Method and system for extracting time domain features of electroencephalogram signals

Abstract

The invention relates to the technical field of electroencephalogram signal processing, and discloses an electroencephalogram signal time domain feature extraction method and system, and the method comprises the steps: receiving an electroencephalogram signal, and carrying out the dynamic segmentation of the electroencephalogram signal through an adaptive window technology; using a deep learning model to extract segmented signal segments; and extracting a final signal feature by using an improved time domain feature extraction algorithm, and outputting a feature extraction result. According to the method, a deep learning model iRNN can be used for evaluating the importance of the features, the features most contributing to classification are recognized and selected from complex EEG signals, the stability of the EEG signal processing flow is improved through the steps of feature selection and normalization, and fluctuation of the signals in different environments is reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of electroencephalogram (EEG) signal processing, and in particular to a method, system, computer equipment and medium for extracting time-domain features of EEG signals. Background Art

[0002] Characteristics and classification of EEG signals Since the EEG signals generated by the cerebral cortex are produced by the joint action of multiple nerve cells, and the brain is a very complex organization, the collected EEG signals have the following characteristics: 1) The signal strength is extremely low, so the anti-interference ability is very weak, and the signal itself is prone to certain fluctuations in different environments. In this case, the first task is to denoise the signal and use some methods to minimize the interference of external factors. 2) The nonlinearity and non-stationarity of EEG signals. Since the external stimuli to the brain are diverse, the types of EEG signals are diverse and extremely complex. 3) EEG signals are not a simple signal wave, but a more complex synthetic wave. Therefore, there are individual differences, that is, the waveform of the brain wave will vary from person to person. Not only that, the frequency components will also be different. At different times, the brain waves will have different forms. Summary of the invention

[0003] In view of the above existing problems, the present invention is proposed.

[0004] Therefore, the present invention provides a time domain feature extraction method for EEG signals, which can use the deep learning model iRNN to evaluate the importance of features, identify and select the features that contribute most to classification from complex EEG signals, and improve the stability of the EEG signal processing process through feature selection and normalization steps, reducing the fluctuations of signals in different environments.

[0005] To solve the above technical problems, the present invention provides the following technical solutions: a method for extracting time domain features of EEG signals, comprising: receiving EEG signals, and dynamically segmenting the EEG signals using adaptive window technology; extracting segmented signal segments using a deep learning model; and extracting final signal features using an improved time domain feature extraction algorithm, and outputting feature extraction results.

[0006] As a preferred solution of the method for extracting time domain features of an EEG signal described in the present invention, wherein: the receiving of the EEG signal includes capturing the electrical activity of the brain through an EEG electrode cap, receiving the EEG activity recorded by the electrode, the electrode position information, the signal quality index, the timestamp and the sampling rate;

[0007] After receiving the signal, preprocessing is performed to improve the signal quality.

[0008] As a preferred solution of the method for extracting time domain features of an EEG signal described in the present invention, the dynamic segmentation includes dynamically adjusting the window size according to the characteristics of the EEG signal, namely, the peak interval of the EEG signal and the coefficient of variation of the signal waveform;

[0009] Set the initial window size W 0 , and a dynamic adjustment factor δ based on signal stability, the following features are extracted from the EEG signal in the current window: the peak interval IPI is calculated, and the coefficient of variation CoV is calculated;

[0010] According to the feature score, dynamically adjust the window size W:

[0011] W new =W old ·(1+δ·S(x)) n

[0012] Among them, W old is the current window size, W new is the new window size, n is the exponential adjustment coefficient, and S(x) represents the feature score.

[0013] As a preferred solution of the method for extracting time domain features of an EEG signal described in the present invention, wherein: the extracted segmented signal segments include: the deep learning model is an improved recurrent neural network iRNN, which captures the long-term dependencies in the time series, converts the signal segments into a format suitable for RNN input, and uses a variant of the gated recurrent unit to dynamically adjust the weights of the forget gate and the input gate according to the characteristics of the signal segments;

[0014] Among them, the variant of the gated recurrent unit is called the adaptive gated unit, which introduces a time-varying weight adjustment factor α t , dynamically calculated based on the statistical properties of the signal fragments:

[0015]

[0016] Among them, std(X t ) is the standard deviation of the signal segment, mean(X t ) is the mean, ∈ is a small positive number to avoid division by zero;

[0017] After each signal segment is processed by iRNN, a feature sequence H = [h 1 ,h 2 ,...,h T ], where h t is the eigenvector at time t, t=1,2,…,T.

[0018] As a preferred solution of the time domain feature extraction method of an EEG signal described in the present invention, the improved time domain feature extraction algorithm includes further improving the feature quality according to the features extracted by the deep learning model, initializing the time series weight vector Λ=[λ 1 ,λ 2 ,...,λ T ], dynamically adjust the timing weight according to the features extracted by iRNN; update the timing weight, the features extracted by iRNN f t Used to calculate the timing weight λ t , which is then used to calculate the fusion feature TWFF.

[0019] As a preferred solution of the method for extracting time-domain features of EEG signals described in the present invention, the improved time-domain feature extraction algorithm further includes selecting a feature subset that contributes most to classification based on feature importance scores of a deep learning model;

[0020] The preprocessed data is trained using the improved iRNN model. For each feature, the importance score is calculated using the trained iRNN model. Based on the feature importance score, the adaptive weight W of each feature is calculated. f ;

[0021] According to the calculated adaptive weight W f , select the K features with the highest weights as the feature subset that contributes most to classification.

[0022] As a preferred solution of the method for extracting time domain features of an EEG signal described in the present invention, wherein: the extraction of the final signal features includes normalizing the features using a method based on signal energy distribution, selecting a feature subset, and then calculating the energy of each selected feature;

[0023] For each feature, calculate the energy E f :

[0024]

[0025] Among them, x f,i is the i-th value of feature f, μ f and σ f are the mean and standard deviation of feature f respectively;

[0026] Using the calculated characteristic energy E f , the selected features are normalized to eliminate the dimensional effects and numerical scale differences between the features.

[0027] Another object of the present invention is to provide a time domain feature extraction system for EEG signals, which can realize the extraction of time domain features of EEG signals and provide effective feature input for EEG signal analysis.

[0028] As a preferred solution of the time domain feature extraction system of an EEG signal described in the present invention, it includes: a signal acquisition and preprocessing module, an adaptive signal segmentation module, a deep learning feature extraction module, a feature optimization and weight adjustment module, and a feature selection and normalization module;

[0029] The signal acquisition and preprocessing module is responsible for capturing brain electrical activity through the EEG electrode cap, receiving and recording EEG signals and related information, and performing preliminary signal preprocessing to improve signal quality;

[0030] The adaptive signal segmentation module uses adaptive window technology to dynamically segment the signal according to the characteristics of the EEG signal, ensuring that the signal features in each segmentation window can be effectively extracted;

[0031] The deep learning feature extraction module uses an improved recurrent neural network iRNN to process the segmented signal segments, capture the long-term dependencies in the time series, output feature sequences, and introduces an adaptive gating unit to dynamically adjust network parameters;

[0032] The feature optimization and weight adjustment module optimizes the feature sequence output by the deep learning model, including dynamically adjusting the time series weight, calculating the importance score of the feature, and updating the time series weight according to these scores to further refine the feature;

[0033] The feature selection and normalization module selects the feature subset that contributes most to classification based on the feature importance score, and normalizes the selected features using a method based on signal energy distribution to ensure the consistency and comparability of the features.

[0034] A computer device comprises a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of a method for extracting time domain features of an electroencephalogram signal when executing the computer program.

[0035] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of a method for extracting time-domain features of an electroencephalogram signal.

[0036] The beneficial effects of the present invention: through the data preprocessing step, the raw data is cleaned and preprocessed, ensuring the quality of the data in the subsequent analysis, which is crucial for the training of deep learning models. Utilizing the improved iRNN model training, the degree of influence of each feature on the classification result can be accurately identified, thereby calculating the feature importance score, which helps to understand the intrinsic structure of the data and the relationship between the features. By calculating adaptive weights and selecting the feature subset with the highest weight, the present invention can adaptively retain the features that have the greatest influence on model predictions, which can not only reduce the feature dimension, but also improve the performance and prediction efficiency of the model. Through the normalization method based on signal energy distribution, the present invention ensures the consistency and comparability of features, eliminates the dimensional influence and numerical scale differences between features, which helps the deep learning model to be more stable and accurate during the training process. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work. Among them:

[0038] Figure 1 A schematic flow chart of a method for extracting time-domain features of an EEG signal provided by an embodiment of the present invention;

[0039] Figure 2 A schematic diagram of working modules of a system for extracting time-domain features of EEG signals provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0040] In order to make the above-mentioned purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in the art without creative work should fall within the scope of protection of the present invention.

[0041] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0042] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.

[0043] The present invention is described in detail with reference to schematic diagrams. When describing the embodiments of the present invention, for the sake of convenience, the cross-sectional diagrams showing the device structure will not be partially enlarged according to the general scale, and the schematic diagrams are only examples, which should not limit the scope of protection of the present invention. In addition, in actual production, the three-dimensional dimensions of length, width and depth should be included.

[0044] At the same time, in the description of the present invention, it should be noted that the directions or positional relationships indicated by the terms "upper, lower, inner and outer" are based on the directions or positional relationships shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific direction, be constructed and operated in a specific direction, and therefore cannot be understood as limiting the present invention. In addition, the terms "first, second or third" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance.

[0045] In the present invention, unless otherwise clearly specified and limited, the terms "install, connect, connect" should be understood in a broad sense, for example: it can be a fixed connection, a detachable connection or an integral connection; it can also be a mechanical connection, an electrical connection or a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0046] Characteristics and classification of EEG signals Since the EEG signals generated by the cerebral cortex are produced by the joint action of multiple nerve cells, and the brain is a very complex organization, the collected EEG signals have the following characteristics: 1) The signal strength is extremely low, so the anti-interference ability is very weak, and the signal itself is prone to certain fluctuations in different environments. In this case, the first task is to denoise the signal and use some methods to minimize the interference of external factors. 2) The nonlinearity and non-stationarity of EEG signals. Since the external stimuli to the brain are diverse, the types of EEG signals are diverse and extremely complex. 3) EEG signals are not a simple signal wave, but a more complex synthetic wave. Therefore, there are individual differences, that is, the waveform of the brain wave will vary from person to person. Not only that, the frequency components will also be different. At different times, the brain waves will have different forms.

[0047] Example 1, reference Figure 1, which is the first embodiment of the present invention, provides a method for extracting time domain features of an EEG signal, comprising:

[0048] S1: Receive EEG signals and dynamically segment them using adaptive window technology.

[0049] Furthermore, receiving the EEG signal includes capturing the electrical activity of the brain via an EEG electrode cap, receiving the EEG activity recorded via the electrodes, electrode position information, signal quality indicators, a timestamp, and a sampling rate;

[0050] After receiving the signal, preprocessing is performed to improve the signal quality.

[0051] The window size is dynamically adjusted according to the characteristics of the EEG signal, i.e., the peak interval of the EEG signal and the coefficient of variation of the signal waveform;

[0052] Set the initial window size W 0 , and a dynamic adjustment factor δ based on signal stability, the following features are extracted from the EEG signal in the current window: the peak interval IPI is calculated, and the coefficient of variation CoV is calculated;

[0053] According to the feature score, dynamically adjust the window size W:

[0054] W new =W old ·(1+δ·S(x)) n

[0055] Among them, W old is the current window size, W new is the new window size, n is the exponential adjustment coefficient, and S(x) represents the feature score.

[0056] It should also be noted that the feature score S(x) is evaluated using the following formula:

[0057]

[0058] Among them, IPI max is the maximum acceptable peak interval, σ is the adjustment parameter, IPI 0 is the reference value of the peak interval, CoV 0 is the reference value of the coefficient of variation and CoV is the coefficient of variation.

[0059] S2: Extract the segmented signal segments using the deep learning model.

[0060] Furthermore, the deep learning model is an improved recurrent neural network iRNN, which captures long-term dependencies in time series, converts signal segments into a format suitable for RNN input, and uses a variant of the gated recurrent unit to dynamically adjust the weights of the forget gate and input gate according to the characteristics of the signal segment;

[0061] A variation of the Gated Recurrent Unit is called the Adaptive Gated Unit:

[0062] h t =(1-z t )⊙h t-1 +z t ⊙tanh(W h ·[h t-1 ,x t ]+b h )

[0063] Among them, h t is the hidden state output, z t is the update gate of the adaptive gating unit, ⊙ represents element-wise multiplication, W h and b h are weights and biases:

[0064] z t =σ(W z ·[h t-1 ,x t ,α t ]+b z )

[0065] Among them, h t-1 is the previous hidden state, W z 、b z is the weight matrix and bias vector of the update gate;

[0066] Introducing time-varying weight adjustment factor α t , dynamically calculated based on the statistical properties of the signal fragments:

[0067]

[0068] Among them, std(X t ) is the standard deviation of the signal segment, mean(X t ) is the mean, ∈ is a small positive number to avoid division by zero;

[0069] After each signal segment is processed by iRNN, a feature sequence H = [h 1 ,h 2 ,...,h T ], where h t is the eigenvector at time t, t=1,2,…,T.

[0070] S3: Apply the improved time domain feature extraction algorithm to extract the final signal features and output the feature extraction results.

[0071] Furthermore, based on the features extracted by the deep learning model, the feature quality is further improved and the time series weight vector Λ=[λ 1 ,λ 2 ,...,λ T ], dynamically adjust the timing weight according to the features extracted by iRNN; update the timing weight:

[0072]

[0073] Among them, λ t is the time series weight at time t, f t is the feature extracted by iRNN at time t, γ is the temperature parameter, LeakyReLU is the linear rectification function with leakage, and the fusion feature TWFF is calculated:

[0074]

[0075] Among them, TWFF is the fused feature vector.

[0076] It should be noted that the improved time-domain feature extraction algorithm also includes selecting the feature subset that contributes most to classification based on the feature importance score of the deep learning model;

[0077] The preprocessed data is trained using the improved iRNN model. For each feature, the importance score is calculated using the trained iRNN model. Based on the feature importance score, the adaptive weight W of each feature is calculated. f :

[0078]

[0079] Among them, α is the tuning parameter and N is the total number of features.

[0080] According to the calculated adaptive weight W f , select the K features with the highest weights as the feature subset that contributes most to classification.

[0081] Furthermore, extracting the final signal features includes normalizing the features using a method based on signal energy distribution, selecting a feature subset, and then calculating the energy of each selected feature;

[0082] For each feature, calculate the energy E f :

[0083]

[0084] Among them, xf,i is the i-th value of feature f, μ f and σ f are the mean and standard deviation of feature f respectively;

[0085] Using the calculated characteristic energy E f , normalize the selected features to eliminate the dimensional effects and numerical scale differences between the features:

[0086]

[0087] Among them, x′ f,i is the normalized eigenvalue.

[0088] Embodiment 2 is an embodiment of the present invention, which provides a method for extracting time-domain features of EEG signals. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through experiments.

[0089] This study aims to verify the effectiveness of a new method for extracting time-domain features from EEG signals. Twenty healthy subjects, half male and half female, aged between 25 and 35 years old, were selected for the experiment. During the experiment, the subjects wore an EEG electrode cap with 19 electrode points placed according to the international 10-20 system. The EEG signal was recorded at a sampling rate of 1,000 Hz for a duration of 5 minutes. After signal acquisition, preprocessing was first performed, including bandpass filtering from 0.5 to 70 Hz and removal of 50 Hz power frequency interference.

[0090] Next, the adaptive window technique is applied to dynamically segment the preprocessed EEG signals. The initial window size is set to 256 sampling points, and the dynamic adjustment factor δ is set to 0.1. The window size is dynamically adjusted according to the peak interval and coefficient of variation of the EEG signal in each window. Subsequently, the segmented signal segments are processed using the improved recurrent neural network iRNN to capture the long-term dependencies in the time series. The iRNN model contains an adaptive gating unit that can dynamically adjust the weights of the forget gate and the input gate according to the statistical characteristics of the signal segment. The experimental results are shown in Table 1.

[0091] Table 1

[0092]

[0093]

[0094] Through the above experimental data, the method of the present invention shows significant advantages in extracting time-domain features of EEG signals. First, the application of adaptive window technology significantly improves the flexibility of signal segmentation. The average adjusted window size is 283 sampling points, which is more adaptable to the dynamic changes of EEG signals compared to the initial window size of 256 sampling points.

[0095] Secondly, the iRNN model effectively captures the long-term dependencies in the time series through adaptive gating units when processing signal fragments. The average value of the feature importance score is 0.84, indicating that the extracted features have a high contribution to the classification task. In addition, the average value of the signal-to-noise ratio is 21dB, indicating that the preprocessing step effectively improves the signal quality.

[0096] In terms of classification accuracy, the average accuracy of the method is 91.5%, which is 6.5 percentage points higher than the average accuracy of the prior art (about 85%). This result shows that the method of the present invention has higher accuracy and reliability in EEG signal classification tasks.

[0097] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

[0098] Embodiment 3, the third embodiment of the present invention, is different from the first two embodiments in that:

[0099] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc., which can store program codes.

[0100] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in conjunction with such instruction execution systems, devices or apparatuses. For the purposes of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in conjunction with such instruction execution systems, devices or apparatuses.

[0101] More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or more wires (electronic device), a portable computer disk case (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be a paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, deciphering or, if necessary, processing in another suitable manner, and then stored in a computer memory.

[0102] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, a plurality of steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0103] Example 4, reference Figure 2 , which is an embodiment of the present invention, provides a time domain feature extraction system for EEG signals, including a signal acquisition and preprocessing module 1, an adaptive signal segmentation module 2, a deep learning feature extraction module 3, a feature optimization and weight adjustment module 4, and a feature selection and normalization module 5;

[0104] The signal acquisition and preprocessing module is responsible for capturing brain electrical activity through the EEG electrode cap, receiving and recording EEG signals and related information, and performing preliminary signal preprocessing to improve signal quality;

[0105] The adaptive signal segmentation module uses adaptive window technology to dynamically segment the signal according to the characteristics of the EEG signal, ensuring that the signal features in each segmentation window can be effectively extracted;

[0106] The deep learning feature extraction module uses an improved recurrent neural network (iRNN) to process the segmented signal segments, capture the long-term dependencies in the time series, output feature sequences, and introduce an adaptive gating unit to dynamically adjust network parameters;

[0107] The feature optimization and weight adjustment module optimizes the feature sequence output by the deep learning model, including dynamically adjusting the time series weights, calculating the importance scores of the features, and updating the time series weights based on these scores to further refine the features;

[0108] The feature selection and normalization module selects the feature subset that contributes most to classification based on the feature importance score, and normalizes the selected features using a method based on signal energy distribution to ensure the consistency and comparability of the features.

[0109] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A method for extracting time domain features of EEG signals, characterized in that: include, Receive EEG signals and dynamically segment them using adaptive window technology; Extract the segmented signal fragments using deep learning models; The improved time domain feature extraction algorithm is applied to extract the final signal features and output the feature extraction results.

2. The method for extracting time domain features of an EEG signal according to claim 1, characterized in that: The receiving of the EEG signal includes capturing the electrical activity of the brain through an EEG electrode cap, receiving the EEG activity recorded through the electrodes, electrode position information, signal quality indicators, timestamps, and sampling rates; After receiving the signal, preprocessing is performed to improve the signal quality.

3. The method for extracting time domain features of an EEG signal as claimed in claim 2, characterized in that: The dynamic segmentation includes dynamically adjusting the window size according to the characteristics of the EEG signal, namely the peak interval of the EEG signal and the coefficient of variation of the signal waveform; Set the initial window size W0 and a dynamic adjustment factor δ based on signal stability, and perform the following feature extraction on the EEG signal in the current window: calculate the peak interval IPI and calculate the coefficient of variation CoV; According to the feature score, dynamically adjust the window size W: W new =W old ·(1+δ·S(x)) n Among them, W old is the current window size, W new is the new window size, n is the exponential adjustment coefficient, and S(x) represents the feature score.

4. The method for extracting time domain features of an EEG signal as claimed in claim 3, characterized in that: The extracting of the segmented signal segments includes: the deep learning model is an improved recurrent neural network iRNN, capturing the long-term dependency in the time series, converting the signal segments into a format suitable for RNN input, using a variant of a gated recurrent unit, and dynamically adjusting the weights of a forget gate and an input gate according to the characteristics of the signal segments; Among them, the variant of the gated recurrent unit is called the adaptive gated unit, which introduces a time-varying weight adjustment factor α t , dynamically calculated based on the statistical properties of the signal fragments: Among them, std(X t ) is the standard deviation of the signal segment, mean(X t ) is the mean, ∈ is a positive number to avoid division by zero; After each signal segment is processed by iRNN, a feature sequence H = [h1,h2,...,h T ], where h t is the eigenvector at time t, t=1,2,…,T.

5. The method for extracting time domain features of an EEG signal as claimed in claim 4, characterized in that: The improved time domain feature extraction algorithm includes: further improving the feature quality according to the features extracted by the deep learning model, initializing the time series weight vector Λ=[λ1,λ2,...,λ T ], dynamically adjust the timing weights according to the features extracted by iRNN; Update the time series weights, and extract the features f from iRNN t Used to calculate the timing weight λ t , which is then used to calculate the fusion feature TWFF.

6. The method for extracting time domain features of an EEG signal according to claim 5, characterized in that: The improved time-domain feature extraction algorithm further includes selecting a feature subset that contributes most to classification based on feature importance scores of a deep learning model; The preprocessed data is trained using the improved iRNN model. For each feature, the importance score is calculated using the trained iRNN model. Based on the feature importance score, the adaptive weight W of each feature is calculated. f ; According to the calculated adaptive weight W f , select the K features with the highest weights as the feature subset that contributes most to classification.

7. The method for extracting time domain features of an EEG signal according to claim 6, characterized in that: The extracting of the final signal features includes normalizing the features using a method based on signal energy distribution, selecting a feature subset, and then calculating the energy of each selected feature; For each feature, calculate the energy E f : Among them, x f,i is the i-th value of feature f, μ f and σ f are the mean and standard deviation of feature f respectively; Using the calculated characteristic energy E f , the selected features are normalized to eliminate the dimensional effects and numerical scale differences between the features.

8. A system using a method for extracting time domain features of an electroencephalogram signal as claimed in any one of claims 1 to 7, characterized in that: It includes signal acquisition and preprocessing module, adaptive signal segmentation module, deep learning feature extraction module, feature optimization and weight adjustment module, feature selection and normalization module; The signal acquisition and preprocessing module is responsible for capturing brain electrical activity through the EEG electrode cap, receiving and recording EEG signals and related information, and performing preliminary signal preprocessing to improve signal quality; The adaptive signal segmentation module uses adaptive window technology to dynamically segment the signal according to the characteristics of the EEG signal, ensuring that the signal features in each segmentation window can be effectively extracted; The deep learning feature extraction module uses an improved recurrent neural network iRNN to process the segmented signal segments, capture the long-term dependencies in the time series, output feature sequences, and introduces an adaptive gating unit to dynamically adjust network parameters; The feature optimization and weight adjustment module optimizes the feature sequence output by the deep learning model, including dynamically adjusting the time series weight, calculating the importance score of the feature, and updating the time series weight according to these scores to further refine the feature; The feature selection and normalization module selects the feature subset that contributes most to classification based on the feature importance score, and normalizes the selected features using a method based on signal energy distribution to ensure the consistency and comparability of the features.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images