Electroencephalogram signal amplification method, system and terminal based on gaussian mixture distribution model
The EEG signal amplification method using the Gaussian mixture distribution model solves the problems of weak aVEP signals and fatigue caused by long-term acquisition, generates efficient artificial signals, and improves the classification accuracy and stability of brain-computer interfaces.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- TIANJIN UNIV
- Filing Date
- 2025-07-10
- Publication Date
- 2026-07-10
AI Technical Summary
Traditional non-invasive brain-computer interface systems face challenges such as weak signals, large variations, and difficulty in stable extraction when using aVEPs. Furthermore, prolonged data collection can lead to mental fatigue in subjects, affecting the accuracy of classification models. Existing data amplification methods may damage original features or require a large number of training samples.
A Gaussian mixture distribution model is adopted. By collecting EEG signals, noise is removed using a 0.5-20Hz Chebyshev filter, extreme points are segmented, a Gaussian mixture model of phase and amplitude is constructed, phase shift and amplitude scaling are performed, and an amplified signal is generated by combining upsampling/undersampling and local weighted smoothing algorithms.
It improves the classification accuracy of single-trial aVEPs, generates a large number of non-repeating artificial signals, changes single-trial features without changing the overall features, and enhances user interaction comfort and the stability of the classification model.
Smart Images
Figure CN120959760B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of data amplification technology, and in particular to a method, system and terminal for amplifying electroencephalogram (EEG) signals based on a Gaussian mixture distribution model. Background Technology
[0002] Brain-computer interfaces (BCIs) bypass the peripheral nervous system and muscles, directly transmitting brain activity to external devices. Compared to BCI systems with implanted electrodes, non-invasive BCIs based on electroencephalography (EEG) offer advantages such as safety, cost-effectiveness, and ease of operation, making them highly promising for applications such as typing, rehabilitation, and robot control. However, the low signal-to-noise ratio and spatial resolution of EEG severely limit the application of BCIs.
[0003] To overcome the interference of background noise on task-evoked potentials, traditional non-invasive visual brain-computer interfaces (v-BCIs) typically use methods such as increasing the stimulus area or intensity to elicit large neuronal clusters and produce obvious EEG features. However, users are only interested in the task performed by the BCI, not in the stimulation of evoked potentials. Furthermore, prolonged exposure to strong visual stimuli can cause visual fatigue, tension, and even headaches in subjects, and large stimuli consume significant visual resources. Reducing the stimulus area and lowering the intensity, however, makes the EEG signal features less prominent, or even submerged in background noise. This significantly degrades the performance of the BCI system, posing a challenge to recognizing user intentions.
[0004] Based on the principle of retinal mapping, the spatial pattern of visual evoked potentials (VEPs) is related to the position of visual stimuli in the user's field of vision. Specifically, lateral stimuli can induce P1-N1 waveforms that differ significantly from those in the ipsilateral hemisphere, i.e., asymmetric visual evoked potentials (aVEPs). Due to their contralateral dominance, brain-computer interface paradigms based on aVEPs can eliminate the visual obstruction caused by flickering stimuli by preventing users from directly looking at the stimulus blocks, thus improving user interaction comfort. Current research includes paradigms and algorithms related to aVEPs. Xu et al. designed a character speller using aVEPs and achieved decoding of 0.5μV level EEG signals. Xiao et al. designed an algorithm that effectively filters background noise and decodes aVEPs by utilizing the bilateral symmetry of brain activity. Zhou et al. improved DCPM and designed a brain-computer interface paradigm based on aVEPs to track gaze points. However, as a type of event-related potentials (ERPs), aVEPs, despite their visually appealing advantages, also suffer from the typical drawbacks of ERPs, such as weak signals, large variations, and difficulty in stable extraction. Traditional methods for this situation typically involve collecting data from multiple trials and directly classifying the time-domain waveforms. With sufficient training samples, satisfactory classification results can be obtained. However, collecting enough EEG data requires subjects to stare at the screen for extended periods, which is not only time-consuming and laborious, but also often leads to mental fatigue, resulting in weakened or altered signal features and affecting the accuracy of the classification model.
[0005] A promising approach is to augment sample data with artificially generated signals, known as data augmentation. Researchers have already applied data augmentation to the field of electroencephalography (EEG), broadly categorized into deep learning methods and geometric methods. Deep learning methods, such as generative adversarial networks (GANs) and variational autoencoders (VAEs), require a certain number of samples for model training. However, these methods are generally used to learn EEG signals with very fixed features, and may even have a counterproductive effect on classification tasks of aVEPs with poor time-locking and phase-locking properties. Geometric methods typically use random pruning and translation to augment signals, but this method may damage some features of the original data. Summary of the Invention
[0006] The purpose of this invention is to provide a method, system, and terminal for amplifying electroencephalogram (EEG) signals based on a Gaussian mixture distribution model. The method involves acquiring asymmetric evoked potential EEG signals with poor time-locking capability and denoising the EEG data using a Chebyshev filter with a frequency range of 0.5-20 Hz. Next, the location of extreme points in each lead is identified, and the data is segmented into smaller segments based on these extreme points. Then, for each extreme point, the possible left or right phase difference is fitted using a Gaussian distribution, and these two Gaussian distributions are weighted and mixed to obtain a phase shift Gaussian mixture model for that extreme point. The phase difference of each extreme point is then extracted from the phase shift Gaussian mixture model, and upsampling / undersampling and local weighted smoothing algorithms are used to shift and smooth the peaks / troughs of each segment. Finally, the same weighted mixture is used to obtain an amplitude-scaled Gaussian mixture model, and the amplitude of each peak / trough segment is scaled based on this model. This provides a new approach in the field of non-time-locking data amplification.
[0007] To achieve the above objectives, the present invention adopts the following technical solution:
[0008] In a first aspect, the present invention provides a method for amplifying electroencephalogram (EEG) signals based on a Gaussian mixture distribution model, comprising the following steps:
[0009] S10. Collect and obtain raw EEG signals, which include lead signals equal to the number of leads;
[0010] S20. Obtain multiple extreme points for each lead signal and segment the lead signal based on the extreme points;
[0011] S30. For each lead signal, a phase-shift Gaussian mixture model is constructed for each extreme point. The phase difference is randomly extracted from the phase-shift Gaussian mixture model corresponding to each extreme point of each lead signal. After applying the phase difference to the corresponding extreme point, the sampling point is adjusted by upsampling / undersampling to achieve peak shift, and the initial amplified signal corresponding to each lead signal is obtained. The initial amplified signal is smoothed and filtered to obtain the filtered amplified signal.
[0012] S40. Construct an amplitude-scaled Gaussian mixture model for the raw EEG signal and sample amplitude variation parameters from it;
[0013] S50. For each lead signal, apply the corresponding amplitude variation parameter to the filtered amplified signal to generate an amplitude variation, and finally obtain the amplified signal.
[0014] As one possible implementation, a phase-shifted Gaussian mixture model for each extreme point is constructed as follows:
[0015] Calculate the mean and variance of the Gaussian distribution for the left and right segments of the extreme point, respectively;
[0016] A corresponding Gaussian distribution is fitted based on the mean and variance of each Gaussian distribution.
[0017] By weighted summation of the Gaussian distributions, a Gaussian mixture model of the extreme points is obtained;
[0018] The weights in the Gaussian mixture model are rewritten as the ratio of the difference between adjacent extreme points to the difference between phase-separated extreme points, thus obtaining the phase-shifted Gaussian mixture model.
[0019] As one possible implementation, the weights corresponding to the two Gaussian distributions are denoted as follows: , , and Further rewritten as follows:
[0020]
[0021]
[0022] in, For the first The first lead signal There are several extreme points; For the first The first lead signal There are several extreme points; For the first The first lead signal There are several extreme points.
[0023] As one possible implementation, the lead signal to the left of the extreme point is upsampled, and the lead signal to the right of the extreme point is undersampled; or,
[0024] Undersample the lead signal on the left side of the extreme point and upsample the lead signal on the right side of the extreme point.
[0025] As one possible implementation, an amplitude-scaled Gaussian mixture model of the original EEG signal is constructed as follows:
[0026] Set the maximum and minimum zoom levels;
[0027] Calculate the expectation and variance of the Gaussian distributions corresponding to the maximum and minimum scaling values, respectively.
[0028] Based on the above two sets of expectations and variances, an amplitude-scaled Gaussian mixture model is fitted.
[0029] As one possible approach, a locally weighted linear regression algorithm is used to smooth and filter the initial amplified signal to obtain the filtered amplified signal.
[0030] Secondly, the present invention provides an EEG signal amplification system based on a Gaussian mixture distribution model, comprising:
[0031] The signal acquisition unit collects and obtains raw EEG signals, which include lead signals equal to the number of leads.
[0032] The segmentation unit acquires multiple extreme points of each lead signal and segments the lead signal based on the extreme points;
[0033] The phase amplification unit constructs a phase-shift Gaussian mixture model for each extreme point of each lead signal. It randomly extracts the phase difference from the phase-shift Gaussian mixture model corresponding to each extreme point of each lead signal. After applying the phase difference to the corresponding extreme point, it adjusts the sampling point by upsampling / undersampling to achieve peak shift, thereby obtaining the initial amplified signal corresponding to each lead signal. The initial amplified signal is then smoothed and filtered to obtain the filtered amplified signal.
[0034] The amplitude amplification unit constructs an amplitude-scaled Gaussian mixture model for the original EEG signal and samples amplitude variation parameters from it.
[0035] The phase amplitude amplification unit applies a corresponding amplitude change parameter to the filtered amplified signal for each lead signal to generate an amplitude change, ultimately obtaining the amplified signal.
[0036] As one possible implementation, the first The first lead signal The Gaussian mixture model of phase shifts at each extreme point is denoted as . , ;in, For the first The first lead signal There are several extreme points; For the first The first lead signal There are several extreme points; For the first The first lead signal There are several extreme points; The distribution is a Gaussian distribution on the left side of the extreme point; It is a Gaussian distribution for the segment to the right of the extreme point.
[0037] As one possible implementation, the magnitude-scaled Gaussian mixture model is denoted as... , ;in, Maximum scaling range; Minimum scaling factor; The Gaussian distribution corresponding to the minimum scaling factor; This represents the Gaussian distribution corresponding to the maximum scaling magnitude.
[0038] Thirdly, the present invention provides a terminal including a processor and a communication interface coupled to the processor, the processor being used to run computer programs or instructions to implement the EEG signal amplification method based on the Gaussian mixture distribution model provided in the first aspect.
[0039] Compared with the prior art, the beneficial effects of the present invention are as follows:
[0040] 1. The EEG signal amplification method based on Gaussian mixture distribution model proposed in this invention is a novel method for generating artificial signals based on original signals. It uses Gaussian Mixed Sampling (GMS) algorithm to expand EEG data samples and improve the classification accuracy of single-trial aVEPs.
[0041] 2. The EEG signal amplification method based on the Gaussian mixture distribution model proposed in this invention is essentially a geometric method and does not require model training.
[0042] 3. The EEG signal amplification method based on Gaussian mixture distribution model proposed in this invention can generate a large amount of data from a small sample. The GMS algorithm involves multiple sampling processes from the Gaussian mixture model, so theoretically it can generate any number of non-repeating artificial signals; the generated signals only change the characteristics between individual trials without changing the overall characteristics. Attached Figure Description
[0043] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this invention, illustrate exemplary embodiments of the invention and are used to explain the invention, but do not constitute an undue limitation of the invention. In the drawings:
[0044] Figure 1 This is a flowchart of the EEG signal amplification method based on the Gaussian mixture distribution model in an embodiment of the present invention;
[0045] Figure 2 The original signal (red) and the signal generated by the Gaussian mixture model (blue) are shown.
[0046] Figure 3 The flowchart shows the asymmetric evoked potential data augmentation algorithm based on the Gaussian mixture distribution model.
[0047] Figure 4 This represents partial resampling of the Gaussian mixture sampling algorithm and the weight instances corresponding to each sampling point;
[0048] Figure 5 Evaluation metrics for the overall dispersion of data across different categories before and after data augmentation, as well as data visualization;
[0049] Figure 6 This is a comparison chart of the accuracy of the four-class classification before and after data amplification. Detailed Implementation
[0050] To facilitate a clear description of the technical solutions in the embodiments of the present invention, the terms "first" and "second" are used to distinguish identical or similar items with essentially the same function and effect. For example, the first threshold and the second threshold are merely used to distinguish different thresholds and do not limit their order. Those skilled in the art will understand that the terms "first" and "second" do not limit the quantity or execution order, and that the terms "first" and "second" are not necessarily different.
[0051] It should be noted that in this invention, the terms "exemplary" or "for example" are used to indicate examples, illustrations, or descriptions. Any embodiment or design described as "exemplary" or "for example" in this invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of terms such as "exemplary" or "for example" is intended to present the relevant concepts in a concrete manner.
[0052] In this invention, "at least one" refers to one or more, and "more than one" refers to two or more. "And / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, or B alone, where A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one" or similar expressions refer to any combination of these items, including any combination of singular or plural items. For example, "at least one of a, b, or c" can represent: a, b, c, a combination of a and b, a combination of a and c, a combination of b and c, or a, b, and c, where a, b, and c can be single or multiple.
[0053] This invention aims to provide a method, system, and terminal for amplifying electroencephalogram (EEG) signals based on a Gaussian mixture distribution model. It applies the Gaussian Mixed Sampling (GMS) algorithm to expand EEG data samples and improve the classification accuracy of single-trial aVEPs. Specific implementation methods are as follows:
[0054] In a first aspect, the present invention provides a method for amplifying electroencephalogram (EEG) signals based on a Gaussian mixture distribution model, see [link to relevant documentation]. Figure 1 This includes the following steps:
[0055] S10. Collect and obtain raw EEG signals, which include lead signals equal to the number of leads;
[0056] For example, collecting the first The EEG signals from each test were recorded as follows: ,in For the number of leads, For the first One-dimensional discrete data sampling points for each lead, where And the EEG data Denoising was performed using a Chebyshev filter in the 0.5-20Hz range; see [link / reference]. Figure 2 The red waveform in the figure is an example EEG waveform of some leads in this invention.
[0057] S20. Obtain multiple extreme points for each lead signal and segment the lead signal based on the extreme points;
[0058] For example, the first Signal of each lead The segmentation method is as follows:
[0059]
[0060] in Representing the The location of each peak time point, among which , and Representing one-dimensional data The positions of the first and last points. Indicates from arrive All sampling points.
[0061] S30. For each lead signal, a phase-shift Gaussian mixture model is constructed for each extreme point. The phase difference is randomly extracted from the phase-shift Gaussian mixture model corresponding to each extreme point of each lead signal. After applying the phase difference to the corresponding extreme point, the sampling points are adjusted using upsampling / undersampling to achieve peak shift, obtaining the initial amplified signal corresponding to each lead signal. The initial amplified signal is then smoothed and filtered to obtain the filtered amplified signal. See [link to relevant documentation] Figure 3 .
[0062] As one possible implementation, a phase-shifted Gaussian mixture model for each extreme point is constructed as follows:
[0063] Calculate the mean and variance of the Gaussian distribution for the left and right segments of the extreme point, respectively;
[0064] For example, based on extreme points Calculate the mean and variance of the two Gaussian distributions for the left and right segments of the location:
[0065]
[0066] in, and According to and The calculated mean and variance, and According to and The calculated mean and variance.
[0067] A corresponding Gaussian distribution is fitted based on the mean and variance of each Gaussian distribution.
[0068] For example, a corresponding Gaussian distribution can be fitted based on these two sets of means and variances. and .
[0069] By weighted summation of the Gaussian distributions, a Gaussian mixture model of the extreme points is obtained;
[0070] As one possible implementation, the weights corresponding to the two Gaussian distributions are denoted as follows: , , and Further rewritten as follows:
[0071]
[0072]
[0073] in, For the first The first lead signal There are several extreme points; For the first The first lead signal There are several extreme points; For the first The first lead signal There are several extreme points.
[0074] For example, see Figure 3 (b) Weighted combination of the two Gaussian distributions to form the extreme points. Phase-shifted Gaussian mixture model:
[0075]
[0076] in, and These are the weights for the corresponding Gaussian distributions. The range of values for each Gaussian distribution is as follows:
[0077] .
[0078] The weights in the Gaussian mixture model are rewritten as the ratio of the difference between adjacent extreme points to the difference between phase-separated extreme points, thus obtaining the phase-shifted Gaussian mixture model.
[0079] For example, to avoid generating a large number of distorted artificial signals, the algorithm needs to sample the phase difference in the opposite direction of a smaller offset interval. The two weight parameters in the phase-shifted Gaussian mixture model are determined by the number of sampling points in the two data segments. The phase-shifted Gaussian mixture model is obtained by weighting and fusing the two Gaussian distributions as follows. The expression is as follows:
[0080]
[0081] ;
[0082] The setting of this weight is described in the Gaussian mixture sampling algorithm. Figure 3 (b) The area of the probability fusion of the blue Gaussian distribution in the figure is smaller than that of the green Gaussian distribution.
[0083] The phase difference is randomly extracted from the Gaussian mixture model corresponding to each extreme point of each lead signal;
[0084] For example, from No. Phase shift Gaussian mixture model of each peak Randomly select possible phase differences .
[0085] After applying the phase difference to the corresponding extreme point, the sampling point is adjusted by upsampling / undersampling to achieve peak shift, and the initial amplified signal corresponding to each lead signal is obtained. The initial amplified signal is then smoothed and filtered to obtain the filtered amplified signal.
[0086] As one possible implementation, the lead signal to the left of the extreme point is upsampled, and the lead signal to the right of the extreme point is undersampled; or,
[0087] Undersample the lead signal on the left side of the extreme point and upsample the lead signal on the right side of the extreme point.
[0088] For example, for The Each peak is applied Peak shift is achieved by adjusting the sampling points using upsampling / undersampling, thus obtaining the initial amplified signal corresponding to the lead signal. However, oversampling / undersampling can cause the waveform to appear stepped and uneven, such as... Figure 4 (a), and step-type non-smoothness such as Figure 4 (b) The results are then smoothed using the LOESS locally weighted linear regression algorithm. See [link to LOESS algorithm]. Figure 3 As shown in (c), the specific steps are as follows:
[0089]
[0090]
[0091]
[0092] in, For the first Local weights of each sampling point This is the sequential sequence of sampling point locations. The number of sampling points. To be based on local weights The coefficients of the fitted polynomial. The hyperparameter used to adjust the smoothness is such that the smaller the parameter, the more high-frequency components of the signal are retained. Substituting these parameters sequentially... Smoothed results are obtained for each sampling point. See also Figure 4 (c) In the figure, red represents the original sampling points, blue represents the waveform without LOESS smoothing, and green represents the waveform with LOESS smoothing. Figure 4 (d) represents the weight of each sampling point. To make the effect clear, a set of weights is drawn every 10 sampling points. 4(e) shows the result after multiplying the original waveform with the weights of each sampling point.
[0093] S40. Construct an amplitude-scaled Gaussian mixture model for the raw EEG signal and sample amplitude variation parameters from it;
[0094] As one possible implementation, an amplitude-scaled Gaussian mixture model of the original EEG signal is constructed as follows:
[0095] Set the maximum and minimum zoom levels;
[0096] For example, the minimum and maximum zoom levels are set as follows: and ;
[0097] Furthermore, when setting and It can cover the changes in amplitude more comprehensively.
[0098] Calculate the expectation and variance of the Gaussian distributions corresponding to the maximum and minimum scaling values, respectively.
[0099] For example, the expression for the expectation and variance of the Gaussian distribution corresponding to the maximum and minimum scaling magnitudes is as follows:
[0100]
[0101] Based on the above two sets of expectations and variances, an amplitude-scaled Gaussian mixture model is fitted.
[0102] For example,
[0103]
[0104] S50. For each lead signal, apply the corresponding amplitude change parameter to the filtered amplified signal to generate an amplitude change, and finally obtain the amplified signal;
[0105] As one possible approach, a locally weighted linear regression algorithm is used to smooth and filter the initial amplified signal to obtain the filtered amplified signal.
[0106] For example, scaling the Gaussian mixture model from magnitude The first sample can be obtained from the middle Parameters of peak amplitude variation ;
[0107] Will Apply amplitude variation parameters This generates amplitude variations, ultimately resulting in an artificial signal with differences in both amplitude and phase. The expression is:
[0108] .
[0109] See the final output results. Figure 2 The blue waveform, and Figure 2 Compared to the original red waveform, each blue generated signal differs from the red input waveform in both peak position and amplitude.
[0110] Next, simulation experiments will be conducted to evaluate the feasibility and effectiveness of the amplification method proposed in this embodiment.
[0111] The evaluation metrics (Angle Squared and Euclidean distances and Variances, ASEV) were designed. The t-SNE dimensionality reduction method was used to reduce the EEG data of all trials to 3D space. Then, the coordinates of the center points of each type of sample in this 3D space were calculated. Finally, the Euclidean distance between these points, the variance in each dimension, and the sum of the squares of the angles between these points and their mean vectors were calculated. These parameters were summed to obtain the dispersion of each category. Figure 5 (a) shows the unamplified data for four-class aVEPs, with ASEV=0.656. After amplification, the dispersion increased significantly, as shown in Figure 1. Figure 5 (b) ASEV = 4.236. Meanwhile... Figure 5(c), (d), (e), and (f) represent the results of calculating the ASEV after randomly dividing the original data into two groups for each of the four categories, as well as the ASEV between the original data and the artificially generated data. Figure 5 The data shows that the ASEV values of the artificial signal and the original signal are not significantly different from those of the original signal and the original signal, indicating that the algorithm proposed in this patent can minimize the changes in the overall characteristics of the signal even if there are differences in a single trial. Figure 6 The four subplots show the classification accuracy of each trial for the four subjects before and after data augmentation. The accuracy increases with the increase in the number of augmented data.
[0112] The EEG signal amplification method based on Gaussian mixture distribution model provided by this invention is essentially a geometric method and does not require model training. Applying this method, a large amount of data can be generated from a small sample. The GMS algorithm involves multiple sampling processes from the Gaussian mixture model, so theoretically, it can generate any number of non-repeating artificial signals. The generated signals only change the features between individual trials without changing the overall features.
[0113] Secondly, the present invention provides an EEG signal amplification system based on a Gaussian mixture distribution model, comprising:
[0114] The signal acquisition unit collects and obtains raw EEG signals, which include lead signals equal to the number of leads.
[0115] The segmentation unit acquires multiple extreme points of each lead signal and segments the lead signal based on the extreme points;
[0116] The phase amplification unit constructs a phase-shift Gaussian mixture model for each extreme point of each lead signal. It randomly extracts the phase difference from the phase-shift Gaussian mixture model corresponding to each extreme point of each lead signal. After applying the phase difference to the corresponding extreme point, it adjusts the sampling point by upsampling / undersampling to achieve peak shift, thereby obtaining the initial amplified signal corresponding to each lead signal. The initial amplified signal is then smoothed and filtered to obtain the filtered amplified signal.
[0117] The amplitude amplification unit constructs an amplitude-scaled Gaussian mixture model for the original EEG signal and samples amplitude variation parameters from it.
[0118] The phase amplitude amplification unit applies a corresponding amplitude change parameter to the filtered amplified signal for each lead signal to generate an amplitude change, ultimately obtaining the amplified signal.
[0119] As one possible way to achieve this, the first The first lead signal The Gaussian mixture model of phase shifts at each extreme point is denoted as . , ;in, For the first The first lead signal There are several extreme points; For the first The first lead signal There are several extreme points; For the first The first lead signal There are several extreme points; The distribution is a Gaussian distribution on the left side of the extreme point; It is a Gaussian distribution for the segment to the right of the extreme point.
[0120] As one possible implementation, the magnitude-scaled Gaussian mixture model is denoted as... , ;in, Maximum scaling range; Minimum scaling factor; The Gaussian distribution corresponding to the minimum scaling factor; This represents the Gaussian distribution corresponding to the maximum scaling magnitude.
[0121] Thirdly, the present invention provides a terminal, including a processor and a communication interface coupled to the processor, wherein the processor is used to run computer programs or instructions to implement the electroencephalogram signal amplification method based on the Gaussian mixture distribution model proposed in this invention.
[0122] Although the invention has been described herein in conjunction with various embodiments, those skilled in the art will understand and implement other variations of the disclosed embodiments by reviewing the accompanying drawings, the disclosure, and the description of the drawings, in carrying out the claimed invention. In this specification, the word "comprising" does not exclude other components or steps, and "a" or "an" does not exclude multiple components. A single processor or other unit can implement several of the functions listed in the specification. While certain measures are described in different embodiments, this does not mean that these measures cannot be combined to produce good results.
[0123] Although the invention has been described in conjunction with specific features and embodiments, it is obvious that various modifications and combinations can be made therein without departing from the spirit and scope of the invention. Accordingly, this specification and drawings are merely illustrative of the invention and are considered to cover any and all modifications, variations, combinations, or equivalents within the scope of the invention. Clearly, those skilled in the art can make various alterations and modifications to the invention without departing from its spirit and scope. Thus, if such modifications and modifications fall within the scope of the invention and its equivalents, the invention is also intended to include such modifications and modifications.
Claims
1. A method for amplifying electroencephalogram (EEG) signals based on a Gaussian mixture distribution model, characterized in that, Includes the following steps: S10. Collect and obtain raw EEG signals, which include lead signals equal to the number of leads; S20. Obtain multiple extreme points for each lead signal and segment the lead signal based on the extreme points; S30. For each lead signal, a phase-shift Gaussian mixture model is constructed for each extreme point. The phase difference is randomly extracted from the phase-shift Gaussian mixture model corresponding to each extreme point of each lead signal. After applying the phase difference to the corresponding extreme point, the sampling point is adjusted by upsampling / undersampling to achieve peak shift, and the initial amplified signal corresponding to each lead signal is obtained. The initial amplified signal is smoothed and filtered to obtain the filtered amplified signal. S40. Construct an amplitude-scaled Gaussian mixture model for the raw EEG signal and sample amplitude variation parameters from it; S50. For each lead signal, apply the corresponding amplitude variation parameter to the filtered amplified signal to generate an amplitude variation, and finally obtain the amplified signal.
2. The EEG signal amplification method based on a Gaussian mixture distribution model according to claim 1, characterized in that, The phase-shifted Gaussian mixture model for each extreme point is constructed as follows: Calculate the mean and variance of the Gaussian distribution for the left and right segments of the extreme point, respectively; A corresponding Gaussian distribution is fitted based on the mean and variance of each Gaussian distribution. By weighted summation of the Gaussian distributions, a Gaussian mixture model of the extreme points is obtained; The weights in the Gaussian mixture model are rewritten as the ratio of the difference between adjacent extreme points to the difference between phase-separated extreme points, thus obtaining the phase-shifted Gaussian mixture model.
3. The EEG signal amplification method based on a Gaussian mixture distribution model according to claim 2, characterized in that, Let the weights corresponding to the two Gaussian distributions be denoted as follows: , , and Further rewritten as follows: in, For the first The first lead signal There are several extreme points; For the first The first lead signal There are several extreme points; For the first The first lead signal There are several extreme points.
4. The EEG signal amplification method based on a Gaussian mixture distribution model according to claim 1, characterized in that, Upsample the lead signal to the left of the extreme point and undersample the lead signal to the right of the extreme point; or, Undersample the lead signal on the left side of the extreme point and upsample the lead signal on the right side of the extreme point.
5. The EEG signal amplification method based on a Gaussian mixture distribution model according to claim 1, characterized in that, An amplitude-scaled Gaussian mixture model of the original EEG signal was constructed as follows: Set the maximum and minimum zoom levels; Calculate the expectation and variance of the Gaussian distributions corresponding to the maximum and minimum scaling values, respectively. Based on the above two sets of expectations and variances, an amplitude-scaled Gaussian mixture model is fitted.
6. The EEG signal amplification method based on a Gaussian mixture distribution model according to claim 1, characterized in that, The initial amplified signal is smoothed and filtered using a locally weighted linear regression algorithm to obtain the filtered amplified signal.
7. A brainwave signal amplification system based on a Gaussian mixture distribution model, characterized in that, include: The signal acquisition unit collects and obtains raw EEG signals, which include lead signals equal to the number of leads. The segmentation unit acquires multiple extreme points of each lead signal and segments the lead signal based on the extreme points; The phase amplification unit constructs a phase-shift Gaussian mixture model for each extreme point of each lead signal. It randomly extracts the phase difference from the phase-shift Gaussian mixture model corresponding to each extreme point of each lead signal. After applying the phase difference to the corresponding extreme point, it adjusts the sampling point by upsampling / undersampling to achieve peak shift, thereby obtaining the initial amplified signal corresponding to each lead signal. The initial amplified signal is then smoothed and filtered to obtain the filtered amplified signal. The amplitude amplification unit constructs an amplitude-scaled Gaussian mixture model for the original EEG signal and samples amplitude variation parameters from it. The phase amplitude amplification unit applies a corresponding amplitude change parameter to the filtered amplified signal for each lead signal to generate an amplitude change, ultimately obtaining the amplified signal.
8. The EEG signal amplification system based on a Gaussian mixture distribution model according to claim 7, characterized in that, No. The first lead signal The Gaussian mixture model of phase shifts at each extreme point is denoted as . , ;in, For the first The first lead signal There are several extreme points; For the first The first lead signal There are several extreme points; For the first The first lead signal There are several extreme points; The distribution is a Gaussian distribution on the left side of the extreme point; It is a Gaussian distribution for the segment to the right of the extreme point.
9. The EEG signal amplification system based on a Gaussian mixture distribution model according to claim 7, characterized in that, Amplitude scaling Gaussian mixture model is denoted as , ;in, This represents the maximum scaling range; Minimum scaling factor; The Gaussian distribution corresponding to the minimum scaling factor; This represents the Gaussian distribution corresponding to the maximum scaling magnitude.
10. A terminal, comprising a processor and a communication interface coupled to the processor, the processor being configured to run a computer program or instructions to implement the electroencephalogram signal amplification method based on a Gaussian mixture distribution model as described in any one of claims 1 to 6.