Method for sensing and reporting danger through brain-computer interface

Abstract

The invention discloses a method for carrying out danger perception and reporting through a brain-computer interface. The method comprises the following steps that S1, electroencephalogram signals of a user are collected in real time through electroencephalogram collection equipment; s2, filtering and de-noising the collected electroencephalogram signals so as to improve the signal quality; s3, key features are extracted from the preprocessed electroencephalogram signals, the key features comprise frequency, amplitude and phase, and the features can describe different electroencephalogram states. The brain activity of the user is monitored in real time to find potential danger, the machine learning algorithm is matched to train the electroencephalogram signals, danger prediction can be accurately carried out, and therefore special crowd users are helped to report danger information in time without depending on an external sensor, and the safety is improved. Danger sensing and reporting functions can be realized only by wearing electroencephalogram acquisition equipment, and the method has the advantages of real-time performance, accuracy, convenience, pertinence and the like.

Description

Technical Field

[0001] The present invention relates to the technical field of brain-computer interface, and in particular to a method for sensing and reporting danger through a brain-computer interface. Background Art

[0002] Danger perception is crucial to the safety of individuals and society. Whether in industrial production, transportation, or the daily lives of special groups such as the disabled or children, timely and accurate perception of potential dangers and taking appropriate measures can effectively avoid accidents. Traditional hazard perception methods mainly rely on human senses and experience, but these methods have certain limitations, such as being easily affected by subjective factors and slow reaction speed. With the development of science and technology, brain-computer interface technology has become an important bridge connecting the human brain and external devices. By analyzing EEG signals, it can provide an important basis for early warning and diagnosis of dangerous conditions.

[0003] EEG signals are electrical signals generated by the activity of neurons in the cerebral cortex and are recorded through electrodes placed on the scalp. They carry rich physiological and pathological information and are widely used in neuroscience, clinical medicine, cognitive science and other fields. EEG signals have characteristics such as non-stationarity, nonlinearity, and time series complexity, which bring challenges to signal processing and analysis. In-depth research and accurate interpretation of EEG signals are of great significance for understanding the working mechanism of the brain, diagnosing brain diseases, and developing technologies based on brain-computer interfaces.

[0004] In some dangerous situations, such as human fainting, driver fatigue driving, and the risk of mental illness, EEG signals will undergo specific changes. By analyzing these changes, we can warn of the occurrence of dangerous conditions in advance and provide a basis for taking corresponding measures. For example, in the field of aviation medicine, the collection of EEG signals can be used to warn of human fainting; in the field of road traffic safety, the use of EEG signals and video monitoring can determine the driver's alertness and detect fatigue driving in time. In addition, for people at high risk of clinical mental illness, EEG signal characteristics can also be used for risk prediction.

[0005] However, the existing safety alarm system mainly relies on external sensors, and has problems such as slow response and high false alarm rate. It is often unable to promptly report danger or seek help when encountering danger. For this reason, we propose a method for danger perception and reporting through brain-computer interface. Summary of the invention

[0006] Based on the technical problems existing in the background technology, the present invention proposes a method for danger perception and reporting through a brain-computer interface.

[0007] The present invention proposes a method for sensing and reporting danger through a brain-computer interface, comprising the following steps:

[0008] S1: Use EEG acquisition equipment to collect the user's EEG signals in real time;

[0009] S2: Filter and denoise the collected EEG signals, select 1 / 5 of the recording frequency as the sampling frequency, obtain the recording curve of the overall data, filter out the data of mains interference, myoelectric interference and electrooculographic interference, calibrate the baseline, prevent baseline drift, and improve signal quality;

[0010] S3: Extract key features from the preprocessed EEG signal, including frequency, amplitude and phase, which can describe different EEG states;

[0011] S4: The features extracted in S3 are used as input data and trained using a machine learning algorithm. During the training process, the machine learning algorithm learns how to distinguish different categories of EEG signal states and optimizes its internal parameters to minimize errors.

[0012] S5: After the training is completed, the machine learning algorithm can predict the corresponding EEG state based on the new EEG signal features, thereby establishing a mapping relationship between EEG signals and dangerous states;

[0013] S6: Monitor the user's EEG signals in real time through an EEG acquisition device, and predict potential dangers based on the mapping relationship established in S5;

[0014] S7: When the user is in danger, a specific EEG signal pattern triggers the reporting mechanism, and sends the danger information to the designated recipient via the wireless network, while sending a distress sound to the surrounding area;

[0015] S8: When it is confirmed that the person has been successfully rescued, the distress signal is turned off.

[0016] Preferably, in S1, the EEG acquisition device may be an EEG headband.

[0017] Preferably, in S2, a 50 Hz notch filter is used to remove 50 Hz mains interference, a 30 Hz low-pass filter is selected to remove electromyographic signals, and electrooculogram artifact calibration is performed using the vertical electrooculogram as a reference level to obtain a signal without the influence of electrooculogram.

[0018] Preferably, in S3, before extracting the key features, the data is first subjected to maximum and minimum normalization processing, and the specific method is as follows:

[0019]

[0020] Among them, x min is the minimum value in the data, x maxis the maximum value in the data, all dimensional values ​​are scaled within the interval of [0,1], and the normalized feature matrix is ​​obtained. Due to the great difference in power values ​​of different individuals and different bands, directly entering the data into the model will affect the distribution and adjustment of weights, resulting in poor model recognition effect, in order to reduce the impact of different data magnitudes.

[0021] Preferably, in S3, when extracting key features, EEG features are extracted from signals in the 8-13 Hz and 14-30 Hz frequency bands respectively, and the specific method is as follows:

[0022] The discrete Fourier transform formula is:

[0023]

[0024] Where x(k) is the EEG signal to be processed, w(T) is the window function, m is the total number of samples, and the length is T = [0, t cut,j ]Prepare the characteristics of the EEG signal, select a time window with a step size of 200ms, perform Fourier transform on the EEG signal, and obtain the amplitude density function f(k) in the frequency domain:

[0025]

[0026] Where N = n 2 , n is the total number of samples,

[0027]

[0028] The average values ​​of energy indexes in the 8-13 Hz and 14-30 Hz frequency bands were calculated, which are the characteristics of EEG signals.

[0029] Preferably, the calculated EEG signal features are subjected to dimensionality reduction processing, and the specific dimensionality reduction steps are:

[0030] S31. The data sample function X is an n×m matrix, where n is the total number of samples and m is the total number of features. The data sample function X is:

[0031]

[0032] S32, normalize the data sample function X:

[0033]

[0034] Where, i = 1, 2…n; j = 1, 2…m;

[0035] S33, calculate the covariance matrix cov of the data sample function X x :

[0036]

[0037] Calculate the eigenvalue λ of the matrix j and the eigenvector v i

[0038] S34, the eigenvalue λ j Sort by size and calculate the contribution rate of each eigenvalue:

[0039]

[0040] P is the number of selected components, and 90% is used as the cumulative contribution rate threshold. The eigenvectors corresponding to p=8 maximum eigenvalues ​​are extracted from the results to form samples, thereby achieving dimensionality reduction of EEG signal features.

[0041] Preferably, in S4, the machine learning algorithm is one of SVM or neural network.

[0042] Preferably, in said S7, the danger information includes location, time and danger type.

[0043] Preferably, in S7, the recipient is a guardian or an emergency rescue center.

[0044] Compared with the prior art, the present invention has the following beneficial effects:

[0045] The present invention detects potential dangers by real-time monitoring of the user's brain activity, and trains EEG signals with machine learning algorithms to accurately predict dangers, thereby helping special groups of users to report danger information in a timely manner. There is no need to rely on external sensors, and the danger perception and reporting functions can be realized by simply wearing EEG acquisition equipment. It has the advantages of real-time, accuracy, convenience and pertinence. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 This is a flow chart of a method for danger perception and reporting through a brain-computer interface proposed by the present invention;

[0047] Figure 2 A schematic diagram of the international 10 / 20 system. DETAILED DESCRIPTION

[0048] The present invention will be further explained below in conjunction with specific embodiments.

[0049] Reference Figure 1 , this embodiment proposes a method for danger perception and reporting through a brain-computer interface, comprising the following steps:

[0050] The following steps are involved:

[0051] S1: Use EEG headband to collect the user's EEG signals in real time. The international 10 / 20 system can be used, such as Figure 2 As shown in the figure, each part has a letter to identify the brain lobe and a number to identify the location of the hemisphere. F stands for frontal lobe, T stands for temporal lobe, C stands for central lobe (although there is no central lobe, the letter C is used for identification), P stands for parietal lobe, and O stands for occipital lobe. z (zero) refers to the electrode placed on the midline. Even numbers refer to the electrode position in the right hemisphere, and odd numbers refer to the left hemisphere. Four anatomical landmarks are used for the correct positioning of the electrode: the nasion (the point between the forehead and the nose), the occipital protuberance (the lowest point on the back of the head), and the point in front of the ear;

[0052] S2: Filter and denoise the collected EEG signals, select 1 / 5 of the recording frequency as the sampling frequency, obtain the recording curve of the overall data, filter out the data of mains interference, myoelectric interference and electrooculographic interference, calibrate the baseline to prevent baseline drift, so as to improve the signal quality. Specifically, remove the 50Hz mains interference through a 50Hz notch filter, select 30Hz for low-pass filtering to remove myoelectric signals, use the vertical electrooculographic level as the reference level to calibrate the electrooculographic artifacts, and obtain a signal without the influence of electrooculographic interference;

[0053] S3: Extract key features from the preprocessed EEG signal. The key features include frequency, amplitude and phase. These features can describe different EEG states. Before extracting the key features, perform maximum and minimum normalization on the data. The specific method is as follows:

[0054]

[0055] Among them, x min is the minimum value in the data, x max is the maximum value in the data, all dimensional values ​​are scaled within the interval of [0,1], and the normalized feature matrix is ​​obtained. Due to the great difference in power values ​​of different individuals and different bands, directly entering the data into the model will affect the distribution and adjustment of weights, resulting in poor model recognition effect, in order to reduce the impact of different data magnitudes.

[0056] Preferably, in S3, when extracting key features, EEG features are extracted from signals in the 8-13 Hz and 14-30 Hz frequency bands respectively, and the specific method is as follows:

[0057] The discrete Fourier transform formula is:

[0058]

[0059] Where x(k) is the EEG signal to be processed, w(T) is the window function, m is the total number of samples, and the length is T = [0, t cut,j ]Prepare the characteristics of the EEG signal, select a time window with a step size of 200ms, perform Fourier transform on the EEG signal, and obtain the amplitude density function f(k) in the frequency domain:

[0060]

[0061] Where N = n 2 , n is the total number of samples,

[0062]

[0063] The average values ​​of energy indexes in the 8-13 Hz and 14-30 Hz frequency bands were calculated, which are the characteristics of EEG signals.

[0064] Preferably, the calculated EEG signal features are subjected to dimensionality reduction processing, and the specific dimensionality reduction steps are:

[0065] S31. The data sample function X is an n×m matrix, where n is the total number of samples and m is the total number of features. The data sample function X is:

[0066]

[0067] S32, normalize the data sample function X:

[0068]

[0069] Where, i = 1, 2…n; j = 1, 2…m;

[0070] S33, calculate the covariance matrix cov of the data sample function X x :

[0071]

[0072] Calculate the eigenvalue λ of the matrix j and the eigenvector υ i

[0073] S34, the eigenvalue λ j Sort by size and calculate the contribution rate of each eigenvalue:

[0074]

[0075] P is the number of selected components, and 90% is used as the cumulative contribution rate threshold. The eigenvectors corresponding to p=8 maximum eigenvalues ​​are extracted from the results to form samples, thereby achieving dimensionality reduction of EEG signal features.

[0076] S4: The features extracted in S3 are used as input data and trained using a machine learning algorithm. The machine learning algorithm is one of the SVM or neural network. During the training process, the machine learning algorithm learns how to distinguish different categories of EEG signal states and optimizes its internal parameters to minimize errors.

[0077] S5: After the training is completed, the machine learning algorithm can predict the corresponding EEG state based on the new EEG signal features, thereby establishing a mapping relationship between EEG signals and dangerous states;

[0078] S6: Monitor the user's EEG signals in real time through an EEG acquisition device, and predict potential dangers based on the mapping relationship established in S5;

[0079] S7: When the user encounters danger, the reporting mechanism is triggered by a specific EEG signal pattern, and the danger information is sent to the designated recipient via the wireless network. At the same time, a distress sound is emitted and a distress signal is sent to the surrounding area. The danger information includes location, time and type of danger. The recipient is either a guardian or an emergency rescue center.

[0080] S8: When it is confirmed that the person has been successfully rescued, the distress signal is turned off.

[0081] The method provided in this embodiment has the following specific application cases:

[0082] 1. Cases of high-risk groups for clinical mental illness

[0083] The ERP data of the functional state are collected as the main variable, and the latency of the classic P300 induced by the oddball paradigm, the amplitude of the novel P300 induced by the novel P300 paradigm, the amplitude of the classic P300 induced by the oddball paradigm, and the mismatch negativity MMN are integrated into a computational model to estimate the risk of psychosis. For example, the prolonged P300 latency indicates that the working memory network of the subject is damaged, resulting in prolonged processing time; the abnormal amplitude of the novel P300 indicates the abnormality of the subject's brain function highlighting network. The integration of these indicators can more comprehensively reflect the functional state of the brain and improve the accuracy of the risk prediction of the disease.

[0084] In terms of hardware, BrainAmp DC EEG amplifier was used to collect 64 channels of EEG data. EEG data collection was recorded by Brain Vision Recorder, with a sampling bandwidth of 0.05-200Hz and a sampling rate of 1000Hz. The impedance of all leads was reduced to below 10k during collection. EEG data preprocessing was completed using BrainVision Analyzer. The reference lead during EEG collection was the tip of the nose. Except for the MMN paradigm, the EEG signals of the P300 and auditory evoked response tasks were re-referenced, using the mean of the bilateral ear protrusions as the reference; the MMN paradigm EEG used the tip of the nose as a reference to ensure effective detection of signals originating from the temporal lobe.

[0085] By constructing a multi-model EEG signal analysis method, we can provide a more effective tool for predicting the risk of developing mental illness in high-risk groups and provide a more accurate basis for clinical intervention and treatment.

[0086] 2. Dangerous traffic scene identification case

[0087] Acquire stimulus objects and experimental variables based on traffic scenarios, complete driving tests on drivers through stimulus objects and experimental variables, and obtain EEG data containing EEG signals of the driver within a preset time period.

[0088] In the identification of dangerous traffic scenes, it is first necessary to determine the appropriate traffic scene stimulus objects and experimental variables. For example, different road conditions, traffic flow, weather conditions, etc. can be set as stimulus objects, and factors such as the driver's fatigue level and driving experience can be considered as experimental variables. Through these stimulus objects and experimental variables, the driver is tested for driving. During the test, professional EEG signal acquisition equipment is used to record the driver's EEG signals in real time to obtain EEG data containing rich EEG information within a preset time.

[0089] EEG signal preprocessing removes signal noise in EEG signals, including notch and bandpass filtering denoising, electrooculogram denoising based on ICA method, experimental variable stimulation segment data extraction, baseline correction, etc.

[0090] Preprocessing the collected EEG signals is a key step to ensure the accuracy of subsequent analysis. Notch and bandpass filtering denoising can remove interference signals of specific frequencies, such as power frequency interference. Electrooculogram denoising based on independent component analysis (ICA) can effectively separate electrooculogram artifacts and improve the purity of EEG signals. Extracting experimental variable stimulus segment data can focus on the EEG signal part related to specific traffic scene stimulation. Baseline correction helps to eliminate baseline drift in the signal and make the EEG signal more stable.

[0091] Extraction and analysis of characteristic indicators: The power spectrum density of several rhythmic waves is extracted from EEG data as characteristic indicators, and the characteristic indicators with significance are analyzed by mathematical statistics. The difference between the predicted value and the true value of the power spectrum density of the significant characteristic indicators is calculated, and the degree of danger of the traffic scene is identified based on the difference.

[0092] From the preprocessed EEG data, the power spectrum density of different rhythmic waves can be extracted as characteristic indicators, such as α wave, β wave, γ wave, δ wave and θ wave. The power spectrum estimation method is used to analyze the driver's EEG signal during driving. For example, the AR power spectrum estimation model is used to calculate the parameters to be identified so that the sum of the power of the prediction errors of the AR model before and after is minimized. The extracted characteristic indicators are tested for normality, and the characteristic indicators that meet the requirements of normal distribution and variance homogeneity are subjected to one-way variance analysis. The characteristic indicators that do not meet the requirements are subjected to non-parametric variance analysis. The characteristic indicators with P(w) less than the preset value are determined to be significant characteristic indicators. Then, the preset time is divided into multiple time windows, and the GM(1,1) model is established based on the grey differential equation. The predicted value of the power spectrum density of the significant characteristic indicators is calculated, and the difference between the predicted value and the true value is calculated. According to this difference, the degree of danger of the traffic scene can be identified. The larger the difference, the greater the deviation between the actual situation and the predicted situation, and the higher the degree of danger of the traffic scene may be.

[0093] 3. Drone Emergency Detection Case

[0094] The user's EEG signals are collected in real time, and non-invasive EEG electrodes are placed on the user's scalp to perform preliminary processing such as signal amplification and analog-to-digital conversion to obtain the original EEG signals.

[0095] Extract the EEG event evoked potential from the EEG signal and obtain its amplitude as the potential amplitude feature. The specific process includes preprocessing the original EEG signal, such as bandpass filtering, using independent component analysis to filter out blink artifacts and electromyographic artifacts, using common average reference and baseline correction to filter out noise, detecting pure EEG signals, obtaining EEG event evoked potentials and outputting them to the decision subsystem.

[0096] Calculate the node information of the brain network in the EEG signal, optimize the relationship between nodes, obtain the adjacency matrix, and use its elements as brain network features. The specific steps are: define the location or brain area corresponding to the EEG electrode as the network node; construct the relationship between network nodes and establish the adjacency matrix; optimize the adjacency matrix to obtain the final adjacency matrix; use the final adjacency matrix elements as brain network features to extract brain network features.

[0097] The compressed potential amplitude characteristics and brain network characteristics are classified through the EEG signal feature extraction unit based on the autoencoder and the regularized linear discriminant analysis model to determine whether it is an emergency situation. If so, the hovering command is output to the drone.

[0098] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A method for danger perception and reporting via a brain-computer interface, characterized in that: The following steps are involved: S1: Use EEG acquisition equipment to collect the user's EEG signals in real time; S2: Filter and denoise the collected EEG signals, select 1 / 5 of the recording frequency as the sampling frequency, obtain the recording curve of the overall data, filter out the data of mains interference, myoelectric interference and electrooculographic interference, calibrate the baseline, prevent baseline drift, and improve signal quality; S3: Extract key features from the preprocessed EEG signal, including frequency, amplitude and phase, which can describe different EEG states; S4: The features extracted in S3 are used as input data and trained using a machine learning algorithm. During the training process, the machine learning algorithm learns how to distinguish different categories of EEG signal states and optimizes its internal parameters to minimize errors. S5: After the training is completed, the machine learning algorithm can predict the corresponding EEG state based on the new EEG signal features, thereby establishing a mapping relationship between EEG signals and dangerous states; S6: Monitor the user's EEG signals in real time through an EEG acquisition device, and predict potential dangers based on the mapping relationship established in S5; S7: When the user is in danger, a specific EEG signal pattern triggers the reporting mechanism, and sends the danger information to the designated recipient via the wireless network, while sending a distress sound to the surrounding area; S8: When it is confirmed that the person has been successfully rescued, the distress signal is turned off.

2. The method for danger perception and reporting through a brain-computer interface according to claim 1, characterized in that: In S1, the EEG acquisition device may be an EEG headband.

3. The method for danger perception and reporting through a brain-computer interface according to claim 1, characterized in that: In S2, a 50 Hz notch filter is used to remove the 50 Hz mains interference, a 30 Hz low-pass filter is selected to remove the electromyographic signal, and an electrooculogram artifact calibration is performed using the vertical electrooculogram as a reference level to obtain a signal without the influence of the electrooculogram.

4. The method for danger perception and reporting through a brain-computer interface according to claim 1, characterized in that: In S3, before extracting the key features, the data is first subjected to maximum and minimum normalization processing, and the specific method is as follows: Among them, x min is the minimum value in the data, x max is the maximum value in the data, scales all dimension values ​​to the interval [0,1], and finds the normalized feature matrix.

5. The method for danger perception and reporting through a brain-computer interface according to claim 4 is characterized in that: In S3, when extracting key features, EEG features are extracted from signals in the 8-13 Hz and 14-30 Hz frequency bands respectively. The specific method is as follows: The discrete Fourier transform formula is: Where x(k) is the EEG signal to be processed, w(T) is the window function, m is the total number of samples, and the length is T = [0, t cut,j ]Prepare the characteristics of the EEG signal, select a time window with a step size of 200ms, perform Fourier transform on the EEG signal, and obtain the amplitude density function f(k) in the frequency domain: Where N = n 2 , n is the total number of samples, The average values ​​of energy indexes in the 8-13 Hz and 14-30 Hz frequency bands were calculated, which are the characteristics of EEG signals.

6. The method for danger perception and reporting through a brain-computer interface according to claim 5, characterized in that: The calculated EEG signal features are processed by dimensionality reduction. The specific dimensionality reduction steps are as follows: S31. The data sample function X is an n×m matrix, where n is the total number of samples and m is the total number of features. The data sample function X is: S32, normalize the data sample function X: Where, i = 1, 2…n; j = 1, 2…m; S33, calculate the covariance matrix cov of the data sample function X x : Calculate the eigenvalue λ of the matrix j and the eigenvector v i S34, the eigenvalue λ j Sort by size and calculate the contribution rate of each eigenvalue: P is the number of selected components, and 90% is used as the cumulative contribution rate threshold. The eigenvectors corresponding to p=8 maximum eigenvalues ​​are extracted from the results to form samples, thereby achieving dimensionality reduction of EEG signal features.

7. The method for danger perception and reporting through a brain-computer interface according to claim 1, characterized in that: In S4, the machine learning algorithm is one of SVM or neural network.

8. The method for danger perception and reporting through a brain-computer interface according to claim 1, characterized in that: In S7, the danger information includes location, time and danger type.

9. The method for danger perception and reporting through a brain-computer interface according to claim 1, characterized in that: In S7, the recipient is a guardian or an emergency rescue center.

Images