Electroencephalogram signal noise reduction method and device based on multi-class noise classification representation and medium
By employing a multi-class noise classification and characterization method, and utilizing a noise identification and estimation network model combined with linear regression fitting coefficients to remove noise, the problem of lost amplitude information in EEG signals in existing technologies is solved, thereby improving the accuracy of data analysis and the practicality of the model.
Patent Information
- Application Number
- CN202410972821.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-19
- Publication Date
- 2024-11-15
AI Technical Summary
Existing EEG signal denoising methods lose the amplitude information of the EEG signal in the output results, which reduces the accuracy of subsequent data analysis and the practicality of the model.
A multi-class noise classification method is adopted, noise type is identified by a noise discrimination network model, and standard noise signal is estimated by the corresponding noise estimation network model. Combined with linear regression fitting coefficients, noise signal is removed from the original EEG signal while retaining amplitude information.
It achieves the preservation of EEG signal amplitude information during the denoising process, improving the accuracy of subsequent data analysis and the practicality of the model, and is applicable to the removal of various types of noise.
Smart Images

Figure CN118965062A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the technical field of electroencephalogram (EEG) signal processing, and in particular to an EEG signal denoising method, device and medium based on multi-class noise classification characterization. Background Art
[0002] EEG signals are weak bioelectric signals generated by the discharge of brain neurons, reflecting brain activity. Because EEG signals are easy to obtain, non-invasive, and have high temporal resolution, they play an increasingly important role in scientific research and disease diagnosis. To achieve the widespread application of EEG technology, the most critical link is to obtain good and reliable EEG signals. Due to the influence of current hardware equipment and the physiological reactions of the human body, the EEG signals collected by current equipment will be interfered by noise. Therefore, in order to obtain reliable EEG signals, it is necessary to design corresponding algorithms to remove noise and obtain clean EEG signals.
[0003] At present, there are EEG signal denoising methods based on neural network models, but these methods generally directly output the denoised EEG signal after denoising the signal. Since these methods need to be trained based on standardized training data during the model training process, the denoised EEG signal directly output is also a standardized signal, and the output result loses the amplitude information of the EEG signal. The amplitude information of the EEG signal plays a very important role in subsequent data analysis; for example, when using the event-related synchronization method to analyze the degree of activation of the local network of the human brain, the difference in the degree of activation of different parts is mainly reflected by the spatial difference in the baseline-free average amplitude of different neural rhythms. Therefore, this will lead to a decrease in the accuracy of the analysis results when the output denoised EEG signal is used for subsequent data analysis, which limits the practicality of the existing model. Summary of the invention
[0004] The embodiments of the present application provide an EEG signal denoising method, device and medium based on multi-class noise classification characterization, which can solve the problem that the output results of the existing EEG signal denoising methods do not retain the amplitude information of the EEG signals, resulting in limited practicality.
[0005] In a first aspect, an embodiment of the present application provides an EEG signal denoising method based on multi-class noise classification representation, comprising the steps of:
[0006] The noise identification network model is used to identify the noise type corresponding to the original EEG signal;
[0007] Determine a matching noise estimation network model according to the noise type; the noise type corresponds to the noise estimation network model one by one;
[0008] estimating a noise signal from the original EEG signal according to the noise estimation network model to obtain a standard noise signal;
[0009] Fitting the standard noise signal to the original EEG signal to determine a fitting coefficient;
[0010] According to the fitting coefficient and the standard noise signal, the noise signal is removed from the original EEG signal to obtain a noise-reduced EEG signal.
[0011] In a possible implementation manner of the first aspect, fitting the standard noise signal with the original EEG signal to determine a fitting coefficient includes:
[0012] According to the linear regression model, linear regression processing is performed on the standard noise signal and the original EEG signal to obtain a fitting coefficient.
[0013] In a possible implementation manner of the first aspect, removing the noise signal from the original EEG signal according to the fitting coefficient and the standard noise signal to obtain the noise-reduced EEG signal includes:
[0014] Multiplying the fitting coefficient and the standard noise signal to obtain an original noise signal;
[0015] The original noise signal is removed from the original EEG signal to obtain a noise-reduced EEG signal.
[0016] In a possible implementation manner of the first aspect, removing the noise signal from the original EEG signal according to the fitting coefficient and the standard noise signal to obtain the noise-reduced EEG signal includes:
[0017] Determine the original noise signal corresponding to the current standard noise signal according to the fitting coefficient corresponding to the current estimated standard noise signal;
[0018] Removing the original noise signal from the original EEG signal to obtain an intermediate EEG signal;
[0019] According to the intermediate EEG signal, the process jumps to the step of: using a noise identification network model to identify the noise type corresponding to the original EEG signal;
[0020] Until the original noise signals corresponding to all noise types are removed from the original EEG signal, a noise-reduced EEG signal is obtained.
[0021] In a possible implementation of the first aspect, the noise identification network model includes: a residual attention network layer, an average pooling layer, a temporal convolution layer, a two-dimensional convolution layer, a regularization layer, and a linear layer; the residual attention network layer and part of the average pooling layer form a plurality of groups of cascade settings; each group includes one residual attention network layer and one average pooling layer;
[0022] The method of using the noise identification network model to identify the noise type corresponding to the original EEG signal includes:
[0023] Using the original EEG signal as a set of input parameters of the residual attention network layer and the average pooling layer, and calculating a first intermediate result;
[0024] The first intermediate result is calculated by sequentially passing the temporal convolution layer and the two-dimensional convolution layer to obtain a second intermediate result;
[0025] The first intermediate result and the second intermediate result are used as input parameters of a subsequent group of the residual attention network layer and the average pooling layer, and after two iterations, a third intermediate result is calculated;
[0026] The original EEG signal is used as an input parameter of a temporal convolution layer for calculation, and then is sequentially calculated through an average pooling layer and a two-dimensional convolution layer to obtain a fourth intermediate result;
[0027] The third intermediate result and the fourth intermediate result are used as input parameters of the regularization layer for calculation, and then are calculated through the linear layer to obtain the noise type corresponding to the original EEG signal.
[0028] In a possible implementation of the first aspect, the noise estimation network model includes a connected encoder and a decoder, the encoder is used to obtain features corresponding to each noise type, and the decoder is used to estimate the standard noise signal corresponding to the noise type based on the features, and the encoder includes a cascaded position encoding module, a multi-head attention module, a residual connection and normalization module, a forward propagation layer, and a residual connection and normalization module.
[0029] In a possible implementation manner of the first aspect, the noise estimation network model is an electrooculogram noise estimation network, an electromyography noise estimation network, a motion noise estimation network, a direct current stimulation noise estimation network, a single-frequency alternating current stimulation noise estimation network, or a dual-frequency alternating current stimulation noise estimation network;
[0030] The electrooculogram noise estimation network is used to estimate electrooculogram noise, the electromyography noise estimation network is used to estimate electromyography noise, the motion noise estimation network is used to estimate motion noise, the direct current stimulation noise estimation network is used to estimate transcranial direct current stimulation noise, the single-frequency alternating current stimulation noise estimation network is used to estimate transcranial alternating current stimulation noise, and the dual-frequency alternating current stimulation noise estimation network is used to estimate transcranial time interference electrical stimulation noise.
[0031] In a second aspect, an embodiment of the present application provides an EEG signal denoising device based on multi-class noise classification representation, which implements the EEG signal denoising method described in the first aspect, including:
[0032] The noise type determination module uses the noise identification network model to identify the noise type corresponding to the original EEG signal;
[0033] A noise estimation network model determination module, which determines a matching noise estimation network model according to the noise type; the noise type corresponds to the noise estimation network model in a one-to-one manner;
[0034] A noise signal estimation module estimates the noise signal from the original EEG signal according to the noise estimation network model to obtain a standard noise signal;
[0035] A fitting coefficient determination module is used to fit the standard noise signal with the original EEG signal to determine the fitting coefficient;
[0036] The denoised EEG signal output module removes the noise signal from the original EEG signal according to the fitting coefficient and the standard noise signal to obtain the denoised EEG signal.
[0037] In a third aspect, an embodiment of the present application provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the method for reducing EEG signal noise as described in any one of the first aspects above is implemented.
[0038] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method for reducing EEG signal noise as described in any one of the above-mentioned first aspects is implemented.
[0039] Compared with the prior art, the embodiments of the present invention have the following beneficial effects:
[0040] The embodiment of the present application first uses a noise identification network model to identify the noise type corresponding to the original EEG signal, and then uses a noise estimation network model corresponding to the noise type to estimate the standard noise signal in the original EEG signal. Considering that the standard noise signal obtained by the above estimation is a standardized signal and is inconsistent with the amplitude information of the original noise signal in the original EEG signal, the present application fits the standard noise signal with the original EEG signal to obtain a fitting coefficient. Based on the fitting coefficient, the noise signal with amplitude information can be removed from the original EEG signal, and finally the denoised EEG signal is output without losing the amplitude information of the EEG signal, thereby ensuring the accuracy of the analysis results and the practicality of the model when the denoised EEG signal is used for subsequent data analysis. BRIEF DESCRIPTION OF THE DRAWINGS
[0041] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.
[0042] Figure 1 It is a flowchart of an EEG signal denoising method based on multi-class noise classification representation provided by an embodiment of the present application;
[0043] Figure 2 is a flowchart of an EEG signal denoising method based on multi-class noise classification characterization provided by another embodiment of the present application;
[0044] Figure 3 It is a structural schematic diagram of a noise identification network model disclosed in an embodiment of the present application;
[0045] Figure 4 is a flowchart of an EEG signal denoising method based on multi-class noise classification characterization provided by another embodiment of the present application;
[0046] Figure 5 It is a structural diagram of a noise estimation network model disclosed in an embodiment of the present application;
[0047] Figure 6 It is a structural schematic diagram of an EEG signal denoising device based on multi-class noise classification characterization provided in an embodiment of the present application;
[0048] Figure 7 It is a schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION
[0049] In the following description, specific details such as specific system structures, technologies, etc. are provided for the purpose of illustration rather than limitation, so as to provide a thorough understanding of the embodiments of the present application. However, it should be clear to those skilled in the art that the present application may also be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to prevent unnecessary details from obstructing the description of the present application.
[0050] It should be understood that when used in the present specification and the appended claims, the term "comprising" indicates the presence of described features, wholes, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, wholes, steps, operations, elements, components and / or combinations thereof.
[0051] It should also be understood that the term “and / or” used in the specification and appended claims refers to any and all possible combinations of one or more of the associated listed items, and includes these combinations.
[0052] As used in the specification and appended claims of this application, the term "if" can be interpreted as "when" or "uponce" or "in response to determining" or "in response to detecting", depending on the context. Similarly, the phrase "if it is determined" or "if [described condition or event] is detected" can be interpreted as meaning "uponce it is determined" or "in response to determining" or "uponce [described condition or event] is detected" or "in response to detecting [described condition or event]", depending on the context.
[0053] In addition, in the description of the present application specification and the appended claims, the terms "first", "second", "third", etc. are only used to distinguish the descriptions and cannot be understood as indicating or implying relative importance.
[0054] References to "one embodiment" or "some embodiments" etc. described in the specification of this application mean that one or more embodiments of the present application include specific features, structures or characteristics described in conjunction with the embodiment. Therefore, the statements "in one embodiment", "in some embodiments", "in some other embodiments", "in some other embodiments", etc. that appear in different places in this specification do not necessarily refer to the same embodiment, but mean "one or more but not all embodiments", unless otherwise specifically emphasized in other ways. The terms "including", "comprising", "having" and their variations all mean "including but not limited to", unless otherwise specifically emphasized in other ways.
[0055] like Figure 1 As shown, an embodiment of the present application discloses a method for reducing EEG signal noise based on multi-class noise classification characterization. The method comprises the steps of:
[0056] S110, using the noise identification network model to identify the noise type corresponding to the original EEG signal. Among them, the noise identification network model can not only identify whether the original EEG signal contains noise components, but also identify the noise type corresponding to the original noise signal in the original EEG signal. The above noise types may include electrooculogram noise, electromyography noise, motion noise, transcranial direct current stimulation noise, transcranial alternating current stimulation noise and transcranial time interference electrical stimulation noise. In this embodiment, the above noise identification network model can be constructed based on convolutional neural networks (CNN), but the present application is not limited to this.
[0057] S120, determine a matching noise estimation network model according to the noise type. Noise types correspond one to one with noise estimation network models. In this embodiment, there are multiple noise estimation network models, and each noise estimation network model is used to estimate a noise signal of a specific type. In this step, after determining the noise type, the noise estimation network model corresponding to the noise type is used to estimate the noise signal corresponding to the noise type. In a possible implementation, if there are 6 types of noise, then correspondingly, there are 6 noise estimation network models. And, one noise estimation network model is used to estimate the noise signal of the corresponding type.
[0058] S130, estimating the noise signal from the original EEG signal according to the noise estimation network model to obtain a standard noise signal. The estimated standard noise signal is a standardized noise signal, i.e., does not carry amplitude information. Both the noise identification network model and the noise estimation network model need to be trained with standardized training data, and the training data includes EEG signal data with various types of noise.
[0059] S140, fitting the standard noise signal with the original EEG signal to determine the fitting coefficient. Specifically, in one possible implementation, linear regression processing is performed on the standard noise signal and the original EEG signal according to a linear regression model to obtain the fitting coefficient. In the specific implementation, the above-mentioned linear regression model can be determined according to the actual situation, and this embodiment does not limit this.
[0060] S150, removing the noise signal from the original EEG signal according to the fitting coefficient and the standard noise signal to obtain a de-noised EEG signal.
[0061] Specifically, if Figure 2 As shown, step S150 includes:
[0062] S1501, the fitting coefficient and the standard noise signal are multiplied to obtain the original noise signal, that is, the noise signal with amplitude information in the original EEG signal is obtained, and the waveform of the noise part contained in the original EEG signal is inferred.
[0063] S1502, remove the original noise signal from the original EEG signal to obtain a denoised EEG signal. That is, the noise signal with amplitude information is removed from the original EEG signal to obtain a denoised EEG signal, which is a denoised EEG signal with amplitude information. This achieves the retention of the amplitude information of the denoised EEG signal, ensuring the accuracy of the analysis results and the practicality of the model when the denoised EEG signal is used for subsequent data analysis.
[0064] In specific implementation, since there are multiple types of noise signals, step S130 and step S140 only target one type of noise each time when performing noise estimation and fitting, so the above step S150 needs to go through multiple iterations until the original noise signals corresponding to all noise types are removed from the original EEG signal to obtain the denoised EEG signal. Specifically, step S150 includes:
[0065] S1503, determining the original noise signal corresponding to the current standard noise signal according to the fitting coefficient corresponding to the currently estimated standard noise signal, that is, multiplying the fitting coefficient and the standard noise signal to obtain the original noise signal corresponding to the current standard noise signal.
[0066] S1504, removing the original noise signal from the original EEG signal to obtain an intermediate EEG signal.
[0067] S1505, jump to step S110 according to the intermediate EEG signal: use the noise identification network model to identify the noise type corresponding to the original EEG signal. At this time, the corresponding step is: use the noise identification network model to identify the noise type corresponding to the intermediate EEG signal.
[0068] S1506, until all original noise signals corresponding to the noise types are removed from the original EEG signal to obtain a de-noised EEG signal. That is, until the noise identification network model fails to identify a noise signal of a specified type, it can be determined that all noise signals of the specified type have been removed.
[0069] If the noise signal has 6 types, then accordingly, steps S110 to S140 and steps S1503 to S1505 need to be iteratively performed 6 times, and each time the iterative execution removes one type of noise signal.
[0070] This embodiment first uses a noise identification network model to identify the noise type corresponding to the original EEG signal, and then uses a noise estimation network model corresponding to the noise type to estimate the standard noise signal in the original EEG signal. Considering that the standard noise signal obtained by the above estimation is a standardized signal and is inconsistent with the amplitude information of the original noise signal in the original EEG signal, this application fits the standard noise signal with the original EEG signal to obtain a fitting coefficient. Based on the fitting coefficient, the noise signal with amplitude information can be removed from the original EEG signal, and finally the denoised EEG signal is output without losing the amplitude information of the EEG signal, thereby ensuring the accuracy of the analysis results and the practicality of the model when the denoised EEG signal is used for subsequent data analysis.
[0071] On the other hand, this embodiment can achieve the removal of various types of noise, and has better practicality. Moreover, this embodiment trains the noise identification network model and the noise estimation network model based on noise characterization learning, and has better noise reduction performance than the solution of directly using deep neural networks to learn the potential characteristics of EEG signals to estimate the EEG signals in the input signal. This is because compared with noise signals, EEG signals have small amplitudes and random features, which are not conducive to the learning and estimation of deep neural networks.
[0072] In some optional embodiments, in the above Figure 1 On the basis of the corresponding embodiment, the noise identification network model includes: a residual attention network (ResNetXtBolck) layer, an average pooling (AvgPooling) layer, a temporal convolution (TemporalConvolutional Network Blocks, TCN Bolcks) layer, a two-dimensional convolution (Conv2d) layer, a regularization (Dropout) layer and a linear (Linear) layer. Figure 3 As shown, the residual attention network layer and the partial average pooling layer form multiple groups of cascade settings, each group includes a residual attention network layer and an average pooling layer. That is, a residual attention network layer and an average pooling layer form a cascade group, and multiple cascade groups are cascaded. Exemplarily, this embodiment shows 3 cascade groups, but the present application is not limited to this.
[0073] In this embodiment, after the above-mentioned residual attention network layer extracts features through three layers of basic two-dimensional convolution modules, channel attention is added to accelerate model convergence, and residual connections are added to alleviate the gradient vanishing problem caused by network depth. The average pooling layer can reduce the dimension of the feature map and improve the generalization ability of the model; it can also enhance the anti-interference ability of the model and make the network model more stable. For deeper neural networks, the introduction of the regularization layer can stabilize the neural network and prevent the occurrence of overfitting. The linear layer completes the linear transformation of the data, and finally outputs the noise type of the noise signal, that is, the classification category.
[0074] The temporal convolution layer extracts the global temporal relationship characteristics of the EEG signal through multi-layer one-dimensional dilated convolution calculation. Y (j) represents the output of the jth layer in the temporal convolutional layer, and L is the sequence length. (j) ∈R K is the convolution kernel of layer j, K is the length of the convolution kernel, k is the position index of the convolution kernel, and t is the sequence index. Then, the calculation process of the dilated convolution is expressed as shown in formula (1):
[0075]
[0076] After sorting, it is shown as formula (2):
[0077]
[0078] Where d is the expansion factor.
[0079] The two-dimensional convolution layer is formed by combining the basic two-dimensional convolution module, batch normalization layer, regularization layer and activation function layer. The calculation process of the two-dimensional convolution layer is shown in formula (3):
[0080]
[0081] Among them, Y and X represent output and input respectively, K is the convolution kernel, b is the bias factor, i and j represent the position index of the output, M represents the size of the convolution kernel in the length direction, N represents the size of the convolution kernel in the width direction, and m and n represent the position index of the convolution kernel.
[0082] like Figure 4 As shown, in this embodiment, step S110 includes:
[0083] S1101, using the original EEG signal as an input parameter of a group of residual attention network layers and average pooling layers, and calculating the first intermediate result. That is, the original EEG signal is cut and combined, and used as an input parameter of the first cascade group, and the first intermediate result is calculated.
[0084] S1102, the first intermediate result is calculated by a temporal convolution layer and a two-dimensional convolution layer in sequence to obtain a second intermediate result.
[0085] S1103, using the first intermediate result and the second intermediate result as input parameters of the next group of residual attention network layers and average pooling layers, and after two iterations, calculating the third intermediate result. Specifically, using the first intermediate result and the second intermediate result as input parameters of the second cascade group, calculating the fifth intermediate result, calculating the sixth intermediate result through the temporal convolution layer and the two-dimensional convolution layer of the fifth intermediate result, using the fifth intermediate result and the sixth intermediate result as input parameters of the third cascade group, calculating the third intermediate result.
[0086] S1104, the original EEG signal is used as the input parameter of the time series convolution layer for calculation, and then is calculated in turn through the average pooling layer and the two-dimensional convolution layer to obtain a fourth intermediate result.
[0087] S1105, the third intermediate result and the fourth intermediate result are used as input parameters of the regularization layer for calculation, and then the linear layer is calculated to obtain the noise type corresponding to the original EEG signal.
[0088] In some optional embodiments, in the above Figure 1 On the basis of the corresponding embodiment, the noise estimation network model is an electrooculogram noise estimation network, an electromyography noise estimation network, a motion noise estimation network, a direct current stimulation noise estimation network, a single-frequency alternating current stimulation noise estimation network, or a dual-frequency alternating current stimulation noise estimation network. Among them, the structures of the above 6 noise estimation networks are the same, only the model parameters are different, and the model parameters are obtained by training with large-scale data corresponding to the respective noise categories.
[0089] The electrooculogram noise estimation network is used to estimate electrooculogram noise, the electromyography noise estimation network is used to estimate electromyography noise, the motion noise estimation network is used to estimate motion noise, the direct current stimulation noise estimation network is used to estimate transcranial direct current stimulation noise, the single-frequency alternating current stimulation noise estimation network is used to estimate transcranial alternating current stimulation noise, and the dual-frequency alternating current stimulation noise estimation network is used to estimate transcranial time interference electrical stimulation noise.
[0090] In this embodiment, the noise estimation network model is implemented based on an autoencoder-decoder, which includes an encoder and a decoder connected to each other. The encoder is used to obtain the features corresponding to each noise type, that is, a low-dimensional compressed representation, and the decoder is used to estimate the corresponding standard noise signal based on the noise features obtained by the encoder. The noise estimation network model is based on the powerful representation learning ability of the autoencoder-decoder structure, and performs representation learning on specific types of noise to achieve estimation and inference of the noise waveform contained in the input EEG signal.
[0091] Among them, Figure 5As shown, the encoder includes a cascaded position encoding module, a multi-head attention module, a residual connection and normalization module, a forward propagation layer, and a residual connection and normalization module. The decoder includes a cascaded multi-head attention module, a residual connection and normalization module, a forward propagation layer, and a residual connection and normalization and linearization module.
[0092] The position encoding module is used to embed the signal position, i.e., calibrate the position, so that the model can directly know the position distribution information of all data without obtaining the global position information of the data through deep training. The position encoding module can implement its functions based on two strategies, namely fixed position encoding and learnable position encoding. Among them, learnable position encoding regards the position encoding as a model parameter and adjusts the position parameter through gradient back propagation.
[0093] The calculation method of fixed position encoding is shown in formula (4):
[0094]
[0095] where t is the input sequence position, is the position encoding, i is the index of the vector, ω k =1 / (10000 2k / d ), d is the dimension of the input, and k is an integer greater than or equal to 0.
[0096] The multi-head attention module can capture the dependencies between different positions in the input sequence. The input vector is first divided into multiple heads through linear mapping, and then the attention weight is calculated for each head. Finally, the attention weights of multiple heads are weighted averaged. For the residual connection and normalization modules, the residual connection prevents the loss of initial features and allows the original input information to be directly passed to deeper layers, which helps to train deep networks. Normalization avoids gradient disappearance and reduces operations. The forward propagation layer maps the data to a high-dimensional space and then to a low-dimensional space through linear transformation, and adds nonlinear transformation to extract deeper features.
[0097] It should be understood that the size of the serial numbers of the steps in the above embodiments does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.
[0098] It should be noted that the above-mentioned method embodiments disclosed in the present application can be freely combined, and the technical solutions obtained after the free combination are also within the protection scope of the present application.
[0099] Corresponding to the EEG signal denoising method based on multi-class noise classification representation described in the above embodiment, Figure 6A structural block diagram of an EEG signal denoising device based on multi-class noise classification characterization provided in an embodiment of the present application is shown. For ease of explanation, only the parts related to the embodiment of the present application are shown.
[0100] Reference Figure 6 , the EEG signal denoising device based on multi-class noise classification representation includes:
[0101] The noise type determination module 61 uses a noise identification network model to identify the noise type corresponding to the original EEG signal.
[0102] The noise estimation network model determination module 62 determines a matching noise estimation network model according to the noise type. Noise types correspond to noise estimation network models one by one.
[0103] The noise signal estimation module 63 estimates the noise signal from the original EEG signal according to the noise estimation network model to obtain a standard noise signal.
[0104] The fitting coefficient determination module 64 fits the standard noise signal with the original EEG signal to determine the fitting coefficient.
[0105] The denoised EEG signal output module 65 removes the noise signal from the original EEG signal according to the fitting coefficient and the standard noise signal to obtain the denoised EEG signal.
[0106] It should be noted that the information interaction, execution process, etc. between the above-mentioned devices / units are based on the same concept as the method embodiment of the present application. Their specific functions and technical effects can be found in the method embodiment part and will not be repeated here.
[0107] The technicians in the relevant field can clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional units and modules is used as an example for illustration. In practical applications, the above-mentioned function allocation can be completed by different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiment can be integrated in a processing unit, or each unit can exist physically separately, or two or more units can be integrated in one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of software functional units. In addition, the specific names of the functional units and modules are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of this application. The specific working process of the units and modules in the above-mentioned system can refer to the corresponding process in the aforementioned method embodiment, which will not be repeated here.
[0108] The present application also provides an electronic device, such as Figure 7As shown, the electronic device 70 includes: at least one processor 701, a memory 702, and a computer program 703 stored in the memory and executable on the at least one processor, and the processor implements the steps in any of the above-mentioned method embodiments when executing the computer program.
[0109] An embodiment of the present application further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps in the above-mentioned method embodiments can be implemented.
[0110] An embodiment of the present application provides a computer program product. When the computer program product runs on a mobile terminal, the mobile terminal can implement the steps in the above-mentioned method embodiments when executing the computer program product.
[0111] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present application implements all or part of the processes in the above-mentioned embodiment method, which can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a computer-readable storage medium. When the computer program is executed by the processor, the steps of the above-mentioned various method embodiments can be implemented. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may at least include: any entity or device that can carry the computer program code to the camera / electronic device, recording medium, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, RandomAccess Memory), electric carrier signal, telecommunication signal and software distribution medium. For example, a USB flash drive, a mobile hard disk, a magnetic disk or an optical disk. In some jurisdictions, according to legislation and patent practice, computer-readable media cannot be electric carrier signals and telecommunication signals.
[0112] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described or recorded in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.
[0113] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.
[0114] In the embodiments provided in the present application, it should be understood that the disclosed devices / network equipment and methods can be implemented in other ways. For example, the device / network equipment embodiments described above are merely schematic. For example, the division of the modules or units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.
[0115] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.
[0116] The embodiments described above are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, a person skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application, and should all be included in the protection scope of the present application.
Claims
1. A method for reducing EEG signal noise based on multi-class noise classification characterization, characterized in that: Includes steps: The noise identification network model is used to identify the noise type corresponding to the original EEG signal; Determine a matching noise estimation network model according to the noise type; the noise type corresponds to the noise estimation network model one by one; estimating a noise signal from the original EEG signal according to the noise estimation network model to obtain a standard noise signal; Fitting the standard noise signal with the original EEG signal to determine a fitting coefficient; According to the fitting coefficient and the standard noise signal, the noise signal is removed from the original EEG signal to obtain a noise-reduced EEG signal.
2. The method for reducing noise of an EEG signal according to claim 1, characterized in that: The step of fitting the standard noise signal with the original EEG signal to determine a fitting coefficient includes: According to the linear regression model, linear regression processing is performed on the standard noise signal and the original EEG signal to obtain a fitting coefficient.
3. The method for reducing noise of an EEG signal according to claim 1, characterized in that: The step of removing the noise signal from the original EEG signal according to the fitting coefficient and the standard noise signal to obtain a noise-reduced EEG signal comprises: Multiplying the fitting coefficient and the standard noise signal to obtain an original noise signal; The original noise signal is removed from the original EEG signal to obtain a noise-reduced EEG signal.
4. The method for reducing noise of an EEG signal according to claim 1, wherein: The step of removing the noise signal from the original EEG signal according to the fitting coefficient and the standard noise signal to obtain a noise-reduced EEG signal comprises: Determine the original noise signal corresponding to the current standard noise signal according to the fitting coefficient corresponding to the current estimated standard noise signal; Removing the original noise signal from the original EEG signal to obtain an intermediate EEG signal; According to the intermediate EEG signal, the process jumps to the step of: using a noise identification network model to identify the noise type corresponding to the original EEG signal; Until the original noise signals corresponding to all noise types are removed from the original EEG signal, a noise-reduced EEG signal is obtained.
5. The method for reducing noise of an EEG signal according to any one of claims 1 to 4, characterized in that: The noise identification network model includes: a residual attention network layer, an average pooling layer, a temporal convolution layer, a two-dimensional convolution layer, a regularization layer and a linear layer; the residual attention network layer and part of the average pooling layer form a plurality of groups of cascade settings; each group includes one residual attention network layer and one average pooling layer; The method of using the noise identification network model to identify the noise type corresponding to the original EEG signal includes: Using the original EEG signal as a set of input parameters of the residual attention network layer and the average pooling layer, and calculating a first intermediate result; The first intermediate result is calculated by sequentially passing the temporal convolution layer and the two-dimensional convolution layer to obtain a second intermediate result; The first intermediate result and the second intermediate result are used as input parameters of a subsequent group of the residual attention network layer and the average pooling layer, and after two iterations, a third intermediate result is calculated; The original EEG signal is used as an input parameter of a temporal convolution layer for calculation, and then is sequentially calculated through an average pooling layer and a two-dimensional convolution layer to obtain a fourth intermediate result; The third intermediate result and the fourth intermediate result are used as input parameters of the regularization layer for calculation, and then are calculated through the linear layer to obtain the noise type corresponding to the original EEG signal.
6. The method for reducing noise of an EEG signal according to any one of claims 1 to 4, characterized in that: The noise estimation network model includes an encoder and a decoder connected to each other, the encoder is used to obtain the features corresponding to each noise type, and the decoder is used to estimate the standard noise signal corresponding to the noise type based on the features. The encoder includes a cascaded position encoding module, a multi-head attention module, a residual connection and normalization module, a forward propagation layer, and a residual connection and normalization module.
7. The method for reducing noise of an EEG signal according to any one of claims 1 to 4, characterized in that: The noise estimation network model is an electrooculogram noise estimation network, an electromyography noise estimation network, a motion noise estimation network, a direct current stimulation noise estimation network, a single-frequency alternating current stimulation noise estimation network, or a dual-frequency alternating current stimulation noise estimation network; The electrooculogram noise estimation network is used to estimate electrooculogram noise, the electromyography noise estimation network is used to estimate electromyography noise, the motion noise estimation network is used to estimate motion noise, the direct current stimulation noise estimation network is used to estimate transcranial direct current stimulation noise, the single-frequency alternating current stimulation noise estimation network is used to estimate transcranial alternating current stimulation noise, and the dual-frequency alternating current stimulation noise estimation network is used to estimate transcranial time interference electrical stimulation noise.
8. An electronic device, characterized in that: The method comprises a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the method according to any one of claims 1 to 7 when executing the computer program.
9. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.
