Cold start method for brain-computer interface

Abstract

The invention provides a cold start method of a brain-computer interface, which relates to the technical field of brain-computer interfaces, and comprises the following steps: collecting an initial signal of a target user by using brain-computer interface equipment; the initial signal comprises electroencephalogram data and magnetoencephalogram data; preprocessing the initial signal to obtain a preprocessed signal; performing feature extraction on the preprocessed signal to obtain a signal feature; inputting the signal features into a trained electroencephalogram activity classification model to obtain a classification result; and generating a feedback result according to the classification result, and sending the feedback result to the target user. The cold start method of the brain-computer interface effectively solves the problems that a system based on transfer learning is poor in adaptive capacity and slow in response in the cold start stage of a new user. According to the method, individuation, real-time feedback and online learning are emphasized, so that a more effective and flexible user adaptation scheme is provided, the interactive experience of the user is smoother, and the overall performance of the system and the user satisfaction degree are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of brain-computer interface (BCI), and in particular to a cold start method of a brain-computer interface. Background Art

[0002] Brain Computer Interface (BCI) refers to a direct connection between the human or animal brain and an external device to achieve information exchange between the brain and the device. This concept has actually been around for a long time, but it was not until the 1990s that phased results began to emerge. In 2008, neurobiologists at the University of Pittsburgh claimed that using a brain-computer interface, monkeys could use a robotic arm to feed themselves.

[0003] Brain-computer interface systems enable users to interact with computers or other devices without mechanical movements by reading and decoding brain signals. However, due to physiological differences between individuals and unstable initial signals, it is a challenge to quickly and effectively initialize the system to achieve efficient interaction.

[0004] The invention patent with application number 202210368393.6 provides a cold start method for a brain-computer interface. After the brain-computer interface is used to obtain the target's EEG signal, the data processing unit runs the data processing model to process the target's EEG signal obtained by the brain-computer interface, and the data processing model outputs information corresponding to the EEG signal; the data processing unit is connected to the target through the brain-computer interface; the data processing model is a transfer learning network model. The existing technology relies on a pre-trained transfer learning model, so for new users or targets, the system may take a long time to adapt, especially when there is a significant difference between the user's EEG signal and the source training data. This means that during the cold start phase of a new user, the system's response and adaptability are poor, which may lead to a poor interactive experience. Summary of the invention

[0005] In order to overcome the shortcomings of the prior art, the purpose of the present invention is to provide a cold start method for a brain-computer interface, which effectively solves the problem of poor adaptability and slow response of the system based on transfer learning during the cold start phase for new users.

[0006] To achieve the above object, the present invention provides the following solutions:

[0007] A cold start method for a brain-computer interface, comprising:

[0008] Using a brain-computer interface device to collect an initial signal from a target user; the initial signal includes: electroencephalogram data and magnetoencephalogram data;

[0009] Preprocessing the initial signal to obtain a preprocessed signal;

[0010] Performing feature extraction on the preprocessed signal to obtain signal features;

[0011] Inputting the signal features into a trained EEG activity classification model to obtain a classification result;

[0012] A feedback result is generated according to the classification result, and the feedback result is sent to the target user.

[0013] Preferably, preprocessing the initial signal to obtain a preprocessed signal includes:

[0014] Performing filtering on the initial signal to obtain a filtered signal;

[0015] Based on the independent component analysis method, the signal source separation is performed on the filtered signal to obtain the preprocessed signal.

[0016] Preferably, filtering the initial signal to obtain a filtered signal comprises:

[0017] Shifting the initial signal to obtain a shifted signal;

[0018] Performing wavelet decomposition on the shifted signal to obtain a plurality of wavelet coefficients;

[0019] Determine the filtering threshold according to the decomposition scale and length of the signal;

[0020] Constructing a denoising function according to the filtering threshold;

[0021] The noise of the translated signal is removed by using the denoising function to obtain the filtered signal.

[0022] Preferably, the formula of the filtering threshold is: Wherein, λ is the filtering threshold, j is the decomposition scale of the translated signal, and d j is the decomposition of the wavelet coefficients with a scale of j, N represents the length of the translated signal, and median represents the median operation.

[0023] Preferably, removing the noise of the translated signal by using the denoising function to obtain the filtered signal comprises:

[0024] The corresponding wavelet coefficients are removed by using the denoising function to obtain a smoothed signal; wherein the denoising function is: Among them, X represents the wavelet coefficient, n represents the adjustment coefficient;

[0025] The smoothed signal is inversely translated to obtain the filtered signal.

[0026] Preferably, based on an independent component analysis method, signal source separation is performed on the filtered signal to obtain the preprocessed signal, including:

[0027] Convert the format of the filtered signal to obtain a two-dimensional array; the rows in the two-dimensional array represent time points, and the columns in the two-dimensional array represent the number of channels;

[0028] Perform signal source separation on the two-dimensional array according to the independent component analysis method to obtain separated independent components;

[0029] Analyzing the separated independent components to evaluate the characteristics of each independent component to obtain the spectral characteristics and the time characteristics;

[0030] Identify components related to the electrocardiogram, electrooculogram, and other artifacts according to the frequency spectrum features and the time features, and obtain analysis results;

[0031] Select components related to brain activity based on the analysis results and mark the components to be excluded;

[0032] Signal reconstruction is performed according to the components related to the brain activity to obtain the preprocessed signal with artifacts removed.

[0033] Preferably, the signal characteristics include: time domain characteristics, frequency domain characteristics and time-frequency domain characteristics.

[0034] Preferably, the EEG activity classification model is constructed by a convolutional neural network; the convolutional neural network includes an input layer, a convolution layer, a pooling layer, a flattening layer, a fully connected layer and an output layer connected in sequence.

[0035] According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0036] The present invention provides a cold start method for a brain-computer interface, comprising: using a brain-computer interface device to collect an initial signal of a target user; the initial signal comprises: electroencephalogram data and magnetoencephalogram data; preprocessing the initial signal to obtain a preprocessed signal; extracting features from the preprocessed signal to obtain signal features; inputting the signal features into a trained EEG activity classification model to obtain a classification result; generating a feedback result based on the classification result, and sending the feedback result to the target user. The cold start method for a brain-computer interface effectively solves the problem of poor adaptability and slow response of a system based on transfer learning during the cold start phase for new users. The present invention emphasizes personalization, real-time feedback and online learning, thereby providing a more effective and flexible user adaptation solution, making the user's interactive experience smoother and improving the overall performance of the system and user satisfaction. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0038] Figure 1 A flow chart of a method provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0039] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0040] The purpose of the present invention is to provide a cold start method for a brain-computer interface, which provides a more effective and flexible user adaptation solution, making the user's interactive experience smoother and improving the overall performance of the system and user satisfaction.

[0041] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0042] Figure 1 A flow chart of a method provided by an embodiment of the present invention, such as Figure 1 As shown, the present invention provides a cold start method for a brain-computer interface, comprising:

[0043] Step 100: Collecting initial signals of a target user using a brain-computer interface device; the initial signals include: electroencephalogram data and magnetoencephalogram data;

[0044] Step 200: preprocessing the initial signal to obtain a preprocessed signal;

[0045] Step 300: extracting features from the preprocessed signal to obtain signal features;

[0046] Step 400: inputting the signal features into a trained EEG activity classification model to obtain a classification result;

[0047] Step 500: Generate a feedback result according to the classification result, and send the feedback result to the target user.

[0048] Specifically, this embodiment allows the user to wear the brain-computer interface device in a relaxed state to perform initial EEG / MEG data collection. During the collection process, the user can watch a quiet video or listen to soothing music to reduce the tension caused by the first use.

[0049] Preferably, preprocessing the initial signal to obtain a preprocessed signal includes:

[0050] Performing filtering on the initial signal to obtain a filtered signal;

[0051] Based on the independent component analysis method, the signal source separation is performed on the filtered signal to obtain the preprocessed signal.

[0052] Secondly, this embodiment performs filtering on the collected signals to remove power frequency, electromyographic artifacts and other environmental noises, and uses independent component analysis (ICA) to separate the brain signal sources of interest and remove electrocardiogram (ECG) and electrooculogram (EOG) interference.

[0053] Preferably, filtering the initial signal to obtain a filtered signal comprises:

[0054] Shifting the initial signal to obtain a shifted signal;

[0055] Performing wavelet decomposition on the shifted signal to obtain a plurality of wavelet coefficients;

[0056] Determine the filtering threshold according to the decomposition scale and length of the signal;

[0057] Constructing a denoising function according to the filtering threshold;

[0058] The noise of the translated signal is removed by using the denoising function to obtain the filtered signal.

[0059] Preferably, the formula of the filtering threshold is: Wherein, λ is the filtering threshold, j is the decomposition scale of the translated signal, and d j is the decomposition of the wavelet coefficients with a scale of j, N represents the length of the translated signal, and median represents the median operation.

[0060] Specifically, in the wavelet threshold denoising process, the quality of threshold selection directly affects the effect of denoising. Since the wavelet coefficient amplitude of noise has the characteristic of decreasing with the increase of decomposition scale, this embodiment introduces the decomposition scale to adaptively adjust the filtering threshold.

[0061] Preferably, removing the noise of the translated signal by using the denoising function to obtain the filtered signal comprises:

[0062] The corresponding wavelet coefficients are removed by using the denoising function to obtain a smoothed signal; wherein the denoising function is: Among them, X represents the wavelet coefficient, n represents the adjustment coefficient;

[0063] The smoothed signal is inversely translated to obtain the filtered signal.

[0064] Specifically, this embodiment uses wavelet coefficients to measure the attenuation deviation, so that the attenuation of wavelet coefficients with larger absolute values ​​decreases as their absolute values ​​increase, thereby effectively avoiding the loss of high-frequency information and improving the signal-to-noise ratio of the signal.

[0065] Furthermore, although the use of a denoising function is a simple and effective denoising method, in some cases, such as in the neighborhood of a signal singular point, the reconstructed signal may have large upper and lower peaks. These peaks are not inherent in the original signal, but due to the lack of translation invariance of the threshold method, the denoised signal may have distortions such as pseudo-Gibbs effect. Therefore, this embodiment can not only effectively suppress the pseudo-Gibbs effect by continuously translating the signal, but also obtain a smaller mean square error than the threshold method denoising, further improving the signal-to-noise ratio.

[0066] Optionally, the purpose of the filtering process of this embodiment is to remove random noise and interference components in the signal, so that the subsequent analysis is more accurate and effective. Exemplarily, this embodiment is used to remove power frequency interference, which is very common noise (such as 50 / 60 Hz interference from alternating current) and needs to be filtered out in signal analysis; and also remove electromyographic artifacts, which may be generated by muscle contraction and other movements, and need to be removed by high-frequency filters; and also remove other environmental noise, which may be low-frequency noise from the external environment.

[0067] Preferably, based on an independent component analysis method, signal source separation is performed on the filtered signal to obtain the preprocessed signal, including:

[0068] Convert the format of the filtered signal to obtain a two-dimensional array; the rows in the two-dimensional array represent time points, and the columns in the two-dimensional array represent the number of channels;

[0069] Perform signal source separation on the two-dimensional array according to the independent component analysis method to obtain separated independent components;

[0070] Analyzing the separated independent components to evaluate the characteristics of each independent component to obtain the spectral characteristics and the time characteristics;

[0071] Identify components related to the electrocardiogram, electrooculogram, and other artifacts according to the frequency spectrum features and the time features, and obtain analysis results;

[0072] Select components related to brain activity based on the analysis results and mark the components to be excluded;

[0073] Signal reconstruction is performed according to the components related to the brain activity to obtain the preprocessed signal with artifacts removed.

[0074] Specifically, the process of using independent component analysis (ICA) to separate signal sources in this embodiment involves multiple steps. The following are the specific logical steps of performing ICA signal source separation on the EEG signal after filtering:

[0075] Step 1: Data Preparation

[0076] Signal recording ensures that the EEG signals (EEG / MEG) collected by the brain-computer interface device have been filtered to remove power frequency noise, electromyographic artifacts and other environmental noise.

[0077] Data formatting converts the filtered signal data into a format suitable for ICA analysis, usually a two-dimensional array, with rows representing time points and columns representing channel (electrode) numbers.

[0078] Step 2: Select an ICA algorithm. Commonly used ICA algorithms include Infomax, FastICA, etc. Specifically, select an algorithm that suits the data characteristics to achieve the best separation effect.

[0079] Step 3: Perform ICA

[0080] Centering: Before performing ICA, each signal channel needs to be zero-mean centered by calculating the mean of each signal and subtracting it from the signal.

[0081] Whitening: Whitening is performed on the centered data to remove the correlation between signals so that the variance of each signal channel is 1. Mathematically, whitening can be achieved through principal component analysis (PCA).

[0082] Perform ICA operation: Apply the selected ICA algorithm to perform signal source separation. ICA decomposes the mixed signal (filtered EEG signal) into multiple independent components (source signals) and generates two matrices:

[0083] The component matrix (S), where each column represents an independent component; the mixing matrix (W), which maps the mixed signal to the weights of the independent components.

[0084] Step 4: Identify and select signal components

[0085] Component analysis, analyzes the separated independent components and evaluates the characteristics of each component through visualization (such as waveform graph and power spectrum).

[0086] Identify artifacts and identify components related to electrocardiogram (ECG), electrooculogram (EOG), and other artifacts based on the spectral and temporal characteristics of the components. Typically, EOG components have specific amplitude and waveform characteristics when the eye moves.

[0087] Select the components of interest, and based on the analysis results, select the components related to brain activity and mark the components to be excluded.

[0088] Step 5: Reconstruct the signal

[0089] Reconstruct the signal, reconstruct the clean EEG signal based on the selected components. Combine the selected independent components together and calculate their composite signal.

[0090] To remove the artifact influence, the identified electrocardiogram (ECG) and electrooculogram (EOG) components are removed from the original mixed signal to obtain an artifact-free EEG signal.

[0091] Step 6: Result Verification

[0092] Results were evaluated to verify the signal quality after artifact removal, and visual inspection and comparative analysis were performed to ensure that artifacts were effectively removed and brain signals of interest were preserved.

[0093] Specifically, through the above steps, ICA can effectively separate the brain signal source of interest from the filtered EEG signal, and remove interference such as electrocardiogram (ECG) and electrooculogram (EOG).

[0094] Furthermore, after filtering and ICA processing, the data finally obtained in this embodiment includes:

[0095] Clean EEG data, which removes power frequency noise, myoelectric interference, and other physiological artifacts, is more suitable for further analysis and feature extraction.

[0096] Independent component data, multiple independent signal components separated by ICA, may contain different brain activity patterns, noise components (such as eye movements, heart movements), etc. Based on these independent component data, more detailed analysis can be performed to find specific EEG signal features or patterns.

[0097] In summary, the filtering process of this embodiment provides clean basic data for ICA separation, and ICA further improves the accuracy of signal analysis, and ultimately obtains more reliable electroencephalogram (EEG) and magnetoencephalogram (MEG) data.

[0098] Preferably, the signal characteristics include: time domain characteristics, frequency domain characteristics and time-frequency domain characteristics.

[0099] Optionally, the feature extraction step of this embodiment includes:

[0100] Step 1: Data preparation, preprocessing Signal preparation, the clean signal after filtering and ICA processing should have been formatted into a form suitable for feature extraction, usually time series data.

[0101] Step 2: Select feature extraction methods in the time domain, frequency domain, and time-frequency domain for specific tasks. Common methods include:

[0102] Time domain analysis: frequency domain analysis (such as Fourier transform), time-frequency analysis (such as short-time Fourier transform, wavelet transform, etc.)

[0103] Step 3: Time domain feature extraction, mean and variance: calculate the mean (indicating the central trend of the signal) and variance (indicating the degree of fluctuation of the signal) of the signal; peak and peak interval: extract the maximum value, minimum value and interval between the peaks of the signal to reflect the abnormal events and frequency of the signal; other statistical features: including skewness, kurtosis, etc., used to describe the distribution pattern of the signal.

[0104] Step 4: Frequency domain feature extraction, fast Fourier transform (FFT): Use FFT to convert the signal from the time domain to the frequency domain to obtain the signal's spectrum; frequency band energy: Calculate the energy of different frequency bands (such as δ, θ, α, β, and γ waves) to understand the characteristics of the brain's electrical activity in different states; main frequency component: Identify the main frequency of the signal and its corresponding power to help analyze the electrical activity characteristics of specific functional states or diseases.

[0105] Step 5: Time-frequency domain feature extraction, wavelet transform: perform wavelet transform on the signal to extract the characteristics of the signal in different time windows and frequencies; time-frequency graph analysis: extract key features from the time-frequency graph, such as local frequency, time domain intensity changes, etc., to capture the characteristics of non-stationary signals.

[0106] Step 6: Feature selection and dimensionality reduction, Feature selection: According to the task requirements, the most meaningful features are selected through methods (such as variance selection, correlation analysis, LASSO, random forest, etc.) to reduce redundant information. Dimensionality reduction: Use techniques such as principal component analysis (PCA) or linear discriminant analysis (LDA) to further reduce the feature dimension and extract key information.

[0107] Step 7: Feature storage and use, Feature storage: Save the extracted signal features into the feature matrix to prepare for subsequent machine learning tasks such as classification, clustering or regression; Quick generation method: Design algorithms for fast feature generation for different tasks or different data types to improve efficiency and convenience.

[0108] Preferably, the EEG activity classification model is constructed by a convolutional neural network; the convolutional neural network includes an input layer, a convolution layer, a pooling layer, a flattening layer, a fully connected layer and an output layer connected in sequence.

[0109] Furthermore, this embodiment uses a convolutional neural network (CNN) as an EEG activity classification model, and the specific structure is as follows:

[0110] 1. Input layer: The input layer receives electroencephalogram (EEG) and magnetoencephalogram (MEG) data from preprocessed signals. Usually the shape of input data is (number of samples, number of channels, number of time points).

[0111] 2. Convolutional Layers, the hierarchical structure is:

[0112] Conv1: The first convolutional layer uses multiple convolutional kernels (filters) to extract local features of the input signal.

[0113] Activation function: ReLU (Rectified Linear Unit) activation function.

[0114] Conv2: The second convolutional layer continues to use multiple convolution kernels to further extract complex features.

[0115] 3. Pooling Layers, the hierarchical structure is:

[0116] MaxPooling: A maximum pooling layer (or average pooling layer) is connected after each convolutional layer to reduce the feature dimension and the amount of computation while retaining important features.

[0117] 4. Flatten Layer: Flattens the high-dimensional data after the pooling layer into one-dimensional data to facilitate subsequent fully connected layer processing.

[0118] 5. Fully Connected Layers: The hierarchical structure is:

[0119] Dense1: The first fully connected layer. The number of nodes can be set according to the specific task, such as 256 neurons.

[0120] Activation function: ReLU activation function.

[0121] Dense2: The second fully connected layer, with the number of nodes set to 128 neurons, further abstracts the data features.

[0122] 6. Output Layer: The hierarchical structure is:

[0123] Dense3: Output layer, the number of nodes is equal to the number of classification categories (such as the number of activity status categories).

[0124] Activation function: Softmax activation function, used for multi-class classification tasks.

[0125] Furthermore, the connection relationship between the layers of this embodiment is as follows:

[0126] Input Layer->Conv1->Pooling1->Conv2->Pooling2->Flatten->Dense1->Dense2->Output Layer

[0127] Specifically, the number and size of the convolution kernels of the convolution layer of this embodiment can be adjusted based on experience, and the size usually used is 3×3 or 5×5. Select a suitable step size in the convolution and pooling layers to control the size of the feature map. In addition, this embodiment can also use a Dropout layer between the fully connected layers to prevent overfitting.

[0128] Optionally, the loss function of the training process of this embodiment is optimized using a cross-entropy loss function (Cross-EntropyLoss), which is suitable for multi-class classification tasks. The optimizer uses an optimizer such as Adam or SGD for model training.

[0129] Furthermore, the Softmax result of the output layer of this embodiment is used as the classification probability, and the feedback result is generated according to the category with the highest probability. The EEG activity classification model of this embodiment adopts a typical convolutional neural network structure, hoping to extract effective EEG features and classify them through a combination of convolution, pooling and fully connected layers. This structure can not only process time series data, but also capture the spatial characteristics of different EEG activity states. Through multi-layer connections and complex feature extraction capabilities, the model can achieve a high classification accuracy and provide a reliable basis for the generation of feedback results.

[0130] Specifically, the process of generating feedback results according to the classification results and sending them to the target user in this embodiment can be divided into several key steps, which are as follows:

[0131] Step 1: Classification result processing

[0132] Get the classification result from the output layer of the trained EEG classification model. The result is usually a probability distribution, indicating the possibility of each category.

[0133] Determine the best category, and select the category with the highest probability as the final classification result by processing the Softmax result of the output layer. This can be achieved through the argmax function, that is, obtaining the category index corresponding to the maximum probability.

[0134] Step 2: Generate feedback content

[0135] Define the feedback content and design the corresponding feedback content according to different classification results. Feedback can be in the form of text, audio, visual prompts, etc. The specific content is determined by the system design goals and user needs. For example, if it is classified as "relaxed state", the feedback may be "You are now in a relaxed state, please continue to maintain it."; if it is classified as "focused", the feedback may be "Good, you are focusing."

[0136] Construct feedback data and generate the corresponding feedback data structure based on the classification results. This can be a simple string or a more complex object, depending on the specific application.

[0137] Step 3: Send feedback

[0138] Select a communication method to determine how to send the feedback result to the target user. This embodiment can be implemented in the following ways:

[0139] Visual feedback: Display feedback information on the screen of the brain-computer interface device.

[0140] Audio Feedback: Feedback is announced through the speaker.

[0141] Mobile device notifications: If connected to a mobile device, push notifications can be sent via the app.

[0142] Vibration feedback: If supported by the device, you can use sensory cues such as vibration.

[0143] Implement sending logic:

[0144] Visual feedback example, showing feedback information through interface updates.

[0145] Audio feedback sample, using the Text-to-Speech (TTS) library to play feedback information.

[0146] Push Notification Sample, which uses the API to send push notifications to the user's device.

[0147] Step 4: User Interaction

[0148] User confirmation or response: Design relevant mechanisms so that users can respond to feedback, such as providing a "confirm" or "understood" button. This step can increase user participation and provide data for subsequent optimization.

[0149] Step 5: Logging and Analysis

[0150] Recording feedback process: In order to improve system performance in the future, feedback records and user reactions can be stored in the database for analysis. This step can improve the classification model and its feedback mechanism.

[0151] Optionally, the process of generating feedback results based on the classification results and sending them to the target user in this embodiment involves obtaining the classification results, generating corresponding feedback content, sending feedback through appropriate communication methods, and recording the interaction process. This process can not only provide real-time feedback to help users adjust their status, but also accumulate important data for system optimization and personalization improvement.

[0152] The beneficial effects of the present invention are as follows:

[0153] (1) The present invention collects the target user's initial electroencephalogram (EEG) and magnetoencephalogram (MEG) signals in real time, and the model can be personalized based on the biosignal characteristics of each user, rather than relying entirely on a general transfer learning model. This method can quickly collect biometric data related to a specific user, allowing the model to better understand the user's status.

[0154] (2) The present invention obtains a clearer preprocessed signal by filtering, denoising and performing independent component analysis (ICA) on the initial signal to remove artifacts and contamination. This process can ensure the effectiveness of subsequent feature extraction, reduce the impact of noise on classification results, and achieve more accurate classification.

[0155] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0156] The principles and implementation methods of the present invention are described in this article using specific examples. The description of the above embodiments is only used to help understand the method and core idea of ​​the present invention. At the same time, for those skilled in the art, according to the idea of ​​the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting the present invention.

Claims

1. A cold start method for a brain-computer interface, characterized in that: include: Using the brain-computer interface device to collect the target user's initial signal; The initial signal includes: electroencephalogram data and magnetoencephalogram data; Preprocessing the initial signal to obtain a preprocessed signal; Performing feature extraction on the preprocessed signal to obtain signal features; Inputting the signal features into a trained EEG activity classification model to obtain a classification result; A feedback result is generated according to the classification result, and the feedback result is sent to the target user.

2. The cold start method of the brain-computer interface according to claim 1, characterized in that: Preprocessing the initial signal to obtain a preprocessed signal includes: Performing filtering on the initial signal to obtain a filtered signal; Based on the independent component analysis method, the signal source separation is performed on the filtered signal to obtain the preprocessed signal.

3. The cold start method of the brain-computer interface according to claim 2, characterized in that: Filtering the initial signal to obtain a filtered signal includes: Shifting the initial signal to obtain a shifted signal; Performing wavelet decomposition on the shifted signal to obtain a plurality of wavelet coefficients; Determine the filtering threshold according to the decomposition scale and length of the signal; Constructing a denoising function according to the filtering threshold; The noise of the translated signal is removed by using the denoising function to obtain the filtered signal.

4. The cold start method of the brain-computer interface according to claim 3, characterized in that: The formula for the filtering threshold is: Wherein, λ is the filtering threshold, j is the decomposition scale of the translated signal, and d j is the decomposition of the wavelet coefficients with a scale of j, N represents the length of the translated signal, and median represents the median operation.

5. The cold start method of the brain-computer interface according to claim 4, characterized in that: Using the denoising function to remove the noise of the translated signal to obtain the filtered signal includes: The corresponding wavelet coefficients are removed by using the denoising function to obtain a smoothed signal; wherein the denoising function is: Among them, X represents the wavelet coefficient, n represents the adjustment coefficient; The smoothed signal is inversely translated to obtain the filtered signal.

6. The cold start method of the brain-computer interface according to claim 2, characterized in that: Based on the independent component analysis method, the signal source separation is performed on the filtered signal to obtain the preprocessed signal, including: Convert the format of the filtered signal to obtain a two-dimensional array; the rows in the two-dimensional array represent time points, and the columns in the two-dimensional array represent the number of channels; Perform signal source separation on the two-dimensional array according to the independent component analysis method to obtain separated independent components; Analyzing the separated independent components to evaluate the characteristics of each independent component to obtain the spectral characteristics and the time characteristics; Identify components related to the electrocardiogram, electrooculogram, and other artifacts according to the frequency spectrum features and the time features, and obtain analysis results; Select components related to brain activity based on the analysis results and mark the components to be excluded; Signal reconstruction is performed according to the components related to the brain activity to obtain the preprocessed signal with artifacts removed.

7. The cold start method of the brain-computer interface according to claim 1, characterized in that: The signal characteristics include: time domain characteristics, frequency domain characteristics and time-frequency domain characteristics.

8. The cold start method of the brain-computer interface according to claim 1, characterized in that: The EEG activity classification model is constructed by a convolutional neural network; the convolutional neural network includes an input layer, a convolution layer, a pooling layer, a flattening layer, a fully connected layer and an output layer connected in sequence.

Images