EEG signal detection method based on separating temporal correlation and auditory interference

Abstract

A method for detecting EEG signals based on separating temporal correlation and auditory interference collects unlabeled EEG data, pre-trains it using a multi-view spectral-spatial-temporal masked autoencoder model to obtain spectral-spatial-temporal features, and then fine-tunes the EEG data using a linear classifier to determine its specific state. This method effectively eliminates temporal and auditory interference, achieving high-precision classification of meditation states. The following describes the implementation of this technical solution in detail, focusing on the innovative experimental design and data collection methods, as well as the accompanying self-supervised learning algorithm.

Description

Technical Field

[0001] The present invention relates to a technology in the field of electroencephalogram (EEG) signal processing, in particular to an EEG signal detection method based on separating time correlation and auditory interference. Background Art

[0002] Meditation state monitoring has been a key research topic in biomedical informatics in recent years. Traditional methods primarily rely on the analysis of electroencephalogram (EEG) data. Existing technologies typically use feature engineering and machine learning methods to distinguish between meditative and non-meditative states. These methods primarily rely on pre-labeled datasets for training and validation. However, traditional meditation state monitoring methods face two significant challenges: first, temporal interference, where EEG data collected at different time points are highly correlated, potentially causing the model to overfit to temporal features rather than the meditation state itself; and second, auditory interference, where audio-guided meditation affects the characteristics of EEG signals, making it difficult to distinguish between EEG signals with and without audio guidance. Summary of the Invention

[0003] To address the aforementioned shortcomings of the existing technology, the present invention proposes an EEG signal detection method based on separating temporal correlation and auditory interference. This method effectively eliminates temporal and auditory interference, enabling high-precision meditation state classification. The following describes the implementation of this technical solution in detail, focusing on the innovative experimental design and data acquisition methods, as well as the accompanying self-supervised learning algorithm.

[0004] The present invention is achieved through the following technical solutions:

[0005] The present invention relates to an EEG signal detection method based on separating temporal association and auditory interference. By collecting unlabeled EEG data, a multi-view spectrum-space-time mask autoencoder model is pre-trained to obtain spectrum-space-time features, and then a linear classifier is fine-tuned to obtain the specific state of the EEG data.

[0006] The present invention relates to a system for implementing the above-mentioned method, comprising: a state labeling unit, an EEG acquisition unit, and a meditation state analysis unit, wherein: the state labeling unit plays meditation prompts in alternation according to a pre-set meditation prompt program, and adds additional data to the collected EEG data to clarify the environment and conditions for collecting EEG at that moment; the EEG acquisition unit is composed of an EEG cap and a signal amplifier and is associated with the state labeling unit, and collects EEG data in real time according to the program shown in the state labeling unit, and collects a section of EEG data containing 64 channels; the meditation state analysis unit performs data analysis on the collected EEG data, performs preprocessing such as noise removal, and then adopts a deep learning method including a self-supervised learning mechanism to obtain the result of real-time detection of meditation state of EEG state.

[0007] Technical Effects

[0008] The present invention introduces multiple data collections and conducts training and testing between different collections; by including meditation experiments with and without audio guidance, analyzing the differences in EEG signals under different conditions, the model can better distinguish different meditation states, thereby improving the robustness of the model under various experimental conditions. Through self-supervised learning technology, the model can be pre-trained on a large amount of unlabeled EEG data, thereby learning a wider range of more general features. Compared with the existing technology, the present invention can effectively reduce the impact of time correlation on EEG signal classification, thereby improving the accuracy of meditation state classification. It reduces the dependence on labeled data and enhances the adaptability of the model under different experimental conditions and different individuals. The emotional EEG dataset SEED, which is unrelated to the meditation classification task, is used for pre-training to improve performance on the meditation detection task. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] Figure 1 This is a data collection flow chart of the present invention;

[0010] Figure 2 This is a flow chart of EEG data processing in the present invention;

[0011] Figure 3 Schematic diagram of the embodiment effect. DETAILED DESCRIPTION

[0012] like Figure 1 and Figure 2 As shown, this embodiment relates to an EEG signal detection method based on separating time correlation and auditory interference, including:

[0013] Step 1: Multiple Data Collection: Each subject participated in three independent experiments, separated by 14 days. During each collection, the subject engaged in audio-guided meditation, non-audio-guided meditation, and a resting state. The time interval between each collection effectively reduced the temporal correlation of EEG signals between different states, preventing the model from overfitting temporal features and thus improving classification accuracy.

[0014] In the data collection process, the subjects alternated between audio-guided meditation and non-audio-guided meditation to explore the impact of auditory interference on EEG signals in the meditation state. Specifically, each collection included 5 minutes of resting state, about 20 minutes of meditation state, and another 5 minutes of resting state. The meditation stage included alternating audio guidance with prompts and audio guidance without prompts. The subjects entered the meditation state through audio instructions, and in the non-audio guidance stage, the subjects were asked to continue meditating in a quiet environment. Through this alternating design, EEG signals under different conditions can be collected separately, so as to more accurately analyze the impact of auditory interference on the meditation state.

[0015] Step 2: Pre-train the multi-view spectral-spatial-temporal masked autoencoder (MV-SSTMA) model on the unlabeled EEG dataset obtained in Step 1 to obtain universal spectral-spatial-temporal features. Specifically, the MV-SSTMA algorithm uses a CNN-Transformer-based architecture. It first randomly masks the unlabeled EEG data during a self-supervised pre-training phase. It then utilizes multi-scale causal convolutional layers and a multi-head self-attention mechanism to extract spectral, spatial, and temporal features. During the pre-training phase, the algorithm uses the unlabeled EEG data to learn universal features. Random masking divides the EEG data into a visible subset and a masked subset. The model reconstructs the masked data to capture spectral, spatial, and temporal information, resulting in features with enhanced generalization. Key operations include projecting the input into a D-dimensional spectral embedding space and partitioning the EEG channels into "patches" to preserve spatial location information. Multi-scale causal convolutional layers are used to process the temporal characteristics of the EEG signals. Causal convolution ensures that the output at time t depends only on the input at time t and before, preventing future information leakage. The multi-scale design (different kernel sizes) enables the model to capture different ranges of time steps to adapt to EEG signals. The multi-head self-attention layer is used to capture the spatial dependencies between EEG channels, helping to extract whole-brain information from multi-channel EEG signals. The convolutional layer and self-attention outputs of each scale (small, medium, and large) are aggregated to obtain complete spatial embedding information.

[0016] The multi-view spectral-spatial-temporal masked autoencoder model (MV-SSTMA) is implemented using, but not limited to, the technology described in "A Multi-view Spectral-Spatial-Temporal Masked Autoencoder for Decoding Emotions with Self-supervised Learning" by Rui Li et al. (https: / / doi.org / 10.1145 / 3503161.3548243).

[0017] The parameters of the multi-view spectral-spatial-temporal mask autoencoder model include:

[0018] Spectral embedding dimension D: used to map input data to a specific spectral feature space.

[0019] Convolution kernel size k s ,k m ,k l : Corresponding to small, medium and large scale convolution layers, used for feature extraction in different time ranges.

[0020] The number of self-attention heads: determines the number of heads in the spatial attention layer, thereby affecting the capture of dependencies between EEG channels.

[0021] Batch Normalization: Used to stabilize the model training process and prevent gradient explosion or disappearance.

[0022] Position encoding: implemented in the spatial position encoding layer to preserve the spatial position information of different EEG channels.

[0023] Step 3: Fine-tune the spectral-spatial-temporal features obtained in Step 2. Specifically, the extracted spectral-spatial-temporal features are input into a linear classifier, ultimately outputting whether the subject is in a meditative or resting state. After pre-training, fine-tune the model on a labeled target dataset to adapt it to specific tasks, such as meditation and resting state classification. By training with a small learning rate and selectively freezing or unfreezing some layers, the model is gradually optimized on the labeled dataset, allowing it to learn the features required for specific tasks. By fine-tuning the model on the labeled meditation dataset and combining the feature extractor with the linear classifier, it achieves high-precision meditation and resting state classification. Through these designs, MV-SSTMA has significant advantages in denoising, improving EEG classification accuracy, and enhancing model generalization capabilities.

[0024] Through specific practical experiments, in experimental settings with both cross-subject and same-subject environments, the method described above was fine-tuned using the self-supervised MV-SSTMA model as its core, using the unlabeled SEED emotional EEG dataset as the pre-training dataset. Compared to existing technologies, this method reduces the interference of temporal correlation on EEG signals by introducing multiple acquisitions and setting a 14-day interval during data collection. Simultaneously, the method alternates between audio-guided and non-audio-guided meditation experimental designs, effectively eliminating the impact of auditory interference on meditation state monitoring. Algorithmically, the method uses a multi-view spectral-spatial-temporal masked autoencoder (MV-SSTMA) model. After self-supervised pre-training and fine-tuning on labeled data, the model can more accurately distinguish between meditation and resting states. Specific experimental results show that in the experiment of the same subject, the F1 score reached 0.7776 and the accuracy was 0.7863; in the experiment across subjects, the F1 score reached 0.7239. Compared with the traditional method, the classification performance has been significantly improved in terms of indicators such as F1 score, accuracy, precision and recall rate, verifying the superiority of this method in eliminating interference factors and improving classification accuracy.

[0025] In terms of methodological improvements, this method improves the data collection design: it introduces an alternating design of audio-guided and non-audio-guided data collection during multiple collections to avoid the influence of time and auditory interference on the classification of meditation states. The principle is to reduce the influence of time correlation on the classification results by introducing an interval through multiple collections. The alternating use of audio-guided and non-audio-guided experimental designs can more accurately learn the EEG signal characteristics under different meditation conditions, thereby avoiding auditory interference. Specifically, each subject participated in three independent experiments, with a 14-day interval between each experiment, and each collection included experiments in the resting state and the meditation state. The meditation state experiment further alternated between audio-guided and non-audio-guided meditation to collect the corresponding EEG signal data.

[0026] This method improves upon the meditation EEG recognition algorithm by introducing a self-supervised learning algorithm and employing a multi-view spectral-spatial-temporal masked autoencoder (MV-SSTMA). This method enhances the model's generalization capabilities through self-supervised pre-training. The principle is to perform self-supervised pre-training on unlabeled data, enabling the model to autonomously learn more general spectral-spatial-temporal features. This approach reduces the reliance on labeled data, making the model broadly adaptable to different experimental conditions and individuals.

[0027] The specific operation involves pre-training the MV-SSTMA model on unlabeled emotion EEG datasets (such as the SEED dataset). This process randomly masks portions of the EEG signal and reconstructs it to learn universal spectral, spatial, and temporal features. The pre-trained model is then fine-tuned using labeled data to optimize its classification performance, making it suitable for meditation and resting state classification tasks.

[0028] Based on the above improvements, this method achieves: 1) Improved classification accuracy: Through multiple acquisitions and interval design, the influence of time and auditory interference is significantly reduced, enabling the model to more accurately distinguish between meditation state and resting state. 2) Enhanced generalization ability: Self-supervised pre-training enables the model to acquire general feature extraction capabilities, improving its adaptability under different experimental conditions and individual differences. 3) Enhanced robustness: The system's evaluation indicators such as accuracy and F1 score in meditation state monitoring are significantly better than traditional technical methods, and show higher robustness across individuals and different meditation conditions. The results are as follows Figure 3 shown.

[0029] In summary, this invention achieves innovation in EEG monitoring during meditation. First, by improving the data acquisition method and introducing multiple acquisitions with a 14-day interval, the interference of temporal correlation on EEG signals is reduced. Furthermore, by adopting an alternating audio-guided and non-audio-guided meditation experimental design, the influence of auditory interference is eliminated. Furthermore, algorithmically, a multi-view spectral-spatial-temporal masked autoencoder (MV-SSTMA) model is proposed. This model extracts spectral-spatial-temporal features from unlabeled EEG data through self-supervised learning and is subsequently fine-tuned on labeled meditation data. This significantly improves classification performance and the model's generalization ability, enabling the model to more accurately distinguish between meditation and resting states.

[0030] The above-mentioned specific implementation can be partially adjusted in different ways by those skilled in the art without departing from the principles and purpose of the present invention. The scope of protection of the present invention shall be based on the claims and shall not be limited by the above-mentioned specific implementation. All implementation schemes within its scope shall be subject to the constraints of the present invention.

Claims

1. A method for detecting EEG signals based on separating temporal correlation and auditory interference, characterized in that: By collecting unlabeled EEG data, pre-training the multi-view spectrum-space-time mask autoencoder model to obtain spectrum-space-time features, and then fine-tuning the linear classifier to obtain the specific state of the EEG data, which is characterized by: Step 1: Multiple Data Collection: Each subject participated in three independent experiments, with a 14-day interval between each experiment. Each collection included audio-guided meditation, non-audio-guided meditation, and resting state experiments. Step 2: Pre-train the multi-view spectral-spatial-temporal mask autoencoder model (MV-SSTMA) on the unlabeled EEG dataset obtained in step 1 to obtain universal spectral-spatial-temporal features; Step 3: Fine-tune the spectrum-space-time features obtained in step 2. Specifically, the extracted spectrum-space-time features are input into a linear classifier, and the final output is whether the subject is in a meditative state or a resting state.

2. The method for detecting EEG signals based on separating temporal correlation and auditory interference according to claim 1, wherein: The time interval between each acquisition can effectively reduce the temporal correlation of EEG signals under different states, avoid overfitting of the model to temporal features, and thus improve the accuracy of classification.

3. The method for detecting EEG signals based on separating temporal correlation and auditory interference according to claim 1, wherein: In the data collection process, the subjects alternated between audio-guided meditation and non-audio-guided meditation to explore the impact of auditory interference on EEG signals in the meditation state. Specifically, each collection included 5 minutes of resting state, about 20 minutes of meditation state, and another 5 minutes of resting state.

4. The method for detecting EEG signals based on separating temporal correlation and auditory interference according to claim 3, wherein: By alternating between audio guidance with and without prompts, the subjects were instructed to enter a meditative state through audio instructions, while in the non-audio guidance phase, the subjects were asked to continue meditating in a quiet environment.

5. The method for detecting EEG signals based on separating temporal correlation and auditory interference according to claim 1, wherein: Step 2 is as follows: the multi-view spectral-spatial-temporal masked autoencoder (MV-SSTMA) algorithm adopts a CNN-Transformer-based architecture, first randomly masks the unlabeled EEG data through a self-supervised pre-training phase, and then uses multi-scale causal convolutional layers and a multi-head self-attention mechanism to extract spectral, spatial, and temporal features; Unlabeled EEG data is used for general feature learning in the self-supervised pre-training stage: the EEG data is divided into a visible subset and a masked subset through random masking, and the spectral, spatial and temporal information is captured by reconstructing the masked data. The extracted features have stronger generalization ability.

6. The method for detecting EEG signals based on separating temporal correlation and auditory interference according to claim 1, wherein: The fine-tuning mentioned above means that the model is gradually optimized on a labeled dataset through training with a small learning rate and selectively freezing some or all layers so that it learns the features required for a specific task.

7. An electroencephalogram (EEG) signal detection system based on separation of temporal correlation and auditory interference, implementing the method according to any one of claims 1 to 6, characterized in that: include: A state labeling unit, an EEG acquisition unit, and a meditation state analysis unit, wherein: the state labeling unit plays meditation prompts in alternation according to a pre-set meditation prompt program, and appends additional data to the collected EEG data to clarify the environment and conditions for collecting EEG at that moment; the EEG acquisition unit consists of an EEG cap and a signal amplifier and is associated with the state labeling unit, and collects EEG data in real time according to the program shown in the state labeling unit, and collects a section of EEG data containing 64 channels; the meditation state analysis unit performs data analysis on the collected EEG data, performs preprocessing such as noise removal, and then adopts a deep learning method including a self-supervised learning mechanism to obtain the result of real-time detection of meditation state of EEG state.

Images