Key channel screening method and system for electroencephalogram cap in power industry

Abstract

The invention discloses a power industry electroencephalogram cap-oriented key channel screening method and system, and the method comprises the following steps: firstly, collecting electroencephalogram signals of a power worker in typical states of waking, fatigue and the like by adopting a 32 or 64 channel electroencephalogram cap, and carrying out the preprocessing operations of band-pass filtering, ICA (Independent Component Analysis), signal-to-noise ratio evaluation and the like; artifacts and inferior channels are removed, and then multi-dimensional features such as the frequency domain, the time domain, the entropy value and statistics of each channel are extracted to comprehensively score the importance of the channels. Constructing a plurality of channel subsets, verifying the recognition performance of the channel subsets through a classification model, and screening out a group of key channels with the highest information expression capability in cognitive state discrimination; according to the method, the number of electroencephalogram channels is reduced, the complexity and calculation overhead of acquisition equipment are reduced, the feasibility of real-time deployment of the model in an electric power field is improved, and the method has relatively high industry adaptability and engineering application value and is suitable for various key scenes with relatively high requirements on personnel state perception.

Description

Technical Field

[0001] This invention relates to the fields of electroencephalogram (EEG) signal processing and power operation safety monitoring, and in particular to a method and system for screening key channels of an EEG cap for power industry workers. Background Technology

[0002] With the widespread application of wearable EEG devices in industrial site safety management, EEG monitoring technology has become an important means for the power industry to monitor the attention, fatigue, and mental state of workers. By collecting and analyzing EEG signals in real time, abnormal states of workers can be detected in a timely manner, preventing safety accidents and ensuring the safe and stable operation of power production. However, existing standard EEG caps are typically equipped with 32 to 64 channels. Although they can provide relatively comprehensive EEG information, they face problems such as serious channel redundancy, inconsistent signal quality, and large data processing computation requirements.

[0003] In practical applications, an excessive number of channels not only increases the complexity and burden of the acquisition equipment, but also leads to low efficiency in signal processing and cognitive state recognition model training and deployment, making it difficult to meet the industrial demands for real-time performance and portability. Furthermore, signal quality differences exist between different channels, and some channels may be affected by power supply interference, electromyographic noise, etc., reducing the overall system's recognition accuracy. Therefore, for EEG data from power workers, effectively selecting key channels that are representative, information-rich, and have a high signal-to-noise ratio has become a key technical challenge for improving the performance and practicality of EEG monitoring systems. Summary of the Invention

[0004] Technical problems to be solved To address the aforementioned issues, this invention proposes a key channel screening method and system for EEG caps in the power industry, enabling automatic screening of the most representative channels of EEG signals and improving the efficiency and stability of subsequent cognitive state recognition algorithms. Technical solution

[0005] This invention proposes a method and system for screening key channels in EEG caps for the power industry, employing the following technical solution: A method for screening key channels in EEG caps for the power industry, comprising the following steps: S1. Use a standard EEG cap to collect EEG signals from power workers in multiple cognitive states; S2. The multi-cognitive state EEG signals described in S1 are preprocessed by bandpass filtering, ICA artifact removal, and signal-to-noise ratio screening. S3. Extract frequency domain, time domain, entropy value and statistical features of each channel; S4. Based on the features extracted in S3, calculate the channel importance score using a multi-dimensional evaluation method to assess the channel importance. S5. Based on the channel importance scores calculated in S4, construct a channel subset to verify the classification performance and perform performance verification. S6. Based on the output of S5, find the minimum set of critical channels that meets the accuracy threshold.

[0006] Preferably, the standard EEG cap is configured with 32 or 64 channels, and the multi-cognitive state EEG signals include at least the awake state and the fatigue state after 3 hours of continuous work. The international 10-20 system electrode layout is used for acquisition, and the sampling frequency is ≥512Hz.

[0007] Preferably, the signal preprocessing specifically includes using a 1-45Hz Butterworth bandpass filter to remove noise, separating eye movement and electromyography artifacts using the FastICA algorithm, and eliminating inferior channels with a signal-to-noise ratio of less than 3dB.

[0008] Preferably, the feature extraction includes calculating four types of feature indicators: relative power spectral density of the band, time-domain mean and standard deviation, sample entropy and spectral entropy, and coefficient of variation and kurtosis.

[0009] Preferably, the channel importance assessment employs one or a combination of the following parallel techniques: statistical significance analysis based on t-test / ANOVA, mutual information calculation of features and labels, feature weight output of SVM / random forest model, and sparse selection of LASSO regression.

[0010] Preferably, when using a combination of multiple evaluation methods, the scores of each item need to be normalized and then weighted and averaged. The weight allocation ratio is 30% for statistical test, 20% for mutual information, 30% for model weight, and 20% for LASSO selection.

[0011] Preferably, the performance verification uses an SVM classifier and 10-fold cross-validation, with the accuracy threshold set at 98% of the full-channel classification accuracy. The final output has 5-15 key channels and must include the three midline electrodes Fz, Cz, and Pz.

[0012] Preferably, the key channel output results simultaneously generate an electrode position optimization scheme, specifically, based on the standard 10-20 system, retaining the selected key channel positions, removing redundant channels, and recalculating the reference electrode positions.

[0013] Preferably, the present invention also provides a key channel screening system for EEG caps in the power industry, including a detachable electrode array module. This module integrates only the screened key channel electrodes, the electrode spacing is scaled proportionally according to the 10-20 system, and has a built-in preamplifier circuit and a wireless transmission module.

[0014] Preferably, the detachable electrode array is connected to the main control unit via a magnetic interface, and the number of electrodes is 5-15, with the layout being the same as the position of the output key channel.

[0015] This invention provides a method and system for screening key channels in EEG caps for the power industry, which has the following beneficial effects: 1. Achieve channel compression and improve system usability: By performing multi-dimensional feature extraction and importance scoring on each channel of the EEG signal, the minimum number of key channels can be selected while ensuring that the classification performance remains basically unchanged. This significantly reduces the number of channels in the EEG acquisition system, reduces equipment complexity and usage costs, and improves the portability and deployment efficiency of practical applications.

[0016] 2. Multi-indicator integrated evaluation, scientific and reasonable channel selection: This invention combines multiple scoring methods such as statistical tests, mutual information analysis, machine learning feature importance and LASSO sparse regression to comprehensively score EEG channels, effectively improving the objectivity and robustness of channel selection and avoiding the bias caused by relying on a single evaluation standard.

[0017] 3. Balancing accuracy and computational efficiency: By ranking and verifying the importance of channels, a "number of channels - classification accuracy" relationship curve can be plotted, which makes it easy to balance recognition accuracy and computational overhead according to actual needs. It is suitable for resource-constrained scenarios such as edge computing and embedded devices.

[0018] 4. Enhanced state recognition capability, applicable to fatigue monitoring: This invention can effectively identify the alertness and fatigue status of power workers, improve the early warning capability of fatigue risks, and has important safety assurance significance. Attached Figure Description

[0019] Figure 1 This is a schematic diagram of a method and system for screening key channels in an EEG cap for the power industry, provided as an embodiment of the present invention. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] refer to Figure 1 As shown, a method and system for screening key channels in EEG caps for the power industry are disclosed. The method for screening key channels in EEG caps for the power industry includes the following steps: S1. Use a standard EEG cap to collect EEG signals from power workers in multiple cognitive states; S2. The multi-cognitive state EEG signals described in S1 are preprocessed by bandpass filtering, ICA artifact removal, and signal-to-noise ratio screening. S3. Extract frequency domain, time domain, entropy value and statistical features of each channel; S4. Based on the features extracted in S3, calculate the channel importance score using a multi-dimensional evaluation method to assess the channel importance. S5. Based on the channel importance scores calculated in S4, construct a channel subset to verify the classification performance and perform performance verification. S6. Based on the output of S5, find the minimum set of critical channels that meets the accuracy threshold.

[0022] Step S1: EEG signal acquisition. Specifically: A standard 32-channel or 64-channel EEG acquisition system (e.g., based on the international 10–20 or 10–10 electrode placement standard) was used to collect EEG data from the subjects. The subjects were several electrical workers, and the data collection scenarios included both awake and fatigued states: awake state: 1 hour before the start of work; fatigued state: 3 hours after continuous high-intensity work. The acquisition frequency was set to 512Hz, and the acquisition time was no less than 5 minutes per round. During the sampling process, subjects were required to sit still, close their eyes, or look at the screen to avoid artifacts caused by large movements. Multiple rounds of data collection were performed on each subject, with a 5-minute interval between each round to ensure the comprehensiveness and stability of the data.

[0023] Step S2 involves preprocessing the acquired EEG signals. Specifically: After acquisition, the EEG data will undergo the following preprocessing operations: First, bandpass filtering will be performed using a Butterworth filter to limit the signal to the range of 1Hz–45Hz, in order to preserve the frequency band of normal brain activity and effectively filter out high-frequency electromyography interference and low-frequency baseline drift; then, independent component analysis (ICA) methods, such as FastICA or Infomax algorithms, will be used to remove artifacts in the signal, decomposing the mixed EEG signal into independent components and identifying non-brain-origin interference components such as eye movement, electrocardiogram, and electromyography, which will be removed or reconstructed; to further improve data quality, the signal-to-noise ratio (SNR) of each channel will be calculated. If the SNR of a certain channel is lower than a preset threshold (e.g., 3dB), it will be judged as a poor-quality channel and removed, and will not be included in subsequent analysis.

[0024] Step S3: Extract channel features. Specifically: Feature extraction is performed on the remaining effective channels, with extraction dimensions including frequency domain, time domain, entropy value, and statistical features: (1) Frequency domain characteristics: The relative power of α wave (8–13Hz), β wave (13–30Hz), and θ wave (4–8Hz) was calculated using the Welch spectrum estimation method; (2) Time-domain characteristics: Calculate the mean and standard deviation of each signal segment; (3) Entropy metrics: including sample entropy and spectral entropy, which reflect signal complexity and spectral distribution; (4) Statistical indicators: including coefficient of variation (CV) and kurtosis, used to measure signal fluctuations and distribution patterns.

[0025] Each channel generates a corresponding feature vector, which serves as the basis for subsequent channel evaluation and classification.

[0026] Step S4: Score the channel importance based on the extracted channel features. Specifically: Assess each channel's ability to distinguish cognitive states (awake and fatigued) using any one or more of the following methods: (1) Single-channel statistical test: Perform a two-sample t-test or analysis of variance (ANOVA) on a certain feature of each channel, calculate its p value, and rank them accordingly; (2) Mutual information analysis: The mutual information between channel features and status labels (awake / fatigued) is calculated to evaluate its information contribution. (3) Machine learning model evaluation: Classification models such as support vector machine (SVM), random forest or gradient boosting tree (such as LightGBM) are used for training, and the feature importance of the model output is used as the channel score; (4) LASSO sparse selection: Construct a LASSO regression model and use L1 regularization term to perform sparse selection of channel dimensions, retaining key channels with non-zero coefficients.

[0027] Finally, the various scoring methods can be normalized, and a weighted average can be used to obtain the comprehensive importance score of each channel, thus forming a channel score ranking.

[0028] Step S5 involves combining channel subsets and verifying performance. Specifically: Based on the channel scores sorted in step S4, channels are progressively combined in descending order of score to construct multiple channel subsets (such as Top-5, Top-10, and Top-15 channel combinations). For each channel subset, its corresponding feature vector is input into a classification model (such as Support Vector Machine (SVM), Linear Discriminant Analysis (LDA), Random Forest, etc.), and its classification performance is evaluated using methods such as 10-fold cross-validation. The classification accuracy of each subset is recorded, and a "channel count – accuracy" curve is plotted. Through comparative analysis, the minimum channel combination required is found while ensuring that the accuracy decrease does not exceed a preset threshold (such as 2%), thus balancing performance and channel count compression.

[0029] Step S6: Output the key channel. Specifically: The final output is a list of key channel IDs obtained through filtering, such as [Fp1, Fz, Cz, P3, O2]. These channels have the highest information expression ability and classification contribution in cognitive state recognition. The selected channels can not only be used as input for the optimized EEG recognition model, but also as a reference for the sensor deployment of customized EEG caps, effectively supporting the engineering implementation of lightweight hardware design and model deployment.

[0030] In addition, this embodiment also provides a key channel screening system for EEG caps in the power industry, including: a detachable electrode array module, which integrates only the screened key channel electrodes, with the electrode spacing scaled proportionally according to the 10-20 system, and has a built-in preamplifier circuit and a wireless transmission module.

[0031] In this embodiment, the detachable electrode array is connected to the main control unit via a magnetic interface, and the number of electrodes is 5-15, with the layout being the same as the key output channel position.

[0032] The above description is merely a preferred embodiment of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principle of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. A method for screening key channels in an EEG cap for the power industry, characterized in that, Includes the following steps: S1. Use a standard EEG cap to collect EEG signals from power workers in multiple cognitive states; S2. The multi-cognitive state EEG signals described in S1 are preprocessed by bandpass filtering, ICA artifact removal, and signal-to-noise ratio screening. S3. Extract frequency domain, time domain, entropy value and statistical features of each channel; S4. Based on the features extracted in S3, calculate the channel importance score using a multi-dimensional evaluation method to assess the channel importance. S5. Based on the channel importance scores calculated in S4, construct a channel subset to verify the classification performance and perform performance verification. S6. Based on the output of S5, find the minimum set of critical channels that meets the accuracy threshold.

2. The key channel screening method for EEG caps in the power industry according to claim 1, characterized in that: The standard EEG cap uses a 32 or 64-channel configuration. The multi-cognitive state EEG signals include at least the awake state and the fatigue state after 3 hours of continuous work. The international 10-20 system electrode layout is used for acquisition, and the sampling frequency is ≥512Hz.

3. The key channel screening method for EEG caps in the power industry as described in claim 1, characterized in that: The signal preprocessing specifically includes using a 1-45Hz Butterworth bandpass filter to remove noise, separating eye movement and electromyography artifacts using the FastICA algorithm, and eliminating inferior channels with a signal-to-noise ratio of less than 3dB.

4. The key channel screening method for EEG caps in the power industry according to claim 1, characterized in that: The feature extraction includes calculating four types of feature indicators: relative power spectral density of the band, time-domain mean and standard deviation, sample entropy and spectral entropy, and coefficient of variation and kurtosis.

5. The key channel screening method for EEG caps in the power industry according to claim 1, characterized in that: The channel importance assessment employs one or a combination of the following parallel techniques: statistical significance analysis based on t-test / ANOVA, mutual information calculation of features and labels, feature weight output of SVM / random forest model, and sparse selection of LASSO regression.

6. The key channel screening method for EEG caps in the power industry according to claim 1, characterized in that: When using a combination of multiple evaluation methods, the scores of each item need to be normalized and then weighted and averaged. The weight allocation ratio is 30% for statistical test, 20% for mutual information, 30% for model weight, and 20% for LASSO selection.

7. The key channel screening method for EEG caps in the power industry according to claim 1, characterized in that: The performance verification uses an SVM classifier and 10-fold cross-validation. The accuracy threshold is set to 98% of the full-channel classification accuracy. The final output has 5-15 key channels and must include the three midline electrodes Fz, Cz, and Pz.

8. The key channel screening method for EEG caps in the power industry according to claim 1, characterized in that: The key channel output results simultaneously generate an electrode position optimization scheme, which, based on the standard 10-20 system, retains the selected key channel positions, removes redundant channels, and recalculates the reference electrode positions.

9. A key channel screening system for an EEG cap for the power industry, implementing the method of any one of claims 1-8, characterized in that: It includes a detachable electrode array module that integrates only selected key channel electrodes with electrode spacing scaled proportionally to a 10-20 system, and incorporates a preamplifier circuit and a wireless transmission module.

10. A key channel screening system for EEG caps in the power industry according to claim 9, characterized in that: The detachable electrode array is connected to the main control unit via a magnetic interface. The number of electrodes is 5-15, and the layout is the same as the position of the key output channel.

Images