Non-invasive EEG signal acquisition methods and devices

Abstract

This invention discloses a non-invasive method and apparatus for acquiring electroencephalogram (EEG) signals, relating to the field of EEG signal acquisition technology. The method includes: reading the sensor distribution scheme of the acquisition sensor group and performing noise signal identification; performing partitioned clustering and establishing partitioned clustering results; acquiring partitioned EEG signal datasets; performing noise intensity discrimination; establishing noise artifacts for the corresponding partitioned EEG signal datasets; configuring hysteresis noise in the partitioned channels using the noise artifacts, performing partitioned suppression of the corresponding partitioned EEG signal datasets, performing cross-channel signal authentication, and outputting the EEG signal acquisition results. This invention solves the technical problems of insufficient accuracy in noise identification and poor suppression effects in the existing technology during EEG signal acquisition, resulting in low accuracy and reliability of EEG signal acquisition. It achieves accurate identification and suppression of noise in EEG signals, improving the accuracy and reliability of EEG signal acquisition.

Description

Technical Field

[0001] This invention relates to the field of electroencephalogram (EEG) signal acquisition technology, specifically to a non-invasive method and apparatus for acquiring EEG signals. Background Technology

[0002] In the field of electroencephalogram (EEG) signal acquisition, non-invasive acquisition techniques are widely used due to their ease of operation and minimal damage to the subject. However, scalp EEG signals are susceptible to interference from environmental noise, physiological artifacts (such as electromyography and electrooculography), and electrode contact noise, resulting in significant shortcomings in noise identification and suppression in existing technologies. Traditional methods often employ fixed filtering or simple noise template matching, which are difficult to adapt to dynamic changes in complex noise environments. This leads to incomplete noise removal or over-filtering that damages the effective signal, thereby affecting the accuracy of EEG signal acquisition and the reliability of subsequent neural activity analysis.

[0003] Existing technologies suffer from insufficient accuracy in noise identification and poor suppression during EEG signal acquisition, resulting in low accuracy and reliability of EEG signal acquisition. Summary of the Invention

[0004] This application provides a non-invasive method and apparatus for acquiring electroencephalogram (EEG) signals, which addresses the technical problem that in the prior art, noise recognition is not accurate enough and the suppression effect is poor, resulting in low accuracy and reliability of EEG signal acquisition.

[0005] In view of the above problems, this application provides a non-invasive method and device for acquiring electroencephalogram (EEG) signals.

[0006] The first aspect of this application provides a non-invasive method for acquiring electroencephalogram (EEG) signals, the method comprising:

[0007] The sensor distribution scheme of the sensor group is read, and noise signal identification is performed to establish a noise source distribution. The sensor distribution scheme and the noise source distribution are used to perform partition clustering to establish partition clustering results. Partition EEG signal datasets are obtained based on the partition clustering results. Noise intensity is determined for the partition EEG signal datasets using the noise source identifiers from the partition clustering results. If the noise intensity determination result satisfies the activation condition, a noise artifact is established for the corresponding partition EEG signal dataset. After temporal alignment of the partition EEG signal datasets, hysteresis noise is configured for the partition channels using the noise artifacts. Partition suppression is then performed on the corresponding partition EEG signal datasets using the partition channels, and cross-channel signal authentication is performed to output the EEG signal acquisition results.

[0008] A second aspect of this application provides a non-invasive electroencephalogram (EEG) signal acquisition device, the device comprising:

[0009] The system includes the following modules: a noise source distribution establishment module, which reads the sensor distribution scheme of the sensor group and performs noise signal identification to establish a noise source distribution; a partition clustering result establishment module, which performs partition clustering using the sensor distribution scheme and the noise source distribution to establish partition clustering results; an EEG signal dataset acquisition module, which acquires partition EEG signal datasets based on the partition clustering results; a noise intensity discrimination module, which uses the noise source identifiers of the partition clustering results to discriminate the noise intensity of the partition EEG signal datasets; a noise artifact establishment module, which establishes noise artifacts for the corresponding partition EEG signal datasets if the noise intensity discrimination result satisfies the activation condition; and an EEG signal acquisition result output module, which aligns the partition EEG signal datasets temporally, configures the hysteresis noise of the partition channels using the noise artifacts, performs partition suppression of the corresponding partition EEG signal datasets using the partition channels, performs cross-channel signal authentication, and outputs the EEG signal acquisition results.

[0010] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0011] The system reads the sensor distribution scheme of the sensor group and performs noise signal identification to establish the noise source distribution; it then performs partitioned clustering to establish partitioned clustering results; based on the partitioned clustering results, it obtains partitioned EEG signal datasets; it performs noise intensity discrimination on the partitioned EEG signal datasets; if the noise intensity discrimination result satisfies the activation condition, it establishes a noise artifact for the corresponding partitioned EEG signal dataset; after temporally aligning the partitioned EEG signal datasets, it configures hysteresis noise for the partitioned channels using the noise artifacts, performs partitioned suppression on the corresponding partitioned EEG signal datasets using the partitioned channels, performs cross-channel signal authentication, and outputs the EEG signal acquisition results. This achieves the technical effect of accurately identifying and suppressing noise in EEG signals, improving the accuracy and reliability of EEG signal acquisition. Attached Figure Description

[0012] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0013] Figure 1 This is a schematic flowchart of a non-invasive electroencephalogram (EEG) signal acquisition method provided in an embodiment of this application.

[0014] Figure 2 This is a schematic diagram of the structure of a non-invasive electroencephalogram (EEG) signal acquisition device provided in an embodiment of this application.

[0015] Figure labeling: Module 10 for establishing noise source distribution, Module 20 for establishing partition clustering results, Module 30 for acquiring EEG signal dataset, Module 40 for noise intensity discrimination, Module 50 for establishing noise artifacts, and Module 60 for outputting EEG signal acquisition results. Detailed Implementation

[0016] This application provides a non-invasive method and device for acquiring electroencephalogram (EEG) signals, which addresses the technical problem that existing technologies suffer from inaccurate noise identification and poor noise suppression during EEG signal acquisition, resulting in low accuracy and reliability of EEG signal acquisition.

[0017] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0018] Example 1, as Figure 1 As shown, this application provides a non-invasive method for acquiring electroencephalogram (EEG) signals, the method comprising:

[0019] Step S100: Read the sensor distribution scheme of the acquisition sensor group and perform noise signal identification to establish the noise source distribution.

[0020] Specifically, the sensor distribution scheme of the sensor group is read to clarify the spatial layout of each sensor on the scalp surface. Then, noise signal identification is performed based on signal feature analysis (such as frequency, amplitude, time-domain waveform, etc.). The original signal is processed by noise extraction network components such as bandpass filter layer and wavelet decomposition layer to locate the spatial location of noise sources such as electromyographic interference and power frequency noise. Then, a noise source distribution including noise type and spatial distribution is established to provide basic data support for subsequent partitioning clustering and noise suppression.

[0021] Step S200: Perform partitioned clustering using the sensor distribution scheme and the noise source distribution to establish partitioned clustering results.

[0022] Specifically, utilizing the spatial location information of each sensor in the read sensor distribution scheme and the established correspondence between noise type and spatial location in the noise source distribution, a clustering algorithm based on spatial distance and noise characteristics is used for partitioned clustering. By calculating the spatial distance between the sensor location and the noise source, and combining the influence of noise source identifiers (such as power frequency noise, electromyographic noise, etc.) on the sensor signal, the sensor array is divided into different functional partitions. This ensures that the sensors in each partition have similar noise characteristics and EEG signal acquisition characteristics. Finally, a partitioned clustering result is established, including the acquisition center location of each partition, noise source identifiers, and sensor members, providing a spatial partitioning basis for subsequent acquisition of partitioned EEG signal datasets and noise suppression.

[0023] Step S300: Obtain the partitioned EEG signal dataset based on the partitioned clustering results.

[0024] Specifically, based on the spatial division, acquisition center, and sensor member information of each partition in the partition clustering results, signal data of the corresponding partitions are extracted from the raw EEG signals to form partition EEG signal datasets. During this process, regional signal anomaly identification is performed simultaneously on each partition dataset. Components such as peak truncation layers are used to detect abnormal signals exceeding the normal range (such as sudden impulse interference), anomaly identification results are established, and abnormal signals are reported. This ensures that the acquired partition EEG signal datasets are free of obvious outliers, providing a valid data foundation for subsequent noise intensity discrimination and noise artifact establishment.

[0025] Step S400: Use the noise source identifier of the partition clustering results to determine the noise intensity of the partitioned EEG signal dataset.

[0026] Specifically, based on the noise source identifiers (such as power frequency noise, electromyographic noise, environmental electromagnetic interference, etc. and spatial location information) of each partition in the partition clustering results, the bandpass filter layer is used to segment the partition EEG signal dataset by frequency band. The wavelet decomposition layer extracts the characteristic frequency band energy values ​​corresponding to different noise sources, and the peak value extraction layer identifies the peak value of the noise amplitude. Combined with the noise model corresponding to the noise source identifier (such as the 50Hz / 60Hz characteristic frequency of power frequency noise and the high-frequency energy distribution of electromyographic noise), the intensity parameters of the noise signal in the time domain and frequency domain are calculated. The noise intensity level is determined by comparing with the preset threshold. For example, when the power frequency noise energy ratio of a certain partition exceeds 30%, it is determined to be high-intensity noise.

[0027] Step S500: If the noise intensity discrimination result is a result that meets the activation condition, then establish the noise artifact of the corresponding partition EEG signal dataset.

[0028] Specifically, when the noise intensity discrimination result meets the preset activation conditions (such as the noise energy ratio exceeding a threshold or the amplitude peak exceeding the normal range), a noise artifact is established for the corresponding partition's EEG signal dataset. First, the noise source partition is located based on the partition clustering results. A noise extraction network containing a bandpass filter layer, a wavelet decomposition layer, and a peak truncation layer is configured, and the noise extraction weights of each layer are set according to the noise source identifier (such as power frequency noise, electromyographic noise, etc.). Then, the network is used to receive the EEG signal dataset of the corresponding partition, and the noise frequency bands are separated by the bandpass filter layer, the multi-scale noise features are extracted by the wavelet decomposition layer, and the peak amplitude peak is located by the peak truncation layer. The noise signal is reconstructed based on the extraction weights to form the noise extraction result. Finally, the noise extraction result is time-series fitted to generate a noise artifact that can characterize the temporal features of the noise in the partition, providing a noise model basis for subsequent hysteresis noise configuration and partition suppression.

[0029] Step S600: After time-aligning the partitioned EEG signal dataset, the hysteresis noise of the partitioned channel is configured using the noise artifacts. Then, partitioned suppression of the corresponding partitioned EEG signal dataset is performed using the partitioned channel, and cross-channel signal authentication is performed to output the EEG signal acquisition results.

[0030] Specifically, the process begins by temporally aligning the EEG signal datasets for each region. The time references of different regions are calibrated using synchronized timestamps to ensure signal temporal consistency. Next, hysteresis noise is configured for the region channels using generated noise artifacts. This involves distance-based noise backtracking identification based on the noise artifacts and localization results to establish source noise artifacts. After removing localization results, signal intensity attenuation and hysteresis fitting are performed based on the positional relationship between the acquisition center of each region's clustering results and the source noise artifacts, generating region noise artifacts with region identifiers. This is then used to configure the hysteresis noise for the corresponding region channels. Then, region suppression is performed on the region datasets using the region channels: if a region channel has hysteresis noise configured, a recursive attenuation factor is configured based on similarity matching results, and a residual dynamic monitoring threshold is activated. After each round of recursive noise removal, the residual noise intensity is calculated. If the threshold is met, a correction factor is configured based on the residual intensity to update the recursive attenuation factor, and noise removal continues. Then, a region filtering layer constructed using supervised learning from historical filtered data is activated for region noise filtering. If no hysteresis noise is configured, the region filtering layer is directly activated to complete the filtering. Finally, the target points for signal acquisition are obtained, and the target point sensitivity analysis is performed based on the target point clustering results. The basic weights of the partitions are established, and cross-channel signal authentication is performed on the inhibition results of each partition. The final EEG signal acquisition results are then output.

[0031] In one possible implementation, step S500 further includes:

[0032] Step S510: After establishing the noise partition, configure the noise extraction network according to the localization results, and set the noise extraction weights of the bandpass filter layer, wavelet decomposition layer, and peak truncation layer in the noise extraction network.

[0033] Step S520: Utilize the noise extraction network to receive the partitioned EEG signal dataset corresponding to the localization result, perform noise extraction through a bandpass filter layer, a wavelet decomposition layer, and a peak truncation layer, respectively, and reconstruct the noise based on the noise extraction weights to establish the noise extraction result.

[0034] Step S530: After performing time-series fitting on the noise extraction results, noise artifacts are generated.

[0035] Specifically, after spatially locating the noise source and establishing partitions, a noise extraction network is configured based on the noise type (e.g., power frequency noise, electromyographic noise) and distribution characteristics obtained from the location. This network includes a bandpass filter layer, a wavelet decomposition layer, and a peak truncation layer. Corresponding noise extraction weights are assigned to each layer for different noise characteristics. The bandpass filter layer weights are used to accurately match the noise frequency band (e.g., the 50Hz power frequency noise band), the wavelet decomposition layer weights are used to extract the energy characteristics of noise in different frequency bands at multiple scales, and the peak truncation layer weights are used to locate the peak amplitude position of the noise signal, thereby constructing a targeted noise feature extraction model.

[0036] The corresponding partitioned EEG signal dataset is input into a pre-configured noise extraction network, which sequentially performs noise extraction through a bandpass filter layer, a wavelet decomposition layer, and a peak truncation layer. The bandpass filter layer filters the input signal based on preset noise extraction weights, accurately separating the noise-dominant frequency band (e.g., the 50Hz power frequency noise band). The wavelet decomposition layer performs multi-resolution decomposition of the signal based on weights, extracting noise features at different scales (e.g., the time-frequency features of high-frequency electromyography noise). The peak truncation layer locates the amplitude peak of the noise signal based on weights, capturing the transient features of sudden noise. Subsequently, the extracted noise features are weighted and reconstructed based on the noise extraction weights of each layer, integrating the scattered noise features into a complete noise signal model, thereby establishing the noise extraction result and achieving accurate isolation and modeling of noise components in the partitioned EEG signal dataset.

[0037] The noise extraction results output by the noise extraction network are subjected to time-series fitting processing. By analyzing the variation patterns of the noise signal in the time dimension (such as periodicity and transient fluctuation characteristics), algorithms such as polynomial fitting and spline interpolation are used to model the time-domain waveform of the noise, transforming discrete noise feature points into continuous time-series signals. Through time-series fitting, the noise extraction results can accurately characterize the amplitude variation characteristics of the noise signal at different time points, ultimately generating a noise artifact that is time-aligned with the original EEG signal. This artifact can serve as a reference model for subsequent partitioned channel hysteresis noise configuration and noise suppression, achieving accurate simulation of real noise signals.

[0038] In one possible implementation, step S600 further includes:

[0039] Step S610: Based on the noise artifacts and the localization results, perform distance-based noise backtracking identification to establish source noise artifacts.

[0040] Step S620: After removing the location result from the partition clustering results, perform signal intensity attenuation and signal hysteresis fitting of the source noise artifacts based on the acquisition center of each partition clustering result and the location of the source noise artifact, and establish partition noise artifacts with partition identifiers.

[0041] Step S630: Configure the hysteresis noise of the corresponding partition channel using the partition noise artifact.

[0042] Specifically, based on the established noise artifacts and their localization results, a distance matrix is ​​constructed by calculating the spatial distance between the noise artifacts and the localization points to perform distance-based noise backtracking identification. Euclidean distance is used to quantify the correlation between the noise artifacts and the localization points, and noise artifacts with distances less than a preset threshold are selected as candidates. Then, Kalman filtering is used to track the spatiotemporal trajectory of the candidate set to determine the source location of noise energy propagation. This establishes source noise artifacts that reflect the true source of the noise, providing accurate source localization for subsequent analysis of noise propagation characteristics.

[0043] After removing the partitions corresponding to the located noise sources from the partition clustering results, the geometric center of each partition clustering result is extracted as the acquisition center. Based on the three-dimensional spatial coordinates of the source noise artifacts, a free space attenuation model (e.g., signal strength is inversely proportional to the square of the distance) is used to calculate the mapping relationship between the distance from the acquisition center to the noise source and the signal strength attenuation. Simultaneously, based on the propagation speed of electromagnetic waves in free space (approximately 3 × 10⁻⁶), the data is further analyzed. 8The actual propagation speed of noise in the EEG signal acquisition scenario (m / s) is used to calculate the time delay from the source noise artifact to each acquisition center. The signal hysteresis is fitted using linear interpolation. The attenuated signal strength parameter and the hysteresis time parameter are mapped to the corresponding partition according to the spatial location of the partition clustering results. A unique partition identifier (such as UUID encoding) is generated for each partition using a hash function, and this identifier is embedded in the data structure of the noise artifact to form a partition noise artifact containing the partition spatial location, signal attenuation coefficient, and time delay parameter. This accurately characterizes the differences in the propagation characteristics of noise in different EEG acquisition partitions, providing a quantitative spatiotemporal characteristic basis for the subsequent hysteresis noise configuration of partition channels.

[0044] Using partition noise artifacts with partition identifiers as input, a transfer function model incorporating time delay and amplitude attenuation factors is constructed for each partition channel based on its hardware characteristics (such as sampling rate and impedance) and noise propagation parameters. By calculating the time delay parameters obtained from the signal hysteresis fitting within the partition noise artifacts, corresponding delay buffers are inserted into the signal processing link. Simultaneously, a gain regulator is configured based on the signal strength attenuation parameters to ensure that the output hysteresis noise accurately simulates the propagation effect of real noise in that channel. For example, in the EEG acquisition channel, based on the hysteresis time and attenuation coefficient of the partition noise artifacts, FIR or IIR filters are used to adjust the delay and amplitude of the noise signal, ultimately configuring hysteresis noise that matches the transmission characteristics of each partition channel.

[0045] In one possible implementation, step S600 further includes:

[0046] Step S640: Determine whether the partition channel is configured with hysteresis noise.

[0047] Step S650: If the partition channel is configured with hysteresis noise, then perform similarity matching of the partition EEG signal dataset according to the hysteresis noise, and configure a recursive attenuation factor according to the similarity matching result.

[0048] Step S660: Perform recursive noise removal on the partitioned EEG signal dataset using the recursive decay factor and the hysteresis noise to establish the recursive noise suppression result.

[0049] Step S670: Complete the partitioned suppression based on the recursive noise suppression results.

[0050] Specifically, for each partition channel, the system determines whether hysteresis noise has been configured for that partition channel by reading the configuration flag bits (such as Boolean parameters) in the partition noise artifact data structure. In practice, it iterates through the partition noise artifacts corresponding to all partition channels, parses the configuration status flags, and if the flag is marked "configured", it confirms that the partition channel has a hysteresis noise model; if the flag is marked "not configured", it determines that the channel has not been configured with hysteresis noise, thus providing a basis for the execution of noise suppression strategies for subsequent partition EEG signal datasets.

[0051] Once it is determined that hysteresis noise has been configured in the partitioned channels, the partitioned EEG signal dataset is matched with the hysteresis noise of the corresponding channels in the time domain. The waveform similarity is measured by calculating the cross-correlation coefficient or using the Dynamic Time Warping (DTW) algorithm. If the cross-correlation coefficient exceeds a preset threshold (e.g., 0.7) or the DTW distance is less than a set value, a strong correlation between the noise and the EEG signal is considered. Based on the noise characteristic parameters (such as noise amplitude, frequency components, and phase shift) in the matching results, a recursive attenuation factor is configured through an adaptive adjustment mechanism: when the similarity is high, the initial value of the attenuation factor is set to 0.8 to enhance noise removal; if the similarity is low, it is set to 0.5 to avoid excessive attenuation of the real EEG signal. This attenuation factor will subsequently be used for weighted attenuation of hysteresis noise during the recursive noise removal process, ensuring that signal integrity is preserved while effectively suppressing noise.

[0052] The configured recursive attenuation factor and hysteresis noise model are enabled to perform recursive noise removal on the partitioned EEG signal dataset. Through iterative calculations, in each round of processing, hysteresis noise is first weighted by the current recursive attenuation factor and then subtracted from the original EEG signal to obtain a preliminary denoised signal. Next, the residual noise intensity is calculated, which is the difference between the original signal and the denoised signal. If the residual noise intensity exceeds the dynamic monitoring threshold (e.g., root mean square value greater than 5 μV), the recursive attenuation factor is updated based on the residual noise intensity (e.g., increasing by 0.1 increments), and the next round of removal begins. This process is repeated until the residual noise intensity meets the threshold requirement or reaches the preset maximum number of recursions (e.g., 10 times), ultimately forming the recursive noise suppression result. This result represents the EEG signal dataset with significantly reduced noise interference after multiple rounds of optimization, laying the foundation for subsequent partitioned suppression.

[0053] Based on the recursive noise suppression results, a convolutional neural network (CNN) regional filtering layer, constructed through supervised learning using historical filtered data, is activated within each partition channel to perform partitioned suppression of the suppressed EEG signal. The recursive noise suppression results are converted into a two-dimensional time-frequency map and input into the CNN. This network contains three convolutional layers (each with 16 3×3 convolutional kernels) and two pooling layers, with weight parameters obtained through training on historically labeled EEG-noise datasets. The convolutional layers utilize local receptive fields to extract noise features (such as periodic patterns of power line interference and high-frequency abrupt changes in EMG artifacts), the pooling layers reduce feature dimensionality and enhance translation invariance, and the fully connected layers output filtering weights based on the spatial location information of the partitioned clustering results. In specific implementation, forward propagation is performed on the time-frequency map of each partition channel, and the probability distribution of noise and EEG signal is calculated using the Softmax function. Time-frequency regions with noise probabilities higher than 0.3 are weighted and removed using a recursive attenuation factor, ultimately generating partitioned EEG signals that have been deeply filtered by the CNN, completing partitioned suppression and providing feature data that meets the signal-to-noise ratio requirements for cross-channel signal authentication.

[0054] In one possible implementation, step S670 further includes:

[0055] Step S671: Activate the regional filtering layer within the partitioned channel, which is constructed through supervised learning of historical filtering data.

[0056] Step S672: Use the region filtering layer to perform region noise filtering on the recursive noise suppression result, and complete the partitioned suppression based on the first region noise filtering result.

[0057] Specifically, a regional filtering layer based on a convolutional neural network (CNN) is activated within the partitioned channel. This filtering layer is trained using supervised learning with historically labeled EEG signals and noise artifact datasets. A CNN architecture consisting of three convolutional layers (each with 16 3×3 convolutional kernels) and two max-pooling layers is employed. Labeled EEG data (including normal EEG components such as alpha and beta waves) and noise data (such as 50Hz power line interference and EMG artifacts) are used as inputs, and the network parameters are optimized through backpropagation. During training, the mean squared error (MSE) loss function is used to measure the difference between the filtered output and the clean EEG signal, enabling the CNN to automatically extract spatial features of noise (such as the synchronous distribution pattern of noise in the electrode array) and frequency domain features (such as the energy concentration area of ​​high-frequency EMG noise). Ultimately, a regional filtering layer with noise recognition capabilities is constructed, providing an adaptive processing module for subsequent regional noise filtering of EEG signals.

[0058] The recursive noise suppression result is input into the activated convolutional neural network (CNN) region filtering layer to perform spatiotemporal feature-based region noise filtering. First, the EEG signal is converted into a time-frequency matrix using a short-time Fourier transform (STFT), which serves as the input data for the CNN. The network's three convolutional layers utilize 16 3×3 convolutional kernels to extract noise features (such as frequency spikes from 50Hz power line interference and high-frequency abrupt changes in EMG artifacts). Two max-pooling layers reduce feature dimensionality and enhance anti-interference capabilities. The fully connected layer generates differentiated filtering weights based on the spatial location information of the partition clustering results (such as the coordinates of prefrontal and parietal lobes). In the specific computation, the time-frequency matrix of each partition channel is forward-propagated, and the probability of each time-frequency point belonging to noise is calculated using the Softmax function. Regions with a probability higher than 0.3 are weighted and removed using a recursive attenuation factor, ultimately yielding the first region noise filtering result. After the result is restored to the time domain signal by inverse transformation, the signal-to-noise ratio of the regional EEG signal can be improved by 15-20 dB, thereby completing the regional suppression and providing low-noise, clean EEG data for subsequent cross-channel signal authentication.

[0059] In one possible implementation, step S640 further includes:

[0060] Step S641: If the partition channel is not configured with hysteresis noise, the regional filtering layer in the partition channel is directly activated to perform filtering processing of the corresponding partition EEG signal dataset and establish the second regional noise filtering result.

[0061] Step S642: Once all partition channels have completed filtering, partition suppression is complete.

[0062] Specifically, if a partition channel is determined to lack hysteresis noise, the regional filtering layer (e.g., using a convolutional neural network (CNN) architecture) built through supervised learning of historical filtering data within that channel is directly activated to perform filtering on the partitioned EEG signal dataset. First, the EEG signal is converted into a time-frequency matrix using a short-time Fourier transform (STFT) and input into a pre-trained CNN. This network contains three convolutional layers (each with 16 3×3 convolutional kernels) to extract the frequency domain features of noise (such as spikes in 50Hz power frequency interference and high-frequency energy accumulation areas of EMG artifacts) and spatial distribution features. Two max-pooling layers are used for dimensionality reduction and feature robustness enhancement. Then, a fully connected layer and a Softmax function are used to calculate the noise probability at each time-frequency point. Regions with probabilities higher than a threshold (e.g., 0.3) are removed. Finally, an inverse transform is used to generate the second regional noise filtering result in the time domain, achieving noise suppression for EEG signals without a hysteresis noise channel.

[0063] After all partition channels (channels corresponding to different EEG acquisition regions based on the sensor distribution scheme, such as channels corresponding to prefrontal and parietal lobes) have completed filtering, the filtering results output from each partition channel (including the first region noise filtering result obtained by recursively removing hysteresis noise and the second region noise filtering result directly generated by the region filtering layer without hysteresis noise) are globally integrated and verified. The signal-to-noise ratio (SNR) of each partition's EEG signal is calculated to ensure it meets a preset standard (e.g., ≥15dB), or the root mean square error (RMSE) of residual noise is below a threshold (e.g., ≤3μV), confirming that noise interference in all partitions' EEG signals has been effectively suppressed. If the verification passes, the partition suppression process is complete, ultimately providing a dataset of EEG signals with consistent noise levels and a satisfactory SNR for subsequent cross-channel signal authentication, ensuring the accuracy and reliability of the acquisition results.

[0064] In one possible implementation, step S660 further includes:

[0065] Step S661: Activate the residual dynamic monitoring threshold.

[0066] Step S662: After each round of recursive noise removal, calculate the residual noise intensity.

[0067] Step S663: If the residual noise intensity meets the residual dynamic monitoring threshold, then a correction factor is configured based on the residual noise intensity, and the recursive attenuation factor is updated using the correction factor before continuing to perform recursive noise removal.

[0068] Specifically, before performing recursive noise removal, a dynamic threshold for monitoring residual noise intensity is activated. This threshold is dynamically set based on the noise source distribution and EEG signal characteristics. For example, initially, the root mean square error (RMSE) threshold can be set to 5 μV, and then adaptively adjusted according to the noise change trend and regional EEG signal characteristics during the historical recursion process to accurately reflect the current noise removal requirements and provide a basis for judgment in each subsequent round of recursive noise removal.

[0069] After each round of recursive noise removal, the residual noise level is quantified by calculating the residual noise intensity. Specifically, the difference between the original EEG signal and the signal after removal in that round is calculated to obtain the residual signal. Then, statistical analysis is performed on the residual signal to calculate its root mean square (RMS) value. First, the difference between the original signal and the removed signal at each sampling point is squared. Then, all squared values ​​are summed and divided by the total number of sampling points. Finally, the square root of the average value is taken. The resulting RMS value accurately reflects the intensity of the residual noise after each round of removal, providing data support for the subsequent adjustment of the recursive attenuation factor.

[0070] If the calculated residual noise intensity meets the residual dynamic monitoring threshold (e.g., root mean square value greater than 3 μV), a correction factor is configured based on the residual noise intensity. For example, when the residual noise intensity is 4 μV, the correction factor is set to 1.2. The recursive attenuation factor is updated using the formula: new recursive attenuation factor = original recursive attenuation factor × correction factor, to enhance the noise removal strength in the next round. After the update, the recursive noise removal operation continues to be performed on the partitioned EEG signal dataset. This process is iterated until the residual noise intensity is below the threshold or the preset maximum number of recursions is reached, ensuring that the noise removal process can be adaptively optimized, thereby effectively improving the quality of EEG signals.

[0071] In one possible implementation, step S600 further includes:

[0072] The task target points for signal acquisition are obtained; target sensitivity analysis of the partition clustering results is performed based on the task target points to establish partition basic weights; cross-channel signal authentication is performed on the inhibition results of partition channels using the partition basic weights, and the EEG signal acquisition results are output.

[0073] Specifically, acquiring the target point for signal acquisition involves identifying the specific neural activity target or region of interest for this EEG signal acquisition. This target point serves as the core guide for the entire signal acquisition and processing process. The setting of the target point must be based on specific clinical needs. For example, when exploring auditory cognition, the target point can be set as the brain region corresponding to the auditory cortex; when performing a motor intention recognition task, the target point can be determined as the relevant region of the primary motor cortex. The target point can be a specific brain region based on neuroanatomical localization (such as the region corresponding to the Fz and Cz electrodes in the international 10-20 electrode system), or it can be a specific EEG signal characteristic (such as the 30-80Hz frequency range of gamma waves or P300 event-related potential components). Once identified, it will be used for subsequent target sensitivity analysis and cross-channel signal authentication.

[0074] Based on the target points, a target sensitivity analysis is performed using the results of the partition clustering. Basic weights for each partition are established; that is, for the identified signal acquisition target points, the sensitivity of each partition's clustering results to the target point is analyzed. The contribution of each partition to the target point is quantified by calculating characteristic parameters (such as signal amplitude, frequency power spectrum, and time-domain waveform correlation) related to the EEG signal of each partition. For example, for visual cortex targets, the change in gamma wave power of the occipital lobe partition's EEG signal under visual stimulation is calculated and compared with other partitions. Partitions closer to the target point or with higher response intensity have higher sensitivity scores. Based on the sensitivity analysis results, corresponding basic weights are assigned to each partition, with weight values ​​ranging from 0 to 1. For example, the weight of a high-sensitivity partition is set to 0.8, and the weight of a low-sensitivity partition is set to 0.2. This establishes basic weights reflecting the differences in importance among partitions, providing a weighted basis for subsequent cross-channel signal authentication.

[0075] The suppressed signal of each partition is multiplied by its corresponding partition base weight, and then cross-channel signal synthesis is performed to form a comprehensive EEG signal. Next, cross-channel signal authentication is completed by verifying the matching degree between this synthesized signal and the task target (e.g., calculating the correlation coefficient between signal features and target features, checking whether the signal meets the preset signal-to-noise ratio threshold, etc.). If authentication is successful, the final output is the noise-suppressed and cross-channel integrated EEG signal acquisition result, ensuring that the result accurately reflects the neural electrical activity related to the task target; if authentication fails, it may be necessary to readjust the partition base weights or return to previous processing steps to optimize the signal suppression effect.

[0076] In one possible implementation, step S300 further includes:

[0077] Step S310: Perform regional signal anomaly identification on the partitioned EEG signal dataset and establish anomaly identification results.

[0078] Step S320: Report regional anomaly signals based on the anomaly identification results.

[0079] Specifically, regional signal anomaly identification is performed on the regional EEG signal dataset to establish anomaly identification results. Using a pre-defined anomalous feature library (such as 50Hz spikes from power frequency interference, high-frequency abrupt changes in electromyography artifacts, and signal baseline drift caused by electrode detachment), the time-domain and frequency-domain features of the EEG signals from each region are compared and analyzed. For example, sudden pulse signals exceeding 3 standard deviations are detected in the time domain, and regions with abnormally elevated power spectral density in specific frequency bands (e.g., 8–13Hz) are identified in the frequency domain, or transient anomalous components in the signal are extracted using wavelet transform. The identified anomalous signal features are matched with a historical anomalous database to determine the anomalous type (e.g., motion artifacts, poor electrode contact) and its location, forming an anomaly identification result that includes the location, type, and intensity of the anomalous signal.

[0080] Based on the regional signal anomaly identification results, detected abnormal signals are reported by region. The anomaly type, location, and intensity information from the anomaly identification results are formatted and output as a visual interface or text report. For example, in the real-time EEG signal monitoring interface, different types of abnormal signals are marked with different colors (e.g., red for electrode detachment, yellow for motion artifacts), and the timestamp and intensity parameters of the anomaly are displayed in the corresponding partition. Simultaneously, an anomaly signal log is generated, including the partition number, anomaly type (e.g., "50Hz power frequency interference," "EMG artifacts"), and suggested handling measures (e.g., "check electrode contact," "remind subject to remain still"). For anomalies that severely affect signal quality (e.g., persistent baseline drift exceeding 50μV), an audible alarm is triggered and the anomaly is highlighted, ensuring that operators can promptly locate and handle the anomaly, guaranteeing the reliability of the acquired EEG signal data.

[0081] Example 2, based on the same inventive concept as the non-invasive EEG signal acquisition method in the foregoing examples, such as... Figure 2 As shown, this application provides a non-invasive electroencephalogram (EEG) signal acquisition device. The device and method embodiments in this application are based on the same inventive concept. The device includes:

[0082] The noise source distribution establishment module 10 is used to read the sensor distribution scheme of the acquisition sensor group, perform noise signal identification, and establish the noise source distribution.

[0083] The partition clustering result establishment module 20 is used to perform partition clustering using the sensor distribution scheme and the noise source distribution, and establish partition clustering results.

[0084] The EEG signal dataset acquisition module 30 is used to acquire a partitioned EEG signal dataset based on the partitioned clustering results.

[0085] The noise intensity discrimination module 40 is used to discriminate the noise intensity of the partitioned EEG signal dataset using the noise source identifier of the partitioned clustering results.

[0086] The noise artifact creation module 50 is used to create noise artifacts for the corresponding partitioned EEG signal dataset if the noise intensity discrimination result is a result that meets the activation condition.

[0087] The EEG signal acquisition result output module 60 is used to perform temporal alignment of the partitioned EEG signal dataset, configure hysteresis noise of the partitioned channel using the noise artifact, perform partitioned suppression of the corresponding partitioned EEG signal dataset using the partitioned channel, perform cross-channel signal authentication, and output the EEG signal acquisition result.

[0088] Furthermore, the device is also used to perform the following functions:

[0089] After establishing noise partitions, a noise extraction network is configured based on the localization results, and noise extraction weights are set for the bandpass filter layer, wavelet decomposition layer, and peak truncation layer in the noise extraction network. The noise extraction network is used to receive the partitioned EEG signal dataset corresponding to the localization results, and noise extraction is performed through the bandpass filter layer, wavelet decomposition layer, and peak truncation layer, respectively. Noise reconstruction is then performed based on the noise extraction weights to establish noise extraction results. After time-series fitting of the noise extraction results, noise artifacts are generated.

[0090] Furthermore, the device is also used to perform the following functions:

[0091] Based on the noise artifacts and the localization results, distance-based noise backtracking identification is performed to establish source noise artifacts; after removing the localization results from the partition clustering results, the signal strength attenuation and signal hysteresis fitting of the source noise artifacts are performed based on the acquisition center of each partition clustering result and the position of the source noise artifacts to establish partition noise artifacts with partition identifiers; the hysteresis noise of the corresponding partition channel is configured using the partition noise artifacts.

[0092] Furthermore, the device is also used to perform the following functions:

[0093] Determine whether the partition channel is configured with hysteresis noise; if the partition channel is configured with hysteresis noise, perform similarity matching of the partition EEG signal dataset based on the hysteresis noise, configure a recursive attenuation factor based on the similarity matching result; use the recursive attenuation factor and the hysteresis noise to perform recursive noise removal of the partition EEG signal dataset, and establish a recursive noise suppression result; complete partition suppression based on the recursive noise suppression result.

[0094] Furthermore, the device is also used to perform the following functions:

[0095] Activate the regional filtering layer within the partitioned channel, which is constructed through supervised learning of historical filtering data; use the regional filtering layer to perform regional noise filtering on the recursive noise suppression result, and complete partitioned suppression based on the first regional noise filtering result.

[0096] Furthermore, the device is also used to perform the following functions:

[0097] If the partition channel is not configured with hysteresis noise, the regional filtering layer in the partition channel is directly activated to perform filtering processing of the corresponding partition EEG signal dataset and establish the second regional noise filtering result; when all partition channels have completed filtering processing, the partition suppression is completed.

[0098] Furthermore, the device is also used to perform the following functions:

[0099] Activate the residual dynamic monitoring threshold; after each round of recursive noise removal, calculate the residual noise intensity; if the residual noise intensity meets the residual dynamic monitoring threshold, configure a correction factor based on the residual noise intensity, update the recursive attenuation factor using the correction factor, and continue to perform recursive noise removal.

[0100] Furthermore, the device is also used to perform the following functions:

[0101] The task target points for signal acquisition are obtained; target sensitivity analysis of the partition clustering results is performed based on the task target points to establish partition basic weights; cross-channel signal authentication is performed on the inhibition results of partition channels using the partition basic weights, and the EEG signal acquisition results are output.

[0102] Furthermore, the device is also used to perform the following functions:

[0103] The region signal anomaly is identified in the partitioned EEG signal dataset, and the anomaly identification results are established; the region abnormal signal is reported based on the anomaly identification results.

[0104] It should be noted that the order of the embodiments described above is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, the above description focuses on specific embodiments of this specification. Additionally, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some implementations, multitasking and parallel processing are possible or may be advantageous.

[0105] The above description is only a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

[0106] This specification and accompanying drawings are merely illustrative examples of this application and are intended to cover any and all modifications, variations, combinations, or equivalents within the scope of this application. Clearly, those skilled in the art can make various alterations and modifications to this application without departing from its scope. Therefore, if such modifications and variations fall within the scope of this application and its equivalents, this application intends to include such modifications and variations.

Claims

1. Non-invasive electroencephalogram signal acquisition method, characterized in that, The method comprises: reading a sensor distribution scheme of a sensor group, and performing noise signal identification to establish a noise source distribution; performing partition clustering using the sensor distribution scheme and the noise source distribution to establish a partition clustering result; obtaining a partitioned electroencephalogram signal data set according to the partition clustering result; performing noise intensity discrimination of the partitioned electroencephalogram signal data set using a noise source identifier of the partition clustering result; if the noise intensity discrimination result is a result satisfying an activation condition, establishing a noise artifact corresponding to the partitioned electroencephalogram signal data set; after time sequence alignment of the partitioned electroencephalogram signal data set, configuring hysteresis noise of a partitioned channel using the noise artifact, performing partitioned suppression of the corresponding partitioned electroencephalogram signal data set using the partitioned channel, and performing cross-channel signal authentication to output an electroencephalogram signal acquisition result; the establishment of the noise artifact corresponding to the partitioned electroencephalogram signal data set comprises: after positioning noise and establishing partitions, configuring a noise extraction network according to the positioning result, and setting noise extraction weights of a band-pass filter layer, a wavelet decomposition layer, and a peak value interception layer in the noise extraction network; receiving the partitioned electroencephalogram signal data set corresponding to the positioning result using the noise extraction network, performing noise extraction through the band-pass filter layer, the wavelet decomposition layer, and the peak value interception layer respectively, and reconstructing noise based on the noise extraction weights to establish a noise extraction result; after time sequence fitting of the noise extraction result, generating a noise artifact; the configuration of hysteresis noise of the partitioned channel using the noise artifact comprises: performing distance-based noise backtracking identification according to the noise artifact and the positioning result to establish a source noise artifact; after excluding the positioning result from the partition clustering result, performing signal strength attenuation and signal hysteresis fitting of the source noise artifact according to the acquisition center of each partition clustering result and the position of the source noise artifact to establish a partitioned noise artifact with a partition identifier; configuring hysteresis noise of the corresponding partitioned channel using the partitioned noise artifact.

2. The non-invasive electroencephal signal acquisition method of claim 1, wherein, the performance of partitioned suppression of the corresponding partitioned electroencephalogram signal data set using the partitioned channel comprises: judging whether the partitioned channel is configured with hysteresis noise; if the partitioned channel is configured with hysteresis noise, performing similar matching of the partitioned electroencephalogram signal data set according to the hysteresis noise, configuring a recursive attenuation factor according to the similar matching result; performing recursive noise elimination of the partitioned electroencephalogram signal data set using the recursive attenuation factor and the hysteresis noise to establish a recursive noise suppression result; completing partitioned suppression according to the recursive noise suppression result.

3. The non-invasive electroencephal signal acquisition method of claim 2, wherein, the completion of partitioned suppression according to the recursive noise suppression result comprises: activating a region filter layer in the partitioned channel, which is constructed through historical filter data supervised learning; performing region noise filtering of the recursive noise suppression result using the region filter layer to obtain a first region noise filtering result, and completing partitioned suppression according to the first region noise filtering result.

4. The non-invasive electroencephal signal acquisition method of claim 3, wherein, the judgment of whether the partitioned channel is configured with hysteresis noise comprises: If the partition channel is not configured with the hysteresis noise, a region filter layer in the partition channel is directly activated to perform filtering processing on the corresponding partitioned electroencephalogram signal dataset, and a second regional noise filtering result is established; When all the partition channels complete the filtering processing, the partition suppression is completed.

5. The non-invasive electroencephal signal acquisition method of claim 2, wherein, The recursive noise removal of the partitioned electroencephalogram signal dataset is performed by using the recursive attenuation factor and the hysteresis noise, including: A residual dynamic monitoring threshold is activated; After each round of recursive noise removal, the residual noise intensity is calculated; If the residual noise intensity meets the residual dynamic monitoring threshold, a correction factor is configured based on the residual noise intensity, and after the recursive attenuation factor is updated by using the correction factor, the recursive noise removal is continued.

6. The non-invasive electroencephal signal acquisition method of claim 1, wherein, The cross-channel signal authentication is performed, including: A task target point of signal acquisition is obtained; A target point sensitivity analysis of the partition clustering result is performed according to the task target point, and a partition basic weight is established; The cross-channel signal authentication of the suppression result of the partition channel is performed by using the partition basic weight, and an electroencephalogram signal acquisition result is output.

7. The non-invasive electroencephal signal acquisition method of claim 1, wherein, The partitioned electroencephalogram signal dataset is obtained according to the partition clustering result, including: A regional signal anomaly recognition is performed on the partitioned electroencephalogram signal dataset, and an anomaly recognition result is established; A regional abnormal signal report is performed according to the anomaly recognition result.

8. Non-invasive electroencephalographic signal acquisition device, characterized in that, The device is used to implement the non-invasive electroencephalogram signal acquisition method of any one of claims 1-7, and the device includes: A noise source distribution establishment module is configured to read a sensor distribution scheme of a sensor group and perform noise signal recognition to establish a noise source distribution; A partition clustering result establishment module is configured to perform partition clustering by using the sensor distribution scheme and the noise source distribution to establish a partition clustering result; An electroencephalogram signal dataset acquisition module is configured to obtain a partitioned electroencephalogram signal dataset according to the partition clustering result; A noise intensity discrimination module is configured to perform noise intensity discrimination of the partitioned electroencephalogram signal dataset by using a noise source identifier of the partition clustering result; A noise artifact establishment module is configured to establish a noise artifact of the corresponding partitioned electroencephalogram signal dataset if the noise intensity discrimination result is a result meeting an activation condition; An electroencephalogram signal acquisition result output module is configured to perform partition suppression of the corresponding partitioned electroencephalogram signal dataset by using the partition channel after the partitioned electroencephalogram signal dataset is time-aligned and the hysteresis noise of the partition channel is configured by using the noise artifact, and perform cross-channel signal authentication to output an electroencephalogram signal acquisition result.

Images