Electroencephalogram signal reconstruction method and system based on artifact removal

Abstract

The invention discloses an electroencephalogram signal reconstruction method and system based on artifact removal. The method comprises the steps that multi-channel electroencephalogram signals are obtained and subjected to filtering preprocessing; blind source separation is carried out through independent component analysis, and statistical independent signal source components and a hybrid matrix are obtained through decomposition; constructing an equivalent current dipole model to calculate a space artifact index, evaluating signal complexity in combination with a time sequence sample entropy, and judging and identifying an artifact component and a neural activity component through weighted fusion and a self-adaptive threshold value; neural activity components are reserved, artifact components are zeroed, and pure electroencephalogram signals are reconstructed through mixed matrix inverse transformation. According to the method, accurate identification and effective removal of the electroencephalogram artifacts are realized, and high-quality data are provided for electroencephalogram analysis.

Description

Technical Field

[0001] This invention relates to the field of signal processing technology, and in particular to a method and system for reconstructing electroencephalogram (EEG) signals based on artifact removal. Background Technology

[0002] Electroencephalography (EEG), as a non-invasive neurophysiological detection method, can record the electrical activity of neuronal populations in the cerebral cortex in real time with millisecond-level temporal resolution, playing a vital role in neuroscience research, clinical diagnosis, and brain-computer interfaces. With the deepening development of brain science research and the widespread application of brain-computer interface technology, the demand for high-quality EEG signals is increasingly urgent. However, EEG signals are inevitably contaminated by various physiological and non-physiological artifacts during the acquisition process, seriously affecting signal quality and the accuracy of subsequent analysis.

[0003] Artifacts in electroencephalogram (EEG) signals mainly include electrooculogram (EOG) artifacts, electromyogram (EMG) artifacts, electrocardiogram (ECG) artifacts, and external electromagnetic interference. The amplitudes of these artifact signals are often much larger than the true EEG signals, completely masking useful neural activity information. EOG artifacts are generated by eye movements and blinking, mainly manifesting as low-frequency, large-amplitude potential changes, most noticeable at the forehead electrodes. EMG artifacts originate from the contraction of head and neck muscles, typically appearing as high-frequency, sudden electrical signals. ECG artifacts are generated by the electric field produced by the heartbeat propagating to the scalp electrodes, exhibiting periodic spike waveforms. These artifacts not only affect the visual observation of EEG signals but, more importantly, lead to erroneous results in EEG feature extraction and pattern recognition.

[0004] Traditional EEG artifact removal methods mainly include frequency domain filtering, time domain filtering, and template matching-based artifact removal techniques. Frequency domain filtering methods suppress artifact signals in specific frequency bands by designing bandpass or notch filters. However, this method has significant limitations; when artifact signals overlap with useful neural signals in the frequency domain, the filtering process inevitably loses important neural activity information. Time domain filtering methods, such as moving average filtering and median filtering, can smooth signals and remove some noise, but their effectiveness is limited for artifact signals with large amplitudes, and they are prone to signal distortion and loss of temporal resolution.

[0005] Template matching-based artifact removal methods require the pre-establishment of standard templates for artifact signals, using pattern matching and subtraction operations to remove artifact components. However, these methods face challenges such as difficulty in template establishment, significant individual differences, and poor adaptability, making them ill-suited for handling complex and varied real-world artifact scenarios. Furthermore, traditional methods often employ fixed parameter settings and uniform processing strategies, failing to adapt to the individual characteristics of different subjects and signal variations under different experimental conditions, resulting in unstable artifact removal performance.

[0006] In recent years, the emerging independent component analysis (ICA) method has provided a new technical approach for EEG artifact removal. Based on the assumption of statistical independence of signals, this method can decompose mixed EEG signals into multiple independent components. However, existing ICA-based artifact removal methods mainly rely on manual visual judgment or simple statistical features in the component identification stage, lacking objective and quantitative automatic identification algorithms. This not only results in low efficiency but also susceptibility to subjective factors, making it difficult to guarantee identification accuracy. Furthermore, existing methods often consider only a single feature dimension when identifying artifact components, such as relying solely on frequency domain features or spatial distribution features, ignoring the multi-dimensional characteristics of artifact signals, leading to low identification accuracy.

[0007] More importantly, existing technologies lack adaptive thresholding mechanisms, mostly relying on fixed or empirical thresholds to identify artifact components. This approach cannot adapt to the differences in feature distribution across different datasets, exhibiting poor robustness and universality in complex real-world application scenarios. Therefore, there is an urgent need to develop a technique that can automatically, accurately, and adaptively identify and remove EEG artifacts to meet the practical needs of high-quality EEG signal processing. Summary of the Invention

[0008] In view of this, the present invention provides a method for reconstructing EEG signals based on artifact removal. The purpose is to establish a time-domain preprocessing mechanism for multi-channel EEG signals, combine it with blind source separation technology to decompose signal components, and construct an adaptive artifact component recognition algorithm based on the fusion of dipole model and time series entropy to achieve accurate identification and effective removal of artifact components in EEG signals, thereby providing high-quality clean signal data for EEG signal analysis.

[0009] To achieve the above objectives, the present invention provides a method for reconstructing electroencephalogram (EEG) signals based on artifact removal, comprising the following steps: B1: Acquire raw multi-channel EEG signal data streams from multiple electrodes on the scalp, perform time-domain preprocessing by setting high-pass and low-pass filters to filter out ultra-high frequency noise and DC drift, and obtain preprocessed multi-channel EEG signals; B2: Input the preprocessed multi-channel EEG signal into the independent component analysis algorithm for blind source separation, and decompose it to obtain statistically independent signal source components and corresponding mixing matrices; B3: Construct an equivalent current dipole model for each independent signal source component to calculate the spatial artifact index, and simultaneously calculate the time series sample entropy to assess signal complexity. Obtain a comprehensive artifact score through weighted fusion and perform adaptive threshold judgment to identify artifact components and neural activity components. B4: Retain neural activity components and remove zero artifact components, then use a hybrid matrix for linear recombination to reconstruct pure multichannel EEG signals.

[0010] Optionally, step B1 includes: A multi-channel EEG signal acquisition system was established, which acquired raw multi-channel EEG signal data streams through multiple electrodes worn on the scalp. The raw multi-channel EEG signals included effective signals from brain neural activity as well as various physiological artifacts and environmental artifacts. The raw multi-channel EEG signals were preprocessed in the time domain. The cutoff frequency of the high-pass filter was set to 0.5Hz to filter out DC drift and ultra-low frequency noise, and the cutoff frequency of the low-pass filter was set to 50Hz to filter out ultra-high frequency interference signals unrelated to brain neural activity. The preprocessed multi-channel EEG signals were obtained through bandpass filtering.

[0011] Optionally, step B2 includes: The preprocessed multi-channel EEG signals are input into an independent component analysis (ICA) algorithm for blind source separation. Based on the statistical independence assumption, the ICA algorithm decomposes the mixed multi-channel EEG signals into a set of statistically independent signal source components. The objective function of ICA is set as maximizing the statistical independence between the independent components, and the separation matrix is ​​solved through an iterative optimization algorithm. This ensures that the independent components of the output satisfy the statistical independence condition; thus, a set of independent component time series is obtained. and the corresponding mixing matrix ,in Indicates the number of independent components. For the first Each independent component The matrix transpose is represented by the mixed matrix. It records how each independent component linearly combines to form the original multichannel EEG signal, satisfying the relational formula. ,in This represents the preprocessed multichannel EEG signal.

[0012] Optionally, step B3 includes: An adaptive algorithm for identifying artifact components based on the fusion of the equivalent current dipole model and time series entropy is constructed, and for each independent component... Artifact identification and judgment are performed; firstly, the spatial artifact index is calculated, including: using independent components. In the mixing matrix The corresponding weight vector An equivalent current dipole source is fitted into a pre-defined three-dimensional human brain head model, and the spatial deviation between the location of the equivalent current dipole source and the brain parenchyma region is calculated. and fitting residual variance The spatial artifact index The calculation formula is: ; in, Indicates the spatial deviation weighting coefficient. This indicates the preset maximum deviation distance threshold. Let represent the residual variance weighting coefficient, and satisfy . , Indicates the first Spatial artifact index of an independent component; Secondly, the time series complexity features are calculated for independent components. Time series calculation of sample entropy The sample entropy The calculation formula is: ; in, Indicates the first The sample entropy of each independent component Indicates independent components The pattern length in the time series is And the distance is less than the tolerance. The number of pattern matches, Indicates independent components The pattern length in the time series is And the distance is less than the tolerance. The number of pattern matches, Indicates the pattern length parameter. This represents the tolerance parameter. Represents the natural logarithm function; Finally, a comprehensive artifact score is calculated and an adaptive threshold is determined. The comprehensive artifact score... The calculation formula is: ; in, Indicates the first The composite artifact score of each independent component, Represents the spatial feature weight coefficients. Represents the time feature weighting coefficient. Let represent the maximum value of the sample entropy among all independent components, and satisfy . Set adaptive threshold ,when Determining independent components As an artifact component, when Determining independent components It is a component of neural activity.

[0013] This step constructs a multi-dimensional fusion artifact recognition framework, capable of accurately identifying artifact components in EEG signals simultaneously from both spatial and temporal dimensions. By analyzing the distribution characteristics of independent components in brain space using an equivalent current dipole model, it is possible to effectively distinguish between genuine signals from intracranial neural activity and interference signals from physiological artifacts such as electromyography, electrooculography, and electrocardiography outside the brain. This is because genuine sources of neural activity are usually located in the cerebral cortex, while the equivalent dipole sources of artifact signals are often located away from the brain parenchyma.

[0014] This step calculates the complexity characteristics of the time series using sample entropy, which can quantify the irregularity and randomness of the signal. Neural activity signals typically exhibit high temporal complexity and irregularity, while artifact signals such as eye movements and heartbeats often have relatively regular periodic patterns. Therefore, sample entropy can serve as an effective time-domain feature for distinguishing neural signals from artifact signals.

[0015] The weighted fusion strategy in this step combines the advantages of spatial dipole analysis and temporal complexity analysis. By adaptively weighting coefficients to balance the contributions of the two features, it avoids misclassification problems that may occur with single-feature judgment. Spatial features mainly assess the physiological and anatomical rationality of the signal source, while temporal features focus on the complexity and randomness of the signal pattern. The two complement each other to form a more robust recognition mechanism.

[0016] Optionally, step B4 includes: Based on the artifact identification results in step B3, construct a clean independent component matrix. All independent components identified as artifacts are set to zero, while all independent components identified as neural activity components are retained; this is the clean independent component matrix. ,in Indicates the first A cleaned, independent component; the reconstructed pure multichannel EEG signal The calculation formula is: ; in, This indicates the reconstructed, clean multichannel EEG signal.

[0017] This step, by directly setting the identified artifact components to zero instead of simply deleting them, ensures that the spatial relationship between channels and the temporal structure of the original signal are completely preserved during signal reconstruction, avoiding the problems of neural signal distortion or loss that may be caused by traditional filtering methods.

[0018] This step leverages the reversibility of independent component analysis (ICA) to achieve a precise mapping from the independent component space to the original electrode space through the inverse transformation of the mixing matrix. This linear recombination method ensures the mathematical rigor and physical rationality of the reconstructed signal, enabling the cleaned EEG signal to accurately reflect the spatial distribution patterns and temporal dynamics of brain neural activity.

[0019] This invention also discloses an EEG signal reconstruction system based on artifact removal, comprising: Preprocessing module: Acquires raw multi-channel EEG signal data streams from multiple electrodes on the scalp, performs time-domain preprocessing by setting high-pass and low-pass filters to filter out ultra-high frequency noise and DC drift; Blind source separation module: The preprocessed multi-channel EEG signal is input into the independent component analysis algorithm for blind source separation, and the statistically independent signal source components and corresponding mixing matrix are obtained. The artifact detection module constructs an equivalent current dipole model for each independent component to calculate the spatial artifact index, while simultaneously calculating the time series sample entropy to assess signal complexity. It obtains a comprehensive artifact score through weighted fusion and performs adaptive threshold judgment to identify artifact components and neural activity components. Signal reconstruction module: Preserves neural activity components and removes zero artifact components, uses the inverse matrix of the hybrid matrix for linear recombination, and reconstructs pure multichannel EEG signals.

[0020] Compared with the prior art, the present invention has at least the following beneficial effects: This invention significantly improves the accuracy and reliability of EEG artifact component identification by constructing a multi-dimensional feature fusion identification framework that combines spatial feature analysis based on the equivalent current dipole model and temporal complexity analysis based on sample entropy. The equivalent current dipole model can assess the physiological rationality of independent components from a neuroanatomical perspective, effectively distinguishing between genuine neural activity originating from within the brain and physiological artifacts originating from outside the brain. Meanwhile, sample entropy analysis quantifies the irregularity of time series from the perspective of signal complexity. Neural activity signals typically exhibit high temporal complexity, while artifact signals often have relatively regular pattern characteristics.

[0021] This invention establishes an adaptive threshold determination mechanism. By analyzing the comprehensive artifact score distribution characteristics of all independent components in the current dataset, it automatically determines the optimal classification boundary without requiring manual setting of a fixed threshold or reliance on prior knowledge. This adaptive mechanism uses a clustering algorithm to perform unsupervised classification of artifact components and neural activity components, using the midpoint of the cluster centroid as a dynamic threshold. It can automatically adapt to individual differences among different subjects, changes in signal characteristics under different experimental conditions, and the characteristics of data collected by different devices.

[0022] This invention achieves complete preservation of useful neural information and thorough removal of artifacts through precise signal reconstruction technology. The reconstructed EEG signal has a higher signal-to-noise ratio and better data quality. The method employs a selective component preservation strategy, setting identified artifact components to zero while retaining neural activity components. Linear recombination is performed using the inverse transformation of the mixing matrix, ensuring the mathematical rigor and physical rationality of the signal reconstruction process. Compared with traditional frequency domain filtering or time domain filtering methods, the reconstruction method of this invention does not cause distortion or loss of useful neural signals and can completely maintain the temporal resolution, spatial distribution pattern, and spectral characteristics of the original EEG signal. Attached Figure Description

[0023] Figure 1 This is a flowchart illustrating an embodiment of the electroencephalogram (EEG) signal reconstruction method based on artifact removal according to the present invention. Figure 2 This is a schematic diagram of the preprocessing effect of EEG signals. (a) Original 64-channel EEG signal; (b) Signal after filtering and preprocessing. Detailed Implementation

[0024] The present invention will be further described below with reference to the accompanying drawings, but this is not intended to limit the present invention in any way. Any modifications or substitutions made based on the teachings of the present invention shall fall within the protection scope of the present invention.

[0025] Example 1: A method for reconstructing EEG signals based on artifact removal, such as Figure 1 As shown, it includes the following steps: The EEG signal reconstruction method based on artifact removal provided by this invention includes the following steps: B1: Acquire raw multi-channel EEG signal data streams from multiple electrodes on the scalp. Perform time-domain preprocessing by setting high-pass and low-pass filters to filter out ultra-high frequency noise and DC drift, obtaining preprocessed multi-channel EEG signals: A multi-channel EEG signal acquisition system was established, acquiring raw multi-channel EEG signal data streams through multiple electrodes worn on the scalp. The raw multi-channel EEG signals include valid signals from brain neural activity as well as various physiological and environmental artifacts. The raw multi-channel EEG signals underwent time-domain preprocessing. In this embodiment, a fourth-order Butterworth filter was used, with the high-pass filter cutoff frequency set to 0.5Hz to filter out DC drift and ultra-low frequency noise, and the low-pass filter cutoff frequency set to 50Hz to filter out ultra-high frequency interference signals unrelated to brain neural activity. The preprocessed multi-channel EEG signals were obtained through band-pass filtering. In this embodiment, the sampling frequency was set to 250Hz, and the filtered signals maintained the original time resolution. Figure 2 As shown, the comparison between the raw EEG signal and the filtered preprocessed signal is illustrated. Figure 2(a) is a raw 64-channel EEG signal segment (5 seconds long) containing DC drift and high-frequency noise, with a signal amplitude range of -150μV to 200μV. Low-frequency drift trend and high-frequency noise components can be clearly observed. Figure 2 (b) The preprocessed signal after bandpass filtering in the 0.5-50Hz range has effectively removed the DC component, significantly reduced high-frequency noise, and stabilized the signal amplitude within the range of -80μV to 80μV. As an alternative implementation, when there is significant interference in the signal, an elliptic filter or a notch filter can be used for filtering. The passband ripple of the elliptic filter is set to 0.1dB, and the stopband attenuation is set to 40dB. For signal acquisition scenarios in extreme noise environments, an adaptive filtering preprocessing module is further set up. By monitoring the signal power spectral density distribution in real time, when the noise power in a specific frequency band exceeds a preset threshold, the filter parameters are automatically adjusted. Specifically, when the power proportion in the 0.1-0.5Hz frequency band exceeds 30%, the cutoff frequency of the high-pass filter is dynamically adjusted to 1Hz, and when the power proportion in the 45-55Hz frequency band exceeds 20%, the 50Hz notch filter is activated.

[0026] B2: Input the preprocessed multi-channel EEG signal into an independent component analysis algorithm for blind source separation, decomposing it to obtain statistically independent signal source components and the corresponding mixing matrix: The preprocessed multi-channel EEG signals are input into an independent component analysis (ICA) algorithm for blind source separation. In this embodiment, the FastICA algorithm is used as the specific implementation method for ICA. The ICA algorithm, based on the statistical independence assumption, decomposes the mixed multi-channel EEG signals into a set of statistically independent signal source components. The objective function of ICA is set to maximize the statistical independence between the independent components. In this embodiment, negative entropy is used as the non-Gaussianity criterion, and the convergence threshold is set to... The maximum number of iterations is set to 1000, and the separation matrix is ​​solved using an iterative optimization algorithm. This ensures that the independent components of the output satisfy the statistical independence condition; thus, a set of independent component time series is obtained. and the corresponding mixing matrix ,in Indicates the number of independent components. For the first Each independent component The matrix transpose is represented by the mixed matrix. It records how each independent component linearly combines to form the original multichannel EEG signal, satisfying the relational formula. ,in This represents the preprocessed multichannel EEG signal. In this embodiment, the mixing matrix... It is a 64×64 square matrix, and the separation matrix W and the mixing matrix A are inverse matrices of each other.

[0027] B3: For each independent signal source component, an equivalent current dipole model is constructed to calculate the spatial artifact index. Simultaneously, the time-series sample entropy is calculated to assess signal complexity. A comprehensive artifact score is obtained through weighted fusion, and an adaptive threshold is applied to identify artifact components and neural activity components. An adaptive algorithm for identifying artifact components based on the fusion of the equivalent current dipole model and time series entropy is constructed, and for each independent component... Artifact identification and judgment are performed; firstly, the spatial artifact index is calculated, including: using independent components. In the mixing matrix The corresponding weight vector An equivalent current dipole source is fitted into a pre-defined three-dimensional human brain head model, and the spatial deviation between the location of the equivalent current dipole source and the brain parenchyma region is calculated. and fitting residual variance In this embodiment, the least squares method is used for dipole localization. When the variance of the fitting residual is greater than 0.15, the dipole position is considered unreliable. The spatial artifact index... The calculation formula is: ; in, Indicates the spatial deviation weighting coefficient. This indicates the preset maximum deviation distance threshold. Let represent the residual variance weighting coefficient, and satisfy . , Indicates the first The spatial artifact index of each independent component, in this embodiment... Set to 0.6. Set to 0.4. Set to 12cm; Secondly, the time series complexity features are calculated for independent components. Time series calculation of sample entropy The sample entropy The calculation formula is: ; in, Indicates the first The sample entropy of each independent component Indicates independent components The pattern length in the time series is And the distance is less than the tolerance. The number of pattern matches, Indicates independent components The pattern length in the time series is And the distance is less than the tolerance. The number of pattern matches, Indicates the pattern length parameter. This represents the tolerance parameter. Representing the natural logarithm function, in this embodiment, the mode length parameter... Set to 2, tolerance parameter The time series standard deviation is set to 0.2, and Chebyshev distance is used as the distance metric for pattern matching during the calculation. When the signal length is short or the signal-to-noise ratio is extremely low, the calculated sample entropy may be numerically unstable. In such cases, approximate entropy is used as an alternative metric for complexity. The formula for approximate entropy is as follows: ,in and These represent pattern lengths of 1 and 2 respectively. and The log-conditional probability mean at time ; Finally, a comprehensive artifact score is calculated and an adaptive threshold is determined. The comprehensive artifact score... The calculation formula is: ; in, Indicates the first The composite artifact score of each independent component, Represents the spatial feature weight coefficients. Represents the time feature weighting coefficient. Let represent the maximum value of the sample entropy among all independent components, and satisfy . In this embodiment, Set to 0.7. Set to 0.3; set adaptive threshold. ,when Determining independent components As an artifact component, when Determining independent components As a component of neural activity, in this embodiment, an adaptive threshold is used. By analyzing the distribution characteristics of the combined artifact scores of all independent components, the k-means clustering algorithm was used to divide the independent components into two clusters: an artifact cluster and a neural activity cluster. The midpoint of the centroids of the two clusters was used as an adaptive threshold. .

[0028] B4: Reconstructing pure multichannel EEG signals by preserving neural activity components and removing zero artifact components, using a hybrid matrix for linear recombination: Based on the artifact identification results in step B3, construct a clean independent component matrix. All independent components identified as artifacts are set to zero, while all independent components identified as neural activity components are retained; this is the clean independent component matrix. ,in Indicates the first After cleaning, each independent component is identified as an artifact component in this embodiment. When the first When an independent component is identified as a component of neural activity The reconstructed pure multichannel EEG signal The calculation formula is: ; in, This indicates the reconstructed, clean multichannel EEG signal.

[0029] Example 2: This invention also discloses an EEG signal reconstruction system based on artifact removal, comprising the following five modules: Preprocessing module: Acquires raw multi-channel EEG signal data streams from multiple electrodes on the scalp, performs time-domain preprocessing by setting high-pass and low-pass filters to filter out ultra-high frequency noise and DC drift; Blind source separation module: The preprocessed multi-channel EEG signal is input into the independent component analysis algorithm for blind source separation, and the statistically independent signal source components and corresponding mixing matrix are obtained. The artifact detection module constructs an equivalent current dipole model for each independent component to calculate the spatial artifact index, while simultaneously calculating the time series sample entropy to assess signal complexity. It obtains a comprehensive artifact score through weighted fusion and performs adaptive threshold judgment to identify artifact components and neural activity components. Signal reconstruction module: Preserves neural activity components and removes zero artifact components, uses the inverse matrix of the hybrid matrix for linear recombination, and reconstructs pure multichannel EEG signals.

[0030] It should be noted that the sequence numbers of the above embodiments of the present invention are merely for descriptive purposes and do not represent the superiority or inferiority of the embodiments. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, apparatus, article, or method that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, apparatus, article, or method. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, apparatus, article, or method that includes that element.

[0031] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, or network device, etc.) to execute the methods described in the various embodiments of the present invention.

[0032] The above are merely preferred embodiments of the present invention and do not limit the scope of the patent. Any equivalent structural or procedural transformations made based on the description and drawings of the present invention, or direct or indirect applications in other related technical fields, are similarly included within the scope of patent protection of the present invention.

Claims

1. A method for reconstructing electroencephalogram (EEG) signals based on artifact removal, characterized in that, Includes the following steps: B1: Acquire raw multi-channel EEG signal data streams from multiple electrodes on the scalp, perform time-domain preprocessing by setting high-pass and low-pass filters to filter out ultra-high frequency noise and DC drift, and obtain preprocessed multi-channel EEG signals; B2: Input the preprocessed multi-channel EEG signal into the independent component analysis algorithm for blind source separation, and decompose it to obtain statistically independent signal source components and corresponding mixing matrices; B3: Construct an equivalent current dipole model for each independent signal source component to calculate the spatial artifact index, and simultaneously calculate the time series sample entropy to assess signal complexity. Obtain a comprehensive artifact score through weighted fusion and perform adaptive threshold judgment to identify artifact components and neural activity components. B4: Retain neural activity components and remove zero artifact components, then use a hybrid matrix for linear recombination to reconstruct pure multichannel EEG signals.

2. The EEG signal reconstruction method based on artifact removal according to claim 1, characterized in that, Step B1 includes: A multi-channel EEG signal acquisition system was established, which acquired raw multi-channel EEG signal data streams through multiple electrodes worn on the scalp. The raw multi-channel EEG signals included effective signals from brain neural activity as well as various physiological artifacts and environmental artifacts. The raw multi-channel EEG signals were preprocessed in the time domain. The cutoff frequency of the high-pass filter was set to 0.5Hz to filter out DC drift and ultra-low frequency noise, and the cutoff frequency of the low-pass filter was set to 50Hz to filter out ultra-high frequency interference signals unrelated to brain neural activity. The preprocessed multi-channel EEG signals were obtained through bandpass filtering.

3. The method for reconstructing EEG signals based on artifact removal according to claim 2, characterized in that, Step B2 includes: The preprocessed multi-channel EEG signals are input into an independent component analysis (ICA) algorithm for blind source separation. Based on the statistical independence assumption, the ICA algorithm decomposes the mixed multi-channel EEG signals into a set of statistically independent signal source components. The objective function of ICA is set as maximizing the statistical independence between the independent components, and the separation matrix is ​​solved through an iterative optimization algorithm. This ensures that the independent components of the output satisfy the statistical independence condition; thus, a set of independent component time series is obtained. and the corresponding mixing matrix ,in Indicates the number of independent components. For the first Each independent component The matrix transpose is represented by the mixed matrix. It records how each independent component linearly combines to form the original multichannel EEG signal, satisfying the relational formula. ,in This represents the preprocessed multichannel EEG signal.

4. The method for reconstructing EEG signals based on artifact removal according to claim 3, characterized in that, Step B3 includes: An adaptive algorithm for identifying artifact components based on the fusion of the equivalent current dipole model and time series entropy is constructed, and for each independent component... Artifact identification and judgment are performed; firstly, the spatial artifact index is calculated, including: using independent components. In the mixing matrix The corresponding weight vector An equivalent current dipole source is fitted into a pre-defined three-dimensional human brain head model, and the spatial deviation between the location of the equivalent current dipole source and the brain parenchyma region is calculated. and fitting residual variance The spatial artifact index The calculation formula is: ； in, Indicates the spatial deviation weighting coefficient. This indicates the preset maximum deviation distance threshold. Let represent the residual variance weighting coefficient, and satisfy . , Indicates the first Spatial artifact index of an independent component; Secondly, the time series complexity features are calculated for independent components. Time series calculation of sample entropy The sample entropy The calculation formula is: ； in, Indicates the first The sample entropy of each independent component Indicates independent components The pattern length in the time series is And the distance is less than the tolerance. The number of pattern matches, Indicates independent components The pattern length in the time series is And the distance is less than the tolerance. The number of pattern matches, Indicates the pattern length parameter. This represents the tolerance parameter. Represents the natural logarithm function; Finally, a comprehensive artifact score is calculated and an adaptive threshold is determined.

5. The method for reconstructing EEG signals based on artifact removal according to claim 4, characterized in that, The determination of the comprehensive artifact score and adaptive threshold includes: Calculate the composite artifact score : ； in, Indicates the first The composite artifact score of each independent component, Represents the spatial feature weight coefficients. Represents the time feature weighting coefficient. Let represent the maximum value of the sample entropy among all independent components, and satisfy . Set adaptive threshold ,when Determining independent components As an artifact component, when Determining independent components It is a component of neural activity.

6. The method for reconstructing EEG signals based on artifact removal according to claim 4, characterized in that, Step B4 includes: Based on the artifact identification results in step B3, construct a clean independent component matrix. All independent components identified as artifacts are set to zero, while all independent components identified as neural activity components are retained; this is the clean independent component matrix. ,in Indicates the first A cleaned, independent component; the reconstructed pure multichannel EEG signal The calculation formula is: ； in, This indicates the reconstructed, clean multichannel EEG signal.

7. A brainwave signal reconstruction system based on artifact removal, characterized in that, include: Preprocessing module: Acquires raw multi-channel EEG signal data streams from multiple electrodes on the scalp, performs time-domain preprocessing by setting high-pass and low-pass filters to filter out ultra-high frequency noise and DC drift; Blind source separation module: The preprocessed multi-channel EEG signal is input into the independent component analysis algorithm for blind source separation, and the statistically independent signal source components and corresponding mixing matrix are obtained. The artifact detection module constructs an equivalent current dipole model for each independent component to calculate the spatial artifact index, while simultaneously calculating the time series sample entropy to assess signal complexity. It obtains a comprehensive artifact score through weighted fusion and performs adaptive threshold judgment to identify artifact components and neural activity components. Signal reconstruction module: Preserves neural activity components and removes zero artifact components, uses the inverse matrix of the mixing matrix for linear recombination, and reconstructs pure multi-channel EEG signals; To achieve the EEG signal reconstruction method based on artifact removal as described in any one of claims 1-6.

Images