A method for classifying electroencephalogram signals based on laurent polynomials

Abstract

This invention relates to a method for classifying electroencephalogram (EEG) signals based on Laurent polynomials. First, feature values ​​corresponding to preset target features are extracted from each segment of the sample EEG signal. Then, a scaling and translation fine-tuning layer, a Laurent polynomial layer, an activation layer, a fully connected layer, a softmax layer, and an output layer are sequentially connected from the input to the output to construct a training network. Finally, the target feature values ​​corresponding to the sample EEG signal are used as input, and the category corresponding to the sample EEG signal is used as output to train the network, obtaining an EEG signal classification model for classifying EEG signals. The design combines Laurent polynomial expansion with a neural network architecture, utilizing its unique negative power term and reference point mechanism to effectively capture complex nonlinear patterns in EEG signals. This method can be widely applied in brain-computer interfaces, health monitoring, and other medical and health fields, providing reliable technical support for real-time monitoring and early warning.

Description

Technical Field

[0001] This invention relates to a brainwave signal classification method based on Laurent polynomials, belonging to the field of signal classification technology. Background Technology

[0002] Electroencephalogram (EEG) signals are electrical potential records generated on the scalp surface by the electrical activity of brain neurons, containing rich information about neurophysiological and psychological states. With the rapid development of sensing technology, computing power, and artificial intelligence algorithms, the precise analysis, identification, and classification of EEG signals have become a cutting-edge research area in brain science and clinical applications. The core objective is to transform raw, complex, and noisy EEG signals into reliable instructions or diagnostic indicators that can characterize specific brain states, user intentions, or pathological features, ultimately achieving widespread application in the fields of brain-computer interfaces and health monitoring.

[0003] Electroencephalography (EEG) signals are one of the gold standards for assessing brain function and diagnosing neurological diseases. In the field of health monitoring, its applications include epilepsy monitoring and early warning, sleep quality analysis and sleep staging, mental state and cognitive load assessment, and early screening for neurological diseases. Despite its promising prospects, EEG analysis and classification technology has the following limitations in health monitoring, especially in long-term, dynamic monitoring scenarios in families and communities:

[0004] Reliance on expert experience: Traditional EEG analysis relies heavily on visual interpretation by clinical neurologists and experienced technicians. The process is time-consuming, highly subjective, and cannot achieve large-scale screening and real-time monitoring.

[0005] Existing automated analysis algorithms suffer from poor adaptability: Many algorithms perform well in controlled laboratory environments but struggle to cope with complex noise introduced by motion artifacts, differences in device wearing, and environmental interference in real-world monitoring environments. Insufficient robustness of the models leads to high false alarm and false negative rates.

[0006] Lack of personalization and context awareness: Most monitoring systems use fixed classification thresholds and models, which cannot adapt to changes in the physiological baseline of individual users and different daily activity scenarios, reducing the accuracy of monitoring results and their clinical reference value.

[0007] Therefore, existing EEG signal analysis, recognition, and classification methods, whether in brain-computer interfaces or health monitoring applications, generally suffer from insufficient recognition accuracy, robustness, and adaptability due to factors such as complex signal characteristics, large individual differences, and environmental noise interference. Summary of the Invention

[0008] The technical problem to be solved by the present invention is to provide a method for classifying EEG signals based on Laurent polynomials. By mapping the time-frequency features in EEG signals to Laurent polynomial expansion terms, defining the coefficients of negative power terms according to the physical characteristics of the signal at pathological singularities, and using the analytical properties of complex functions to construct an interpretable diagnostic model, EEG signal classification can be achieved.

[0009] To solve the above-mentioned technical problems, the present invention adopts the following technical solution: The present invention designs an EEG signal classification method based on Laurent polynomials, comprising the following steps:

[0010] Step A. For each segment of EEG signal corresponding to a preset category, extract the feature values ​​of each preset target feature corresponding to each segment of EEG signal, and normalize and update each target feature value in each segment of EEG signal, and then proceed to Step B;

[0011] Step B. Concatenate the scaling and translation fine-tuning layer, Laurent polynomial layer, activation layer, fully connected layer, softmax layer, and output layer in sequence from the input to the output to construct the network to be trained, and then proceed to step C;

[0012] Step C. Based on the EEG signals of each sample, with the target feature values ​​corresponding to the sample EEG signals as input and the category corresponding to the sample EEG signals as output, train the network to be trained to obtain an EEG signal classification model, which is used to classify EEG signals.

[0013] As a preferred embodiment of the present invention: the scaling and translation fine-tuning layer receives the target feature values ​​corresponding to the sample EEG signals according to the following formula:

[0014] ;

[0015] The fine-tuning values ​​of each target feature corresponding to the sample EEG signal are obtained and output to the Laurent polynomial layer, where, , Indicates the number of EEG signals in the sample. , Indicates the number of target features. Indicates the first The corresponding EEG signal of the sample One target feature value, Indicates the first The scaling parameters corresponding to each target feature Indicates the first The translation parameters corresponding to each target feature Indicates the first The corresponding EEG signal of the sample Each target feature is fine-tuned.

[0016] As a preferred embodiment of the present invention: the fine-tuning values ​​of each target feature corresponding to the received sample EEG signal in the Laurent polynomial layer are calculated according to the following formula:

[0017] ;

[0018] Obtain the eigenvector corresponding to the sample EEG signal after processing by Laurent polynomial layer. , The number of eigenvalues ​​is And output to the activation layer, where, Indicates the first Each sample EEG signal corresponds to an eigenvector after processing with Laurent polynomial layers. This indicates the order of the negative power expansion. Indicates the order of a positive power expansion. Represents the positive power term in the Laurent expansion. order coefficient tensor, Represents the negative power term in the Laurent expansion. order coefficient tensor, and All dimensions are , This represents the dimension of the feature vector corresponding to the sample EEG signal after processing by the Laurent polynomial layer. , This indicates that the corresponding dimension in the Laurent expansion is The reference point vector, Represents the tensor product. Represents the tensor inner product. ,in, Represents the negative power term in the Laurent expansion. The order coefficient tensors correspond to respectively Data from various dimensions This indicates that the corresponding dimension in the Laurent expansion is The reference point vector of the first One target feature value.

[0019] As a preferred embodiment of the present invention: the activation layer is a non-linear activation layer, and the non-linear activation layer receives the feature vector corresponding to the sample EEG signal after processing by the Laurent polynomial layer, according to the following formula:

[0020] ;

[0021] ;

[0022] Obtain the feature vector corresponding to the sample EEG signal after activation layer processing. And output to the fully connected layer, where, , Indicates the first The EEG signals of the sample correspond to the first sample after processing by Laurent polynomial layer. 1 eigenvalue, Indicates the first Each sample EEG signal corresponds to a feature value vector after processing by the activation layer. Indicates the first The EEG signal of the sample corresponds to the first sample after processing by the activation layer. Each feature value.

[0023] As a preferred embodiment of the present invention: the fully connected layer receives the feature vector corresponding to the sample EEG signal after processing by the activation layer, as follows:

[0024] ;

[0025] The analysis results of the sample EEG signals corresponding to each category were obtained. And output to the softmax layer, where, , Indicates the number of categories, Indicates the first The weights corresponding to each category Indicates the first The bias corresponding to each category Indicates the first The EEG signal of the sample corresponds to the first The analysis results values ​​for each category.

[0026] As a preferred embodiment of the present invention: the softmax layer receives the analysis result values ​​of the sample EEG signals corresponding to each category, according to the following formula:

[0027] ;

[0028] The probabilities of sample EEG signals corresponding to different categories are obtained and output to the output layer. Indicates the first The EEG signal of the sample corresponds to the first The probability of each category, Indicates the first The EEG signal of the sample corresponds to the first The analysis results of each category are used to perform an exponential calculation.

[0029] As a preferred technical solution of the present invention: the output layer receives the probability of the sample EEG signal corresponding to each category, selects the category corresponding to the highest probability, and constitutes the category corresponding to the sample EEG signal, that is, the output of the network to be trained.

[0030] As a preferred embodiment of the present invention: in step C, during the training process of the network to be trained, the training is completed when the cross-entropy loss function converges or reaches a preset maximum number of iterations, wherein the cross-entropy loss function is as follows:

[0031] ;

[0032] in, This represents the result of the loss function, where, Indicates the first The true category corresponding to each sample of EEG signal Indicates the first The EEG signal of the sample corresponds to the first The probability of each category, This represents the set of parameters to be trained in the network to be trained. Represents the regularization coefficient. Indicates the parameters to be trained. This represents the sum of squares of all parameters to be trained.

[0033] As a preferred embodiment of the present invention: In step C, during the training process of the network to be trained, based on a preset initial learning rate, the Adam optimizer combined with the BP algorithm is used to adjust the parameters according to the following steps. , Reference point vector Calculate the gradient and update it;

[0034] Step a. Based on the forward propagation of the Laurent polynomial layer, use the following formula:

[0035] ;

[0036] ;

[0037] Calculate the loss function result during the training process of the network to be trained. For parameters The gradient;

[0038] According to the following formula:

[0039] ;

[0040] ;

[0041] Calculate the loss function result during the training process of the network to be trained. For parameters The gradient;

[0042] According to the following formula:

[0043] ;

[0044] ;

[0045] Calculate the loss function result during the training process of the network to be trained. For reference point vector The gradient; then proceed to step b; where, , Indicates the number of EEG signals in the sample. Indicates the first Each sample EEG signal corresponds to an eigenvector after processing with Laurent polynomial layers. This indicates the order of the negative power expansion. Indicates the order of a positive power expansion. Represents the positive power term in the Laurent expansion. order coefficient tensor, Represents the negative power term in the Laurent expansion. order coefficient tensor, and All dimensions are , This represents the dimension of the feature vector corresponding to the sample EEG signal after processing by the Laurent polynomial layer. Indicates the number of target features. , Indicates the first The corresponding EEG signal of the sample Each target feature is fine-tuned. This indicates that the corresponding dimension in the Laurent expansion is The reference point vector;

[0046] Step b. Based on the loss function results during the training of the network to be trained. For the parameters respectively ,parameter Reference point vector The gradients are calculated by the Adam optimizer for each parameter. ,parameter Reference point vector Update.

[0047] As a preferred technical solution of the present invention: In step A, for the EEG signals of each sample corresponding to each preset category, the abnormal values ​​and noise in the EEG signals of each sample are first removed to achieve the update, and then the feature values ​​of each sample EEG signal corresponding to each preset target feature are extracted.

[0048] The electroencephalogram (EEG) signal classification method based on Laurent multinomials described in this invention has the following technical advantages compared with existing technologies:

[0049] This invention designs an EEG signal classification method based on Laurent polynomials. First, feature values ​​corresponding to preset target features are extracted from each segment of the sample EEG signal. Then, a scaling and translation fine-tuning layer, a Laurent polynomial layer, an activation layer, a fully connected layer, a softmax layer, and an output layer are sequentially connected from the input to the output to construct a training network. Finally, the target feature values ​​corresponding to the sample EEG signal are used as input, and the category corresponding to the sample EEG signal is used as output to train the network, obtaining an EEG signal classification model for classifying EEG signals. The design combines Laurent polynomial expansion with a neural network architecture, utilizing its unique negative power term and reference point mechanism to effectively capture complex nonlinear patterns in EEG signals. This method can be widely applied in brain-computer interfaces, health monitoring, and other medical and health fields, providing reliable technical support for real-time monitoring and early warning. Attached Figure Description

[0050] Figure 1 This is a real-time block diagram of the EEG signal classification method based on Laurent polynomials designed in this invention;

[0051] Figure 2 This is a graph showing the experimental results of the ten-fold cross-validation in the design of this invention. Detailed Implementation

[0052] The specific embodiments of the present invention will be further described in detail below with reference to the accompanying drawings.

[0053] This invention designs a brainwave signal classification method based on Laurent multinomials. In practical applications, such as... Figure 1 As shown, the specific design is carried out through steps A to C.

[0054] Step A. For each segment of EEG signal corresponding to a preset category, first remove outliers and noise from each segment of EEG signal to update. Then extract the feature values ​​of each preset target feature corresponding to each segment of EEG signal. Normalize and update each target feature value in each segment of EEG signal. Then proceed to Step B.

[0055] Step B. Construct the network to be trained by sequentially connecting the scaling and translation fine-tuning layer, Laurent polynomial layer, activation layer, fully connected layer, softmax layer, and output layer from the input end to the output end, and then proceed to step C.

[0056] In the network to be trained constructed above, the input of the scaling and translation fine-tuning layer constitutes the input of the network to be trained, and the output of the output layer constitutes the output of the network to be trained. The scaling and translation fine-tuning layer is used to receive the target feature values ​​corresponding to the sample EEG signals, according to the following formula:

[0057] ;

[0058] The fine-tuning values ​​of each target feature corresponding to the sample EEG signal are obtained and output to the Laurent polynomial layer, where, , Indicates the number of EEG signals in the sample. , Indicates the number of target features. Indicates the first The corresponding EEG signal of the sample One target feature value, Indicates the first The scaling parameters corresponding to each target feature Indicates the first The translation parameters corresponding to each target feature Indicates the first The corresponding EEG signal of the sample Each target feature is fine-tuned.

[0059] Laurent polynomial layers receive the fine-tuned values ​​of each target feature corresponding to the sample EEG signal, and apply them according to the following formula:

[0060] ;

[0061] Obtain the eigenvector corresponding to the sample EEG signal after processing by Laurent polynomial layer. , The number of eigenvalues ​​is And output to the activation layer, where, Indicates the first Each sample EEG signal corresponds to an eigenvector after processing with Laurent polynomial layers. This indicates the order of the negative power expansion. Indicates the order of a positive power expansion. Represents the positive power term in the Laurent expansion. order coefficient tensor, Represents the negative power term in the Laurent expansion. order coefficient tensor, and All dimensions are , This represents the dimension of the feature vector corresponding to the sample EEG signal after processing by the Laurent polynomial layer. , This indicates that the corresponding dimension in the Laurent expansion is The reference point vector, Represents the tensor product. Represents the tensor inner product. ,in, Represents the negative power term in the Laurent expansion. The order coefficient tensors correspond to respectively Data from various dimensions This indicates that the corresponding dimension in the Laurent expansion is The reference point vector of the first One target feature value.

[0062] The activation layer is a non-linear activation layer. It receives the feature vector corresponding to the sample EEG signal after processing by a Laurent polynomial layer, and then processes it according to the following formula:

[0063] ;

[0064] ;

[0065] Obtain the feature vector corresponding to the sample EEG signal after activation layer processing. And output to the fully connected layer, where, , Indicates the first The EEG signals of the sample correspond to the first sample after processing by Laurent polynomial layer. 1 eigenvalue, Indicates the first Each sample EEG signal corresponds to a feature value vector after processing by the activation layer. Indicates the first The EEG signal of the sample corresponds to the first sample after processing by the activation layer. Each feature value.

[0066] The fully connected layer receives the feature vector corresponding to the sample EEG signal after processing by the activation layer, and calculates it according to the following formula:

[0067] ;

[0068] The analysis results of the sample EEG signals corresponding to each category were obtained. And output to the softmax layer, where, , Indicates the number of categories, Indicates the first The weights corresponding to each category Indicates the first The bias corresponding to each category Indicates the first The EEG signal of the sample corresponds to the first The analysis results values ​​for each category.

[0069] The softmax layer receives the analysis results of the sample EEG signals corresponding to each category, and calculates them according to the following formula:

[0070] ;

[0071] The probabilities of sample EEG signals corresponding to different categories are obtained and output to the output layer. Indicates the first The EEG signal of the sample corresponds to the first The probability of each category, Indicates the first The EEG signal of the sample corresponds to the first The analysis results of each category are used to perform an exponential calculation.

[0072] The output layer receives the probability of the sample EEG signal corresponding to each category, selects the category corresponding to the highest probability, and forms the category corresponding to the sample EEG signal, which is the output of the network to be trained.

[0073] Based on the above network structure design, the EEG signal obtained by discrete wavelet decomposition is mapped to the expansion terms of Laurent polynomials; a reference point mechanism is introduced to focus on the feature changes in specific regions of the signal; the smooth change pattern of the signal is captured by positive power terms, and the behavior of the signal near singularities is effectively modeled by negative power terms; activation functions are used to enhance nonlinear expressive power; fully connected layers are applied to realize the dimensional mapping from features to classification targets; and the softmax function is used in the output layer to generate the classification probability distribution, thereby realizing the classification of EEG signals.

[0074] Step C. Based on the EEG signals of each sample, with the target feature values ​​corresponding to the sample EEG signals as input and the category corresponding to the sample EEG signals as output, train the network to be trained to obtain an EEG signal classification model, which is used to classify EEG signals.

[0075] In the actual training process of the network to be trained, based on a preset initial learning rate, the Adam optimizer combined with the backpropagation algorithm is used to adjust the parameters according to steps a to b as follows. , Reference point vector Calculate the gradient and update it.

[0076] Step a. Based on the forward propagation of the Laurent polynomial layer, use the following formula:

[0077] ;

[0078] ;

[0079] Calculate the loss function result during the training process of the network to be trained. For parameters The gradient;

[0080] According to the following formula:

[0081] ;

[0082] ;

[0083] Calculate the loss function result during the training process of the network to be trained. For parameters The gradient;

[0084] According to the following formula:

[0085] ;

[0086] ;

[0087] Calculate the loss function result during the training process of the network to be trained. For reference point vector The gradient; then proceed to step b; where, , Indicates the number of EEG signals in the sample. Indicates the first Each sample EEG signal corresponds to an eigenvector after processing with Laurent polynomial layers. This indicates the order of the negative power expansion. Indicates the order of a positive power expansion. Represents the positive power term in the Laurent expansion. order coefficient tensor, Represents the negative power term in the Laurent expansion. order coefficient tensor, and All dimensions are , This represents the dimension of the feature vector corresponding to the sample EEG signal after processing by the Laurent polynomial layer. Indicates the number of target features. , Indicates the first The corresponding EEG signal of the sample Each target feature is fine-tuned. This indicates that the corresponding dimension in the Laurent expansion is The reference point vector.

[0088] Step b. Based on the loss function results during the training of the network to be trained. For the parameters respectively ,parameter Reference point vector The gradients are calculated by the Adam optimizer for each parameter. ,parameter Reference point vector Update.

[0089] Regarding the training objective of the network to be trained, in practical applications, training is completed when the cross-entropy loss function converges or reaches a preset maximum number of iterations. The cross-entropy loss function is as follows:

[0090] ;

[0091] in, This represents the result of the loss function, where, Indicates the first The true category corresponding to each sample of EEG signal Indicates the first The EEG signal of the sample corresponds to the first The probability of each category, This represents the set of parameters to be trained in the network to be trained. Represents the regularization coefficient. Indicates the parameters to be trained. This represents the sum of squares of all parameters to be trained.

[0092] After obtaining the trained network, i.e., the EEG signal classification model, in practical applications, a ten-fold cross-validation method can be further designed. The final result is the average of 10 experiments. Performance indicators such as accuracy, sensitivity, and specificity are used to evaluate the EEG signal classification model. Figure 2 As shown in the figure, the red line represents the accuracy, reflecting the overall classification correctness of the model. The curve fluctuates little in multiple rounds of validation, indicating that the model is robust. The blue line represents the sensitivity, reflecting the model's ability to identify signals during epileptic seizures. This curve remains stable at a level close to 100%, indicating that the model can accurately capture the signal characteristics during epileptic seizures. The green line represents the specificity, representing the model's ability to distinguish signals during non-seizure periods. This curve remains above 90%, indicating that the model can effectively eliminate interference from normal EEG signals.

[0093] The aforementioned technical solution, based on Laurent polynomials for EEG signal classification, first extracts feature values ​​corresponding to preset target features for each segment of the sample EEG signal. Then, it sequentially strings together scaling and translation fine-tuning layers, Laurent polynomial layers, activation layers, fully connected layers, softmax layers, and output layers from the input to the output to construct a training network. Finally, it trains the network using the target feature values ​​corresponding to the sample EEG signal as input and the category corresponding to the sample EEG signal as output, obtaining an EEG signal classification model for classifying EEG signals. The design combines Laurent polynomial expansion with a neural network architecture, utilizing its unique negative power term and reference point mechanism to effectively capture complex nonlinear patterns in EEG signals. This method can be widely applied in brain-computer interfaces, health monitoring, and other medical and health fields, providing reliable technical support for real-time monitoring and early warning.

[0094] The embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the present invention is not limited to the above embodiments. Within the scope of knowledge possessed by those skilled in the art, various changes can be made without departing from the spirit of the present invention.

Claims

1. A brainwave signal classification method based on Laurent polynomials, characterized in that, Includes the following steps: Step A. For each segment of EEG signal corresponding to a preset category, extract the feature values ​​of each preset target feature corresponding to each segment of EEG signal, and normalize and update each target feature value in each segment of EEG signal, and then proceed to Step B; Step B. Concatenate the scaling and translation fine-tuning layer, Laurent polynomial layer, activation layer, fully connected layer, softmax layer, and output layer in sequence from the input to the output to construct the network to be trained, and then proceed to step C; Step C. Based on the EEG signals of each sample, take the target feature values ​​corresponding to the sample EEG signals as input and the category corresponding to the sample EEG signals as output, train the network to be trained to obtain an EEG signal classification model, which is used to classify EEG signals. In step B, the scaling and translation fine-tuning layer receives the target feature values ​​corresponding to the sample EEG signals, according to the following formula: ； The fine-tuning values ​​of each target feature corresponding to the sample EEG signal are obtained and output to the Laurent polynomial layer, where, , Indicates the number of EEG signals in the sample. , Indicates the number of target features. Indicates the first The corresponding EEG signal of the sample One target feature value, Indicates the first The scaling parameters corresponding to each target feature Indicates the first The translation parameters corresponding to each target feature Indicates the first The corresponding EEG signal of the sample Each target feature is fine-tuned; The fine-tuning values ​​of each target feature corresponding to the received EEG signal sample in the Laurent polynomial layer are calculated according to the following formula: ； Obtain the eigenvector corresponding to the sample EEG signal after processing by Laurent polynomial layer. , The number of eigenvalues ​​is And output to the activation layer, where, Indicates the first Each sample EEG signal corresponds to an eigenvector after processing with Laurent polynomial layers. This indicates the order of the negative power expansion. Indicates the order of a positive power expansion. Represents the positive power term in the Laurent expansion. order coefficient tensor, Represents the negative power term in the Laurent expansion. order coefficient tensor, and All dimensions are , This represents the dimension of the feature vector corresponding to the sample EEG signal after processing by the Laurent polynomial layer. , This indicates that the corresponding dimension in the Laurent expansion is The reference point vector, Represents the tensor product. Represents the tensor inner product. ,in, Represents the negative power term in the Laurent expansion. The order coefficient tensors correspond to respectively Data from various dimensions This indicates that the corresponding dimension in the Laurent expansion is The reference point vector of the first One target feature value; The activation layer is a non-linear activation layer. The non-linear activation layer receives the feature vector corresponding to the sample EEG signal after processing by the Laurent polynomial layer, according to the following formula: ； ； Obtain the feature vector corresponding to the sample EEG signal after activation layer processing. And output to the fully connected layer, where, , Indicates the first The EEG signals of the sample correspond to the first sample after processing by Laurent polynomial layer. 1 eigenvalue, Indicates the first Each sample EEG signal corresponds to a feature value vector after processing by the activation layer. Indicates the first The EEG signal of the sample corresponds to the first sample after processing by the activation layer. One eigenvalue; The fully connected layer receives the feature vector corresponding to the sample EEG signal after processing by the activation layer, as shown in the following formula: ； The analysis results of the sample EEG signals corresponding to each category were obtained. And output to the softmax layer, where, , Indicates the number of categories, Indicates the first The weights corresponding to each category Indicates the first The bias corresponding to each category Indicates the first The EEG signal of the sample corresponds to the first Analysis results values ​​for each category; The softmax layer receives the analysis results of the sample EEG signals for each category, according to the following formula: ； The probabilities of sample EEG signals corresponding to different categories are obtained and output to the output layer. Indicates the first The EEG signal of the sample corresponds to the first The probability of each category, Indicates the first The EEG signal of the sample corresponds to the first The analysis results of each category are used for exponential calculation; In step C, during the training process of the network to be trained, based on the preset initial learning rate, the Adam optimizer combined with the BP algorithm is used to adjust the parameters according to the following steps. , Reference point vector Calculate the gradient and update it; Step a. Based on the forward propagation of the Laurent polynomial layer, use the following formula: ； ； Calculate the loss function result during the training process of the network to be trained. For parameters The gradient; According to the following formula: ； ； Calculate the loss function result during the training process of the network to be trained. For parameters The gradient; According to the following formula: ； ； Calculate the loss function result during the training process of the network to be trained. For reference point vector The gradient; then proceed to step b; where, , Indicates the number of EEG signals in the sample. Indicates the first Each sample EEG signal corresponds to an eigenvector after processing with Laurent polynomial layers. This indicates the order of the negative power expansion. Indicates the order of a positive power expansion. Represents the positive power term in the Laurent expansion. order coefficient tensor, Represents the negative power term in the Laurent expansion. order coefficient tensor, and All dimensions are , This represents the dimension of the feature vector corresponding to the sample EEG signal after processing by the Laurent polynomial layer. Indicates the number of target features. , Indicates the first The corresponding EEG signal of the sample Each target feature is fine-tuned. This indicates that the corresponding dimension in the Laurent expansion is The reference point vector; Step b. Based on the loss function results during the training of the network to be trained. For the parameters respectively ,parameter Reference point vector The gradients are obtained by the Adam optimizer for each parameter. ,parameter Reference point vector Update.

2. The EEG signal classification method based on Laurent polynomials according to claim 1, characterized in that: The output layer receives the probability of the sample EEG signal corresponding to each category, selects the category corresponding to the highest probability, and forms the category corresponding to the sample EEG signal, which is the output of the network to be trained.

3. The EEG signal classification method based on Laurent polynomials according to claim 1, characterized in that: In step C, during the training process of the network to be trained, training is completed when the cross-entropy loss function converges or reaches a preset maximum number of iterations. The cross-entropy loss function is as follows: ； in, This represents the result of the loss function, where, Indicates the first The true category corresponding to each sample of EEG signal Indicates the first The EEG signal of the sample corresponds to the first The probability of each category, This represents the set of parameters to be trained in the network to be trained. Represents the regularization coefficient. Indicates the parameters to be trained. This represents the sum of squares of all parameters to be trained.

4. The EEG signal classification method based on Laurent polynomials according to claim 1, characterized in that: In step A, for each segment of EEG signal corresponding to a preset category, the abnormal values ​​and noise in each segment of EEG signal are first removed to update the signal. Then, the feature values ​​of each segment of EEG signal corresponding to each preset target feature are extracted.

Images