Electroencephalogram signal decoding method and device, electronic equipment and storage medium

Abstract

The application provides an electroencephalogram signal decoding method and device, electronic equipment and storage medium, wherein the method comprises: obtaining to-be-decoded electroencephalogram data; performing channel compression on the to-be-decoded electroencephalogram data to obtain compressed electroencephalogram data; performing strategy selection on the compressed electroencephalogram data based on a strategy model to obtain a compression strategy of the to-be-decoded electroencephalogram data and a corresponding decoding model light-weight strategy; compressing the to-be-decoded electroencephalogram data based on the compression strategy to obtain target compression data, and performing intention decoding on the target compression data based on a target decoding model selected based on the decoding model light-weight strategy to obtain an intention decoding result. The method, device, electronic equipment and storage medium provided by the application reduce the amount of calculation required for electroencephalogram decoding by compressing the to-be-decoded electroencephalogram data and light-weighting the decoding model, improve the decoding efficiency of the decoding model, and make it possible to deploy on many resource-limited terminals (such as rehabilitation robots).

Description

Technical Field

[0001] The present invention relates to the field of electroencephalogram (EEG) processing technology, and in particular to an EEG signal decoding method, device, electronic equipment, and storage medium. Background Art

[0002] Brain-computer interface (BCI) technology based on electroencephalogram (EEG) provides a new way for people with disabilities to interact, allowing them to control external devices such as wheelchairs or robotic arms through their thoughts. With the rapid development of artificial intelligence technology, the widespread application of neural network models has brought new opportunities and challenges to EEG signal decoding.

[0003] In the existing technology, although deep learning-based EEG signal decoding methods have achieved good classification results on commonly used public data sets, the EEG classification accuracy obtained by these models usually increases proportionally with the complexity of the decoding model and the amount of data, that is, the higher the dimension of the input data, the more complex the decoding model, and the higher the EEG classification accuracy.

[0004] However, the above factors result in existing EEG signal decoding technologies requiring high computing power from the terminal, which severely restricts the deployment of these technologies on many resource-limited terminals (such as rehabilitation robots). Summary of the Invention

[0005] The present invention provides an EEG signal decoding method, device, electronic device and storage medium to address the defect in the existing EEG signal decoding technology that the computing power required by the terminal is high, which seriously restricts the deployment of these technologies on many resource-limited terminals (such as rehabilitation robots).

[0006] The present invention provides an electroencephalogram (EEG) signal decoding method, comprising:

[0007] Obtaining EEG data to be decoded;

[0008] Performing channel compression on the EEG data to be decoded to obtain compressed EEG data;

[0009] Based on the strategy model, a strategy is selected for the compressed EEG data to obtain a compression strategy for the EEG data to be decoded and a corresponding lightweight decoding model strategy;

[0010] The EEG data to be decoded is compressed based on the compression strategy to obtain target compressed data, and the target compressed data is intentionally decoded based on the target decoding model selected by the decoding model lightweight strategy to obtain an intention decoding result.

[0011] According to a method for decoding EEG signals provided by the present invention, the target decoding model includes a time domain convolution layer, a spatial convolution layer, a pooling layer and a classification layer connected in sequence. The target decoding model is lightweight by changing the number of convolution kernels of the time domain convolution layer and the spatial convolution layer.

[0012] According to an EEG signal decoding method provided by the present invention, the training step of the target decoding model includes two parts: a pre-training step of a candidate target decoding model and a collaborative training step of a strategy model and the target decoding model; the collaborative training step includes a forward propagation step and a gradient back propagation step;

[0013] The candidate target decoding model pre-training step includes: pre-training the candidate target decoding model corresponding to each compression strategy based on the sample compressed EEG data under each compression strategy and the intention label of the sample compressed EEG data;

[0014] The forward propagation step includes: selecting a strategy for the EEG sample data to be decoded based on the strategy model to obtain a candidate compression strategy and a corresponding decoding model lightweight strategy; compressing the EEG sample data to be decoded based on the candidate compression strategy to obtain candidate sample compressed data, and performing intent decoding on the candidate sample compressed data based on the candidate target decoding model selected by the decoding model lightweight strategy to obtain an intent decoding prediction result;

[0015] The gradient back propagation step includes: based on the intention decoding prediction result and the classification label of the EEG sample data to be decoded, as well as the number of floating-point operations on the candidate sample compressed data input into the candidate target decoding model, performing parameter iteration on the strategy model and the candidate target decoding model through gradient back propagation, and obtaining the target decoding model based on the candidate target decoding model after parameter iteration.

[0016] According to an EEG signal decoding method provided by the present invention, the method includes performing parameter iteration on the strategy model and the candidate target decoding model based on the intention decoding prediction result and the classification label of the EEG sample data to be decoded, as well as the number of floating-point operations on the candidate sample compressed data input into the candidate target decoding model through gradient back propagation, and obtaining the target decoding model based on the candidate target decoding model after parameter iteration, including:

[0017] Determining a decoding accuracy loss function based on the intent decoding prediction result and the classification label;

[0018] Determining a decoding efficiency loss function based on averaging the number of floating-point operations of the candidate sample compressed data input into the candidate target decoding model;

[0019] Based on the decoding accuracy loss function and the decoding efficiency loss function, the policy model and the candidate target decoding model are iterated by gradient back propagation, and the target decoding model is obtained based on the candidate target decoding model after parameter iteration.

[0020] According to an EEG signal decoding method provided by the present invention, the strategy model includes a strategy feature extraction model and a strategy decision model connected to the strategy feature extraction model;

[0021] The strategy model is based on which the strategy is selected for the compressed EEG data to obtain the compression strategy of the EEG data to be decoded and the corresponding lightweight strategy of the decoding model, including:

[0022] Inputting the compressed EEG data into the strategy feature extraction model, and extracting the strategy features in the compressed EEG data by the strategy feature extraction model;

[0023] Inputting the policy features into the policy decision model, and having the policy decision model output prediction probabilities of multiple data compression policies and prediction probabilities of multiple decoding model policies;

[0024] Based on the predicted probabilities of the multiple data compression strategies and the predicted probabilities of the multiple decoding model strategies, a compression strategy for the to-be-decoded EEG data and a corresponding lightweight decoding model strategy are determined.

[0025] According to an EEG signal decoding method provided by the present invention, the method of determining the compression strategy of the EEG data to be decoded and the corresponding lightweight decoding model strategy based on the prediction probabilities of the multiple data compression strategies and the prediction probabilities of the multiple decoding model strategies includes:

[0026] Based on the maximum value among the predicted probabilities of the multiple data compression strategies and the maximum value among the predicted probabilities of the multiple decoding model strategies, the compression strategy of the to-be-decoded EEG data and the corresponding lightweight decoding model strategy are determined.

[0027] According to an EEG signal decoding method provided by the present invention, channel compression is performed on the EEG data to be decoded to obtain compressed EEG data, including:

[0028] Based on a preset channel conversion rule, channel compression is performed on the EEG data to be decoded to obtain compressed EEG data; the preset channel conversion rule specifically includes a difference rule, a mean rule, and a selective activation rule;

[0029] The difference rule achieves channel data compression by calculating the difference between the EEG data collected by the corresponding electrodes on both sides of the brain area;

[0030] The mean rule achieves channel data compression by averaging the EEG data of adjacent channels;

[0031] The selective activation rule achieves channel data compression by removing channels corresponding to EEG data that are irrelevant to the task to be classified.

[0032] The present invention also provides an electroencephalogram signal decoding device, comprising:

[0033] An acquisition unit, used for acquiring EEG data to be decoded;

[0034] a channel compression unit, configured to perform channel compression on the EEG data to be decoded to obtain compressed EEG data;

[0035] A strategy selection unit, configured to select a strategy for the compressed EEG data based on a strategy model, to obtain a compression strategy for the EEG data to be decoded and a corresponding lightweight decoding model strategy;

[0036] The intention decoding unit is used to compress the EEG data to be decoded based on the compression strategy to obtain target compressed data, and to perform intention decoding on the target compressed data based on the target decoding model selected by the decoding model lightweight strategy to obtain an intention decoding result.

[0037] The present invention also 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 program, any of the above-described methods for decoding electroencephalogram signals is implemented.

[0038] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements any of the above-described methods for decoding EEG signals.

[0039] The present invention also provides a computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the computer program implements any of the above-mentioned methods for decoding electroencephalogram signals.

[0040] The EEG signal decoding method, device, electronic device and storage medium provided by the present invention obtain EEG data to be decoded, perform channel compression on the EEG data to be decoded to obtain compressed EEG data, and then select a strategy for the compressed EEG data based on a strategy model to obtain a compression strategy for the EEG data to be decoded and a corresponding lightweight decoding model strategy. Finally, the EEG data to be decoded is compressed based on the compression strategy to obtain target compressed data, and the target compressed data is intent-decoded based on the target decoding model selected by the lightweight decoding model strategy to obtain an intent decoding result. Therefore, by compressing the EEG data to be decoded and lightweighting the decoding model, the amount of computation required for EEG decoding is reduced, the decoding efficiency of the decoding model is improved, and deployment on many resource-limited terminals (such as rehabilitation robots) is possible. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] In order to more clearly illustrate the technical solutions in the present invention or the prior art, a brief introduction is given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. 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 the FasterEEG method for decoding EEG signals provided by the present invention;

[0043] Figure 2 It is a schematic structural diagram of the target decoding model provided by the present invention;

[0044] Figure 3 It is a flowchart of various methods of EEG signal decoding methods provided by the present invention;

[0045] Figure 4 This is a diagram of EEG signal decoding efficiency and decoding accuracy provided by the present invention;

[0046] Figure 5 is a schematic diagram of the preset channel conversion rule provided by the present invention;

[0047] Figure 6 It is a structural diagram of the EEG signal decoding device provided by the present invention;

[0048] Figure 7 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0049] To make the objectives, technical solutions, and advantages of the present invention more clear, the technical solutions of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the embodiments described are only some of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0050] The terms "first," "second," and the like in the specification and claims of the present invention are used to distinguish similar objects, and are not used to describe a particular order or precedence. It should be understood that the terms used in this manner are interchangeable where appropriate, so that the embodiments of the present application can be implemented in orders other than those illustrated or described herein, and that the objects distinguished by "first," "second," and the like are generally of the same type.

[0051] The brain is one of the most important organs in our body, possessing crucial functions such as perception, cognition, and movement. It is the cornerstone of our understanding and transformation of the world. The brain's operation primarily relies on spike activity within neural networks, which can be recorded using multi-channel EEG technology, such as scalp and intracranial recordings. These recordings contain rich spatial and detailed temporal information, revealing the functional and pathological mechanisms of the brain.

[0052] EEG-based brain-computer interface (BCI) technology provides a new way of interaction for people with disabilities, enabling them to control external devices such as wheelchairs or robotic arms through their thoughts.

[0053] In addition, BCI technology is also widely used in the field of neurorehabilitation, helping patients with stroke or brain injury to restore their motor skills. In the field of neuroscience, BCI technology is also used to study cognitive and behavioral neuroscience, promoting the understanding of the brain's internal information processing mechanisms.

[0054] With the rapid development of artificial intelligence (AI) technology, the widespread application of neural network models has brought new opportunities and challenges to EEG signal decoding. Neural network models can be trained to identify EEG patterns associated with specific tasks, such as controlling the imagined movements of a robotic arm. Furthermore, neural network models can perform operations such as classification, feature extraction, and pattern recognition on EEG data, providing a more accurate and efficient means for EEG signal decoding.

[0055] Although deep learning-based EEG signal decoding methods have achieved good classification results on commonly used public datasets, the EEG classification accuracy obtained by these models usually increases proportionally with the complexity of the decoding model and the amount of data. That is, the higher the dimension of the input data, the more complex the decoding model, and the higher the EEG classification accuracy.

[0056] However, the above factors result in existing EEG signal decoding technologies requiring high computing power from the terminal, which severely restricts the deployment of these technologies on many resource-limited terminals (such as rehabilitation robots).

[0057] In summary, the amount of input data and the number of decoding model parameters are key factors affecting the decoding efficiency of the BCI model.

[0058] First, in terms of input data volume, to ensure decoding accuracy, a typical EEG-based BCI system generally contains 32 to 128 EEG channels. Although the use of multiple channels can improve decoding accuracy to a certain extent, the resulting high amount of EEG data and high computing resource requirements make BCI decoding inefficient. In addition, the information obtained from high-density EEG channels is generally highly redundant (cross-correlated), and the key EEG channels involved in identifying different types of brain intentions are also different. For example, for a BCI paradigm based on steady-state visual evoked signals, the information collected by EEG channels distributed in the occipital lobe is crucial for intention decoding. For a motor imagery BCI paradigm, only the CZ, C3, and C4 channels distributed in the motor cortex may be required to achieve efficient and accurate decoding.

[0059] Second, for the decoding model, previous work often uses a fixed-size classification model for all samples, which can easily lead to resource waste, especially for simple samples. For example, it is easier to distinguish upper limb movement intentions ("left hand" and "right hand") than lower limb movement intentions ("left foot" and "right foot").

[0060] Based on the above problems, the present invention provides an EEG signal decoding method. Figure 1 FIG. 1 is a flow chart of the electroencephalogram signal decoding method FasterEEG provided by the present invention, as shown in FIG. Figure 1 As shown, the method includes:

[0061] Step 110: Obtain the EEG data to be decoded.

[0062] Specifically, the EEG data to be decoded can be obtained. The EEG data to be decoded here is the EEG data that needs to be decoded later. The EEG data to be decoded can be used Indicates that each trial X0 contains N0 EEG channels and T time series samples.

[0063] Step 120 : performing channel compression on the EEG data to be decoded to obtain compressed EEG data.

[0064] Specifically, after obtaining the brain electrical data to be decoded, channel compression can be performed on the brain electrical data to be decoded to obtain compressed brain electrical data. The compressed brain electrical data refers to brain electrical data obtained after channel compression is performed on the brain electrical data to be decoded. The compressed brain electrical data can be EEG data.

[0065] Here, the channel compression performed on the brain electrical data to be decoded can be based on a preset channel conversion rule. The preset channel conversion rule specifically includes a difference rule, a mean rule, and a selective activation rule.

[0066] That is, the full-channel brain electrical data X0 to be decoded is compressed into a data format X K of a preset minimum data size.

[0067] Channel compression mainly compresses the number of channels. The compressed brain electrical data can be sorted from large to small as X1, X2,..., X K , where N1>N2>…>N K

[0068] In order to convert the full-channel brain electrical data X0 into compressed data with different numbers of channels, a series of corresponding data compression strategies f k are defined. k The original full-channel data (brain electrical data to be decoded) X0 can be compressed into a data format containing N

[0069] X k =f k X0 Formula (1)

[0070] Where X * represents brain electrical data containing N input channels.

[0071] The compressed brain electrical data can be 1 / 5-channel brain electrical data, which is not limited in the embodiments of the application.

[0072] In step 130, based on the strategy model, the compressed brain electrical data is selected based on the strategy model to obtain the compression strategy of the brain electrical data to be decoded and the corresponding decoding model lightweight strategy.

[0073] Specifically, in order to reduce the additional calculation amount generated by the strategy model, the input of the strategy model is the compressed data X K with the smallest number of channels, so as to minimize the calculation amount of the strategy model.

[0074] Based on the strategy model, the compressed brain electrical data X K is selected based on the strategy model to obtain the compression strategy of the brain electrical data to be decoded X0 and the corresponding decoding model lightweight strategy.

[0075] The strategy model here is used to select the optimal decoding method for the EEG data to be decoded, thereby selecting the compression strategy of the EEG data to be decoded and the corresponding lightweight strategy of the decoding model.

[0076] It can be understood that the number of channels of the EEG data to be decoded is compressed to the minimum number of channels data, that is, redundant, useless channels are removed to achieve a reduction in the amount of initial input data, and then the strategy features related to the optimal number of channels are extracted from the minimum number of channels data. By extracting the strategy features from the minimum number of channels data, it is possible to use less channel data to obtain more comprehensive strategy features covering all channels in the EEG data to be decoded, which significantly reduces the amount of calculation in the process of extracting the strategy features and effectively improves the decoding rate.

[0077] Here, the compression strategy can be p chl 、p mdl Indicates that p chl represents the compression strategy of the EEG data to be decoded, p mdl Indicates the corresponding decoding model lightweight strategy, which selects a target decoding model.

[0078] Figure 2 It is a schematic diagram of the structure of the target decoding model provided by the present invention, such as Figure 2 As shown, the target decoding model can be a lightweight ShallowConvNet model, and the target decoding model can include a temporal convolution layer, a spatial convolution layer, a pooling layer and a classification layer connected in sequence. The temporal convolution layer here can be a temporal convolutional network (TCN), the spatial convolution layer here can be a spatial convolution network (SCN), the pooling layer here can be a maximum pooling layer or an average pooling layer, etc. The classification layer here can be a linear classification layer, which is not specifically limited in the embodiment of the present invention.

[0079] It is understandable that most of the computational cost of the target decoding model comes from the temporal and spatial convolution layers, where the default number of convolution kernels is 40. By changing the number of convolution kernels, the computational cost of the target decoding model can be significantly reduced.

[0080] That is, the target decoding model is lightweight by changing the number of convolution kernels in the temporal convolution layer and the spatial convolution layer.

[0081] It is understandable that for each target decoding model of different sizes, from small to large, it is defined as The larger M is, the larger the target decoding model is, and the corresponding GFLOPs (Giga Floating-point Operations Per Second) is larger.

[0082] Taking into account the difficulty of decoding the EEG data to be decoded, the prior art uses a fixed-size decoding model. In the embodiment of the present invention, a large decoding model is applied to difficult EEG signal recognition, while a small decoding model is applied to simple EEG signal recognition. This avoids waste of resources and improves resource utilization.

[0083] GFLOPS here refers to 1 billion floating-point operations per second, which is often used as a GPU performance parameter but does not necessarily represent the actual performance of the GPU, because the specific splitting of polygons and pixels, as well as texture filling, must also be considered. In theory, the higher the value, the better.

[0084] Here, there may be multiple compression strategies for the EEG data to be decoded, and there may be multiple corresponding lightweight strategies for the decoding model.

[0085] Step 140 , compressing the EEG data to be decoded based on the compression strategy to obtain target compressed data, and performing intent decoding on the target compressed data based on the target decoding model selected by the decoding model lightweight strategy to obtain an intent decoding result.

[0086] Specifically, Figure 3 This is a flow chart of various EEG signal decoding methods provided by the present invention, such as Figure 3 Compared with the commonly used “one size fits all” approach ( Figure 3 Compared with (a1), the embodiment of the present invention is more inclined to adaptively perform a certain degree of data compression on the EEG data to be decoded to reduce the amount of input data according to the difference of the EEG data to be decoded, and at the same time select a target decoding model of appropriate size for classification, so as to improve the decoding efficiency without reducing the decoding accuracy.

[0087] That is, after obtaining the compression strategy of the EEG data to be decoded and the corresponding lightweight strategy of the decoding model, the EEG data to be decoded can be compressed based on the compression strategy to obtain the target compressed data, and the target compressed data can be intentionally decoded based on the target decoding model selected by the lightweight strategy of the decoding model to obtain the intention decoding result.

[0088] The target compressed data here refers to the compressed data ultimately used for intended decoding, and the target decoding model here refers to the decoding model ultimately used for intended decoding.

[0089] The intention decoding results here may include upper limb movement intentions (eg, "left hand" and "right hand") and lower limb movement intentions (eg, "left foot" and "right foot").

[0090] The method provided by the embodiment of the present invention obtains EEG data to be decoded, performs channel compression on the EEG data to be decoded to obtain compressed EEG data, and then selects a strategy for the compressed EEG data based on a strategy model to obtain a compression strategy for the EEG data to be decoded and a corresponding lightweight strategy for the decoding model. Finally, the EEG data to be decoded is compressed based on the compression strategy to obtain target compressed data, and the target compressed data is intent-decoded based on the target decoding model selected by the lightweight strategy for the decoding model to obtain an intent decoding result. Thus, by compressing the EEG data to be decoded and lightweighting the decoding model, the amount of computation required for EEG decoding is reduced, the decoding efficiency of the decoding model is improved, and deployment on many resource-limited terminals (such as rehabilitation robots) is possible.

[0091] Based on the above embodiment, the training step of the target decoding model includes two parts: a candidate target decoding model pre-training step and a strategy model and a target decoding model collaborative training step; the collaborative training step includes a forward propagation step and a gradient back propagation step;

[0092] The candidate target decoding model pre-training step includes: pre-training the candidate target decoding model corresponding to each compression strategy based on the sample compressed EEG data under each compression strategy and the intention label of the sample compressed EEG data;

[0093] The forward propagation step includes: selecting a strategy for the EEG sample data to be decoded based on the strategy model to obtain a candidate compression strategy and a corresponding decoding model lightweight strategy; compressing the EEG sample data to be decoded based on the candidate compression strategy to obtain candidate sample compressed data, and performing intent decoding on the candidate sample compressed data based on the candidate target decoding model selected by the decoding model lightweight strategy to obtain an intent decoding prediction result;

[0094] The gradient back propagation step includes: based on the intention decoding prediction result and the classification label of the EEG sample data to be decoded, as well as the number of floating-point operations on the candidate sample compressed data input into the candidate target decoding model, performing parameter iteration on the strategy model and the candidate target decoding model through gradient back propagation, and obtaining the target decoding model based on the candidate target decoding model after parameter iteration.

[0095] Specifically, in order to better obtain the target decoding model, the training steps of the target decoding model can include two parts: the candidate target decoding model pre-training step and the strategy model and target decoding model collaborative training step. The collaborative training step here includes the forward propagation step and the gradient backpropagation step.

[0096] The candidate target decoding model pre-training steps include: pre-collecting sample compressed EEG data and intention labels of the sample compressed EEG data under each compression strategy, and pre-building the initial candidate target decoding model corresponding to each compression strategy.

[0097] After obtaining the candidate target decoding models corresponding to each compression strategy, the pre-collected sample compressed EEG data and the intent labels of the sample compressed EEG data under each compression strategy can be used to train the initial candidate target decoding models:

[0098] First, the sample compressed EEG data under each compression strategy is input into the initial candidate target decoding model, and the initial candidate target decoding model obtains and outputs the predicted intention of the sample compressed EEG data under each compression strategy.

[0099] After obtaining the predicted intention based on the initial candidate target decoding model, the predicted intention can be compared with the intention label of the pre-collected sample compressed EEG data, and the first loss function value is calculated according to the degree of difference between the two. The parameters of the initial candidate target decoding model are iterated based on the first loss function value, and the initial candidate target decoding model after the parameter iteration is completed is recorded as the candidate target decoding model.

[0100] It can be understood that the greater the difference between the predicted intention and the intention label of the pre-collected sample compressed EEG data, the greater the value of the first loss function; the smaller the difference between the predicted intention and the intention label of the pre-collected sample compressed EEG data, the smaller the value of the first loss function.

[0101] After pre-training the candidate target decoding models corresponding to each compression strategy, the strategy model and the target decoding model can be collaboratively trained.

[0102] The EEG sample data to be decoded and the classification labels of the EEG sample data to be decoded may be collected in advance, and an initial strategy decoding model including an initial strategy model and a candidate target decoding model may be constructed in advance.

[0103] After obtaining the initial strategy decoding model including the initial strategy model and the candidate target decoding model, the initial strategy decoding model can be trained by applying the pre-collected EEG sample data to be decoded and the classification labels of the EEG sample data to be decoded:

[0104] First, the EEG sample data to be decoded is input into the initial strategy decoding model, and the initial strategy model in the initial strategy decoding model performs strategy selection on the EEG sample data to be decoded, and obtains candidate compression strategies and corresponding decoding model lightweight strategies. Then, the EEG sample data to be decoded can be compressed based on the candidate compression strategies to obtain candidate sample compressed data, and the candidate sample compressed data can be intentionally decoded based on the candidate target decoding model selected by the decoding model lightweight strategy to obtain intention decoding prediction results.

[0105] After obtaining the intention decoding prediction result based on the initial strategy decoding model, the intention decoding prediction result can be compared with the classification label of the pre-collected EEG sample data to be decoded, as well as the number of floating-point operations on the candidate sample compressed data input into the candidate target decoding model. Through gradient backpropagation, the parameters of the strategy model and the candidate target decoding model are iterated, and the target decoding model is obtained based on the candidate target decoding model after parameter iteration.

[0106] The initial policy decoding model after parameter iteration has the same structure as the initial policy decoding model. Therefore, the initial policy decoding model can be divided into two parts: the initial policy model after parameter iteration and the candidate target decoding model after parameter iteration. The candidate target decoding model after parameter iteration can be directly used as the target decoding model.

[0107] Based on the above embodiment, the gradient back propagation step includes:

[0108] Step 210: determining a decoding accuracy loss function based on the intention decoding prediction result and the classification label;

[0109] Step 220, determining a decoding efficiency loss function based on averaging the number of floating-point operations of the candidate sample compressed data input into the candidate target decoding model;

[0110] Step 230, based on the decoding accuracy loss function and the decoding efficiency loss function, perform parameter iteration on the strategy model and the candidate target decoding model through gradient back propagation, and obtain the target decoding model based on the candidate target decoding model after parameter iteration.

[0111] Specifically, after obtaining the intent decoding prediction result and the classification label, the decoding accuracy loss function can be determined based on the intent decoding prediction result and the classification label. The formula of the decoding accuracy loss function is as follows:

[0112]

[0113] Among them, (X, y act ) represents the original input EEG data sample and its actual label; represents the FasterEEG model, and Θ represents the model parameters.

[0114] For different inputs, the number of channels in the classification model's input data varies, as does the target decoding model size, based on the policy model's behavioral decisions. Consequently, the computational effort (GFLOPs) required to complete the EEG decoding also varies. In this embodiment of the present invention, the GFLOPs of the training data are averaged to form the final decoding loss function.

[0115]

[0116] in, It represents the GFLOPs table corresponding to different pre-built data compression strategies and model lightweight strategies. When calculating the decoding loss, the actual decision of the strategy model (p chl , p mdl ), by looking up the table, we can get L eff .

[0117] The final loss function constructed is the weighted sum of decoding accuracy loss and decoding efficiency loss:

[0118] L=L acc +αL eff

[0119] Here, α represents a weight coefficient, which is used to balance decoding efficiency and decoding accuracy. That is, the larger α is, the higher the proportion of decoding efficiency loss is.

[0120] After obtaining the final constructed loss function, the policy model and the candidate target decoding model are iterated through gradient backpropagation, and the target decoding model is obtained based on the candidate target decoding model after parameter iteration.

[0121] It can be understood that the greater the difference between the intent decoding prediction result and the classification label, the greater the decoding accuracy loss function; the smaller the difference between the intent decoding prediction result and the classification label, the smaller the decoding accuracy loss function.

[0122] In addition, taking the data source BCI competition IV dataset 2a (BCIC IV 2a) as an example, the channel conversion rule of the present invention is used to perform channel compression. Nine subjects are tested with different values ​​of the weight coefficient α to test the performance of the target decoding model. The results are shown in Table 1:

[0123] Table 1. Comparison of target decoding model performance when weight coefficient α takes different values

[0124]

[0125] In Table 1, L, M, and S represent the sizes of candidate decoding models, and 22, 12, and 6 represent the number of channels after data compression.

[0126] Figure 4 This is a diagram of the EEG signal decoding efficiency and decoding accuracy provided by the present invention, such as Figure 4 As shown, Figure 4 Corresponding to Table 1.

[0127] Compared with SOTA models: Compared with EEGNet, DeepConvNet, and ShallowConvNet, FasterEEG (the method proposed in this paper) has the highest EEG decoding accuracy and efficiency.

[0128] Compared with ShallowConvNet, FasterEEG (the method proposed in this invention) can improve the EEG classification accuracy by 1.89%, and the amount of computation required is only 28% of that of ShallowConvNet.

[0129] Based on the above embodiment, the policy model includes a policy feature extraction model and a policy decision model connected to the policy feature extraction model;

[0130] Step 130 includes:

[0131] Step 131: input the compressed EEG data into the strategy feature extraction model, and extract the strategy features in the compressed EEG data by the strategy feature extraction model;

[0132] Step 132: input the policy features into the policy decision model, and the policy decision model outputs prediction probabilities of multiple data compression policies and prediction probabilities of multiple decoding model policies;

[0133] Step 133 : Determine the compression strategy of the to-be-decoded EEG data and the corresponding lightweight decoding model strategy based on the prediction probabilities of the multiple data compression strategies and the prediction probabilities of the multiple decoding model strategies.

[0134] Specifically, unlike using fixed-size input data and target decoding models for EEG classification, the present invention learns a strategy to minimize the use of input data and target decoding models while ensuring correct classification. Specifically, each experiment involves selecting from a total of L = K × M different GFLOPs of computational cost, where K is the candidate compressed data format and M is the candidate target decoding model size. The output of the strategy model determines the degree of data compression and model lightweighting.

[0135] The policy model here can include a policy feature extraction model and strategy feature extraction models The connected strategy decision model ζ(·,θζ), where θ* represents the learnable parameters in the corresponding module. Specifically, the compressed EEG data X with the maximum compression degree K , fed into the strategy feature extraction model Extracting strategic features.

[0136] Then, the policy features can be input into the policy decision model, and the policy decision model outputs the predicted probabilities of multiple data compression strategies and the predicted probabilities of multiple decoding model strategies.

[0137] Finally, based on the prediction probabilities of multiple data compression strategies and multiple decoding model strategies, the compression strategy of the EEG data to be decoded and the corresponding lightweight decoding model strategy can be determined.

[0138] The output F of the policy decision model ζ(·,θζ) pol_fea The formula is as follows:

[0139]

[0140] Among them, F pol_fea The output is the probability of different data compression and model lightweight strategies being selected.

[0141] Finally, the compression strategy of the EEG data to be decoded and the corresponding lightweight decoding model strategy can be determined based on the maximum value among the predicted probabilities of multiple data compression strategies and the maximum value among the predicted probabilities of multiple decoding model strategies.

[0142] That is, which strategy corresponds to the highest output probability, and which candidate data compression and model lightweight strategy is selected.

[0143] like Figure 3 As shown in (b), the output of the policy model is mainly p chl and p mdl , where p chl represents the compression strategy of the EEG data to be decoded, p mdl Indicates the corresponding decoding model lightweight strategy, where p in the FasterEEG framework chl =1, such as Figure 3 (a2) shows the FasterEEG framework, as shown in Figure 3 (c) shows that in the FasterEEG framework, p chl =1&p mdl =1.

[0144] Based on the above embodiments, Figure 5 is a schematic diagram of the preset channel conversion rules provided by the present invention, such as Figure 5 As shown, step 120 includes:

[0145] Based on a preset channel conversion rule, channel compression is performed on the EEG data to be decoded to obtain compressed EEG data; the preset channel conversion rule specifically includes a difference rule, a mean rule, and a selective activation rule;

[0146] The difference rule achieves channel data compression by calculating the difference between the EEG data collected by the corresponding electrodes on both sides of the brain area;

[0147] The mean rule achieves channel data compression by averaging the EEG data of adjacent channels;

[0148] The selective activation rule achieves channel data compression by removing channels corresponding to EEG data that are irrelevant to the task to be classified.

[0149] Specifically, channel compression can be performed on the EEG data to be decoded based on preset channel conversion rules to obtain compressed EEG data. The preset channel conversion rules here specifically include a difference rule, a mean rule, and a selective activation rule.

[0150] Brain activity, especially brain signals stimulated by movement or motor imagery, generally shows an increase in the amplitude of EEG signals in the ipsilateral brain area of ​​the moving (motor imagery) limb and an inhibition of EEG signals in the contralateral brain area.

[0151] Therefore, channel compression can be achieved by calculating the difference between the EEG signals collected by the corresponding electrodes on both sides of the brain, that is, the difference rule; and considering the similarity and redundancy of adjacent EEG information, channel compression can be achieved based on the averaging principle of adjacent channel signals, that is, the mean rule; at the same time, the EEG signals of some brain areas may be irrelevant to the task to be classified. At this time, the channel data can be directly removed, so the selective activation rule is used to achieve channel compression.

[0152] The following describes the EEG signal decoding device provided by the present invention. The EEG signal decoding device described below and the EEG signal decoding method described above can be referenced to each other.

[0153] Based on any of the above embodiments, the present invention provides an EEG signal decoding device, Figure 6 Schematic diagram of the structure of the EEG signal decoding device provided by the present invention. Figure 6 As shown, the device includes:

[0154] An acquisition unit 610 is used to acquire EEG data to be decoded;

[0155] a channel compression unit 620, configured to perform channel compression on the EEG data to be decoded to obtain compressed EEG data;

[0156] A strategy selection unit 630 is configured to select a strategy for the compressed EEG data based on a strategy model, and obtain a compression strategy for the EEG data to be decoded and a corresponding lightweight decoding model strategy;

[0157] The intention decoding unit 640 is used to compress the EEG data to be decoded based on the compression strategy to obtain target compressed data, and to perform intention decoding on the target compressed data based on the target decoding model selected by the decoding model lightweight strategy to obtain an intention decoding result.

[0158] The device provided by the embodiment of the present invention obtains the EEG data to be decoded, performs channel compression on the EEG data to be decoded to obtain compressed EEG data, and then selects a strategy for the compressed EEG data based on the strategy model to obtain the compression strategy of the EEG data to be decoded and the corresponding lightweight strategy of the decoding model. Finally, the EEG data to be decoded is compressed based on the compression strategy to obtain target compressed data, and the target compressed data is intent-decoded based on the target decoding model selected by the lightweight strategy of the decoding model to obtain the intent decoding result. Therefore, by compressing the EEG data to be decoded and lightweighting the decoding model, the computational amount required for EEG decoding is reduced, the decoding efficiency of the decoding model is improved, and the deployment on many resource-limited terminals (such as rehabilitation robots) is possible.

[0159] Based on any of the above embodiments, the target decoding model includes a time domain convolution layer, a spatial convolution layer, a pooling layer and a classification layer connected in sequence, and the target decoding model is lightweight by changing the number of convolution kernels of the time domain convolution layer and the spatial convolution layer.

[0160] Based on any of the above embodiments, the training step of the target decoding model includes two parts: a pre-training step of a candidate target decoding model and a collaborative training step of a strategy model and the target decoding model; the collaborative training step includes a forward propagation step and a gradient back propagation step;

[0161] The candidate target decoding model pre-training step includes: pre-training the candidate target decoding model corresponding to each compression strategy based on the sample compressed EEG data under each compression strategy and the intention label of the sample compressed EEG data;

[0162] The forward propagation step includes: selecting a strategy for the EEG sample data to be decoded based on the strategy model to obtain a candidate compression strategy and a corresponding decoding model lightweight strategy; compressing the EEG sample data to be decoded based on the candidate compression strategy to obtain candidate sample compressed data, and performing intent decoding on the candidate sample compressed data based on the candidate target decoding model selected by the decoding model lightweight strategy to obtain an intent decoding prediction result;

[0163] The gradient back propagation step includes: based on the intention decoding prediction result and the classification label of the EEG sample data to be decoded, as well as the number of floating-point operations on the candidate sample compressed data input into the candidate target decoding model, performing parameter iteration on the strategy model and the candidate target decoding model through gradient back propagation, and obtaining the target decoding model based on the candidate target decoding model after parameter iteration.

[0164] Based on any of the above embodiments, the intention decoding prediction result and the classification label of the EEG sample data to be decoded, as well as the number of floating-point operations on the candidate sample compressed data input into the candidate target decoding model, perform parameter iteration on the strategy model and the candidate target decoding model through gradient backpropagation, and obtain the target decoding model based on the candidate target decoding model after parameter iteration, specifically for:

[0165] Determining a decoding accuracy loss function based on the intent decoding prediction result and the classification label;

[0166] Determining a decoding efficiency loss function based on averaging the number of floating-point operations of the candidate sample compressed data input into the candidate target decoding model;

[0167] Based on the decoding accuracy loss function and the decoding efficiency loss function, the policy model and the candidate target decoding model are iterated by gradient back propagation, and the target decoding model is obtained based on the candidate target decoding model after parameter iteration.

[0168] Based on any of the above embodiments, the policy model includes a policy feature extraction model and a policy decision model connected to the policy feature extraction model;

[0169] The strategy model is based on which the strategy is selected for the compressed EEG data to obtain the compression strategy of the EEG data to be decoded and the corresponding lightweight strategy of the decoding model, including:

[0170] a feature extraction unit, configured to input the compressed EEG data into the strategy feature extraction model, and extract strategy features from the compressed EEG data by the strategy feature extraction model;

[0171] A decision unit, configured to input the policy features into the policy decision model, and output the predicted probabilities of a plurality of data compression policies and a plurality of decoding model policies from the policy decision model;

[0172] A strategy determination unit is used to determine the compression strategy of the to-be-decoded EEG data and the corresponding lightweight decoding model strategy based on the prediction probabilities of the multiple data compression strategies and the prediction probabilities of the multiple decoding model strategies.

[0173] Based on any of the above embodiments, the determination strategy unit is specifically configured to:

[0174] Based on the maximum value among the predicted probabilities of the multiple data compression strategies and the maximum value among the predicted probabilities of the multiple decoding model strategies, the compression strategy of the to-be-decoded EEG data and the corresponding lightweight decoding model strategy are determined.

[0175] Based on any of the above embodiments, the channel compression unit 620 is specifically configured to:

[0176] Based on a preset channel conversion rule, channel compression is performed on the EEG data to be decoded to obtain compressed EEG data; the preset channel conversion rule specifically includes a difference rule, a mean rule, and a selective activation rule;

[0177] The difference rule achieves channel data compression by calculating the difference between the EEG data collected by the corresponding electrodes on both sides of the brain area;

[0178] The mean rule achieves channel data compression by averaging the EEG data of adjacent channels;

[0179] The selective activation rule achieves channel data compression by removing channels corresponding to EEG data that are irrelevant to the task to be classified.

[0180] Figure 7 An example of a physical structure diagram of an electronic device is shown below. Figure 7 As shown, the electronic device may include: a processor 710, a communication interface 720, a memory 730 and a communication bus 740, wherein the processor 710, the communication interface 720 and the memory 730 communicate with each other via the communication bus 740. The processor 710 may call the logic instructions in the memory 730 to execute the EEG signal decoding method, which includes: obtaining EEG data to be decoded; performing channel compression on the EEG data to be decoded to obtain compressed EEG data; performing strategy selection on the compressed EEG data based on a strategy model to obtain a compression strategy for the EEG data to be decoded and a corresponding lightweight decoding model strategy; compressing the EEG data to be decoded based on the compression strategy to obtain target compressed data, and performing intent decoding on the target compressed data based on the target decoding model selected by the lightweight decoding model strategy to obtain an intent decoding result.

[0181] In addition, the logic instructions in the above-mentioned memory 730 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when sold or used as an independent product. 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 and includes several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0182] On the other hand, the present invention also provides a computer program product, which includes a computer program, which can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the EEG signal decoding method provided by the above methods, which includes: obtaining EEG data to be decoded; performing channel compression on the EEG data to be decoded to obtain compressed EEG data; based on a strategy model, performing strategy selection on the compressed EEG data to obtain a compression strategy for the EEG data to be decoded and a corresponding lightweight decoding model strategy; compressing the EEG data to be decoded based on the compression strategy to obtain target compressed data, and performing intent decoding on the target compressed data based on the target decoding model selected by the lightweight decoding model strategy to obtain an intent decoding result.

[0183] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to execute the EEG signal decoding method provided by the above-mentioned methods, the method comprising: obtaining EEG data to be decoded; performing channel compression on the EEG data to be decoded to obtain compressed EEG data; performing strategy selection on the compressed EEG data based on a strategy model to obtain a compression strategy for the EEG data to be decoded and a corresponding lightweight decoding model strategy; compressing the EEG data to be decoded based on the compression strategy to obtain target compressed data, and performing intent decoding on the target compressed data based on the target decoding model selected by the lightweight decoding model strategy to obtain an intent decoding result.

[0184] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, i.e., they may be located in one location or distributed across multiple network units. Some or all of the modules may be selected based on actual needs to achieve the objectives of the present embodiment. Persons of ordinary skill in the art will be able to understand and implement the present invention without inventive effort.

[0185] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, or of course, by hardware. Based on this understanding, the essence of the above technical solution or the part that contributes to the existing technology can be embodied in the form of a software product. The computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, an optical disk, etc., and includes a number of instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or certain parts of the embodiments.

[0186] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A method for decoding an electroencephalogram signal, characterized in that: include: Obtaining EEG data to be decoded; Performing channel compression on the EEG data to be decoded to obtain compressed EEG data; Based on the strategy model, a strategy is selected for the compressed EEG data to obtain a compression strategy for the EEG data to be decoded and a corresponding lightweight decoding model strategy; Compressing the to-be-decoded EEG data based on the compression strategy to obtain target compressed data, and performing intent decoding on the target compressed data based on the target decoding model selected by the decoding model lightweight strategy to obtain an intent decoding result; The target decoding model training step includes two parts: a candidate target decoding model pre-training step and a strategy model and the target decoding model collaborative training step; The collaborative training step includes a forward propagation step and a gradient back propagation step; The candidate target decoding model pre-training step includes: pre-training the candidate target decoding model corresponding to each compression strategy based on the sample compressed EEG data under each compression strategy and the intention label of the sample compressed EEG data; The forward propagation step includes: selecting a strategy for the EEG sample data to be decoded based on the strategy model to obtain a candidate compression strategy and a corresponding decoding model lightweight strategy; compressing the EEG sample data to be decoded based on the candidate compression strategy to obtain candidate sample compressed data, and performing intent decoding on the candidate sample compressed data based on the candidate target decoding model selected by the decoding model lightweight strategy to obtain an intent decoding prediction result; The gradient back propagation step includes: based on the intention decoding prediction result and the classification label of the EEG sample data to be decoded, as well as the number of floating-point operations on the candidate sample compressed data input into the candidate target decoding model, performing parameter iteration on the strategy model and the candidate target decoding model through gradient back propagation, and obtaining the target decoding model based on the candidate target decoding model after parameter iteration.

2. The method for decoding EEG signals according to claim 1, wherein: The target decoding model includes a time domain convolution layer, a spatial convolution layer, a pooling layer and a classification layer connected in sequence. The target decoding model is lightweight by changing the number of convolution kernels of the time domain convolution layer and the spatial convolution layer.

3. The EEG signal decoding method according to claim 1, wherein: The method comprises: performing parameter iteration on the policy model and the candidate target decoding model based on the intention decoding prediction result and the classification label of the EEG sample data to be decoded, and the number of floating-point operations on the candidate sample compressed data input into the candidate target decoding model through gradient back propagation, and obtaining the target decoding model based on the candidate target decoding model after parameter iteration, including: Determining a decoding accuracy loss function based on the intent decoding prediction result and the classification label; Determining a decoding efficiency loss function based on averaging the number of floating-point operations of the candidate sample compressed data input into the candidate target decoding model; Based on the decoding accuracy loss function and the decoding efficiency loss function, the policy model and the candidate target decoding model are iterated by gradient back propagation, and the target decoding model is obtained based on the candidate target decoding model after parameter iteration.

4. The method for decoding EEG signals according to claim 1, wherein: The policy model includes a policy feature extraction model and a policy decision model connected to the policy feature extraction model; The strategy model is based on which the strategy is selected for the compressed EEG data to obtain the compression strategy of the EEG data to be decoded and the corresponding lightweight strategy of the decoding model, including: Inputting the compressed EEG data into the strategy feature extraction model, and extracting the strategy features in the compressed EEG data by the strategy feature extraction model; Inputting the policy features into the policy decision model, and having the policy decision model output prediction probabilities of multiple data compression policies and prediction probabilities of multiple decoding model policies; Based on the predicted probabilities of the multiple data compression strategies and the predicted probabilities of the multiple decoding model strategies, a compression strategy for the to-be-decoded EEG data and a corresponding lightweight decoding model strategy are determined.

5. The method for decoding EEG signals according to claim 4, wherein: The step of determining the compression strategy for the to-be-decoded EEG data and the corresponding lightweight decoding model strategy based on the predicted probabilities of the multiple data compression strategies and the predicted probabilities of the multiple decoding model strategies includes: Based on the maximum value among the predicted probabilities of the multiple data compression strategies and the maximum value among the predicted probabilities of the multiple decoding model strategies, the compression strategy of the to-be-decoded EEG data and the corresponding lightweight decoding model strategy are determined.

6. The method for decoding an electroencephalogram signal according to any one of claims 1 to 5, characterized in that: The step of performing channel compression on the to-be-decoded EEG data to obtain compressed EEG data includes: Based on a preset channel conversion rule, channel compression is performed on the EEG data to be decoded to obtain compressed EEG data; the preset channel conversion rule specifically includes a difference rule, a mean rule, and a selective activation rule; The difference rule achieves channel data compression by calculating the difference between the EEG data collected by the corresponding electrodes on both sides of the brain area; The mean rule achieves channel data compression by averaging the EEG data of adjacent channels; The selective activation rule achieves channel data compression by removing channels corresponding to EEG data that are irrelevant to the task to be classified.

7. An electroencephalogram signal decoding device, characterized in that: include: An acquisition unit, used for acquiring EEG data to be decoded; a channel compression unit, configured to perform channel compression on the EEG data to be decoded to obtain compressed EEG data; A strategy selection unit, configured to select a strategy for the compressed EEG data based on a strategy model, to obtain a compression strategy for the EEG data to be decoded and a corresponding lightweight decoding model strategy; an intention decoding unit, configured to compress the to-be-decoded EEG data based on the compression strategy to obtain target compressed data, and perform intention decoding on the target compressed data based on the target decoding model selected by the decoding model lightweight strategy to obtain an intention decoding result; It also includes a training unit, the training unit includes a candidate target decoding model pre-training unit, a strategy model and a target decoding model collaborative training unit; the collaborative training unit includes a forward propagation unit and a gradient back propagation unit; The candidate target decoding model pre-training unit is specifically used to: pre-train the candidate target decoding model corresponding to each compression strategy based on the sample compressed EEG data under each compression strategy and the intention label of the sample compressed EEG data; The forward propagation unit is specifically configured to: select a strategy for the EEG sample data to be decoded based on the strategy model to obtain a candidate compression strategy and a corresponding decoding model lightweight strategy; compress the EEG sample data to be decoded based on the candidate compression strategy to obtain candidate sample compressed data; and perform intent decoding on the candidate sample compressed data based on the candidate target decoding model selected by the decoding model lightweight strategy to obtain an intent decoding prediction result; The gradient back propagation unit is specifically used to: based on the intention decoding prediction result and the classification label of the EEG sample data to be decoded, as well as the number of floating-point operations on the candidate sample compressed data input into the candidate target decoding model, perform parameter iteration on the strategy model and the candidate target decoding model through gradient back propagation, and obtain the target decoding model based on the candidate target decoding model after parameter iteration.

8. 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 program, the electroencephalogram signal decoding method according to any one of claims 1 to 6 is implemented.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the EEG signal decoding method according to any one of claims 1 to 6 is implemented.

Images