Attention recognition method based on electroencephalogram signals
By downsampling, notch filtering, and bandpass filtering of EEG signals, combined with a U-shaped network model and a multilayer perceptron, the accuracy and real-time performance issues of EEG attention recognition methods were solved, enabling efficient attention recognition in consumer wearable devices.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-01-09
- Publication Date
- 2026-04-17
AI Technical Summary
Existing EEG attention recognition methods suffer from low recognition accuracy due to individual differences and environmental noise, and are difficult to meet the requirements of real-time performance and cross-subject adaptability, especially limiting their application in consumer wearable devices.
By collecting EEG signals from users when performing tasks, and performing downsampling, notch filtering, and bandpass filtering, combined with a U-shaped network model and a multilayer perceptron, attention features can be identified.
It improves the accuracy and real-time performance of attention recognition, reduces computational load, adapts to different subjects, and is suitable for consumer wearable devices.
Smart Images
Figure CN121867786A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of pattern recognition, and more particularly to an attention recognition method based on electroencephalogram (EEG) signals. Background Technology
[0002] EEG attention recognition refers to the process of identifying and judging a user's attention level (e.g., high attention, low attention) or attention direction through collected EEG signals. Related technologies typically include three types of EEG recognition methods: alpha-beta ratio method, machine learning methods, and deep learning methods. However, the alpha-beta ratio method has weak anti-interference capabilities, is easily affected by individual EEG differences and environmental noise, and has low recognition accuracy. Machine learning methods rely on manual feature engineering based on domain knowledge; manually constructed features may not fully reflect the complex information of attention states, and the model's generalization ability is limited by the quantity and diversity of training data. Deep learning methods have complex model structures, high computational resource requirements, require a large number of labeled samples during training, and have slow inference speeds, making them difficult to meet the needs of real-time applications. Therefore, existing EEG recognition methods suffer from significant differences in EEG signals among individuals. Models trained for specific subjects show a sharp drop in recognition accuracy when applied to new subjects, lacking adaptability across subject tasks and failing to meet the needs of large-scale, diverse users in real-world scenarios. Summary of the Invention
[0003] The main objective of this application is to provide an attention recognition method based on electroencephalogram (EEG) signals, which aims to reduce the computational load of attention recognition methods and improve their adaptability to different subjects.
[0004] In a first aspect, this application provides an attention recognition method based on electroencephalogram (EEG) signals, the attention recognition method based on EEG signals comprising the following steps: The user's initial EEG signal is obtained by collecting the EEG signals during the user's execution of a preset task. The initial EEG signal is downsampled based on a preset extraction factor to obtain a first initial EEG signal; the first initial EEG signal is notched based on a preset power frequency interference frequency to obtain a second initial EEG signal; the second initial EEG signal is bandpass filtered based on preset electromyography interference frequency and electrooculography interference frequency to obtain a target EEG signal. The target EEG signal is input into the encoder of a preset U-shaped network model for downsampling to obtain the context features corresponding to the target EEG signal; the context features are input into the decoder of the U-shaped network model for upsampling to obtain the reconstructed pseudo EEG signal corresponding to the target EEG signal. The reconstructed pseudo-EEG signal is convolved to obtain the attention feature vector corresponding to the reconstructed pseudo-EEG signal; the attention feature vector is input into at least one multilayer perceptron to obtain the signal recognition result.
[0005] In some embodiments, the step of notch filtering the first initial EEG signal based on a preset power frequency interference frequency to obtain a second initial EEG signal includes: The downsampling rate of the downsampling process is determined based on the ratio of the initial sampling rate of the initial EEG signal to the extraction factor. The normalized center angular frequency is determined based on the ratio of the power frequency interference frequency to the downsampled sampling rate. The interference transfer function of the power frequency interference frequency is calculated based on the normalized center angular frequency and the preset damping coefficient. The first initial EEG signal and the interference transfer function are convolved to obtain a second initial EEG signal after filtering out the power frequency interference.
[0006] In some embodiments, the step of bandpass filtering the second initial EEG signal based on preset electromyography interference frequencies and electrooculography interference frequencies to obtain the target EEG signal includes: Obtain the high-pass cutoff frequency and low-pass cutoff frequency of electromyography (EMG) corresponding to the EMG interference frequency, and the high-pass cutoff frequency and low-pass cutoff frequency of electrooculography (EOG) corresponding to the EOG interference frequency. The electromyography high-pass cutoff frequency and the electromyography low-pass cutoff frequency are converted into electromyography high-pass cutoff angular frequency and electromyography low-pass cutoff angular frequency; the electrooculography high-pass cutoff frequency and the electrooculography low-pass cutoff frequency are converted into electrooculography high-pass cutoff angular frequency and electrooculography low-pass cutoff angular frequency. Based on a preset quality factor, the electromyography high-pass transfer function and the electromyography low-pass transfer function are determined according to the electromyography high-pass cutoff angular frequency and the electromyography low-pass cutoff angular frequency; the electrooculography high-pass transfer function and the electrooculography low-pass transfer function are determined according to the electrooculography high-pass cutoff angular frequency and the electrooculography low-pass cutoff angular frequency. The second initial EEG signal is convolved with the electromyography high-pass transfer function, the electromyography low-pass transfer function, the electrooculography high-pass transfer function, and the electrooculography low-pass transfer function to obtain the target EEG signal after filtering out the electromyography interference frequency and the electrooculography interference frequency.
[0007] In some embodiments, the power frequency interference frequency is 50Hz; the electromyography interference frequency has a high-pass cutoff frequency of 500Hz and a low-pass cutoff frequency of 50Hz; the electrooculography interference frequency has a high-pass cutoff frequency of 3Hz and a low-pass cutoff frequency of 0.1Hz.
[0008] In some embodiments, the step of downsampling the target EEG signal into an encoder of a preset U-shaped network model to obtain contextual features corresponding to the target EEG signal includes: The target EEG signal is input into multiple downsampling blocks of the encoder to obtain the contextual features, wherein each downsampling block includes: a first convolutional layer, an activation function, and a max pooling layer.
[0009] In some embodiments, after inputting the target EEG signal into multiple downsampling blocks of the encoder to obtain the contextual features, the method further includes: The contextual features are input into the bottleneck layer to obtain the target features, and the target features are input into the decoder. The bottleneck layer includes a third convolutional layer.
[0010] In some implementations, the step of upsampling the contextual features into the decoder of the U-shaped network model to obtain the reconstructed pseudo-EEG signal corresponding to the target EEG signal includes: The target features are input into multiple upsampling blocks of the decoder to obtain the reconstructed pseudo-EEG signal. Each upsampling block includes: a transposed convolutional layer, a feature splicing layer, a second convolutional layer, and an activation function. The feature splicing layer is connected to the corresponding first convolutional layer in the downsampling block.
[0011] In some implementations, the step of performing convolution processing on the reconstructed pseudo-EEG signal to obtain the attention feature vector corresponding to the reconstructed pseudo-EEG signal includes: The reconstructed EEG signal is temporally convolved based on a preset temporal convolution kernel to obtain temporal convolution output features, wherein the temporal convolution kernel is a 1×m convolution kernel; The temporal convolution output features are spatially convolved based on a preset spatial convolution kernel to obtain spatial convolution output features, wherein the spatial convolution kernel is an n×1 convolution kernel; The temporal convolutional output features and the spatial convolutional output features are subjected to batch normalization, activation, pooling, and random discarding to obtain the attention feature vector.
[0012] In some implementations, inputting the attention feature vector into at least one multilayer perceptron to obtain the signal recognition result includes: The first intermediate vector corresponding to the attention feature vector is determined based on the first weight matrix and the first bias. The first intermediate vector is input into the ReLU activation function to obtain the second intermediate vector corresponding to the first intermediate vector. The third intermediate vector corresponding to the second intermediate vector is determined based on the second weight matrix and the second bias. The classification probability of the attention feature vector is obtained by inputting the third intermediate vector into the Softmax function; The attention type corresponding to the initial EEG signal is determined based on the classification probability, and the signal recognition result is determined.
[0013] In some embodiments, the acquisition of EEG signals during the user's execution of a preset task to obtain the user's initial EEG signals includes: During the user's performance of high-attention and low-attention tasks, EEG signals from at least one of the user's left and right mastoid regions were collected using 2-4 lead electrodes.
[0014] This application provides an attention recognition method based on electroencephalogram (EEG) signals. The method involves collecting EEG signals from a user during the execution of a preset task to obtain an initial EEG signal; downsampling the initial EEG signal based on a preset extraction factor to obtain a first initial EEG signal; notch filtering the first initial EEG signal based on a preset power frequency interference frequency to obtain a second initial EEG signal; bandpass filtering the second initial EEG signal based on preset electromyography (EMG) and electrooculography (EOG) interference frequencies to obtain a target EEG signal; downsampling the target EEG signal using an encoder of a preset U-shaped network model to obtain contextual features corresponding to the target EEG signal; upsampling the contextual features using a decoder of the U-shaped network model to obtain a reconstructed pseudo-EEG signal corresponding to the target EEG signal; convolution processing the reconstructed pseudo-EEG signal to obtain an attention feature vector corresponding to the reconstructed pseudo-EEG signal; and inputting the attention feature vector into at least one multilayer perceptron to obtain a signal recognition result. By processing the initial EEG signal, the computational load of the attention recognition method is reduced, and its adaptability to different subjects is improved. Attached Figure Description
[0015] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0016] Figure 1 A flowchart illustrating an attention recognition method based on electroencephalogram (EEG) signals provided in an embodiment of this application; Figure 2A schematic block diagram of a U-shaped network provided in an embodiment of this application; Figure 3 This is a schematic block diagram of the structure of a computer device according to an embodiment of this application. Detailed Implementation
[0017] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.
[0018] The flowchart shown in the attached diagram is for illustrative purposes only and does not necessarily include all content and operations / steps, nor does it necessarily have to be performed in the order described. For example, some operations / steps can be broken down, combined, or partially merged, so the actual execution order may change depending on the actual situation.
[0019] This application provides an attention recognition method based on electroencephalogram (EEG) signals.
[0020] The following detailed description of some embodiments of this application is provided in conjunction with the accompanying drawings. Unless otherwise specified, the following embodiments and features can be combined with each other.
[0021] Please refer to Figure 1 , Figure 1 This is a flowchart illustrating an attention recognition method based on electroencephalogram (EEG) signals, provided as an embodiment of this application. This EEG-based attention recognition method can be used in a terminal or server. The terminal can be an electronic device such as a mobile phone, tablet, laptop, desktop computer, personal digital assistant, or wearable device; the server can be a standalone server, a server cluster, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms.
[0022] like Figure 1 As shown, the attention recognition method based on EEG signals includes steps S101 to S104.
[0023] Step S101: Collect the EEG signals of the user during the execution of the preset task to obtain the user's initial EEG signals.
[0024] For example, brain-computer interfaces (BCIs), as an emerging technology connecting the brain to external devices, have become a research hotspot in the field of human-computer interaction. The basic principle of brain-computer interfaces is to collect electroencephalograms (EEGs) generated by the activity of neurons in the brain through implanted or non-implanted sensors, process these EEGs, and finally convert them into control commands that can be recognized by external devices, thereby realizing direct human-computer interaction without the involvement of peripheral nerves and muscles.
[0025] When a person is in a state of focused or distracted attention, the intensity of neuronal activity varies in different brain regions (such as the prefrontal and parietal lobes), leading to differences in electroencephalogram (EEG) signal responses. To process these EEG signals, it is necessary to perform pattern recognition based on the specific changes in EEG signals under different attentional states to determine the level of attention. For example, when attention is focused, the amplitude of alpha waves (8-13 Hz) in the parietal and occipital lobes of the brain significantly decreases, while the amplitude of beta waves (14-30 Hz) relatively increases.
[0026] In related technologies, the alpha-beta ratio method is a commonly used attention recognition method, which determines the attention state by calculating the amplitude ratio of alpha waves to beta waves in the EEG signal. A threshold is usually set; a ratio below the threshold is considered a state of focused attention, while a ratio above the threshold is considered a state of distracted attention. However, the alpha-beta ratio method has weak anti-interference capabilities and is easily affected by individual EEG differences and environmental noise, resulting in low recognition accuracy. Machine learning methods are also widely used in attention recognition. These methods first extract time-domain features (such as mean, variance, peak value, etc.), frequency-domain features (such as power spectral density), or time-frequency-domain features (such as wavelet transform coefficients) of the EEG signal through manual feature engineering. The extracted features are then input into a classification model for training and recognition. Commonly used algorithms include Support Vector Machine (SVM), Random Forest, and K-Nearest Neighbors (KNN). However, machine learning methods rely on domain-specific knowledge for manual feature engineering, and manually designed features may not fully reflect the complex information of the attention state. Furthermore, the model's generalization ability is limited by the quantity and diversity of training data. With the development of deep learning technology, deep learning methods have also been applied to attention recognition. Attention recognition methods based on models such as convolutional neural networks (CNN), recurrent neural networks (RNN), and long short-term memory networks (LSTM) have gradually emerged. These methods do not require manual feature design and can automatically learn deep features in EEG signals through the network. They are particularly suitable for processing multi-channel, long-term EEG data and have better recognition accuracy than traditional machine learning methods in complex scenarios. However, deep learning methods have complex model structures, high requirements for computing resources, require a large number of labeled samples during training, and have slow inference speed, making it difficult to meet the needs of real-time application scenarios. On the other hand, most existing models rely on whole-brain EEG for recognition. Collecting whole-brain EEG signals requires the deployment of multiple sensors, which not only increases the complexity and cost of equipment but may also affect the stability of recognition due to multi-channel signal synchronization issues.
[0027] The attention recognition method based on EEG signals provided in this application can achieve attention recognition based on a small amount of initial EEG signals by processing the initial EEG signals. This solves the problems in related technologies that rely on whole-brain multi-lead EEG, resulting in high device complexity, weak cross-subject generalization ability, and difficulty in balancing accuracy and real-time performance. It enables high-precision attention recognition with a small number of EEG leads, while improving the model's cross-subject adaptability and recognition real-time performance.
[0028] In some embodiments, the acquisition of EEG signals during the user's execution of a preset task to obtain the user's initial EEG signals includes: During the user's performance of high-attention and low-attention tasks, EEG signals from at least one of the user's left and right mastoid regions were collected using 2-4 lead electrodes.
[0029] For example, the attention recognition based on EEG signals provided in this application breaks through the limitations of traditional whole-brain or prefrontal multi-lead acquisition by selecting the area behind the ear as the EEG signal acquisition point. This area is not only neurally associated with attention-related brain regions such as the temporal lobe and parietal lobe, which can effectively capture the EEG characteristics corresponding to changes in attention state, but also has smooth skin and less hair interference behind the ear, resulting in high signal acquisition stability. At the same time, only 2-4 leads of electrodes are needed, which can be integrated into the ear hooks or earmuffs of wearable devices such as headphones and headbands. When wearing them, there is no need to contact the scalp, which greatly reduces the burden on users and improves the comfort of wearing them for a long time. In addition, the small number of electrodes is conducive to the miniaturization and low power consumption design of the device, making it suitable for consumer-grade wearable device scenarios.
[0030] For example, the initial EEG signal acquisition was performed using a portable EEG acquisition device. This device is small in size, low in power consumption, supports flexible multi-channel configuration, and is suitable for wearable scenarios. During acquisition, 2-4 lead electrodes were placed in the mastoid region behind the left and right ears of the subject, respectively, with the electrodes in contact with the skin. The EEG signals behind the ears were continuously acquired during the subject's high-attention tasks (such as working attentively or watching a movie) and low-attention tasks (such as observing a landscape) as the initial EEG signals.
[0031] Step S102: Downsample the initial EEG signal based on a preset extraction factor to obtain a first initial EEG signal; perform notch filtering on the first initial EEG signal based on a preset power frequency interference frequency to obtain a second initial EEG signal; perform bandpass filtering on the second initial EEG signal based on preset electromyography interference frequency and electrooculography interference frequency to obtain a target EEG signal.
[0032] For example, the initial EEG signal collected in step S101 is preprocessed to reduce interference signals in the initial EEG signal and improve the signal quality of the EEG signal, thereby improving the accuracy of attention recognition.
[0033] For example, since attention-related features in EEG are mainly concentrated in the low-frequency range, to reduce the computational load caused by excessively high sampling rates, an integer decimation method is used to downsample the acquired raw EEG signal. One data point is extracted every few sampling points to obtain the first initial downsampled EEG signal; the number of sampling points at intervals is a preset decimation factor. The downsampling process can be represented by the following formula: in, N represents the discrete-time sequence of the first initial EEG signal after downsampling, where N is the total number of data points in the downsampled signal.
[0034] In some embodiments, the step of notch filtering the first initial EEG signal based on a preset power frequency interference frequency to obtain a second initial EEG signal includes: The downsampling rate of the downsampling process is determined based on the ratio of the initial sampling rate of the initial EEG signal to the extraction factor. The normalized center angular frequency is determined based on the ratio of the power frequency interference frequency to the downsampled sampling rate. The interference transfer function of the power frequency interference frequency is calculated based on the normalized center angular frequency and the preset damping coefficient. The first initial EEG signal and the interference transfer function are convolved to obtain a second initial EEG signal after filtering out the power frequency interference.
[0035] For example, the frequency of environmental noise that the initial EEG signal may encounter is set as the power frequency interference frequency, and a second-order IIR notch filter is used to filter out the power frequency interference frequency, wherein the normalized center angular frequency is... , This indicates the preset power frequency interference frequency. This represents the sampling rate after downsampling; the interference transfer function can be calculated using the following formula: r=0.9 is the damping coefficient. Then, the initial EEG signal and the interference transfer function are convolved based on the following formula: in, Indicates the second initial EEG signal. This represents signal convolution.
[0036] In some embodiments, the step of bandpass filtering the second initial EEG signal based on preset electromyography interference frequencies and electrooculography interference frequencies to obtain the target EEG signal includes: Obtain the high-pass cutoff frequency and low-pass cutoff frequency of electromyography (EMG) corresponding to the EMG interference frequency, and the high-pass cutoff frequency and low-pass cutoff frequency of electrooculography (EOG) corresponding to the EOG interference frequency. The electromyography high-pass cutoff frequency and the electromyography low-pass cutoff frequency are converted into electromyography high-pass cutoff angular frequency and electromyography low-pass cutoff angular frequency; the electrooculography high-pass cutoff frequency and the electrooculography low-pass cutoff frequency are converted into electrooculography high-pass cutoff angular frequency and electrooculography low-pass cutoff angular frequency. Based on a preset quality factor, the electromyography high-pass transfer function and the electromyography low-pass transfer function are determined according to the electromyography high-pass cutoff angular frequency and the electromyography low-pass cutoff angular frequency; the electrooculography high-pass transfer function and the electrooculography low-pass transfer function are determined according to the electrooculography high-pass cutoff angular frequency and the electrooculography low-pass cutoff angular frequency. The second initial EEG signal is convolved with the electromyography high-pass transfer function, the electromyography low-pass transfer function, the electrooculography high-pass transfer function, and the electrooculography low-pass transfer function to obtain the target EEG signal after filtering out the electromyography interference frequency and the electrooculography interference frequency.
[0037] For example, electromyography (EMG) and electrooculography (EOG) interference can mask EEG attentional characteristics. A fourth-order Butterworth bandpass filter is used to filter these signals, preserving the frequency bands relevant to attention. This bandpass filter is composed of a cascaded low-pass and high-pass filter. Specifically, the EMG interference frequency is a range between the EMG high-pass and low-pass cutoff frequencies; the EOG interference frequency is also a range between the EOG high-pass and low-pass cutoff frequencies.
[0038] For example, the Qualcomm cutoff angular frequency ,in High-pass cutoff frequency; low-pass cutoff angular frequency ,in This represents the low-pass cutoff frequency. Based on the above formula, the high-pass cutoff angular frequencies of electromyography (EMG) and electrooculography (EOG) are determined according to their respective high-pass and high-pass cutoff frequencies, and the low-pass cutoff angular frequencies of EMG and EOG are determined according to their respective low-pass and low-pass cutoff frequencies.
[0039] For example, based on a preset quality factor Q, the high-pass transfer function is determined using the following formulas. Low-pass transfer function : Wherein, the quality factor Q can be Based on the above formulas, the high-pass transfer functions of electromyography (EMG) and electrooculography (EOG) are determined according to the high-pass cutoff angular frequencies of EMG and EOG, respectively, and the low-pass transfer functions of EMG and EOG are determined according to the low-pass cutoff angular frequencies of EMG and EOG.
[0040] For example, based on the following formula, the second initial EEG signal is... The signal is convolved with the electromyographic high-pass transfer function, the electromyographic low-pass transfer function, the electrooculography high-pass transfer function, and the electrooculography low-pass transfer function to obtain the preprocessed target EEG signal. : .
[0041] In some embodiments, the power frequency interference frequency is 50Hz; the electromyography interference frequency has a high-pass cutoff frequency of 500Hz and a low-pass cutoff frequency of 50Hz; the electrooculography interference frequency has a high-pass cutoff frequency of 3Hz and a low-pass cutoff frequency of 0.1Hz.
[0042] For example, the mains power frequency in my country is 50Hz, so the frequency of power frequency interference generated by electrical appliances in the environment is approximately 50Hz. Therefore, the power frequency interference frequency for notch filtering is set to 50Hz. The main frequency range for electromyography (EMG) interference is 50-500Hz, and the main frequency range for electrooculography (EOG) interference is 0.1-3Hz. Therefore, the high-pass cutoff frequency for EMG interference is set to 500Hz, the low-pass cutoff frequency for EMG interference is set to 50Hz, the high-pass cutoff frequency for EOG interference is set to 3Hz, and the low-pass cutoff frequency for EOG interference is set to 0.1Hz.
[0043] Step S103: Input the target EEG signal into the encoder of a preset U-shaped network model for downsampling to obtain the context features corresponding to the target EEG signal; input the context features into the decoder of the U-shaped network model for upsampling to obtain the reconstructed pseudo EEG signal corresponding to the target EEG signal.
[0044] For example, the target EEG signal is reconstructed using a U-Net, and a reconstructed pseudo-EEG signal containing spatial correlation information of multiple brain regions is generated based on the core features of the target EEG signal, thus making up for the feature dimension defects that may exist in a small number of target EEG signals.
[0045] Please refer to Figure 2 , Figure 2 This is a schematic diagram of the structure of a U-shaped network provided in an embodiment of this application.
[0046] like Figure 2 As shown, in some embodiments, the step of downsampling the target EEG signal into an encoder of a preset U-shaped network model to obtain the contextual features corresponding to the target EEG signal includes: The target EEG signal is input into multiple downsampling blocks of the encoder to obtain the contextual features, wherein each downsampling block includes: a first convolutional layer, an activation function, and a max pooling layer.
[0047] For example, Figure 2 In this embodiment, the encoder contains three downsampling blocks to minimize the computational load of the U-shaped network while ensuring reconstruction accuracy. In practice, the encoder may contain more downsampling blocks, which is not limited here.
[0048] The target EEG signal is input into multiple contiguous downsampling blocks, which convert the target EEG signal into contextual features. The decoder then upsamples the contextual features to reconstruct the target EEG signal and obtain a reconstructed pseudo-EEG signal.
[0049] like Figure 2 As shown, each downsampling block contains a first convolutional layer (Conv), a ReLU activation function (not shown), and a max pooling layer (MaxPool). The feature map output by the i-th downsampling block in the encoder is determined according to the following formula: in, For the input feature map, The kernel size is [size]. for Number of output channels per layer. Understandably, the input feature map for the first downsampling block is... .
[0050] In some embodiments, after inputting the target EEG signal into multiple downsampling blocks of the encoder to obtain the contextual features, the method further includes: The contextual features are input into the bottleneck layer to obtain the target features, and the target features are input into the decoder. The bottleneck layer includes a third convolutional layer.
[0051] For example, the encoder and decoder are connected through a bottleneck layer, which includes a third convolutional layer. The convolutional operations of the third convolutional layer further extract feature representations from the context features, transforming the context features of the encoder into target features for the decoder to process.
[0052] In some implementations, the step of upsampling the contextual features into the decoder of the U-shaped network model to obtain the reconstructed pseudo-EEG signal corresponding to the target EEG signal includes: The target features are input into multiple upsampling blocks of the decoder to obtain the reconstructed pseudo-EEG signal. Each upsampling block includes: a transposed convolutional layer, a feature splicing layer, a second convolutional layer, and an activation function. The feature splicing layer is connected to the corresponding first convolutional layer in the downsampling block.
[0053] For example, Figure 2In this embodiment, the decoder contains three upsampling blocks to minimize the computational cost of the U-shaped network while ensuring reconstruction accuracy. In practice, the decoder can contain more upsampling blocks, which is not limited here. It is understood that the feature concatenation layer achieves skip connections through vector addition. The feature concatenation layer in the upsampling block adds the second convolutional layer to the first convolutional layer of the downsampling block. Therefore, the number of upsampling blocks is the same as the number of downsampling blocks.
[0054] like Figure 2 As shown, each upsampling block contains a transposed convolutional layer (DeConv), a feature concatenation layer, a second convolutional layer, and a ReLU activation function (not shown in the figure). The feature map output by the j-th upsampling block in the encoder is determined according to the following formula: Indicates the first The number of output channels per layer, with the output of the last upsampling block being the reconstructed pseudo-EEG signal. .
[0055] Step S104: Perform convolution processing on the reconstructed pseudo-EEG signal to obtain the attention feature vector corresponding to the reconstructed pseudo-EEG signal; input the attention feature vector into at least one multilayer perceptron to obtain the signal recognition result.
[0056] For example, the reconstructed pseudo-EEG signals obtained from U-Net are input into EEGNET, and depthwise separable convolutions are used to capture temporal dynamic features and spatial distribution features respectively. Finally, a multilayer perceptron outputs the signal recognition results, which include: focused attention and distracted attention. Of course, it is not limited to this; the signal recognition results can also be used to represent the level of focused attention.
[0057] For example, the attention recognition method based on EEG signals provided in this application not only retains the convenience of collecting data behind the ear, but also achieves recognition accuracy comparable to the whole-brain lead method through the synergy of signal reconstruction and a dedicated classification network.
[0058] In some implementations, the step of performing convolution processing on the reconstructed pseudo-EEG signal to obtain the attention feature vector corresponding to the reconstructed pseudo-EEG signal includes: The reconstructed EEG signal is temporally convolved based on a preset temporal convolution kernel to obtain temporal convolution output features, wherein the temporal convolution kernel is a 1×m convolution kernel; The temporal convolution output features are spatially convolved based on a preset spatial convolution kernel to obtain spatial convolution output features, wherein the spatial convolution kernel is an n×1 convolution kernel; The temporal convolutional output features and the spatial convolutional output features are subjected to batch normalization, activation, pooling, and random discarding to obtain the attention feature vector.
[0059] For example, the reconstructed pseudo-EEG signal is sequentially input into a convolutional layer, a depthwise convolutional layer, and a depthwise separable convolutional layer. Temporal convolution and spatial convolution are performed sequentially in the depthwise convolutional layer to extract attention-related features. Specifically, the temporal convolution uses a 1×m convolutional kernel to capture the temporal dynamic features of the EEG signal, while the spatial convolution uses an n×1 convolutional kernel to mine the spatial distribution features between multiple leads.
[0060] For example, the reconstructed EEG signal is temporally convolved based on the following formula to obtain the temporal convolution output features: in, This represents the output features of temporal convolution. This represents the number of output channels for the temporal convolution. This represents the temporal convolution kernel.
[0061] For example, spatial convolution is performed on the temporal convolution output features based on the following formula to obtain the spatial convolution output features: in, This represents the output features of spatial convolution. This represents the number of output channels of the spatial convolution. This represents the number of output channels of the spatial convolution. Represents the spatial convolution kernel.
[0062] For example, the temporal and spatial convolutional output features are processed by subsequent batch normalization (BN), ELU activation, pooling, and dropout to obtain the attention feature vector. .
[0063] In some implementations, inputting the attention feature vector into at least one multilayer perceptron to obtain the signal recognition result includes: The first intermediate vector corresponding to the attention feature vector is determined based on the first weight matrix and the first bias. The first intermediate vector is input into the ReLU activation function to obtain the second intermediate vector corresponding to the first intermediate vector. The third intermediate vector corresponding to the second intermediate vector is determined based on the second weight matrix and the second bias. The classification probability of the attention feature vector is obtained by inputting the third intermediate vector into the Softmax function; The attention type corresponding to the initial EEG signal is determined based on the classification probability, and the signal recognition result is determined.
[0064] For example, embodiments of this application employ a two-layer multilayer perceptron (MLP) network to focus on the attention feature vector. Classification is performed using a two-layer MLP. The processing can be expressed by the following formula: in, , These represent the first weight matrix and the second weight matrix, respectively. , These represent the first bias and the second bias, respectively. The classification probability is used to determine the category corresponding to the maximum value among the classification probabilities, which is then taken as the recognition result.
[0065] For example, to verify the actual performance of the attention recognition method based on EEG signals provided in the embodiments of this application, a controlled experiment was designed and the system's performance was evaluated at the system level. This experiment included various types of tasks to stimulate the subjects' attention in visual, auditory, and cognitive states, while rest tasks were included to allow subjects to recover their energy and perform model classification comparisons. Specific experimental tasks included: d2 elimination task, auditory task, mental arithmetic task, jigsaw puzzle task, and Raven's Reasoning task.
[0066] Specifically, the d2 elimination task is a classic visual selective attention test. Participants are presented with a matrix of target characters and distracting characters, and are required to quickly and accurately identify and eliminate the target characters within a specified time, while ignoring the distracting characters. The auditory task primarily assesses auditory attention. An audio clip containing various sound information is played, including target sounds (such as pure tones of specific frequencies, specific words, etc.) and distracting sounds (such as ambient noise, sounds of other frequencies, etc.). Participants are required to listen attentively to the audio content and respond accordingly. The mental arithmetic task aims to stimulate cognitive attention and working memory. A series of mathematical calculation problems (such as addition, subtraction, multiplication, and division of two-digit numbers) are presented sequentially. Participants are required to mentally calculate and report the results within a specified time without using paper, pen, calculators, or other auxiliary tools. The jigsaw puzzle task combines spatial attention and visual cognition. Participants are presented with a scrambled jigsaw puzzle and asked to assemble the pieces into a complete picture within a specified time. During the puzzle-solving process, participants need to continuously observe the shape, color, and texture of the pieces and analyze the spatial relationships between them. This requires strong spatial attention, visual discrimination, and problem-solving skills. Raven's Reasoning Task is a non-verbal intellectual reasoning task that primarily assesses participants' fluid intelligence and abstract reasoning abilities. It also requires a high level of attention. Participants are presented with a series of reasoning questions composed of different shapes. Each question contains a matrix of missing shapes, and participants are required to select the correct answer from the given options to fill in the missing parts based on the patterns and logical relationships between the shapes.
[0067] The experimental parameters were acquired using the Neuroscan SynAmps2 EEG acquisition system. Electrodes were placed at two locations behind the left and right ears, with the bridge of the nose as the reference point to simulate the real-world scenario of a portable wearable device. The front-end amplification circuit employed a dedicated high-gain differential amplifier module from Neuroscan, effectively suppressing 50Hz power frequency interference and noise from electromyography (EMG) and electrooculography (EOG), significantly improving the signal-to-noise ratio of the EEG signal. The system integrates online impedance monitoring and real-time signal quality assessment functions, allowing real-time monitoring of the electrode contact status in each channel to ensure that the acquired EEG data quality meets the requirements of subsequent algorithm processing and analysis. The sampling rate was set to 500Hz.
[0068] For example, the experimental procedure of this application embodiment is as follows: (a) Experimental training: Before the experiment begins, each subject is informed of the overall experimental procedure to ensure that the subject can complete the overall experiment smoothly and in accordance with the requirements. (b) Electrode placement and quality detection: After the electrodes are placed, the system performs impedance detection and signal quality assessment. The experiment can only begin after all channels have passed the multi-dimensional index check. (c) Attention task: The system will randomly provide any attention task to the subject. After the task, the subject fills in the attention state during the task through a questionnaire to determine the effectiveness of the task. After the task, the subject will enter a rest period to wait for the next task. (d) Data recording: The experiment records the EEG signals of each subject during the task and rest periods, classifies the attention state, and calculates the overall classification accuracy.
[0069] The experiment used to verify the system of this invention collected data from ten subjects. To verify the cross-subject effectiveness of the invention, a leave-one-out method was adopted, meaning the test subject did not participate in training at all. Furthermore, to improve the model's recognition ability, a small number of test subject data were used to fine-tune the cross-subject model to test the invention's learning ability on new data. The results are as follows: Table 1: As shown in Table 1, the system of the present invention achieves an accuracy of 0.66 in the cross-subject attention recognition task, and the accuracy can be increased to 0.76 after fine-tuning with a small amount of data. This indicates that the system has the ability to perform attention recognition using EEG signals with a small number of leads, which meets the usage requirements of wearable products.
[0070] For example, the above method can be implemented as a computer program, which can be used in, for example... Figure 3 It runs on the computer device shown.
[0071] Please see Figure 3 , Figure 3 This is a schematic block diagram illustrating the structure of a computer device provided in an embodiment of this application. The computer device may be a server or a terminal.
[0072] like Figure 3 As shown, the computer device includes a processor, a memory, and a network interface connected via a system bus, wherein the memory may include a storage medium and internal memory.
[0073] The storage medium may store an operating system and a computer program. The computer program includes program instructions that, when executed, cause the processor to perform any attention recognition method based on electroencephalogram (EEG) signals.
[0074] The processor provides computing and control capabilities, supporting the operation of the entire computer device.
[0075] Internal memory provides an environment for the execution of computer programs stored in the storage medium. When the computer program is executed by the processor, it enables the processor to perform any attention recognition method based on electroencephalogram (EEG) signals.
[0076] This network interface is used for network communication, such as sending assigned tasks. Those skilled in the art will understand that... Figure 3 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.
[0077] It should be understood that the processor can be a Central Processing Unit (CPU), but it can also be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among these, a general-purpose processor can be a microprocessor or any conventional processor.
[0078] In one embodiment, the processor is configured to run a computer program stored in memory to perform the following steps: The user's initial EEG signal is obtained by collecting the EEG signals during the user's execution of a preset task. The initial EEG signal is downsampled based on a preset extraction factor to obtain a first initial EEG signal; the first initial EEG signal is notched based on a preset power frequency interference frequency to obtain a second initial EEG signal; the second initial EEG signal is bandpass filtered based on preset electromyography interference frequency and electrooculography interference frequency to obtain a target EEG signal. The target EEG signal is input into the encoder of a preset U-shaped network model for downsampling to obtain the context features corresponding to the target EEG signal; the context features are input into the decoder of the U-shaped network model for upsampling to obtain the reconstructed pseudo EEG signal corresponding to the target EEG signal. The reconstructed pseudo-EEG signal is convolved to obtain the attention feature vector corresponding to the reconstructed pseudo-EEG signal; the attention feature vector is input into at least one multilayer perceptron to obtain the signal recognition result.
[0079] In some embodiments, the step of notch filtering the first initial EEG signal based on a preset power frequency interference frequency to obtain a second initial EEG signal includes: The downsampling rate of the downsampling process is determined based on the ratio of the initial sampling rate of the initial EEG signal to the extraction factor. The normalized center angular frequency is determined based on the ratio of the power frequency interference frequency to the downsampled sampling rate. The interference transfer function of the power frequency interference frequency is calculated based on the normalized center angular frequency and the preset damping coefficient. The first initial EEG signal and the interference transfer function are convolved to obtain a second initial EEG signal after filtering out the power frequency interference.
[0080] In some embodiments, the step of bandpass filtering the second initial EEG signal based on preset electromyography interference frequencies and electrooculography interference frequencies to obtain the target EEG signal includes: Obtain the high-pass cutoff frequency and low-pass cutoff frequency of electromyography (EMG) corresponding to the EMG interference frequency, and the high-pass cutoff frequency and low-pass cutoff frequency of electrooculography (EOG) corresponding to the EOG interference frequency. The electromyography high-pass cutoff frequency and the electromyography low-pass cutoff frequency are converted into electromyography high-pass cutoff angular frequency and electromyography low-pass cutoff angular frequency; the electrooculography high-pass cutoff frequency and the electrooculography low-pass cutoff frequency are converted into electrooculography high-pass cutoff angular frequency and electrooculography low-pass cutoff angular frequency. Based on a preset quality factor, the electromyography high-pass transfer function and the electromyography low-pass transfer function are determined according to the electromyography high-pass cutoff angular frequency and the electromyography low-pass cutoff angular frequency; the electrooculography high-pass transfer function and the electrooculography low-pass transfer function are determined according to the electrooculography high-pass cutoff angular frequency and the electrooculography low-pass cutoff angular frequency. The second initial EEG signal is convolved with the electromyography high-pass transfer function, the electromyography low-pass transfer function, the electrooculography high-pass transfer function, and the electrooculography low-pass transfer function to obtain the target EEG signal after filtering out the electromyography interference frequency and the electrooculography interference frequency.
[0081] In some embodiments, the power frequency interference frequency is 50Hz; the electromyography interference frequency has a high-pass cutoff frequency of 500Hz and a low-pass cutoff frequency of 50Hz; the electrooculography interference frequency has a high-pass cutoff frequency of 3Hz and a low-pass cutoff frequency of 0.1Hz.
[0082] In some embodiments, the step of downsampling the target EEG signal into an encoder of a preset U-shaped network model to obtain contextual features corresponding to the target EEG signal includes: The target EEG signal is input into multiple downsampling blocks of the encoder to obtain the contextual features, wherein each downsampling block includes: a first convolutional layer, an activation function, and a max pooling layer.
[0083] In some embodiments, after inputting the target EEG signal into multiple downsampling blocks of the encoder to obtain the contextual features, the method further includes: The contextual features are input into the bottleneck layer to obtain the target features, and the target features are input into the decoder. The bottleneck layer includes a third convolutional layer.
[0084] In some implementations, the step of upsampling the contextual features into the decoder of the U-shaped network model to obtain the reconstructed pseudo-EEG signal corresponding to the target EEG signal includes: The target features are input into multiple upsampling blocks of the decoder to obtain the reconstructed pseudo-EEG signal. Each upsampling block includes: a transposed convolutional layer, a feature splicing layer, a second convolutional layer, and an activation function. The feature splicing layer is connected to the corresponding first convolutional layer in the downsampling block.
[0085] In some implementations, the step of performing convolution processing on the reconstructed pseudo-EEG signal to obtain the attention feature vector corresponding to the reconstructed pseudo-EEG signal includes: The reconstructed EEG signal is temporally convolved based on a preset temporal convolution kernel to obtain temporal convolution output features, wherein the temporal convolution kernel is a 1×m convolution kernel; The temporal convolution output features are spatially convolved based on a preset spatial convolution kernel to obtain spatial convolution output features, wherein the spatial convolution kernel is an n×1 convolution kernel; The temporal convolutional output features and the spatial convolutional output features are subjected to batch normalization, activation, pooling, and random discarding to obtain the attention feature vector.
[0086] In some implementations, inputting the attention feature vector into at least one multilayer perceptron to obtain the signal recognition result includes: The first intermediate vector corresponding to the attention feature vector is determined based on the first weight matrix and the first bias. The first intermediate vector is input into the ReLU activation function to obtain the second intermediate vector corresponding to the first intermediate vector. The third intermediate vector corresponding to the second intermediate vector is determined based on the second weight matrix and the second bias. The classification probability of the attention feature vector is obtained by inputting the third intermediate vector into the Softmax function; The attention type corresponding to the initial EEG signal is determined based on the classification probability, and the signal recognition result is determined.
[0087] In some embodiments, the acquisition of EEG signals during the user's execution of a preset task to obtain the user's initial EEG signals includes: During the user's performance of high-attention and low-attention tasks, EEG signals from at least one of the user's left and right mastoid regions were collected using 2-4 lead electrodes.
[0088] It should be noted that those skilled in the art will understand that, for the sake of convenience and brevity, the specific working process of the computer device described above can be referred to the corresponding process in the aforementioned embodiment of the attention recognition method based on EEG signals, and will not be repeated here.
[0089] This application also provides a computer-readable storage medium storing a computer program, the computer program including program instructions, and the method implemented when the program instructions are executed can refer to various embodiments of the attention recognition method based on electroencephalogram signals in this application.
[0090] The computer-readable storage medium may be an internal storage unit of the computer device described in the foregoing embodiments, such as the hard disk or memory of the computer device. The computer-readable storage medium may also be an external storage device of the computer device, such as a plug-in hard disk, SmartMedia Card (SMC), Secure Digital (SD) card, or Flash Card equipped on the computer device.
[0091] It should be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the scope of the application. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.
[0092] It should also be understood that the term "and / or" as used in this specification and the appended claims refers to any combination and all possible combinations of one or more of the associated listed items, and includes such combinations. It should be noted that, herein, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or system that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or system. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or system that includes that element.
[0093] The sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments. The above descriptions are merely specific implementations of this application, but the scope of protection of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and these modifications or substitutions should all be covered within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.
Claims
1. A method for attention recognition based on electroencephalogram signals, characterized in that, include: The user's initial EEG signal is obtained by collecting the EEG signals during the user's execution of a preset task. The initial EEG signal is downsampled based on a preset extraction factor to obtain a first initial EEG signal. The first initial EEG signal is notched and filtered based on a preset power frequency interference frequency to obtain a second initial EEG signal. The second initial EEG signal is bandpass filtered based on the preset electromyography interference frequency and electrooculography interference frequency to obtain the target EEG signal; The target EEG signal is input into the encoder of a preset U-shaped network model for downsampling to obtain the context features corresponding to the target EEG signal; the context features are input into the decoder of the U-shaped network model for upsampling to obtain the reconstructed pseudo EEG signal corresponding to the target EEG signal. The reconstructed pseudo-EEG signal is convolved to obtain the attention feature vector corresponding to the reconstructed pseudo-EEG signal; The attention feature vector is input into at least one multilayer perceptron to obtain the signal recognition result.
2. The attention recognition method based on electroencephalogram (EEG) signals according to claim 1, characterized in that, The second initial EEG signal is obtained by notch filtering the first initial EEG signal based on a preset power frequency interference frequency, including: The downsampling rate of the downsampling process is determined based on the ratio of the initial sampling rate of the initial EEG signal to the extraction factor. The normalized center angular frequency is determined based on the ratio of the power frequency interference frequency to the downsampled sampling rate. The interference transfer function of the power frequency interference frequency is calculated based on the normalized center angular frequency and the preset damping coefficient. The first initial EEG signal and the interference transfer function are convolved to obtain a second initial EEG signal after filtering out the power frequency interference.
3. The attention recognition method based on electroencephalogram (EEG) signals according to claim 1, characterized in that, The method of bandpass filtering the second initial EEG signal based on preset electromyography interference frequencies and electrooculography interference frequencies to obtain the target EEG signal includes: Obtain the high-pass cutoff frequency and low-pass cutoff frequency of electromyography (EMG) corresponding to the EMG interference frequency, and the high-pass cutoff frequency and low-pass cutoff frequency of electrooculography (EOG) corresponding to the EOG interference frequency. The electromyography (EMG) high-pass cutoff frequency and the EMG low-pass cutoff frequency are converted into EMG high-pass cutoff angular frequency and EMG low-pass cutoff angular frequency; the electrooculography (EOG) high-pass cutoff frequency and the EOG low-pass cutoff frequency are converted into EOG high-pass cutoff angular frequency and EOG low-pass cutoff angular frequency. Based on a preset quality factor, the high-pass transfer function and low-pass transfer function of electromyography are determined according to the high-pass cutoff angular frequency and the low-pass cutoff angular frequency of electromyography; the high-pass transfer function and low-pass transfer function of electrooculography are determined according to the high-pass cutoff angular frequency and the low-pass cutoff angular frequency of electrooculography. The second initial EEG signal is convolved with the electromyography high-pass transfer function, the electromyography low-pass transfer function, the electrooculography high-pass transfer function, and the electrooculography low-pass transfer function to obtain the target EEG signal after filtering out the electromyography interference frequency and the electrooculography interference frequency.
4. The attention recognition method based on electroencephalogram (EEG) signals according to any one of claims 1-3, characterized in that, The power frequency interference frequency is 50Hz; the electromyography interference frequency has a high-pass cutoff frequency of 500Hz and a low-pass cutoff frequency of 50Hz; the electrooculography interference frequency has a high-pass cutoff frequency of 3Hz and a low-pass cutoff frequency of 0.1Hz.
5. The attention recognition method based on electroencephalogram (EEG) signals according to claim 1, characterized in that, The step of downsampling the target EEG signal into the encoder of a preset U-shaped network model to obtain the contextual features corresponding to the target EEG signal includes: The target EEG signal is input into multiple downsampling blocks of the encoder to obtain the contextual features, wherein each downsampling block includes: a first convolutional layer, an activation function, and a max pooling layer.
6. The attention recognition method based on electroencephalogram (EEG) signals according to claim 5, characterized in that, After inputting the target EEG signal into multiple downsampling blocks of the encoder to obtain the contextual features, the method further includes: The contextual features are input into the bottleneck layer to obtain the target features, and the target features are input into the decoder. The bottleneck layer includes a third convolutional layer.
7. The attention recognition method based on electroencephalogram (EEG) signals according to claim 6, characterized in that, The step of upsampling the contextual features into the decoder of the U-shaped network model to obtain the reconstructed pseudo-EEG signal corresponding to the target EEG signal includes: The target features are input into multiple upsampling blocks of the decoder to obtain the reconstructed pseudo-EEG signal. Each upsampling block includes: a transposed convolutional layer, a feature splicing layer, a second convolutional layer, and an activation function. The feature splicing layer is connected to the corresponding first convolutional layer in the downsampling block.
8. The attention recognition method based on electroencephalogram (EEG) signals according to claim 1, characterized in that, The step of performing convolution processing on the reconstructed pseudo-EEG signal to obtain the attention feature vector corresponding to the reconstructed pseudo-EEG signal includes: The reconstructed pseudo-EEG signal is temporally convolved based on a preset temporal convolution kernel to obtain temporal convolution output features, wherein the temporal convolution kernel is a 1×m convolution kernel; The temporal convolution output features are spatially convolved based on a preset spatial convolution kernel to obtain spatial convolution output features, wherein the spatial convolution kernel is an n×1 convolution kernel; The temporal convolutional output features and the spatial convolutional output features are subjected to batch normalization, activation, pooling, and random discarding to obtain the attention feature vector.
9. The attention recognition method based on electroencephalogram (EEG) signals according to claim 1, characterized in that, The step of inputting the attention feature vector into at least one multilayer perceptron to obtain the signal recognition result includes: The first intermediate vector corresponding to the attention feature vector is determined based on the first weight matrix and the first bias. The first intermediate vector is input into the ReLU activation function to obtain the second intermediate vector corresponding to the first intermediate vector. The third intermediate vector corresponding to the second intermediate vector is determined based on the second weight matrix and the second bias. The classification probability of the attention feature vector is obtained by inputting the third intermediate vector into the Softmax function; The attention type corresponding to the initial EEG signal is determined based on the classification probability, and the signal recognition result is determined.
10. The attention recognition method based on electroencephalogram (EEG) signals according to claim 1, characterized in that, The process of collecting EEG signals during the user's execution of a preset task to obtain the user's initial EEG signals includes: During the user's performance of high-attention and low-attention tasks, EEG signals from at least one of the user's left and right mastoid regions were collected using 2-4 lead electrodes.