Motion intention analysis method and device

Abstract

The invention discloses a motion intention analysis method and device which are used for performing motion intention recognition on electroencephalogram signals. The method comprises the following steps: acquiring an electroencephalogram signal at a human brain electrode position, and performing information compression processing on the electroencephalogram signal to obtain potential state information of the electroencephalogram signal; the electroencephalogram signals are obtained based on a plurality of electrode signals; the potential state information is used for representing spatial-temporal characteristics of the electroencephalogram signals, the spatial-temporal characteristics are used for representing collection moments and collection channels of the multiple electrode signals, and each electrode signal in the multiple electrode signals corresponds to one collection channel; then, feature data of the electroencephalogram signals are extracted from the potential state information, the motion intention corresponding to the electroencephalogram signals is analyzed according to the feature data, it is not needed to judge whether cognitive distraction exists or not on the electroencephalogram signals, motion intention recognition can be directly conducted on the electroencephalogram signals, and therefore the motion intention recognition efficiency can be improved.

Description

Technical Field

[0001] The embodiments of the present application relate to the technical field of bioelectric signal processing, and in particular, to a method and device for analyzing motion intention. Background Art

[0002] In recent years, human-machine collaborative technology has been developing rapidly in the fields of human-machine interaction, human enhancement and rehabilitation, and has gradually gained more and more attention. Movement intention is the subjective idea and instruction of controlling body movement, so decoding movement intention is one of the most important technologies in human-machine collaborative technology. How to efficiently and accurately realize the information exchange between humans and machines is the key to human-machine collaborative technology, which determines the completion effect and overall safety of human-machine system tasks. Human movement intention can be identified by various means, among which parsing human movement intention based on electroencephalogram (EEG) is a method with broad application prospects.

[0003] Currently, before analyzing the movement intention based on EEG, it is necessary to determine whether there is cognitive distraction and select the corresponding decoding model based on the judgment result. However, in actual use, since the machine cannot obtain prior information about the cognitive state of a person, it is impossible to adaptively adjust the model parameters according to the cognitive state of a person.

[0004] Therefore, there is an urgent need for a method for analyzing movement intentions with and without cognitive distraction. Summary of the invention

[0005] The embodiments of the present application provide a method and device for analyzing movement intention, which are used to identify movement intention of EEG signals without the need for cognitive distraction judgment.

[0006] In the first aspect, the present application provides a method for analyzing movement intention, which is described below with a processor for identifying movement intention as the execution subject. In this method, the processor obtains the EEG signal at the electrode position of the human brain, and performs information compression processing on the EEG signal to obtain the potential state information of the EEG signal. Among them, the EEG signal is obtained based on multiple electrode signals, and the potential state information is used to characterize the spatiotemporal characteristics of the EEG signal, and the spatiotemporal characteristics are used to characterize the acquisition time and acquisition channel of multiple electrode signals, and each electrode signal in the multiple electrode signals corresponds to an acquisition channel. The processor extracts the characteristic data of the EEG signal from the potential state information, and analyzes the movement intention corresponding to the EEG signal based on the characteristic data.

[0007] In this method, after obtaining the EEG signal, the processor can extract the characteristic data of the EEG signal from the latent state information of the EEG signal, and analyze the movement intention corresponding to the EEG signal based on the characteristic data, without having to prejudge whether cognitive distraction exists. In this way, the movement intention of EEG signals with cognitive distraction and EEG signals without cognitive distraction can be identified, thereby improving the applicability of the movement intention analysis method.

[0008] In a possible design, before performing information compression processing on the EEG signal to obtain the potential state information of the EEG signal, the processor can also pre-process the EEG signal, wherein the pre-processing includes re-referencing, baseline removal, filtering, independent component analysis and down-sampling.

[0009] Through this design, the processor can remove interference components in EEG signals.

[0010] In a possible design, the processor can input the EEG signal into a latent variable model to obtain potential state information; wherein the latent variable model can be a factor model.

[0011] Through this design, the processor can determine the potential state information of the EEG signal.

[0012] In a possible design, the processor can obtain sample EEG signals according to a set update cycle. The sample EEG signals are historical EEG signals. The processor can process the sample EEG signals based on the maximum expectation algorithm to obtain model parameters of the factor model; wherein the model parameters include a load matrix and independent factor covariances. The processor can also update the factor model according to the model parameters.

[0013] Through this design, the processor can update the factor model regularly, thereby ensuring the accuracy of the factor model.

[0014] In one possible design, the processor can also perform spectral decomposition on the sample covariance of the latent variable model to obtain multiple neural activity patterns and determine the eigenvalues ​​of the multiple neural activity patterns; wherein the eigenvalue of any neural activity pattern is used to characterize the dominance of the neural activity pattern. The processor can also perform linear decomposition on the latent state information to obtain multiple latent components; wherein each latent component is used to characterize the spatiotemporal characteristics of an electrode signal, and the multiple latent components correspond one-to-one to the multiple neural activity patterns. The processor can select multiple target latent components from the multiple latent components based on the eigenvalues ​​of the multiple neural activity patterns. The processor can obtain feature data based on the multiple target latent components.

[0015] Through this design, the processor can extract feature data from the latent state information based on the dominance of neural activity.

[0016] In one possible design, the processor can determine the dominance of each neural activity pattern according to the characteristic value of each neural activity pattern. The processor can also select multiple target neural activity patterns from multiple neural activity patterns, and the dominance of the multiple target neural activity patterns is higher than the dominance of neural activity patterns other than the multiple target neural activity patterns in the multiple neural activity patterns. The processor selects multiple potential components corresponding to the multiple target neural activity patterns from the multiple potential components as multiple target potential components.

[0017] Through this design, the processor can select multiple target latent components that can represent movement intention from multiple latent components based on the dominance of multiple neural activity patterns.

[0018] In one possible design, the processor can standardize the feature data and input the processed feature data into a classification model; the processor classifies the processed feature data based on the classification model to obtain the movement intention corresponding to the EEG signal.

[0019] Through this design, the processor can identify the feature data based on the classification model and obtain the movement intention corresponding to the EEG signal.

[0020] In a second aspect, the present application further provides a motion intention analysis device, the device comprising:

[0021] An acquisition unit, used to acquire an electroencephalogram signal at a position of a human brain electrode; the electroencephalogram signal is obtained based on a plurality of electrode signals;

[0022] A processing unit, used for performing information compression processing on the EEG signal to obtain potential state information of the EEG signal; the potential state information is used to characterize the spatiotemporal characteristics of the EEG signal; the spatiotemporal characteristics are used to characterize the acquisition time and acquisition channel of the multiple electrode signals; each electrode signal in the multiple electrode signals corresponds to an acquisition channel; and feature data of the EEG signal is extracted from the potential state information;

[0023] An analysis unit is used to analyze the movement intention corresponding to the EEG signal according to the characteristic data.

[0024] According to a third aspect, an electronic device is provided, comprising a processor and a memory, wherein one or more computer programs are stored in the memory, and the one or more computer programs include instructions. When the processor calls the instructions, the electronic device executes the motion intention analysis method of the first aspect.

[0025] In a fourth aspect, the present application provides a computer-readable storage medium having a computer program or instructions stored thereon, which, when executed, enables a computer to execute the method in the above-mentioned first aspect or any possible implementation of the first aspect.

[0026] In a fifth aspect, the present application provides a computer program product. When a computer executes the computer program product, the computer executes the method in the above-mentioned first aspect or any possible implementation manner of the first aspect.

[0027] In a sixth aspect, an embodiment of the present application further provides a chip, which is used to read a computer program stored in a memory and execute the method in the above-mentioned first aspect and any possible design of the first aspect.

[0028] For the beneficial effects of the second to sixth aspects mentioned above, please refer to the description of the beneficial effects of the first aspect mentioned above, and will not be repeated here. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 A schematic diagram of an application scenario of motion intention analysis provided in an embodiment of the present application;

[0030] Figure 2 A schematic diagram of the position of electrodes in the human brain provided in an embodiment of the present application;

[0031] Figure 3 A schematic diagram of a motion intention analysis method provided in an embodiment of the present application;

[0032] Figure 4 A schematic diagram of a use case of motion intention recognition provided in an embodiment of the present application;

[0033] Figure 5 A complete flow chart of a motion intention analysis method provided in an embodiment of the present application;

[0034] Figure 6 A schematic diagram of an experimental paradigm provided in an embodiment of the present application;

[0035] Figure 7 A schematic diagram of a method for dividing a training-testing data set provided in an embodiment of the present application;

[0036] Figure 8 A schematic diagram of the performance results of a classification model provided in an embodiment of the present application;

[0037] Fig. 9 A schematic diagram of the change of processor memory usage over time when performing motion intent analysis provided by an embodiment of the present application;

[0038] Fig.10A schematic diagram of the structure of a motion intention analysis device provided in an embodiment of the present application;

[0039] Fig.11 A schematic diagram of the structure of an electronic device according to the solution provided in an embodiment of the present application. DETAILED DESCRIPTION

[0040] The embodiment of the present application provides a method and device for analyzing motion intention. The method and device are based on the same concept. Since the method and device solve the problem in a similar way, the implementation of the device and the method can refer to each other, and the repeated parts will not be repeated.

[0041] In order to make the purpose, technical scheme and advantages of the embodiments of the present application clearer, the embodiments of the present application will be further described in detail below in conjunction with the accompanying drawings. Among them, in the description of the embodiments of the present application, the terms "first" and "second" are used only for descriptive purposes and cannot be understood as indicating or implying relative importance or implicitly indicating the number of indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include one or more of the features.

[0042] To facilitate understanding, exemplary descriptions of concepts related to the present application are provided for reference.

[0043] 1) Electroencephalogram (EEG) is the spontaneous, rhythmic activity of brain electrical cell groups recorded by electrodes.

[0044] 2) Independent component analysis (ICA) is a method of calculation based on statistical principles. It is a linear transformation that separates data or signals into linear combinations of statistically independent non-Gaussian signal sources.

[0045] 3) Factor analysis (FA) is a multivariate statistical method for discovering potential patterns from raw data. Its core is to describe the relationship between the covariances of a large number of variables through a small number of random variables "factors" that reflect the essence but cannot be directly observed.

[0046] 5) The n-back task is a delayed digital memory task with auditory stimulation and verbal feedback designed by the Age Laboratory at the Massachusetts Institute of Technology in 2006. It has now become a widely used standard method to stimulate quantitative cognitive workload.

[0047] In the description of the embodiments of the present application, "and / or" describes the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B can represent three situations: A exists alone, A and B exist at the same time, and B exists alone. The character " / " generally indicates that the objects associated before and after are in an "or" relationship. At least one referred to in the present application refers to one or more; multiple refers to two or more. In addition, it should be understood that in the description of the present application, words such as "first" and "second" are only used to distinguish the purpose of description, and cannot be understood as indicating or implying relative importance, nor can they be understood as indicating or implying order. The embodiments of the present application will be described in detail below in conjunction with the accompanying drawings.

[0048] At present, before the EEG signal is used to identify the motor intention, it is necessary to judge the cognitive distraction of the EEG signal, and select the corresponding intention analysis model based on the judgment result to identify the motor intention of the EEG signal. Among them, this method relies on the prior information of the person's cognitive state. However, when using this method, the processor cannot obtain the prior information of the person's cognitive state, that is, it is impossible to adjust the model parameters according to the person's cognitive state, resulting in low recognition accuracy of the motor intention. Therefore, there is an urgent need for a method for analyzing the motor intention of EEG signals with and without cognitive distraction.

[0049] In order to improve the accuracy of motor intention analysis, an embodiment of the present application provides a motor intention analysis method. The following is an explanation using a processor that identifies motor intention as the execution subject: In this method, the processor obtains an EEG signal at the electrode position of a human brain; wherein the EEG signal is obtained based on multiple electrode signals. The processor performs information compression processing on the EEG signal to obtain potential state information of the EEG signal; wherein the potential state information is used to characterize the spatiotemporal characteristics of the EEG signal, and the spatiotemporal characteristics are used to characterize the acquisition time and acquisition channel of multiple electrode signals, and each electrode signal in the multiple electrode signals corresponds to an acquisition channel. Then, the processor can extract the characteristic data of the EEG signal from the potential state information, and identify the motor intention corresponding to the EEG signal based on the characteristic data, without having to pre-judge whether there is cognitive distraction, so that the motor intention can be identified for EEG signals in both cases of cognitive distraction and no cognitive distraction, thereby improving the applicability and accuracy of the motor intention analysis method.

[0050] The embodiments of the present application will be described in detail below with reference to the accompanying drawings.

[0051] Figure 1 A schematic diagram of a possible application scenario of motion intention analysis applicable to the solution provided in the embodiment of the present application. Figure 1As shown, the application scenario includes an EEG collector 10 and a processor 20. The EEG collector 10 and the processor 20 may be in the same electronic device or in different electronic devices. Exemplarily, when the EEG collector 10 and the processor 20 are in the same electronic device, the electronic device may be a non-invasive brain-computer interface acquisition device. When the EEG collector 10 and the processor 20 are in different electronic devices, the EEG collector 10 may be a head cap, and the processor 20 may be a processor in a terminal device or a server.

[0052] The EEG acquisition instrument 10 acquires EEG signals at the electrode positions of the human brain in real time, and transmits the acquired EEG signals to the processor 20. Exemplarily, after acquiring the EEG signals, the EEG acquisition instrument 10 can amplify and convert the EEG signals into analog-to-digital signals in real time through a wireless amplifier, and transmit the converted EEG signals to the processor 20.

[0053] In some examples, since the brain needs to control upper limb movements and bear a certain cognitive workload at the same time, perception, movement, and cognitive related brain areas are involved. Therefore, according to the 10-20 international standard lead rule, the electrode positions of the EEG collector 10 can be selected as Fpz, Fp1, Fp2, AFz, AF3, AF4, AF7, AF8, Fz, F1, F2, F3, F4, F5, F6, F7, F8, FCz, FC1, FC2, FC3, FC4, FC5, FC6, FT7, FT8, Cz, C1, C2, C3, C4, C5, C6, T7, T8, CPz, CP1, CP2, CP3, CP4, CP5, CP6, TP7, TP8, Pz, P3, P4, P5, P6, P7, P8, POz, PO3, PO4, PO5, PO6, PO7, PO8, Oz, O1, and O2 positions of the human brain, such as Figure 2 As shown. The reference electrode is set at the CPz position, and the ground electrode is set at the AFz position. In addition, the embodiment of the present application can also select other electrode positions that comply with the 10-20 international standard lead rule, which is not limited here.

[0054] The EEG acquisition instrument 10 acquires multiple electrode signals based on the electrode positions, and obtains EEG signals based on the multiple electrode signals. After the EEG acquisition instrument 10 sends the EEG signals to the processor 20, the processor 20 can perform information compression processing on the EEG signals to obtain the potential state information of the EEG signals. Among them, the potential state information is used to characterize the spatiotemporal characteristics of the EEG signals, and the spatiotemporal characteristics are used to characterize the acquisition time and acquisition channel of the multiple electrode signals, and each of the multiple electrode signals corresponds to an acquisition channel.

[0055] The processor 20 extracts characteristic data of the EEG signal from the potential state information, and analyzes the movement intention corresponding to the EEG signal based on the characteristic data. In the above manner, the processor 20 can directly identify the movement intention of the collected EEG signal without the need to make a cognitive distraction judgment in advance, thereby reducing the impact of the person's cognitive state on the movement intention recognition, and further improving the efficiency and accuracy of the movement intention recognition.

[0056] The solution provided by this application is introduced below in conjunction with specific embodiments.

[0057] Example 1

[0058] Figure 3 This is a schematic diagram of a method for analyzing movement intention provided by an embodiment of the present application. The following description is based on a processor that analyzes movement intention of EEG signals as the execution subject. Figure 3 As shown, the method includes:

[0059] S301: The processor obtains an electroencephalogram (EEG) signal at a position of an electrode in a human brain, wherein the EEG signal is obtained based on signals from multiple electrodes.

[0060] In some embodiments, after collecting the EEG signal, the EEG acquisition device can transmit the EEG signal to the processor. In other embodiments, the processor can also collect EEG signals at the position of the human brain electrode. The sampling rate of the EEG signal can be 1000 Hz or other frequencies, which are not limited here.

[0061] S302: The processor performs information compression processing on the EEG signal to obtain potential state information of the EEG signal.

[0062] The latent state information is used to characterize the spatiotemporal characteristics of the EEG signal, and the spatiotemporal characteristics are used to characterize the acquisition time and acquisition channel of multiple electrode signals, and each electrode signal in the multiple electrode signals corresponds to an acquisition channel. Exemplarily, the latent state information can also characterize the coordination law between multiple electrode signals in the EEG signal.

[0063] In some embodiments, before the processor performs information compression processing on the EEG signal, it can also pre-process the EEG signal to remove artifacts in the EEG signal and obtain the frequency part containing the information required for motion intention recognition. The pre-processing includes re-referencing, baseline removal, filtering, independent component analysis and downsampling.

[0064] In some examples, the processor may pre-process the EEG signal through the following steps A1-A6:

[0065] A1: The processor can re-reference the EEG signal to the average of the bilateral mastoid processes behind the ears to obtain a first corrected signal.

[0066] In some embodiments, the embodiments of the present application may set two re-reference channels at the mastoid positions behind the ears to obtain mastoid signals. After obtaining the mastoid signals collected by the two re-reference channels, the processor performs mean processing on them to obtain the mean of the bilateral mastoids behind the ears. The processor re-references the EEG signal to the mean of the bilateral mastoids behind the ears to obtain a first corrected signal.

[0067] A2: The processor may perform baseline removal on the first correction signal to obtain a second correction signal.

[0068] In some embodiments, the processor may eliminate the baseline drift of the first corrected signal by detrending to obtain a second corrected signal.

[0069] A3: The processor may filter the second corrected signal to obtain a third corrected signal.

[0070] In some embodiments, the processor may perform bandpass filtering on the second correction signal using an ideal Fourier transform to obtain a third correction signal.

[0071] Exemplarily, since the neural activity information related to movement is mainly concentrated in the lower frequencies of the EEG signal, the processor can use an ideal Fourier transform to perform a 0.5-10 Hz bandpass filter on the second corrected signal to obtain a third corrected signal.

[0072] A4: The processor may also perform filtering processing and independent component analysis on the second corrected signal to obtain a mixing matrix and an unmixing matrix.

[0073] In some embodiments, after the processor performs bandpass filtering on the second corrected signal, it performs independent component analysis on the filtered second corrected signal to obtain a mixing matrix and an unmixing matrix.

[0074] Exemplarily, in order to prevent ultra-low frequency information (e.g., 0-2 Hz) in the EEG signal from affecting the separation results of other frequency bands, the processor may pre-band-pass filter the second correction signal using an ideal Fourier transform. For example, when the ultra-low frequency information is 0-2 Hz, the processor performs a 2-10 Hz band-pass filter on the second correction signal to prevent it from affecting the separation results of other frequency bands.

[0075] The processor may also perform independent component analysis on the filtered second corrected signal to obtain a mixing matrix and an unmixing matrix.

[0076] A5: The processor may also apply the unmixing matrix to the third corrected signal to obtain independent components. The processor may apply the mixing matrix to the independent components to obtain a fourth corrected signal.

[0077] In some embodiments, after obtaining the independent component of the third corrected signal, the processor may calculate the correlation coefficient between the independent component and the electrooculogram signal, wherein the electrooculogram signal may be collected by four electrodes, and the four electrodes are respectively arranged below the lower eyelids and outside the outer canthus of the left and right eyes.

[0078] Exemplarily, after collecting the electrooculogram signal, the processor can calculate the correlation coefficient between the independent component and the electrooculogram signal. The processor can regard the independent component with a correlation coefficient greater than 0.4 as an electrooculogram artifact, and set the independent component to 0 to achieve the purpose of removing the electrooculogram artifact. After removing the electrooculogram artifact, the processor can use the mixing matrix for the independent component to obtain a fourth corrected signal.

[0079] A6: The processor may downsample the fourth corrected signal to obtain a preprocessed EEG signal.

[0080] In some embodiments, the processor may downsample the fourth corrected signal to 100 Hz to obtain a preprocessed EEG signal.

[0081] After preprocessing the EEG signal, the processor can perform information compression processing on the preprocessed EEG signal, visualize multiple independent electrode signals constituting the EEG signal, and obtain potential state information of the EEG signal.

[0082] In some embodiments, the processor can use factor analysis to compress and extract information from the EEG signal to obtain the potential state information of the EEG signal. Since factor analysis has a strong ability to identify and separate covariant components from multiple observations, a latent variable model is used in the processor in the non-invasive brain-computer interface, based on the assumption that "the observation of complex neural activity inside the brain at the electrode is the superposition of the projections of several linearly independent neural activities at the corresponding electrode positions", so that the processor in this application can more effectively compress and abstract the potential state information from the complex and mixed EEG signal, and visualize the information that was originally unobservable but hidden in nature. Exemplarily, the processor can abstract and separate the electrode signals of mutually independent neural activities from the preprocessed EEG signal based on factor analysis to obtain the potential state information of the neural activity.

[0083] The processor can input the EEG signal into a latent variable model to obtain potential state information of the EEG signal, wherein the latent variable model can be a factor model.

[0084] In some embodiments, the processor needs to update the parameters of the latent variable model before inputting the EEG signal into the latent variable model to obtain the potential state information. The embodiment of the present application can construct the latent variable model in the following manner:

[0085] B1: The processor can establish a factor model of EEG signals based on factor analysis.

[0086] Among them, the factor model can be initially defined as:

[0087] y=LF+∈

[0088] Among them, y represents the EEG signal, L represents the load matrix, F represents the initial potential state information, and ∈ represents random noise.

[0089] In some embodiments, when establishing the factor model, the processor may also calculate the mean value of the electrode signal of each acquisition channel in the EEG signal.

[0090] B2: The processor can estimate the model parameters of the factor model by using the maximum expectation algorithm, wherein the model parameters of the factor model include the load matrix L and the independent factor covariance ψ.

[0091] B3: The processor can calculate the initial potential state information based on the model parameters.

[0092] Exemplarily, the initial potential state information can be expressed by the following formula:

[0093] F=L T (LL T +ψ) -1 y

[0094] Among them, y represents the EEG signal, F represents the initial potential state information, L represents the loading matrix, and ψ represents the independent factor covariance.

[0095] B4: The processor performs noise-free estimation on the EEG signal based on the initial potential state information to obtain an idealized EEG signal, wherein the idealized EEG signal refers to an EEG signal without interference factors such as noise.

[0096] For example, an idealized EEG signal can be defined as:

[0097] y clean =LL T (LL T +ψ) -1 y

[0098] Among them, y clean represents the idealized EEG signal, y represents the EEG signal, L represents the loading matrix, and ψ represents the covariance of independent factors.

[0099] The processor can define the factor model as:

[0100] y clean =LL T (LL T +ψ) -1 y

[0101] Among them, yclean represents the idealized EEG signal, y represents the EEG signal, L represents the loading matrix, and ψ represents the covariance of independent factors.

[0102] At this time, the processor can replace the collected EEG signal with an idealized EEG signal and observe the idealized EEG signal, thereby eliminating the error introduced by the non-ideal experimental conditions and obtaining the idealized potential state information.

[0103] Among them, the sample covariance of the established factor model is the idealized sample covariance cov(y clean )=LL T .

[0104] In some embodiments, after acquiring the EEG signal, the processor may input the EEG signal into a factor model, and obtain the potential state information of the EEG signal based on the factor analysis method. The potential state information of the EEG signal is the idealized potential state information F clean =L T (LL T ) -1 y clean .

[0105] In some embodiments, the processor also needs to regularly update the parameters of the factor model to ensure the accuracy of the factor model. The processor can update the factor model through the following steps C1-C3:

[0106] C1: The processor obtains sample EEG signals according to a set update cycle, wherein the sample EEG signals may be historical EEG signals.

[0107] In some examples, the processor may select a portion of the EEG signals from historical EEG signals as sample EEG signals.

[0108] C2: The processor can process the sample EEG signal based on the maximum expectation algorithm to obtain the model parameters of the factor model.

[0109] The processor may substitute the sample EEG signal into the initial definition y=LF+∈ of the factor model based on the maximum expectation algorithm to obtain the model parameters of the factor model.

[0110] C3: The processor may update the factor model according to the determined model parameters.

[0111] Exemplarily, after obtaining the model parameters, the processor may substitute the obtained model parameters into y clean =LL T (LL T +ψ) -1 y, the update of the factor model is completed.

[0112] S303: The processor extracts characteristic data of the EEG signal from the potential state information.

[0113] In some embodiments, the processor may extract characteristic data of the EEG signal from the potential state information through the following steps D1-D4:

[0114] D1: The processor can perform spectral decomposition on the sample covariance of the latent variable model to obtain multiple neural activity patterns and determine the eigenvalues ​​of the multiple neural activity patterns, wherein the eigenvalue of any neural activity pattern is used to characterize the dominance of the neural activity pattern.

[0115] In the embodiment of the present application, the sample covariance of the factor model directly characterizes the correlation between the various acquisition channels, reflecting a large amount of overlapping information when complex neural activities occur in the brain.

[0116] In some embodiments, the processor performs spectral decomposition on the sample covariance matrix to obtain a set of linearly superimposed neural activity patterns, and determines eigenvalues ​​of the plurality of neural activity patterns.

[0117] The processor can determine the contribution of the neural activity pattern in the sample covariance according to the size of the eigenvalue corresponding to each neural activity pattern. For example, the larger the eigenvalue, the stronger the representation dominance, the larger the central deviation of the neural activity pattern, and the stronger the synergy law.

[0118] The processor can determine the dominance of each neural activity pattern according to the characteristic value of each neural activity pattern. This application uses the dominance analysis of neural activity patterns as a method for obtaining prior information to mine and determine the most dominant part of the neural activity pattern.

[0119] The processor may select multiple target neural activity patterns from multiple neural activity patterns. The dominance of the multiple target neural activity patterns is higher than the dominance of neural activity patterns other than the multiple target neural activity patterns in the multiple neural activities. For example, the processor may screen out the target neural activity patterns with a dominance higher than a set threshold, or select the first n target neural activity patterns with the highest dominance in order.

[0120] After determining the dominance of the plurality of neural activity patterns, the processor may arrange the neural activity patterns in descending order based on the dominance. The processor may select the first K neural activity patterns in the neural activity pattern sequence as target neural activity patterns. For example, K may be 5.

[0121] The embodiment of the present application can use spectral decomposition to further analyze the sample covariance, and reconstruct a priori EEG information representation model based on the dominance of the decomposed neural activity pattern over the overall neural activity.

[0122] D2: The processor can linearly decompose the latent state information into multiple latent components.

[0123] Among them, each potential component is used to characterize the spatiotemporal characteristics of an electrode signal, and multiple potential components correspond one-to-one to multiple neural activity patterns.

[0124] In some embodiments, the linear decomposition may be a rectangular decomposition or other decomposition methods, which is not limited in the present application.

[0125] D3: The processor may select a plurality of target latent components from the plurality of latent components based on feature values ​​of the plurality of neural activity patterns.

[0126] The processor can select multiple latent components corresponding to multiple target neural activity patterns from multiple latent components as multiple target latent components.

[0127] D4: The processor can obtain feature data based on multiple target potential components.

[0128] In some embodiments, the processor may connect and fuse multiple target potential components to obtain characteristic data of the EEG signal.

[0129] S304: The processor analyzes the movement intention corresponding to the EEG signal based on the feature data.

[0130] The processor can also perform standardization on the obtained feature data and input the processed feature data into the classification model. The processor can also classify the processed feature data based on the classification model to obtain the movement intention corresponding to the EEG signal.

[0131] Exemplarily, the processor may perform z-score normalization processing on the obtained feature data to obtain one-dimensional feature data. The processor inputs the one-dimensional feature data into a classification model, classifies the one-dimensional feature data based on the classification model, and obtains the movement intention of the EEG signal.

[0132] In some embodiments, the classification model may be a support vector machine or other classification models, which are not limited here.

[0133] In an embodiment of the present application, the processor can determine the characteristic data of the EEG signal based on the potential state information of the EEG signal, and analyze the movement intention of the EEG signal according to the characteristic data without the need to judge cognitive distraction. That is, the movement intention recognition method provided by the present application does not rely on prior information on cognitive state, and is generally applicable regardless of whether there is cognitive distraction, thereby improving applicability.

[0134] like Figure 4As shown, the embodiment of the present application provides a use case for motion intention recognition. After acquiring the EEG signal y, the processor preprocesses the EEG signal y. The preprocessing includes re-referencing, baseline removal, bandpass filtering, independent component analysis, correlation de-electrooculography and downsampling processing. Figure 4 The specific processing process of the processor in preprocessing the EEG signal y is the same as the processing process A1-A6 in the above embodiment, which will not be repeated here.

[0135] The processor inputs the processed EEG signal y into the factor model: y clean =LL T (LL T +ψ) -1 y; the processor obtains the potential state information F of the EEG signal y based on factor analysis clean =L T (LL T ) -1 y clean The processor has potential state information F clean Perform rectangular decomposition to obtain the potential components F1(λ1),…,F corresponding to multiple neural activity patterns n (λ n ).

[0136] In some embodiments, after the processor has established the factor model, it can perform spectral decomposition on the sample covariance of the factor model to obtain multiple neural activity patterns mode(λ1), ···, mode(λ n ), and determine the characteristic values ​​λ1, ···, λ of multiple neural activity patterns n The processor may also calculate the degree of dominance of the neural activity pattern based on the characteristic value λ of the neural activity pattern.

[0137] like Figure 4 As shown, the processor can calculate the dominance of the neural activity pattern by the following formula:

[0138]

[0139] The processor may select K most dominant target neural activity patterns from the plurality of neural activity patterns. After obtaining the plurality of latent components, the processor selects latent components corresponding to the K target neural activity patterns from the plurality of latent components as target latent components.

[0140] The processor can connect the obtained K target potential components to obtain the characteristic data of the EEG signal and perform standardization on the characteristic data. The processor can also input the processed characteristic data into the classification model, classify the processed characteristic data, and output the movement intention corresponding to the EEG signal y.

[0141] In some embodiments, the processor may also periodically update the factor model. The processor obtains a sample EEG signal and preprocesses the sample EEG signal. The processor may also perform factor analysis on the preprocessed EEG signal based on the EM algorithm to obtain model parameters L and ψ of the factor model. The factor analysis process is based on the initial definition of the factor model y=LF+∈. The processor may update the factor model according to the obtained model parameters.

[0142] based on Figure 4 As shown in the content, after the processor pre-processes the acquired EEG signal, it can obtain the latent state information of the EEG signal according to the factor model, and extract feature data from the latent state information based on the dominance of the neural activity pattern, without the need to make a judgment of cognitive distraction, thereby reducing the reliance on prior information on the cognitive state of the person in the process of motor intention recognition, constructing new prior information based on the dominance of the neural activity pattern, and further improving the accuracy of motor intention recognition.

[0143] Figure 5 A complete flow chart of a motion intention analysis method provided in an embodiment of the present application; wherein, taking a processor as the execution subject, executing the motion intention recognition process as an example, Figure 5 As shown, the method includes:

[0144] S501: The processor obtains an electroencephalogram (EEG) signal at a position of an electrode in a human brain, wherein the EEG signal is obtained based on signals from multiple electrodes.

[0145] In some embodiments, the processor can obtain EEG signals at the locations of electrodes on the human brain through an EEG cap.

[0146] S502: The processor preprocesses the EEG signal to obtain a preprocessed EEG signal, wherein the preprocessing includes re-referencing, baseline removal, filtering, independent component analysis and downsampling.

[0147] S503: The processor may input the preprocessed EEG signal into a latent variable model to obtain latent state information. The latent variable model is a factor model, and the latent state information is used to characterize the spatiotemporal characteristics of the EEG signal; the spatiotemporal characteristics are used to characterize the acquisition time and acquisition channel of multiple electrode signals; each of the multiple electrode signals corresponds to an acquisition channel.

[0148] In some embodiments, the processor may also obtain sample EEG signals according to a set update cycle. The sample EEG signals are historical EEG signals. The processor may process the sample EEG signals based on a maximum expectation algorithm to obtain model parameters of the factor model. The model parameters include a load matrix and independent factor covariances. The processor may update the factor model according to the model parameters.

[0149] S504: The processor performs spectral decomposition on the sample covariance of the latent variable model to obtain multiple neural activity patterns, and determines eigenvalues ​​of the multiple neural activity patterns.

[0150] S505: The processor determines the dominance of each neural activity pattern according to the characteristic value of each neural activity pattern.

[0151] S506: The processor selects a plurality of target neural activity patterns from the plurality of neural activity patterns, wherein the dominance of the plurality of target neural activity patterns is higher than the dominance of the neural activity patterns other than the plurality of target neural activity patterns in the plurality of neural activity patterns.

[0152] S507: The processor linearly decomposes the potential state information to obtain multiple potential components, wherein each potential component is used to characterize the spatiotemporal characteristics of an electrode signal, and the multiple potential components correspond to the multiple neural activity patterns one by one.

[0153] S508: The processor selects, from among the multiple potential components, multiple potential components corresponding to the multiple target neural activity patterns as multiple target potential components.

[0154] S509: The processor obtains feature data according to the multiple target potential components.

[0155] S510: The processor performs standardization processing on the feature data, and inputs the processed feature data into the classification model.

[0156] S511: The processor classifies the processed feature data based on the classification model to obtain the movement intention corresponding to the EEG signal.

[0157] based on Figure 5 As shown in the content, after obtaining the EEG signal, the processor can obtain the potential state information of the EEG signal based on factor analysis, and extract feature data from the potential state information based on the dominance of the neural activity pattern, which can improve the accuracy of information extraction. In addition, after obtaining the EEG signal, the processor can identify the movement intention of the EEG signal with cognitive distraction and the EEG signal without cognitive distraction without the need to judge cognitive distraction, reducing the dependence on the prior information of the person's cognitive state, thereby improving the recognition accuracy of the movement intention.

[0158] Example 2

[0159] In order to verify the experimental effect of the motion intention analysis method described in Example 1, the present application embodiment adopts two groups of different experimental methods to confirm the experimental effect. The specific experimental contents are as follows:

[0160] Step 1: Show the subjects the stimuli of the motor task and cognitive distraction task, and ask them to complete the experimental tasks as required.

[0161] This application designs a dual-task experimental paradigm consisting of a motor main task and a cognitive distraction auxiliary task, wherein the task requirements are displayed on the screen, and the subjects complete the specified tasks according to the prompts on the screen.

[0162] In some examples, such as Figure 6 As shown, the present application provides an experimental paradigm. The subject sits in front of the experimental table in a comfortable sitting position, with both hands placed on the experimental table, and a screen is placed behind the experimental table. During the experiment, the subject is required to complete the necessary upper limb movement tasks and try to stay still.

[0163] In each experiment, the subjects were asked to complete a motor task. Figure 6 As shown in the figure, at the beginning of the experiment, a solid square appears in the center of the screen, representing the current position of the subject's right hand. After 3 seconds, another solid square appears in one of the four positions: front, back, left, or right, representing the end position of the upper limb movement. At the same time, the subjects are asked to stay relaxed as much as possible and stop thinking. After 3 seconds, the square representing the terminal position turns hollow as a visual cue for the start of the upper limb movement task. After the visual cue appears, the subjects need to move the solid square from the starting position to the terminal within 6 seconds, and the subjects' right hand is required to remain in the plane of the experimental table from beginning to end.

[0164] In order to impose cognitive distraction, the subjects were asked to perform a cognitive auxiliary task while completing the main motor task: the n-back task. In the task, a number sequence containing 10 numbers from 0 to 9, arranged in random order, was presented in the form of auditory stimulation. The subjects were asked to provide immediate feedback when hearing each number and verbally say the nth number before the current number. The number n determines the number of numbers that need to be remembered at the end of the number sequence, which determines the difficulty of the task. This application uses the most difficult 2-back task to ensure that a sufficiently obvious cognitive distraction effect is imposed on the subjects. The start time of the auditory presentation of the digital sequence is the same as that of the visual motor task. The broadcast time of each number is 0.75s, and the interval between each two numbers is 2s. Such a configuration can ensure that the time for the auxiliary task can completely cover the time of the main motor task.

[0165] The experiment designed in this application consists of two parts: the first is a distraction-free stage, in which the subjects only need to complete the main motor task; the second is a distraction stage, in which the subjects need to complete both the main motor task and the cognitive distraction auxiliary task. Each stage is divided into four parts, and the end points of the main motor task in each part are different, namely front, back, left, and right. Each part contains 50 experiments, and after completing each part, the subjects are required to take a 10-minute break.

[0166] Step 2: When performing the experimental task, the processor can collect the EEG signal at the designated electrode position on the subject's scalp. The processor can be a processor in the experimental host. The experimental host can be a terminal device or a server.

[0167] In some embodiments, the processor may collect EEG signals in real time through an EEG collector, which may amplify and convert the collected EEG signals into analog-to-digital signals in real time through a wireless amplifier, and transmit the converted EEG signals to the processor.

[0168] Step 3: When performing the experimental task, the processor collects the motion trajectory of the subject's upper limbs and aligns it with the EEG signal. The motion trajectory of the upper limbs can be collected by a motion tracking sensor connected to the subject's upper limbs.

[0169] In some examples, when the original sampling rate of the motion tracking sensor is 60Hz, the processor then upsamples the motion trajectory to 100Hz to synchronize with the EEG signal.

[0170] In each experiment, when the distance from a specific sampling point to the initial position is greater than a preset threshold, this point is taken as the actual starting time of the subject's movement. The processor uses this moment as the timing zero point and selects the EEG signal from -0.5S to 0.5S for subsequent processing.

[0171] Step 4: The processor may preprocess the collected EEG signal to obtain a preprocessed EEG signal.

[0172] In some embodiments, the processor can pre-process the EEG signal collected from the specified electrode position, remove artifacts in the collected EEG signal and obtain the frequency part containing the information required for movement intention recognition, thereby obtaining the pre-processed EEG signal.

[0173] The preprocessing process of the EEG signal by the processor in step 4 is the same as the preprocessing process in A1-A6 in embodiment 1, and will not be repeated here.

[0174] Step 5: The processor can perform information compression processing on the preprocessed EEG signal based on factor analysis to obtain potential state information of the subject's EEG signal.

[0175] The process of extracting the latent state information in step 5 is the same as the process of extracting the latent state information in S302 in embodiment 1, and will not be described in detail here.

[0176] Step 6: The processor can perform spectral decomposition on the sample covariance to obtain multiple neural activity patterns.

[0177] The processor may also arrange the multiple neural activity patterns in descending order according to the contribution of the neural activity patterns to the overall sample covariance. The processor may also select several patterns that have the greatest impact on the overall neural activity from the arranged multiple neural activity patterns.

[0178] In some examples, the processor can determine the dominance of multiple neural activity patterns based on the characteristic values ​​of multiple neural activity patterns. The processor selects from the multiple neural activity patterns, several neural activity patterns with the greatest dominance as multiple target neural activity patterns. Exemplarily, in the EEG signals obtained under the present experimental conditions, the number of neural activity patterns with the most dominant brain neural activity can be K=5.

[0179] Step 7: The processor may decompose the potential state information into a plurality of potential components corresponding to the plurality of neural activity patterns according to the plurality of neural activity patterns. The processor may select the most dominant potential component from the plurality of potential components to form the feature data.

[0180] In some examples, the processor may select latent components corresponding to multiple target neural activity patterns from the multiple latent components as multiple target latent components. The processor may obtain feature data based on the multiple target latent components.

[0181] For example, when the number of the most dominant neural activity patterns is 5, the dimension of the obtained feature data is 1×500.

[0182] Step 8: The processor may split the collected raw data of the subject into training / test data sets, and train and test the classification model.

[0183] In some embodiments, the processor may use a random cross-validation method to split the original data. Figure 7 As shown, the processor can randomly select 80% of the data of each subject under each condition for training and the remaining 20% ​​for testing. Repeat 50 times to obtain the average decoding accuracy of each subject under each condition.

[0184] To be more realistic, samples from both distraction and non-distraction conditions were used to train a joint model. The model can simultaneously decode the upper limb movement intentions of subjects with and without cognitive distraction without prior judgment of attention status.

[0185] The training and test datasets contain an equal number of samples with and without cognitive distraction. In addition, the number of samples of the two upper limb movement directions in each condition is also the same, ensuring the balance of the model for cognitive distraction and movement direction.

[0186] After obtaining the feature data, the processor can construct a classification model with the feature data in the training data set as features and the upper limb movement direction as a label. Exemplarily, the processor can train the data of the two groups of movement directions, front-back and left-right, respectively, to obtain corresponding model parameters, thereby constructing a classification model for identifying the two groups of movement directions, front-back and left-right. The kernel function of the classification model can be a Gaussian kernel function.

[0187] After the original data with and without cognitive distraction conditions are mixed to obtain a training data set, the model in the present application can be learned to identify the dominant neural activity pattern common to the two types of data: movement-related neural activity. Therefore, the method provided by the present application can achieve a huge improvement in decoding performance and is also extremely stable against cognitive distraction.

[0188] After the classification model training is completed, the processor can test the accuracy of the classification model based on the test data set.

[0189] For each subject's EEG data under each condition, the training-testing was repeated 50 times to obtain the average decoding accuracy of each subject under each condition, such as Figure 8 shown.

[0190] The processor can also determine the mean of the decoding accuracy for all subjects. Under conditions of cognitive distraction (distracted state), the average decoding accuracy reached 86.59% (front-back) and 86.00% (left-right); without cognitive distraction (concentrated state), the average decoding accuracy reached 87.43% (front-back) and 86.36% (left-right). In the distracted state, the decoding accuracy decreased by 0.84% ​​(front-back) and 0.36% (left-right), respectively. The results show that 1) this method improves the decoding performance of upper limb movement intention; 2) the decoding performance of this method in the distracted state is on par with that in the concentrated state, and it is very robust to cognitive distraction.

[0191] In some embodiments, the processor may also verify the time-based memory usage of the motion intention analysis method and record the memory usage change trend.

[0192] The processor can record the changes in its memory usage over time at a sampling frequency of 10Hz, such as Fig. 9 shown. Fig. 9The changes in CPU memory usage can clearly show:

[0193] 1) In the preprocessing stage, the processor continuously reads in multiple experimental data, re-references and removes the baseline; concatenates the experimental data and performs two bandpass filters; thereafter, the processor performs independent component analysis and downsampling.

[0194] 2) In the factor analysis and pattern decomposition stage, the processor first performs factor analysis on the data, and the memory usage changes continuously between two close extreme values. Secondly, after obtaining the factor model parameters, the processor decomposes the potential state information into multiple potential component outputs, and its memory size is maintained at a constant value. After the processor decomposes the training data, concentrated state test data, and dispersed state test data respectively, when outputting according to the pattern, the processor's memory usage benchmark remains at the same level.

[0195] Based on the experimental content described in the above-mentioned Example 2, the decoding performance of the motor intention analysis method provided by the present application is equal to that of the concentrated state in the decentralized state. Therefore, the motor intention analysis method provided by the present application is extremely robust to cognitive distraction. In addition, the present application constructs a priori information that does not depend on the cognitive state, which is generally used in the case of cognitive distraction or not. The same decoding model that is robust to cognitive distraction overcomes the disadvantage that the traditional single model cannot adaptively adjust the decoding model parameters according to the cognitive state of the person, making the use scenario of the upper limb motor intention neural decoding model of the person closer to reality.

[0196] Based on the same technical concept, the embodiment of the present application also provides a motion intention analysis device, such as Fig.10 As shown, the motion intention analysis device 1000 may include:

[0197] The acquisition unit 1001 is used to acquire the EEG signal of the electrode position of the human brain; the EEG signal is obtained based on multiple electrode signals;

[0198] The processing unit 1002 is used to perform information compression processing on the EEG signal to obtain the latent state information of the EEG signal; the latent state information is used to characterize the spatiotemporal characteristics of the EEG signal; the spatiotemporal characteristics are used to characterize the acquisition time and acquisition channel of multiple electrode signals; each electrode signal in the multiple electrode signals corresponds to an acquisition channel; and feature data of the EEG signal is extracted from the latent state information;

[0199] The analyzing unit 1003 is used to analyze the movement intention corresponding to the EEG signal according to the characteristic data.

[0200] In a possible design, before performing information compression processing on the EEG signal to obtain the potential state information of the EEG signal, the processing unit 1002 is further used to:

[0201] Preprocess the EEG signal; the preprocessing includes re-referencing, baseline removal, filtering, independent component analysis and downsampling.

[0202] In a possible design, the processing unit 1002 is specifically configured to:

[0203] The EEG signal is input into the latent variable model to obtain the potential state information; the latent variable model is a factor model.

[0204] In one possible design, the processing unit 1002 is further configured to:

[0205] According to the set update cycle, a sample EEG signal is obtained; the sample EEG signal is a historical EEG signal;

[0206] Based on the maximum expectation algorithm, the sample EEG signals are processed to obtain the model parameters of the factor model; the model parameters include the load matrix and the covariance of independent factors;

[0207] Update the factor model based on the model parameters.

[0208] In a possible design, the processing unit 1002 is specifically configured to:

[0209] Perform spectral decomposition on the sample covariance of the latent variable model to obtain multiple neural activity patterns, and determine the eigenvalues ​​of the multiple neural activity patterns; the eigenvalue of any neural activity pattern is used to characterize the dominance of the neural activity pattern;

[0210] The latent state information is linearly decomposed to obtain multiple latent components; each latent component is used to characterize the spatiotemporal characteristics of an electrode signal; multiple latent components correspond one to one with multiple neural activity patterns;

[0211] selecting a plurality of target latent components from the plurality of latent components according to feature values ​​of the plurality of neural activity patterns;

[0212] According to multiple target latent components, feature data is obtained.

[0213] In a possible design, the processing unit 1002 is specifically configured to:

[0214] According to the characteristic value of each neural activity pattern, the dominance degree of each neural activity pattern is determined respectively;

[0215] selecting multiple target neural activity patterns from multiple neural activity patterns; the dominance of the multiple target neural activity patterns is higher than the dominance of the neural activity patterns other than the multiple target neural activity patterns in the multiple neural activity patterns;

[0216] A plurality of latent components corresponding to a plurality of target neural activity patterns are selected from the plurality of latent components as a plurality of target latent components.

[0217] In a possible design, the analyzing unit 1003 is specifically used for:

[0218] Standardize the feature data and input the processed feature data into the classification model;

[0219] Based on the classification model, the processed feature data is classified to obtain the movement intention corresponding to the EEG signal.

[0220] Based on the above content and the same concept, Fig.11 As shown, the present application provides an electronic device. Fig.11 A schematic structural diagram of an electronic device 1100 is shown.

[0221] The electronic device 1100 includes at least one processor 1110, which is used to implement or support the electronic device 1100 to implement the function of the motion intention analysis device in the method provided in the embodiment of the present application. Exemplarily, the processor 1110 can perform motion intention recognition on the acquired EEG signal, and the detailed description in the method example is specifically referred to, which will not be repeated here.

[0222] The electronic device 1100 may also include at least one memory 1120 for storing program instructions and / or data. The memory 1120 is coupled to the processor 1110. The coupling in the embodiment of the present application is an indirect coupling or communication connection between devices, units or modules, which may be electrical, mechanical or other forms, and is used for information exchange between devices, units or modules. The processor 1110 may operate in conjunction with the memory 1120. The processor 1110 may execute program instructions stored in the memory 1120. At least one of the at least one memory may be included in the processor.

[0223] The electronic device 1100 may further include an interface 1130 for communicating with the processor 1110, or for communicating with an EEG acquisition device or other device via a transmission medium. For example, the other device may be a client. The processor 1110 may use the interface 1130 to send and receive data.

[0224] The specific connection medium between the interface 1130, the processor 1110 and the memory 1120 is not limited in the embodiment of the present application. Fig.11 In the embodiment, the memory 1120, the processor 1110 and the interface 1130 are connected via a bus 1140. Fig.11The connections between the other components are shown in bold lines, which are only for illustration and are not intended to be limiting. The bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Fig.11 Only one thick line is used in the diagram, but this does not mean that there is only one bus or only one type of bus.

[0225] In the embodiment of the present application, the processor 1110 may be a general-purpose processor, a digital signal processor, an application-specific integrated circuit, a field programmable gate array or other programmable logic device, a discrete gate or transistor logic device, or a discrete hardware component, and may implement or execute the methods, steps, and logic block diagrams disclosed in the embodiment of the present application. The general-purpose processor may be a microprocessor or any conventional processor, etc. The steps of the method disclosed in the embodiment of the present application may be directly embodied as being executed by a hardware processor, or may be executed by a combination of hardware and software modules in the processor.

[0226] In an embodiment of the present application, the memory 1120 may be a non-volatile memory, such as a hard disk drive (HDD) or a solid-state drive (SSD), etc., or a volatile memory (volatile memory), such as a random-access memory (RAM). The memory is any other medium that can be used to carry or store a desired program code in the form of an instruction or data structure and can be accessed by a computer, but is not limited thereto. The memory in the embodiment of the present application may also be a circuit or any other device that can implement a storage function, for storing program instructions and / or data.

[0227] It should be understood that the above-mentioned electronic device can be a chip, and the processor can be implemented by hardware or by software. When implemented by hardware, the processor can be a logic circuit, an integrated circuit, etc.; when implemented by software, the processor can be a general-purpose processor, which is implemented by reading the software code stored in the memory. The memory can be integrated in the processor, or can be located outside the processor and exist independently.

[0228] Based on the above content and the same concept, the present application provides a computer-readable storage medium having a computer program or instructions stored thereon. When the computer program or instructions are executed, the computing device executes the method in the above method embodiment.

[0229] Based on the above content and the same concept, the present application provides a computer program product. When a computer executes the computer program product, the computing device executes the method in the above method embodiment.

[0230] It should be understood that the term "at least one" in the embodiments of the present application refers to one or more, and "plurality" refers to two or more. "And / or" describes the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone, where A and B can be singular or plural. The character " / " generally indicates that the associated objects before and after are in an "or" relationship. "At least one of the following" or similar expressions refers to any combination of these items, including any combination of single items or plural items. For example, at least one of a, b or c can represent: a, b, c, a and b, a and c, b and c, or a, b and c, where a, b, c can be single or multiple.

[0231] Furthermore, unless otherwise specified, the ordinal numbers such as "first" and "second" mentioned in the embodiments of the present application are used to distinguish multiple objects, and are not used to limit the order, timing, priority or importance of multiple objects. For example, the first information and the second information are only used to distinguish different indication information, and do not indicate the difference in priority or importance of the two types of information.

[0232] It should be understood that in the various embodiments of the present application, the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0233] In addition, in the embodiments of the present application, the word "exemplary" is used to indicate an example or illustration. Any embodiment or implementation described as an "example" in the embodiments of the present application should not be interpreted as being more preferred than other embodiments or implementations. That is, the use of the word "example" is intended to present concepts in a specific way.

[0234] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code.

[0235] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0236] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0237] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0238] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalents, the present application is also intended to include these modifications and variations.

Claims

1. A motion intention analysis method, characterized in that: The method comprises: Obtaining an electroencephalogram (EEG) signal at a human brain electrode location; the EEG signal is obtained based on multiple electrode signals; Performing information compression processing on the EEG signal to obtain potential state information of the EEG signal; the potential state information is used to characterize the spatiotemporal characteristics of the EEG signal; the spatiotemporal characteristics are used to characterize the acquisition time and acquisition channel of the multiple electrode signals; each of the multiple electrode signals corresponds to an acquisition channel; Extracting characteristic data of the EEG signal from the potential state information; The movement intention corresponding to the EEG signal is analyzed according to the characteristic data.

2. The method according to claim 1, characterized in that Before performing information compression processing on the EEG signal to obtain potential state information of the EEG signal, the method further includes: The EEG signal is preprocessed; the preprocessing includes re-referencing, baseline removal, filtering, independent component analysis and downsampling.

3. The method according to claim 1 or 2, characterized in that: The step of performing information compression processing on the EEG signal to obtain potential state information of the EEG signal includes: The EEG signal is input into a latent variable model to obtain the potential state information; the latent variable model is a factor model.

4. The method according to claim 3, characterized in that The method further comprises: Acquire sample EEG signals according to a set update cycle; the sample EEG signals are historical EEG signals; Based on the maximum expectation algorithm, the sample EEG signal is processed to obtain the model parameters of the factor model; the model parameters include the load matrix and the independent factor covariance; The factor model is updated according to the model parameters.

5. The method according to claim 3 or 4, characterized in that: The step of extracting characteristic data of the EEG signal from the potential state information comprises: Performing spectral decomposition on the sample covariance of the latent variable model to obtain multiple neural activity patterns, and determining eigenvalues ​​of the multiple neural activity patterns; the eigenvalue of any neural activity pattern is used to characterize the dominance of the neural activity pattern; Linearly decomposing the latent state information to obtain a plurality of latent components; each latent component is used to characterize the spatiotemporal characteristics of an electrode signal; the plurality of latent components correspond one-to-one to the plurality of neural activity patterns; selecting a plurality of target latent components from the plurality of latent components according to the feature values ​​of the plurality of neural activity patterns; The feature data is obtained according to the multiple target potential components.

6. The method according to claim 5, characterized in that The step of selecting a plurality of target latent components from the plurality of latent components according to the feature values ​​corresponding to the plurality of neural activity patterns comprises: According to the characteristic value of each neural activity pattern, the dominance degree of each neural activity pattern is determined respectively; Selecting a plurality of target neural activity patterns from the plurality of neural activity patterns; the dominance of the plurality of target neural activity patterns is higher than the dominance of the neural activity patterns other than the plurality of target neural activity patterns in the plurality of neural activity patterns; A plurality of latent components corresponding to the plurality of target neural activity patterns are selected from the plurality of latent components as the plurality of target latent components.

7. The method according to any one of claims 1 to 6, characterized in that: The step of analyzing the movement intention corresponding to the EEG signal according to the characteristic data includes: Performing standardization processing on the feature data, and inputting the processed feature data into a classification model; Based on the classification model, the processed feature data is classified to obtain the movement intention corresponding to the EEG signal.

8. A motion intention analysis device, characterized in that: include: An acquisition unit, used for acquiring an electroencephalogram signal of a human brain electrode position; The EEG signal is obtained based on multiple electrode signals; A processing unit, used for performing information compression processing on the EEG signal to obtain potential state information of the EEG signal; The potential state information is used to characterize the spatiotemporal characteristics of the EEG signal; the spatiotemporal characteristics are used to characterize the acquisition time and acquisition channel of the multiple electrode signals; each electrode signal in the multiple electrode signals corresponds to an acquisition channel; characteristic data of the EEG signal is extracted from the potential state information; An analysis unit is used to analyze the movement intention corresponding to the EEG signal according to the characteristic data.

9. The device according to claim 8, characterized in that Before performing information compression processing on the EEG signal to obtain potential state information of the EEG signal, the processing unit is further used to: The EEG signal is preprocessed; the preprocessing includes re-referencing, baseline removal, filtering, independent component analysis and downsampling.

10. The device according to claim 8 or 9, characterized in that The processing unit is specifically used for: The EEG signal is input into a latent variable model to obtain the potential state information; the latent variable model is a factor model.

11. The device according to claim 10, characterized in that The processing unit is also used for: Acquire sample EEG signals according to a set update cycle; the sample EEG signals are historical EEG signals; Based on the maximum expectation algorithm, the sample EEG signal is processed to obtain the model parameters of the factor model; the model parameters include the load matrix and the independent factor covariance; The factor model is updated according to the model parameters.

12. The device according to claim 10 or 11, characterized in that The processing unit is specifically used for: Performing spectral decomposition on the sample covariance of the latent variable model to obtain multiple neural activity patterns, and determining eigenvalues ​​of the multiple neural activity patterns; the eigenvalue of any neural activity pattern is used to characterize the dominance of the neural activity pattern; Linearly decomposing the latent state information to obtain a plurality of latent components; each latent component is used to characterize the spatiotemporal characteristics of an electrode signal; The plurality of latent components correspond one-to-one to the plurality of neural activity patterns; selecting a plurality of target latent components from the plurality of latent components according to the feature values ​​of the plurality of neural activity patterns; The feature data is obtained according to the multiple target potential components.

13. The device according to claim 12, characterized in that The processing unit is specifically used for: According to the characteristic value of each neural activity pattern, the dominance degree of each neural activity pattern is determined respectively; Selecting a plurality of target neural activity patterns from the plurality of neural activity patterns; the dominance of the plurality of target neural activity patterns is higher than the dominance of the neural activity patterns other than the plurality of target neural activity patterns in the plurality of neural activity patterns; A plurality of latent components corresponding to the plurality of target neural activity patterns are selected from the plurality of latent components as the plurality of target latent components.

14. The device according to any one of claims 8 to 13, characterized in that: The analysis unit is specifically used for: Performing standardization processing on the feature data, and inputting the processed feature data into a classification model; Based on the classification model, the processed feature data is classified to obtain the movement intention corresponding to the EEG signal.

15. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-executable instructions, and when the computer-executable instructions are called by an electronic device, the electronic device executes the method according to any one of claims 1 to 7.

16. A chip, characterized in that: The chip is coupled to a memory, and the chip reads a computer program stored in the memory to execute the method according to any one of claims 1 to 7.

17. A computer program product, characterized in that When the computer program product is run on a computer, the computer is caused to execute the method according to any one of claims 1 to 7.

Images