Emotion detection method for electroencephalogram signals based on multi-stage feature fusion

Abstract

The present application relates to the field of medical signal processing and the field of artificial intelligence, in particular to an electroencephalogram signal emotion detection method based on multi-stage feature fusion, which is to construct a first electroencephalogram signal feature extraction network to extract frequency domain features with time sequence electroencephalogram signals as input; a second electroencephalogram signal feature extraction network is constructed with differential entropy features as input, which is to extract spatial and time domain features, spatial and channel features and time domain features by introducing a spatial attention mechanism with EEGNet as the backbone network; a third electroencephalogram signal feature extraction network is constructed to extract second statistical features with first statistical features as input; after obtaining the fusion features by fusing the five kinds of features, an electroencephalogram signal recognition network is used for recognition to obtain the emotion detection result output. The present application solves the problem that the existing electroencephalogram signal emotion detection cannot fully utilize the information of space, time domain, frequency domain, channel and the like in the electroencephalogram signal.

Description

Technical Field

[0001] This invention relates to the fields of medical signal processing and artificial intelligence, specifically to a method for emotion detection based on multi-stage feature fusion of electroencephalogram (EEG) signals. Background Technology

[0002] Emotion, as a comprehensive expression, integrates human will and behavior, largely reflecting a person's attitude and cognition towards things, and also playing a significant role in influencing human behavior and reactions. In advanced human-computer interaction systems, accurate emotion recognition has become a key research area for achieving harmonious and personalized human-computer interaction, sparking a research boom in neuroscience, psychology, and clinical applications. In the field of intelligent driving, because emotions have a significant impact on driver behavior and driving safety, and because the signals collected during driving have unique characteristics, driver emotion detection has become a rather special part of the field of emotion recognition.

[0003] In the specific field of driver emotion recognition, it can be mainly divided into two categories: one is recognition based on non-physiological signals, and the other is recognition based on physiological signals. Recognition based on non-physiological signals primarily focuses on facial expressions and voice tone. However, these signals suffer from poor reliability because they can be deliberately obscured, and some individuals have expression difficulties in this area, all of which lead to inaccurate recognition. Recognition based on physiological signals utilizes the detection of human physiological signals such as electroencephalogram (EEG), electrooculogram (EOG), electromyogram (EMG), and electrocardiogram (ECG) signals to further study their correlation with human emotions. Among these, EEG signals are particularly special, considered the "gold standard" for emotion recognition because they encompass a wealth of information about a person's mental activity and physiological state, providing strong evidence for emotion detection. However, current research largely relies on traditional machine learning methods, such as Support Vector Machines (SVM) and K-Nearest Neighbors (KNN), or generalized deep learning networks such as Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM) networks for emotion detection. This type of research ignores the unique spatial structure of EEG signals—the spatial discontinuity between channels and the strong coupling relationship, which wastes a large amount of data information contained in EEG signals and reduces the effectiveness of recognition.

[0004] In the prior art, Chinese patent application CN116509392A discloses an emotion detection and recognition method based on EEG and convolutional neural networks. This method uses a GCN-NGCU structure to extract differential entropy features and power spectral density features from manually extracted EEG signals, and finally outputs emotion classification through a fully connected layer. Chinese patent application CN117204856A discloses a method for detecting emotions using popular features. After calculating differential entropy features from EEG signals, features are extracted through channel feature extraction and a spatial popular mapping module, ultimately outputting an emotion classification. However, in these two inventions, research on the channel features of EEG signals is still lacking. Either the extraction of channel features is neglected, or research is only conducted on information within a single channel, ignoring the mutual coupling and influence between different channels in the EEG signal, failing to fully utilize the time-domain, frequency-domain, and channel information contained in the EEG signal. Furthermore, the current approach to driver emotion detection relies solely on classification, neglecting the objective fact that only extreme emotions influence driving behavior, leading to practical limitations. To address these issues, this invention designs a specialized network structure for extracting features specific to different domains. Through feature scale matching and fusion modules, these features collectively achieve a continuous assessment of the driver's emotions. Summary of the Invention

[0005] In view of this, the present invention provides a method for emotion detection of EEG signals based on multi-stage feature fusion, in order to overcome the above problems.

[0006] The specific technical solution of the present invention is as follows:

[0007] A method for emotion detection based on multi-stage feature fusion of electroencephalogram (EEG) signals includes the following steps:

[0008] Step 1: Acquire and process EEG signal data to obtain differential entropy features, temporal EEG signals, and first statistical features used as input vectors;

[0009] Step 2: Use an EEG signal feature extraction network to extract and fuse EEG signal features:

[0010] The EEG signal feature extraction network includes a first EEG signal feature extraction network, a second EEG signal feature extraction network, a third EEG signal feature extraction network, and a feature scale matching and fusion module, wherein:

[0011] The input of the first EEG signal feature extraction network is a time-series EEG signal, and the output is frequency domain features;

[0012] The input to the second EEG signal feature extraction network is the differential entropy feature used as the input vector, and the output is spatial and channel features, as well as temporal features.

[0013] The third EEG signal feature extraction network takes the first statistical feature as input and outputs the second statistical feature.

[0014] The feature scale matching and fusion module receives frequency domain features, spatial features, time domain features, and second statistical features, and fuses them to obtain multi-dimensional fused features;

[0015] Step 3: The EEG signal recognition network processes the fusion features to obtain the emotion detection result. The emotion detection result is a continuous value from -1 to 1 for the subject's emotional state from negative emotion, neutral emotion to positive emotion; where -1 indicates that the subject is in a negative emotion, 1 indicates that the subject is in a positive emotion, and 0 indicates that the subject is in a neutral emotion. Negative emotions are fear or sadness, and positive emotions are happiness or excitement.

[0016] Furthermore, the EEG signal emotion detection method based on multi-stage feature fusion also includes steps 4 and 5;

[0017] Step 4: Train and optimize the EEG signal feature extraction network and the EEG signal recognition network to obtain the EEG signal emotion detection model;

[0018] Step 5: Input the EEG signal to be detected into the EEG signal emotion detection model for detection, and output the emotion detection result of the EEG signal.

[0019] Furthermore, in step 2, when the EEG signal feature extraction network performs the fusion operation on frequency domain signal features, spatial and channel features, temporal features, and second statistical features:

[0020] If the difference in the length of the feature vectors of the five types of features—time-domain and frequency-domain signal features, second statistical features, spatial features, and channel features—is within two orders of magnitude, feature fusion is performed. If the difference in the length of the feature vectors of the time-domain and frequency-domain signal features, spatial features, channel features, and second statistical features exceeds two orders of magnitude, then dimensionality reduction processing is required for the corresponding feature vectors.

[0021] Furthermore, the method for obtaining differential entropy features by acquiring EEG signal data and processing it in step 1 is as follows:

[0022] The sampling rate of EEG signal data is standardized, and non-overlapping sliding windows are used to segment the EEG signal data after standardization. Then, normalization is performed on each segment.

[0023] Within a single slice, a fifth-order Butterworth filter is used to filter signals in five frequency bands: δ (1Hz~4Hz), θ (4Hz~8Hz), α (8Hz~14Hz), β (14Hz~31Hz), and γ (31Hz~50Hz), to obtain signals in five frequency bands.

[0024] The differential entropy features of each frequency band signal are calculated and extracted separately. All differential entropy features extracted from a single channel are then concatenated and combined to generate differential entropy features that can be used as input vectors.

[0025] Furthermore, the first statistical features extracted in step 1 include: mean amplitude (mean), kurtosis (K), standard deviation (std), waveform factor (wf), impulse factor (f), and frequency domain power spectral density (PSD) of the time-series signal.

[0026] Furthermore, the second EEG signal feature extraction network in step 2 is a second EEG signal feature extraction network structure composed of EEGNet as the backbone network and by introducing a spatial attention mechanism and a temporal feature extraction module.

[0027] Furthermore, in the EEG signal feature extraction network:

[0028] The first EEG signal feature extraction network consists of a frequency domain feature calculation module and a frequency domain feature extraction module. The frequency domain feature calculation module receives time-series EEG signals and transforms them to the frequency domain to obtain frequency domain EEG signals. The frequency domain feature extraction module is used to extract frequency domain features from the received frequency domain EEG signals and send the frequency domain features to the feature fusion module.

[0029] The second EEG signal feature extraction network consists of an EEGNet module, a spatial attention module, and a temporal feature extraction module. Specifically: the EEGNet module receives the differential entropy used as the input vector and performs dimensionality reduction on it, then sends the dimensionality-reduced differential entropy to both the spatial attention module and the temporal feature extraction module; the spatial attention feature extraction module uses a spatial attention mechanism to extract spatial and channel features from the dimensionality-reduced differential entropy, then sends these features to the feature fusion module; and the temporal feature extraction module extracts temporal features from the dimensionality-reduced differential entropy, then sends these temporal features to the feature fusion module.

[0030] The third EEG signal feature extraction network consists of a statistical feature calculation module and a statistical feature extraction module. The statistical feature calculation module performs statistics on the received first statistical features and sends the statistical feature results to the statistical feature extraction module. The statistical feature extraction module extracts the second statistical features from the received statistical feature results and sends the second statistical features to the feature fusion module.

[0031] Furthermore, the EEG signal recognition network includes a vector flattening layer, a fully connected layer, an activation function layer, and a fully connected layer.

[0032] Furthermore, in step 4, when training the EEG signal feature extraction network and the EEG signal recognition network, the input training data is based on the EEG data collected in step 1. The data is expanded by adding noise signals actually collected in the public dataset using an overlapping sliding window with a length of 3s and a sliding distance of 1s.

[0033] This invention presents a multi-stage feature fusion-based EEG signal emotion detection method. By processing the acquired EEG signals, it obtains differential entropy features, temporal EEG signals, and a first statistical feature, which serve as the input vector. The differential entropy features are extracted by slicing the acquired EEG signal data into a sliding window, filtering and normalizing the data within each slice. Then, a first EEG signal feature extraction network is constructed using the temporal EEG signal as input to extract frequency domain features; a second EEG signal feature extraction network is constructed using the differential entropy features as input. This second network, with EEGNet as its backbone, extracts spatial, channel, and temporal features by introducing a spatial attention mechanism. After fusing these five features to obtain a fused feature, an EEG signal recognition network is used to recognize the fused feature, resulting in an emotion detection output. This invention addresses the problem that existing EEG signal emotion detection methods fail to fully utilize the spatial, temporal, frequency, and channel information within EEG signals.

[0034] Due to the adoption of the above-mentioned design scheme, the present invention has the following beneficial effects:

[0035] 1. The differential entropy data used as the input vector in this invention employs a sliding window method to slice the acquired EEG signal data, and removes outliers and normalizes the data within the slices, thus preserving the data distribution characteristics while reducing the impact of noise.

[0036] 2. In the EEG signal feature extraction network, multiple feature extraction branches and the addition of statistical features improve the overall performance of the model. In the second EEG signal feature extraction network, a spatial attention mechanism is introduced into EEGNE to extract channel and spatial features, and a temporal feature extraction module is introduced to extract temporal features. Combined with the idea of ​​multi-stage feature fusion, the extracted channel features, spatial features, temporal features, frequency domain features, and statistical features are fused. This better preserves the effective information of the data, thereby improving the accuracy of recognition.

[0037] 3. In the feature fusion process, the feature scale matching and fusion module uses the similarity of features in terms of scale and dimension to promote feature matching by applying normalization, one-dimensional convolution and other methods, thereby avoiding the "long tail" problem in multi-feature fusion recognition and improving algorithm performance.

[0038] 4. During the training process, this invention uses a partially overlapping sliding window method to expand the training data and improve the robustness of the model. Attached Figure Description

[0039] Figure 1 Here is a flowchart of an example of an EEG signal emotion detection method based on multi-stage feature fusion;

[0040] Figure 2 This is a flowchart of the differential entropy feature extraction process used as the input vector in the embodiment;

[0041] Figure 3 The diagram shows the structure of the EEG signal recognition network in this embodiment. Detailed Implementation

[0042] The following detailed description of each module of the present invention is provided with reference to the accompanying drawings, but the embodiments of the present invention are not limited thereto.

[0043] like Figure 1 As shown in the figure, this embodiment provides a method for emotion detection based on multi-stage feature fusion of electroencephalogram (EEG) signals, including step 1, acquiring and processing EEG signals; step 2, constructing an EEG signal feature extraction network to extract features from the EEG signals; and step 3, using an EEG signal recognition network to perform emotion recognition on the features of the EEG signal input, thereby obtaining the emotion detection result. The specific descriptions of each step are as follows:

[0044] Step 1: Acquiring and Processing EEG Signal Data: In this embodiment, the EEG signal acquired by the driver is used as the data. During signal acquisition, the subject is placed in a simulated driving environment created by a 6D driving simulator. Emotions are induced in the subject by playing videos, and simultaneously, an EEG acquisition device with a sampling frequency of 1000Hz and 32 channels is used to acquire the EEG signals. However, this embodiment is also applicable to emotion recognition using driver EEG signals acquired through other acquisition methods.

[0045] Differential entropy features, temporal EEG signals, and first statistical features are obtained from the acquired EEG signals and used as input vectors; where:

[0046] Differential entropy features used as input vectors are obtained as follows Figure 2 As shown, it includes the following steps:

[0047] The acquired EEG signal data were uniformly adjusted to 200Hz, and samples with a sampling rate higher than 200Hz were downsampled, as shown in the following formula:

[0048] x dn (n)=x(nM) (1)

[0049] In equation (1), x dn (n) represents the data of the nth sample point of the downsampled signal, and x(nM) represents the data of the nMth sample point of the original signal, where M is the downsampling factor.

[0050] After the same sampling rate, a sliding window of size 3s is used to perform non-overlapping segmentation. The maximum and minimum values ​​are recorded within each segment, and normalization is performed on each individual segment. The normalization formula is shown in (2):

[0051]

[0052] In equation (2), after represents the normalized data, before represents the original data, and max and min represent the maximum and minimum values ​​of the data in a single data slice, respectively.

[0053] Within a single slice, a fifth-order Butterworth filter is used to filter signals in five frequency bands: δ (1Hz–4Hz), θ (4Hz–8Hz), α (8Hz–14Hz), β (14Hz–31Hz), and γ (31Hz–50Hz). The Butterworth filter can be expressed by the following formula:

[0054]

[0055] In equation (3), |G(ω)| 2 Let ω be the square of the amplitude of the filtered signal, ω be the frequency, and n be the filter order. C This is the cutoff frequency.

[0056] After obtaining signals from five frequency bands, the differential entropy of the signal in each band was calculated every 0.5 seconds. A total of six differential entropy features were calculated within a 3-second time window for a single signal recognition, resulting in 30 differential entropy features across the five frequency bands. These 30 differential entropy features extracted from a single channel were concatenated to generate a final input vector of size N*32*30, where N is the number of samples per epoch. Sampling was performed using a 32-channel EEG sampling device with internationally recognized 10-20 EEG sampling electrodes.

[0057] This study is the first to extract temporal EEG signal features from EEG signals using a time-domain analysis method.

[0058] The first statistical features are obtained, including: mean amplitude (mean), kurtosis (K), standard deviation (std), waveform factor (wf), impulse factor (f), and power spectral density (PSD) of the time-series signal. Their main calculation methods are as follows:

[0059]

[0060] In the formula, N is the number of sample points, and x(i) is the sample point value.

[0061]

[0062]

[0063] In the formula, f is the sampling frequency.

[0064] Step 2: Use an EEG signal feature extraction network to extract and fuse EEG signal features:

[0065] The EEG signal feature extraction network includes a first EEG signal feature extraction network, a second EEG signal feature extraction network, a third EEG signal feature extraction network, and a feature scale matching and fusion module, wherein:

[0066] The first EEG signal feature extraction network takes a time-series EEG signal as input and outputs frequency domain features. The network consists of a frequency domain feature calculation module and a frequency domain feature extraction module. The frequency domain feature calculation module receives the time-series EEG signal and transforms it to the frequency domain to obtain a frequency domain EEG signal. The frequency domain feature extraction module extracts frequency domain features from the received frequency domain EEG signal and sends these features to the feature fusion module.

[0067] The second EEG signal feature extraction network takes differential entropy features as input vectors as input and outputs spatial and channel features, as well as temporal features. The network consists of an EEGNet module, a spatial attention module, and a temporal feature extraction module. Specifically: the EEGNet module receives the differential entropy used as input vectors and performs dimensionality reduction on it, sending the reduced entropy to both the spatial attention module and the temporal feature extraction module; the spatial attention feature extraction module uses a spatial attention mechanism to extract spatial and channel features from the reduced entropy and sends these features to the feature fusion module; and the temporal feature extraction module extracts temporal features from the reduced entropy and sends these features to the feature fusion module.

[0068] The third EEG signal feature extraction network takes a first statistical feature as input and outputs a second statistical feature. The network consists of a statistical feature calculation module and a statistical feature extraction module. The statistical feature calculation module performs statistical analysis on the received first statistical feature and sends the result to the statistical feature extraction module. The statistical feature extraction module extracts the second statistical feature from the received result and sends it to the feature fusion module.

[0069] The feature fusion module is used to fuse temporal features, frequency features, spatial features, statistical information features, and spatial channel features to obtain fused features, which are then output to the EEG signal recognition network for emotion detection.

[0070] Table 1. Actual structure and parameters of each part of the EEG signal feature extraction network

[0071]

[0072]

[0073] It should be noted that in the EEG signal feature extraction network, its feature scale matching and fusion module consists of a feature dimension determination and fusion layer. By judging the magnitude of the vector features input to this layer and performing splicing operations, the splicing and fusion of EEG signal features extracted at different stages are achieved, improving recognition accuracy. That is, the feature fusion module in this embodiment can achieve adaptive feature fusion. Feature fusion is performed when the difference in the length of the feature vectors input to the feature fusion module is within two orders of magnitude. When the difference in the length of the feature vectors input to the feature fusion module is greater than two orders of magnitude, the effect of a certain type of feature is too low or cancels out, and the dimensionality of the input feature vectors is reduced. It should be noted that the measurement of magnitude is based on the dimension of all features before final fusion; in this embodiment, the lowest dimension is used as the standard for measurement.

[0074] Step 3: The EEG signal recognition network processes the fused features to obtain the emotion detection result. The implementation method is as follows:

[0075] A brainwave signal recognition network was constructed. The network consists of a vector flattening layer, a first fully connected (FC1) layer, an activation function layer (ReLU), and a second fully connected (FC2) layer; wherein:

[0076] The vector flattening layer is used to expand the fused features output by the EEG feature extraction network into one-dimensional features, so that the subsequent two fully connected layers can perform operations.

[0077] The first fully connected layer has parameters of 806*60, and the second fully connected layer has parameters of 60*1. ReLU is chosen as the activation function to achieve the network's non-linear performance.

[0078] This embodiment uses an EEG signal recognition network based on the results of an EEG signal feature extraction network to accurately identify five features: spatial, channel, temporal, frequency, and statistical features, thereby achieving continuous assessment of the driver's emotions.

[0079] Regarding the selection of the loss function, this embodiment uses the RMSE loss function and the Adam optimizer to train the EEG signal feature extraction network and the EEG signal recognition network to obtain the EEG signal emotion detection model. The RMSE loss function is calculated as follows:

[0080]

[0081] The network was trained using the Adam optimizer, with a maximum training epoch of 128 and an initial learning rate of 1e-3. The loss function (Loss) was calculated and backpropagated into the network for optimization. Training stopped when the loss function converged or the maximum number of iterations was reached.

[0082] The EEG signal emotion detection model includes a trained EEG signal feature extraction network and an EEG signal recognition network. The EEG signal emotion detection model can be used to detect the driver's emotions.

[0083] It is understood that the present invention has been described through some embodiments, and those skilled in the art will recognize that various changes or equivalent substitutions can be made to these features and embodiments without departing from the spirit and scope of the invention. Furthermore, under the teachings of the present invention, these features and embodiments can be modified to adapt to specific situations and materials without departing from the spirit and scope of the invention. Therefore, the present invention is not limited to the specific embodiments disclosed herein, and all embodiments falling within the scope of the claims of this application are within the protection scope of the present invention.

Claims

1. A method for emotion detection based on multi-stage feature fusion of electroencephalogram (EEG) signals, characterized in that, The method includes: Step 1: Acquire and process EEG signal data to obtain differential entropy features, temporal EEG signals, and first statistical features used as input vectors; Step 2: Use an EEG signal feature extraction network to extract and fuse EEG signal features: The EEG signal feature extraction network includes a first EEG signal feature extraction network, a second EEG signal feature extraction network, a third EEG signal feature extraction network, and a feature scale matching and fusion module, wherein: The first EEG signal feature extraction network takes time-series EEG signals as input and outputs frequency domain features. It consists of a frequency domain feature calculation module and a frequency domain feature extraction module. The frequency domain feature calculation module receives the time-series EEG signals and transforms them to the frequency domain to obtain frequency domain EEG signals. The frequency domain feature extraction module is used to extract frequency domain features from the received frequency domain EEG signals and send the frequency domain features to the feature fusion module. The second EEG signal feature extraction network takes differential entropy features as input vectors as input and outputs spatial and channel features, as well as temporal features. It consists of an EEGNet module, a spatial attention module, and a temporal feature extraction module. Specifically: the EEGNet module receives the differential entropy used as the input vector and performs dimensionality reduction on it, sending the dimensionality-reduced differential entropy to the spatial attention module and the temporal feature extraction module respectively; the spatial attention feature extraction module uses a spatial attention mechanism to extract spatial and channel features from the dimensionality-reduced differential entropy and sends these features to the feature fusion module; the temporal feature extraction module extracts temporal features from the dimensionality-reduced differential entropy and sends these temporal features to the feature fusion module. The input to the third EEG signal feature extraction network is the first statistical feature, and the output is the second statistical feature; The feature scale matching and fusion module receives frequency domain features, spatial features, temporal features, and a second statistical feature, and fuses them to obtain multi-dimensional fused features. It consists of a statistical feature calculation module and a statistical feature extraction module, wherein: the statistical feature calculation module performs statistics on the received first statistical feature and sends the statistical feature result to the statistical feature extraction module; the statistical feature extraction module is used to extract the second statistical feature from the received statistical feature result and sends the second statistical feature to the feature fusion module. When the EEG signal feature extraction network performs fusion operations on frequency domain signal features, spatial and channel features, time domain features, and second statistical features: if the difference in the length of the feature vectors of the five types of features (time domain and frequency domain signal features, second statistical features, spatial and channel features, and second statistical features) is within two orders of magnitude, feature fusion is performed; if the difference in the length of the feature vectors of the time domain and frequency domain signal features, spatial and channel features, and second statistical features exceeds two orders of magnitude, then dimensionality reduction processing is required for the corresponding feature vectors. Step 3: The EEG signal recognition network processes the fusion features to obtain the emotion detection result. The emotion detection result is a continuous value from -1 to 1 for the subject's emotional state from negative emotion, neutral emotion to positive emotion; where -1 indicates that the subject is in a negative emotion, 1 indicates that the subject is in a positive emotion, and 0 indicates that the subject is in a neutral emotion. Negative emotions are fear or sadness, and positive emotions are happiness or excitement.

2. The method for emotion detection based on multi-stage feature fusion of EEG signals according to claim 1, characterized in that: The EEG signal emotion detection method based on multi-stage feature fusion also includes steps 4 and 5; Step 4: Train and optimize the EEG signal feature extraction network and the EEG signal recognition network to obtain the EEG signal emotion detection model; Step 5: Input the EEG signal to be detected into the EEG signal emotion detection model for detection, and output the emotion detection result of the EEG signal.

3. The method for emotion detection based on multi-stage feature fusion of EEG signals according to claim 1, characterized in that: The method for obtaining differential entropy features by acquiring EEG signal data and processing it in step 1 is as follows: The sampling rate of EEG signal data is standardized, and non-overlapping sliding windows are used to segment the EEG signal data after standardization. Then, normalization is performed on each segment. Within each slice, a fifth-order Butterworth filter is used for each... as well as Filtering of five frequency band signals to obtain signals in five frequency bands; The differential entropy features of each frequency band signal are calculated and extracted separately. All differential entropy features extracted from a single channel are then concatenated and combined to generate differential entropy features that can be used as input vectors. The first statistical features extracted in step 1 include: the average amplitude, kurtosis, standard deviation, waveform factor, impulse factor, and frequency domain power spectral density of the time-series signal.

4. The method for emotion detection based on multi-stage feature fusion of EEG signals according to claim 1, characterized in that: The second EEG signal feature extraction network in step 2 is a second EEG signal feature extraction network structure composed of EEGNet as the backbone network and the introduction of spatial attention mechanism and temporal feature extraction module.

5. The method for detecting emotion using EEG signals based on multi-stage feature fusion according to claim 1, characterized in that: The EEG signal recognition network includes a vector flattening layer, a fully connected layer, an activation function layer, and a fully connected layer.

6. The method for emotion detection based on multi-stage feature fusion of EEG signals according to claim 2, characterized in that: In step 4, when training the EEG signal feature extraction network and the EEG signal recognition network, the input training data is based on the EEG data collected in step 1. The data is expanded by adding noise signals actually collected in the public dataset using an overlapping sliding window with a length of 3s and a sliding distance of 1s.

Images