abnormal EEG amplitude features are used to characterize baseline drift, spike peaks, or sustained high amplitude changes. Abnormal features of the electroencephalogram (EEG) spectrum are used to characterize abnormal energy enhancement in the low-frequency band, near the main motor frequency, or near harmonics. Abnormal features of EEG channel correlation are used to characterize the deviation of related structures between the target channel and adjacent channels; Local outlier features in EEG are used to characterize the degree of abnormality of a target channel relative to other channels within the same window.

4. The method for identifying and classifying EEG motion artifacts according to claim 1, characterized in that, In S2, the motion prior features include at least one of the following: The time-domain motion energy is calculated based on triaxial acceleration and / or triaxial angular velocity. Acceleration rate of change, angular velocity rate of change, and jerk feature calculated from adjacent sampling points or adjacent windows; The main motion frequency obtained based on IMU spectrum estimation; The harmonic distribution characteristics are calculated based on the harmonic position of the main motion frequency.

5. The method for identifying and classifying EEG motion artifacts according to claim 1, characterized in that, In S3, the cross-modal coupling features between motor prior features and EEG abnormality features, and between EEG window data and IMU window data, include: Temporal correlation between EEG window data and IMU window data within a preset time lag range; Frequency domain coherence between EEG window data and IMU window data within a preset frequency range; The degree of matching between the spectral peaks of EEG window data and the main motion frequencies and harmonics of IMU window data.

6. The method for identifying and classifying EEG motion artifacts according to claim 1, characterized in that, In step S3, a pseudo-trace confidence map is constructed, comprising channel-level confidence, time-level confidence, and frequency band-level confidence, including: Based on abnormal EEG amplitude, local outlier characteristics, temporal correlation, and motion dominant frequency matching characteristics, calculate the channel-level confidence of the target channel within the target window; The time-level confidence of the target window is calculated based on the channel-level confidence of multiple channels, the IMU time-domain motion energy, motion mutation characteristics, and harmonic distribution characteristics. Based on frequency domain coherence, degree of overlap within the main motion frequency band, and frequency band anomalous energy, calculate the frequency band-level confidence of the target channel in the target window and target frequency band. The channel-level confidence, time-level confidence, and frequency band-level confidence are weighted and fused to obtain a comprehensive artifact confidence map.

7. The method for identifying and classifying EEG motion artifacts according to claim 1, characterized in that, In step S4, the contamination level of the target EEG region is determined based on the artifact confidence map, and a graded removal process is performed based on the contamination level, including: When the confidence level of artifacts in the target EEG region is less than the first threshold, the target EEG region is identified as a slightly contaminated region, and reference signal constraint suppression processing is performed. When the artifact confidence level of the target EEG region is greater than or equal to the first threshold and less than the second threshold, the target EEG region is identified as a moderately contaminated region and decomposition and screening are performed. When the confidence level of artifacts in the target EEG region is greater than or equal to the second threshold, the target EEG region is identified as a heavily contaminated region, and local restoration processing is performed.

8. The method for identifying and classifying EEG motion artifacts according to claim 7, characterized in that, In step S4, the reference signal constraint suppression process is performed, including: Select an IMU dimension or combination thereof from the IMU window data that meets the preset conditions for coupling with the target EEG region, and use it as a reference input; Motion coupling components in the target EEG region are estimated using dynamic subspace projection. Low-damage suppression is applied to the motor coupling components, preserving the original EEG core structure of the mildly contaminated region in the output signal.

9. The method for identifying and classifying EEG motion artifacts according to claim 7, characterized in that, In step S4, the decomposition and filtering process is performed, including: Wavelet decomposition was performed on the moderately polluted region in the target EEG to obtain multiple candidate components; Based on the correlation, coherence, and main motion frequency matching degree between the candidate components and the IMU window data, motion coupling scores are performed on the candidate components. Candidate components that meet the preset conditions for motor coupling scores are suppressed or eliminated, and the EEG signals corresponding to moderately polluted areas are reconstructed using the retained candidate components.

10. The method for identifying and classifying EEG motion artifacts according to claim 7, characterized in that, In step S4, a partial recovery process is performed, including: The heavily polluted EEG local blocks, adjacent low-pollution EEG local blocks, corresponding IMU motion prior features and artifact confidence maps are input into the local recovery model; The local recovery model generates the recovered EEG signal corresponding to the heavily contaminated EEG local block; The recovered EEG signal is subjected to time-domain boundary continuity constraints and frequency-domain smoothing constraints, and the processed recovered EEG signal is then backfilled into the original EEG sequence.

11. A device for identifying and classifying brainwave motion artifacts, characterized in that, include: The data acquisition and preprocessing module is used to simultaneously acquire multi-channel EEG signals and IMU signals, and preprocess the multi-channel EEG signals and IMU signals to obtain EEG window data and IMU window data. The EEG abnormality feature and motor prior module is used to extract EEG abnormality features and motor prior features based on EEG window data and IMU window data, respectively. The artifact confidence map generation module is used to construct artifact confidence maps containing channel-level confidence, time-level confidence, and frequency band-level confidence based on the cross-modal coupling features between motion prior features and EEG abnormal features, as well as between EEG window data and IMU window data. The graded removal module is used to determine the contamination level of the target EEG region based on the artifact confidence map, and to perform graded removal processing based on the contamination level to obtain candidate EEG signals; The confidence reconstruction module is used to generate confidence reconstruction weights based on the artifact confidence map, and to fuse the original EEG signal and candidate EEG signal using the confidence reconstruction weights to obtain the target EEG output signal.

12. A computing device, comprising a memory and a processor, characterized in that, The memory stores computer-executable instructions, and the processor executes the computer-executable instructions to implement the steps of the method according to any one of claims 1 to 10.

13. A computer-readable storage medium having computer-executable instructions stored thereon, characterized in that, When the computer-executable instructions are executed by a processor, they implement the steps of the method according to any one of claims 1 to 10.

Images