A method and system for active and passive intention recognition based on electroencephalogram signals

Abstract

The application provides a kind of active and passive intention recognition method and system based on electroencephalogram signal: the method comprises: obtaining the electroencephalogram signal of test personnel;Extract the electroencephalogram feature of multiple frequency bands based on electroencephalogram signal;And based on the electroencephalogram signal power spectrum density of preset frequency band in the electroencephalogram feature of multiple frequency bands, obtain active feature and passive feature;Multiple frequency bands of electroencephalogram feature, active feature and passive feature are input into pre-trained active and passive intention recognition model, and the active and passive intention categories of test personnel are obtained by identification;Active and passive intention recognition model is trained based on the electroencephalogram signal obtained from active intention experiment and passive intention experiment.The application solves the problem that the processing of electroencephalogram signal in the prior art does not consider the influence of active and passive intention, resulting in poor accuracy of control instruction output by brain-computer interface.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of electroencephalogram signal recognition, and particularly relates to a method and system for recognizing active and passive intentions based on electroencephalogram signals. BACKGROUND

[0002] Brain-computer interface technology is a technology for realizing interaction by directly connecting a human brain and an external device. The use of electroencephalogram signals in brain-computer interface technology is mostly reflected in the processing and classification of electroencephalogram signals in brain areas such as motor areas, sensory areas, and language areas. It enables people to control computers or other external devices through thoughts or electroencephalogram activities, or to recognize the action intentions and control instructions of people by analyzing and processing electroencephalogram signals corresponding to limb movements.

[0003] However, human intentions are divided into active intentions and passive intentions. Currently, the processing of electroencephalogram signals does not take into account the influence of active and passive intentions, which is not conducive to obtaining the active intentions of the brain to effectively assist in the interaction between electroencephalogram and other devices, and affects the accuracy of the control instruction output of the brain-computer interface system. SUMMARY

[0004] In view of the above analysis, the present application aims to provide a method and system for recognizing active and passive intentions based on electroencephalogram signals, to solve the problem that the processing of electroencephalogram signals in the prior art does not take into account the influence of active and passive intentions, resulting in poor accuracy of the control instruction output of the brain-computer interface.

[0005] The main purpose of the present application is achieved by the following technical solutions:

[0006] On the one hand, the present application provides a method for recognizing active and passive intentions based on electroencephalogram signals, which comprises:

[0007] obtaining the electroencephalogram signals of the test subject;

[0008] extracting electroencephalogram features of multiple frequency bands based on the electroencephalogram signals, and obtaining active features and passive features based on the power spectral density of the electroencephalogram signals of a preset frequency band among the electroencephalogram features of multiple frequency bands;

[0009] inputting the electroencephalogram features of multiple frequency bands, the active features, and the passive features into a pre-trained active and passive intention recognition model to recognize the active and passive intention categories of the test subject; the active and passive intention recognition model is trained based on the electroencephalogram signals obtained from active intention experiments and passive intention experiments.

[0010] Further, the electroencephalogram features of multiple frequency bands include alpha waves, beta waves, gamma waves, delta waves, and theta waves in the frontal lobe region; and the preset frequency band includes alpha waves, beta waves, and theta waves.

[0011] 3. The electroencephalogram-based active-passive intention recognition method according to claim 2, wherein the active feature and the passive feature are obtained based on the power spectral densities of the alpha wave, the beta wave and the theta wave through the following formula:

[0012]

[0013] wherein F p is the active feature, F n is the passive feature, P α , P β and P θ are the power spectral densities of the alpha wave, the beta wave and the theta wave, respectively. p n α β θ

[0014] Further, the active-passive intention recognition model is based on a random forest model and is obtained through the following method:

[0015] A plurality of electroencephalogram signals are obtained through active intention experiments and passive intention experiments, and active intention and passive intention labels are labeled.

[0016] The plurality of electroencephalogram signals are preprocessed, electroencephalogram features of a plurality of frequency bands are extracted, and the active feature and the passive feature of each electroencephalogram signal are calculated.

[0017] A training sample set is constructed based on the electroencephalogram features of a plurality of frequency bands corresponding to a plurality of electroencephalogram signals, the active feature, the passive feature and the labeled label.

[0018] The training sample set is divided into k training subsets, each training subset is trained based on a preset training number, and an active-passive intention recognition model is obtained. In one training process, a plurality of sample data are randomly selected from each training subset, m features are randomly selected based on the plurality of sample data, a decision tree is obtained through information gain method based on the m features, and the output of the decision tree is the active-passive intention category corresponding to the decision tree. The active-passive intention recognition model is obtained based on the plurality of decision trees obtained based on the k training subsets.

[0019] Further, the recognition result of the active-passive intention recognition model is obtained based on the output results of the k decision trees through absolute majority voting method. The absolute majority voting method is represented as:

[0020]

[0021] wherein c j represents the jth classification category, h k represents the output result of the jth decision tree, and k represents the number of decision trees. j ​​​​​​​​​​​​​​​​​​​​The kth prediction output represents the prediction result of the sample x as the jth label, T is the total number of prediction outputs, N is the total number of prediction labels, and in the embodiment, two labels of active intention and passive intention are included, that is, if a certain label has more than half of the votes, the prediction is of this category, otherwise, the prediction is refused.

[0022] Further, the active intention experiment includes: the subject keeps standing, gazes at the screen, keeps the state of concentration, and actively reacts according to the instructions of the screen, while collecting multiple groups of brain electrical signals of active intention, labeling the active intention label and saving.

[0023] Further, the passive intention experiment includes: the subject keeps standing, does not gaze at the screen, keeps the state of relaxation; the arm of the subject is pulled by an auxiliary experimental personnel, and the subject reacts according to the instructions of the screen to simulate the action execution in the passive state, while collecting multiple groups of brain electrical signals of passive intention, labeling the passive intention label and saving.

[0024] In another aspect, the embodiment also provides a passive intention recognition system based on brain electrical signals, which includes a data acquisition module, a data transmission module and a data processing module.

[0025] The data acquisition module acquires brain electrical signals through an electrode brain cap, and sends the brain electrical signals to the data processing module for processing through the USB interface of the data transmission module;

[0026] The data processing module is used for pre-processing the brain electrical signals, and using the feature data obtained after pre-processing, the brain electrical signals corresponding to the active and passive intention categories are recognized through a pre-trained active and passive intention recognition model; the active and passive intention recognition model is based on a random forest model, and is trained through the brain electrical signals acquired through the active intention experiment and the passive intention experiment.

[0027] Further, when the brain electrical signals are collected by the data acquisition module, the working electrode of the electrode brain cap is arranged at the position corresponding to FP1, FP2 and FP2 of the frontal lobe of the subject. z The reference electrode is arranged at the right ear lobe, and the grounding electrode is arranged at the left ear lobe.

[0028] Further, the data processing module is used for pre-processing the brain electrical signals, including:

[0029] After the collected brain electrical signals are denoised, amplified and A / D converted, the brain electrical features of multiple frequency bands are extracted; the brain electrical features of multiple frequency bands include alpha wave, beta wave, gamma wave, delta wave and theta wave.

[0030] The power spectral density of the alpha wave, the beta wave and the theta wave is obtained by the Welch method.

[0031] Based on the power spectrum density of the alpha wave, beta wave and theta wave, the active feature and passive feature corresponding to the electroencephalogram signal are calculated by the following formula:

[0032]

[0033] Wherein, F p is the active feature, F n is the passive feature, P α , P β and P θ are the power spectrum densities of the alpha wave, beta wave and theta wave respectively.

[0034] The beneficial effects of the technical solution are as follows:

[0035] The present application is based on the characteristics of different wave bands of electroencephalogram signals, especially the alpha wave detected when in a relaxed or closed eye state, and the beta wave easily detected when in a state of concentration and active logical thinking. First, the active feature and passive feature corresponding to the electroencephalogram signal are obtained, and based on the active feature, passive feature and all electroencephalogram features, they are used together to analyze and judge the active intention and passive intention of the brain. An active intention experiment and a passive intention experiment are constructed to obtain the electroencephalogram signal, and the active feature and passive feature corresponding to the electroencephalogram signal of the active intention experiment and passive intention experiment are added to the model training, which can provide more information and obtain better classification effect. The method can effectively assist medical devices and daily wearable external devices to monitor the brain state.

[0036] Other features and advantages of the present application will be set forth in the following description, and in part will become apparent to those skilled in the art from the description, or can be learned by practice of the present application. The objects and other advantages of the present application will be realized and attained by the structure particularly pointed out in the written description and claims hereof as well as the appended drawings. BRIEF DESCRIPTION OF DRAWINGS

[0037] The accompanying drawings, which are included to provide a further understanding of the application and are incorporated in and constitute a part of this application, illustrate embodiments of the application and together with the description serve to explain the principles of the application. In the drawings:

[0038] Figure 1 is the active feature and passive feature recognition method flowchart based on electroencephalogram signals of the embodiment of the present application;

[0039] Figure 2 is the setting position schematic diagram of the electrode of the electroencephalogram acquisition device of the embodiment of the present application. DETAILED DESCRIPTION

[0040] The preferred embodiments of the present invention are described in detail below in conjunction with the accompanying drawings, wherein the accompanying drawings constitute a part of this application and are used together with the implementation cases of the present invention to illustrate the principles of the present invention, and are not used to limit the scope of the present invention.

[0041] One embodiment of the present invention provides a method for identifying active and passive intentions based on EEG signals. Figure 1 As shown, the method includes the following steps:

[0042] Step S1: obtaining the EEG signal of the subject;

[0043] Specifically, this embodiment uses a gold cup electrode EEG cap to collect EEG signals, and collects EEG signals of three channels in total. Since active intention and passive intention belong to the category of attention, Figure 2 As shown ( Figure 2 101 is the nose position, 102 is the right ear position), the electrodes are placed in the prefrontal cortex area that performs cognitive functions and thinking, and the corresponding numbers are FP1, FP z and FP2; the reference electrode is set at the A2 position of the right earlobe, and the ground electrode is set at the A1 position of the left earlobe; to obtain EEG signals related to cognitive function and thinking for active and passive intention recognition.

[0044] Step S2: extracting EEG features of multiple frequency bands based on the EEG signal; and obtaining active features and passive features based on the EEG signal power spectrum density of a preset frequency band among the EEG features of the multiple frequency bands;

[0045] Specifically, EEG signals, also known as brain waves, are bioelectric signals generated when information is transmitted between neurons. This embodiment uses the five main brain waves that can be emitted by the human brain: α waves, β waves, γ waves, δ waves, and θ waves, among which:

[0046] The frequency range of alpha waves is 8Hz to 12Hz. It is the most obvious wave among rhythmic brain waves, and the amplitude is higher on the dominant side. When a person is awake and relaxed or has their eyes closed, alpha waves can be detected in the EEG detected in the occipital lobe area.

[0047] The frequency range of beta waves is 12Hz to 30Hz, which are low-amplitude waves. When brain waves in this frequency band appear, people tend to be in a state of focused attention, active logical thinking, emotional fluctuations, alertness or anxiety;

[0048] Gamma waves have a frequency range of 30 to 100 Hz. When these waves appear, people are usually in a state of excitement or arousal, or experiencing strong stimulation. They also appear in cross-modal perceptual processing, such as object recognition and short-term memory of sound or touch.

[0049] Delta waves have a frequency range of less than 4 Hz, high amplitude waves, usually occur when adults are in deep sleep;

[0050] Theta waves have a frequency range of 4-8 Hz, occur in the brain of infants, and in adults and adolescents when drowsy (or early sleep) occurs; also occur when the human brain is idle or meditating.

[0051] After obtaining the electroencephalogram signal, first extract the electroencephalogram features of the above 5 frequency bands, and obtain the power spectral density features of each frequency band electroencephalogram feature by using the Welch method;

[0052] Welch power spectral density estimation is a frequency spectrum estimation method based on signal segmentation average, which can perform frequency spectrum analysis on time domain signals to obtain the energy distribution of signals at different frequencies. The power spectrum density mean of the above 5 frequency bands is calculated to obtain the power spectrum of each frequency band. First, the length N of the electroencephalogram data x(n) is divided into L segments, each segment has M data, the window function w(n) is multiplied by each segment of data, and the power spectrum of the i-th segment of electroencephalogram data is obtained as:

[0053]

[0054] Where n represents the nth data in each segment of data, e -jαn is an imaginary function, and U is a normalization factor:

[0055]

[0056] Finally, the power spectral density is obtained by the following formula:

[0057]

[0058] After obtaining the power spectral density of the above 5 frequency bands, the active feature and passive feature are calculated based on the preset frequency band electroencephalogram power spectral density, and the preset frequency band includes alpha wave, beta wave and theta wave. The calculation formula is:

[0059]

[0060] Where F p is the active feature, F n is the passive feature, P α , P β and P θ are the power spectral densities of alpha wave, beta wave and theta wave, respectively.

[0061] It should be noted that, since the proportion of delta waves and gamma waves in the brain waves of a general adult is small, the influence of the two waves is not considered when calculating the active and passive features. When the human brain is in a state of high concentration or alertness, the brain electrical signals mainly show signals in the beta frequency band, and when the human brain is in a resting or sleeping state, the brain electrical signals mainly show signals in the alpha frequency band or lower frequency bands. Therefore, the active features and passive features are extracted based on the alpha waves and beta waves in this embodiment, which are used for active and passive intention recognition, can better highlight the influence of the active features and passive features in the brain electrical signals in the active and passive intention states, can provide more information related to the active and passive intentions, and can obtain better classification results.

[0062] Step S3: inputting the brain electrical features in multiple frequency bands, the active features and the passive features into a pre-trained active and passive intention recognition model to identify the active and passive intention categories of the test subject; the active and passive intention recognition model is trained based on the brain electrical signals obtained through the active intention experiment and the passive intention experiment. Compared with a conventional model that is not trained using the active features and the passive features, the model inputting the active features and the passive features can better distinguish the active intention and the passive intention of the test subject, and can obtain higher accuracy and recall rate in the classification effect.

[0063] Specifically, the active and passive intention recognition model is based on a random forest model and is trained through the following method:

[0064] A plurality of brain electrical signals are obtained through the active intention experiment and the passive intention experiment, and active intention and passive intention labels are labeled.

[0065] The plurality of brain electrical signals are preprocessed, brain electrical features in multiple frequency bands are extracted, and the active features and the passive features of each brain electrical signal are calculated.

[0066] A training sample set is constructed based on the brain electrical features in multiple frequency bands corresponding to the plurality of brain electrical signals, the corresponding active features, the passive features, and the labeled labels.

[0067] The training sample set is divided into k training subsets, each training subset is trained based on a preset training number, and an active and passive intention recognition model is obtained; in one training process, a plurality of sample data are randomly selected from each training subset, m features are randomly selected based on the plurality of sample data, a decision tree is obtained by splitting based on the m features through an information gain method; the output of the decision tree is the active and passive intention categories corresponding to the decision tree; the active and passive intention recognition model composed of a plurality of decision trees is trained based on the k training subsets.

[0068] Before the formal start of the electroencephalogram signal collection experiment, the test personnel need to pre-memorize the gesture actions and their names used. The gesture actions include but are not limited to static finger actions, static arm actions, dynamic finger actions, dynamic arm actions, etc. The specific gesture actions can be determined according to actual needs. According to needs, the experimental actions can also be lower limb actions, etc.

[0069] Among them, the active intention experiment includes: the test personnel stand, look at the screen, keep the attention concentrated state, and actively make a response according to the screen indication, while collecting multiple groups of active intention electroencephalogram signals, labeling active intention labels and saving. The screen indication can be various gesture action related indications.

[0070] The passive intention experiment includes: the test personnel stand, do not look at the screen, keep relaxed state; the test personnel are pulled by the auxiliary experiment personnel, and the test personnel make a response according to the screen indication to simulate the action execution in the passive state, while collecting multiple groups of passive intention electroencephalogram signals, labeling passive intention labels and saving.

[0071] When collecting electroencephalogram signals each time, first, the electroencephalogram collection device is initialized, and the electroencephalogram collection program is run; a name of a gesture action pre-memorized by the test personnel appears randomly on the screen, the test personnel make the action according to the action name, and the electroencephalogram signals are collected for 3 seconds, during which the action is kept unchanged; a rest prompt appears on the screen, lasting for 1 second, during which the initial standing state is restored; the next random gesture action name appears, and the action-rest-action process is repeated until the collection experiment of this group ends.

[0072] It should be noted that the duration of the electroencephalogram signal collection experiment and the number of electroencephalogram signals required can be determined according to actual conditions; but the duration of each electroencephalogram signal collection experiment needs to be consistent, and the number of original electroencephalogram signals obtained by the active intention experiment and the passive intention experiment needs to be consistent.

[0073] The five frequency band electroencephalogram features, active features and passive features are all input into the classification model for training to obtain the active and passive intention recognition model. The training algorithm can also be determined according to needs, such as random forest algorithm, support vector machine algorithm, convolutional neural network algorithm, etc. In this embodiment, the random forest algorithm is preferred.

[0074] Random forest algorithm is an ensemble learning method that improves the accuracy and stability of the model by constructing multiple decision trees and combining their prediction results. Each tree uses different data subsets and feature subsets during training to reduce overfitting and improve the generalization ability of the model.

[0075] During training of each tree, a subset of the total training samples is selected for training. The remaining data is used for evaluation, and the error is assessed. At each node, a subset of all features is randomly selected for use in computing the best split.

[0076] The training process is as follows: the total number of training samples used is N, and a single decision tree randomly selects n training samples from the N training sets with replacement as the training samples of the current single tree. Let the number of input features of the training samples be M, and when splitting at each node of each decision tree, m input features are randomly selected from the M input features, and m is much smaller than M. Then, a best feature is selected from the m input features based on information entropy and information gain for splitting. m remains constant during the construction of the decision tree. Each tree continues to split until all training samples of the node belong to the same class, and pruning is not needed. Finally, the class with the most classification results is taken as the classification result of the entire random forest.

[0077] The classification result can be obtained by voting, and in the embodiment, the absolute majority voting method is preferred, and the formula is as follows:

[0078]

[0079] wherein, c j represents the jth classification category, represents the kth prediction output of the prediction result of the sample x as the jth label, T is the total number of prediction outputs, and N is the total number of prediction labels. In the embodiment, two labels of active intention and passive intention are included, that is, if a certain label gets more than half of the votes, the prediction is of this category, otherwise, the prediction is refused.

[0080] The present application can better learn the EEG signal features under active intention and passive intention by constructing active intention experiments and passive intention experiments to obtain the EEG signals under active intention and passive intention for model training, and greatly improves the accuracy of active and passive intention recognition.

[0081] Another embodiment of the present application also discloses a system for recognizing active and passive intention based on EEG signals, which comprises a data acquisition module, a data transmission module and a data processing module.

[0082] The data acquisition module acquires EEG signals through an electrode EEG cap, and sends the EEG signals to the data processing module for processing through the USB interface of the data transmission module.

[0083] The data processing module is used for pre-processing the electroencephalogram signals, and using the feature data obtained after pre-processing to recognize the active and passive intention categories corresponding to the electroencephalogram signals through a pre-trained active and passive intention recognition model.

[0084] Specifically, the present application provides an electroencephalogram acquisition device for obtaining and preliminarily processing electroencephalogram signals, which is composed of a data transmission module, a data acquisition module and a data preliminary processing module. The sampling frequency of the data acquisition module is set to 250 Hz, the data acquisition module is connected to the electrode wire, the acquisition channel is opened, and the collected original electroencephalogram signals are collected. The data preliminary processing module preliminarily processes the collected electroencephalogram signals, including but not limited to noise reduction, amplification, A / D conversion, etc. of the electroencephalogram signals. The electroencephalogram signals after preliminary processing are transmitted to the host computer for subsequent processing and classification through the data transmission module. The electroencephalogram acquisition device is connected to the host computer through a USB interface, and the host computer can be a computer in this embodiment. The data transmission module converts the computer power supply to supply power to the electroencephalogram acquisition device as a whole by using the USB interface.

[0085] Further, when the data acquisition module is used to collect electroencephalogram signals, the working electrodes of the electrode electroencephalogram cap are arranged at the positions corresponding to FP1, FP2 and FPz of the frontal lobe of the subject. z The reference electrode is arranged at the right ear lobe, and the grounding electrode is arranged at the left ear lobe.

[0086] Further, the data processing module includes a data preliminary processing module and a host computer, and the pre-processing of the electroencephalogram signals includes preliminary processing of the electroencephalogram data by the data preliminary processing module and extraction of electroencephalogram features by the host computer, specifically including:

[0087] The collected electroencephalogram signals are sent to the host computer after noise reduction, amplification and A / D conversion by the data preliminary processing module;

[0088] The electroencephalogram features of multiple frequency bands are extracted from the electroencephalogram signals after preliminary processing by the host computer; the electroencephalogram features of multiple frequency bands include alpha waves, beta waves, gamma waves, delta waves and theta waves;

[0089] The power spectral densities of the alpha waves, beta waves and theta waves are obtained by the Welch method;

[0090] Based on the power spectral densities of the alpha waves, beta waves and theta waves, the active features and passive features corresponding to the electroencephalogram signals are calculated by the following formula:

[0091]

[0092] Wherein, F p is an active feature, F n is a passive feature, P α , P β and P θ are power spectral densities of alpha waves, beta waves and theta waves respectively.

[0093] In summary, the embodiment of the present application provides a method for recognizing active and passive intentions based on electroencephalogram signals, which fully utilizes the characteristics of different wave bands of electroencephalogram signals, especially the alpha waves that are easy to detect when in a relaxed or closed-eye state, and the beta waves that are easy to detect when in a state of attention concentration and active logical thinking. The active and passive features corresponding to the electroencephalogram signals are calculated, and the active and passive intentions of the brain are analyzed and judged based on the active features, passive features and all electroencephalogram features. More information can be provided, and better classification results can be obtained. The method can effectively assist medical devices and daily wearable external devices in monitoring the brain state. The present application constructs an active intention experiment and a passive intention experiment to obtain electroencephalogram signals under active and passive intentions for model training, so that the model can better learn the electroencephalogram signal features under active and passive intentions, analyze the active or passive intention of the brain, and obtain the upper limb movement intention of the brain, greatly improving the accuracy of active and passive intention recognition. The method can effectively assist medical devices and daily wearable devices in monitoring the brain state.

[0094] Those skilled in the art can understand that all or part of the processes of the above-mentioned embodiments can be completed by a computer program instructing related hardware. The program can be stored in a computer readable storage medium, such as a magnetic disk, an optical disk, a read-only memory or a random access memory.

[0095] The above description is only a preferred embodiment of the present application, and the protection scope of the present application is not limited thereto. Any changes or replacements within the technical scope disclosed by the present application can be easily thought of by those skilled in the art, and should be covered within the protection scope of the present application.

Claims

1. A method for active and passive intention recognition based on EEG signals, characterized in that: include: Obtaining the EEG signals of the subjects; Extracting EEG features of multiple frequency bands based on the EEG signal, wherein the EEG features of the multiple frequency bands include α wave, β wave, γ wave, δ wave and θ wave in the prefrontal lobe area; Based on the power spectrum density of the EEG signal in a preset frequency band among the EEG features of multiple frequency bands, the preset frequency band includes α wave, β wave and θ wave, the active feature and the passive feature are obtained by the following formula: Among them, F p is an active feature, F n is a passive feature, P α 、P β and P θ are the power spectral densities of α, β, and θ waves, respectively; The EEG features, active features and passive features of multiple frequency bands are input into a pre-trained active and passive intention recognition model to identify the active and passive intention categories of the subjects; the active and passive intention recognition model is trained based on the EEG signals obtained from the active intention experiment and the passive intention experiment.

2. The active and passive intention recognition method based on EEG signals according to claim 1 is characterized in that: The active and passive intent recognition model is based on a random forest model and is trained using the following method: Through active intention experiments and passive intention experiments, multiple EEG signals are obtained and labeled with active intention and passive intention labels; Preprocessing the plurality of EEG signals, extracting EEG features of multiple frequency bands, and calculating active features and passive features of each EEG signal; A training sample set is constructed based on EEG features of multiple frequency bands corresponding to multiple EEG signals, corresponding active features, passive features, and annotated labels; Dividing the training sample set into k training subsets, and performing training based on each training subset and a preset number of training times to obtain an active and passive intention recognition model; In a training process, a plurality of sample data are randomly selected from each training subset, and m features are randomly selected based on the plurality of sample data. Based on the m features, splitting is performed using the information gain method to obtain a decision tree; the output of the decision tree is the active and passive intent categories corresponding to the decision tree; An active and passive intention recognition model consisting of multiple decision trees is obtained based on k training subsets.

3. The method for active and passive intention recognition based on EEG signals according to claim 2, characterized in that: Based on the output results of k decision trees, the recognition results of the active and passive intention recognition model are obtained by using the absolute majority voting method; the absolute majority voting method is expressed as: Among them, c j represents the jth classification category, It means that the prediction result on sample x is the k-th prediction output of the j-th class label, T is the total number of prediction outputs, and N is the total number of predicted labels. That is, if a label gets more than half of the votes, it is predicted to be that category, otherwise the prediction is rejected.

4. The method for active and passive intention recognition based on EEG signals according to claim 2, characterized in that: The active intention experiment includes: the subject stands, looks at the screen, maintains a focused state, and actively responds according to the instructions on the screen. At the same time, multiple groups of active intention EEG signals are collected, marked with active intention labels and saved.

5. The method for active and passive intention recognition based on EEG signals according to claim 2, characterized in that: The passive intention experiment includes: the subject remains standing, not looking at the screen, and remains in a relaxed state; an auxiliary experimenter pulls the subject's arm and assists the subject in responding to instructions on the screen to simulate the execution of actions in a passive state, and at the same time collects multiple sets of passive intention EEG signals, annotates them with passive intention labels, and saves them.

6. An active and passive intention recognition system based on EEG signals, characterized in that: The system includes a data acquisition module, a data transmission module and a data processing module; The data acquisition module acquires EEG signals through the electrode EEG cap and sends the EEG signals to the data processing module for processing through the USB interface of the data transmission module; The data processing module is used to pre-process the EEG signal, including: After performing noise reduction, amplification, and A / D conversion on the collected EEG signals, EEG features of multiple frequency bands are extracted; the EEG features of the multiple frequency bands include α waves, β waves, γ waves, δ waves, and θ waves; The power spectral density of α, β, and θ waves was obtained by the Welch method; Based on the power spectral density of α waves, β waves, and θ waves, the active and passive features corresponding to the EEG signal are calculated using the following formula: Among them, F p is an active feature, F n is a passive feature, P α 、P β and P θ are the power spectral densities of α, β, and θ waves, respectively; The feature data obtained after preprocessing is used to identify the active and passive intention categories corresponding to the EEG signal through a pre-trained active and passive intention recognition model; the active and passive intention recognition model is based on a random forest model and is trained through EEG signals obtained through active intention experiments and passive intention experiments.

7. The active and passive intention recognition system based on EEG signals according to claim 6 is characterized in that: When the data acquisition module is used to collect EEG signals, the working electrodes of the EEG cap are set at the prefrontal lobe area of ​​the subject, which are numbered FP1, FP2, and FP3. z and FP2; the reference electrode was placed on the right earlobe, and the ground electrode was placed on the left earlobe.

Images