Feedback method and system for patient intention brain wave recognition based on brain-computer interface

Abstract

The invention discloses a feedback method and system for patient intention brain wave recognition based on a brain-computer interface, and the method comprises the steps: collecting an electroencephalogram signal and associated data of the brain-computer interface of a patient, and carrying out the preprocessing of the electroencephalogram signal and the associated data; performing pattern classification on the electroencephalogram signals to obtain pattern data, and performing intention correction on the pattern data by adopting the associated data to obtain corrected data; electroencephalogram signal recognition is carried out on the correction data to obtain recognition data, and a patient intention brain wave recognition feedback model is constructed according to the recognition data; and optimizing the patient intention brain wave recognition feedback model according to the feedback error, inputting data to be fed back into the optimized patient intention brain wave recognition feedback model, and outputting a feedback result. The method not only can improve the feedback precision of patient intention brain wave recognition based on the brain-computer interface, but also has good interpretability, and can be directly applied to a feedback system of patient intention brain wave recognition.

Description

Technical Field

[0001] The present invention relates to the field of intention recognition, and in particular to a feedback method and system for patient intention brain wave recognition based on a brain-computer interface. Background Art

[0002] With the rapid development of neuroscience and information technology, brain-computer interface technology has become a new bridge connecting the human brain with external devices or systems, especially in the fields of medical rehabilitation, assisted communication, neuroscience research, etc. It has shown great application potential. Traditionally, it is a major challenge for groups such as paralyzed patients, people with language disorders, and patients with cognitive dysfunction to effectively express their wishes and intentions. Therefore, it is particularly important to develop technology that can accurately identify the intentions contained in the patient's EEG signals and provide real-time feedback accordingly.

[0003] At present, brain-computer interface technology mainly decodes individual intentions by collecting and analyzing electrical signals generated by the cerebral cortex (i.e., electroencephalogram signals, EEG). However, EEG signals are complex and changeable, easily interfered by noise, and there are significant differences between different individuals. These factors increase the difficulty of intention recognition. Existing EEG signal recognition methods mostly focus on the extraction and classification of signal features, but often ignore the importance of associated data such as external environment, emotional state, historical behavior, etc. related to intention, resulting in limited recognition accuracy and generalization ability.

[0004] In addition, most brain-computer interface systems lack an effective feedback mechanism to optimize and adjust the recognition model, making it difficult for the system to maintain long-term stability and accuracy when facing complex and changeable practical application scenarios. Therefore, developing a method that can comprehensively consider EEG signals and related data, continuously optimize the recognition model through a feedback mechanism, and thus achieve high-precision and high-stability patient intention brainwave recognition has become a hot topic and difficulty in current research. Summary of the invention

[0005] The purpose of the present invention is to provide a feedback method for patient intention brain wave recognition based on brain-computer interface.

[0006] To achieve the above object, the present invention is implemented according to the following technical solutions:

[0007] The present invention comprises the following steps:

[0008] Collecting EEG signals and associated data of the patient's brain-computer interface, and preprocessing the EEG signals and the associated data;

[0009] Performing pattern classification on the EEG signal to obtain pattern data, and performing intention correction on the pattern data using the associated data to obtain correction data;

[0010] Performing electroencephalogram signal recognition on the correction data to obtain recognition data, and constructing a patient intention electroencephalogram recognition feedback model according to the recognition data;

[0011] The patient intention brain wave recognition feedback model is optimized according to the feedback error, the data to be fed back is input into the optimized patient intention brain wave recognition feedback model, and the feedback result is output.

[0012] Furthermore, the method of performing pattern classification on the EEG signal to obtain pattern data includes:

[0013] Add particles as a search tool for EEG signals and update the search speed and position of the particles. The expression is:

[0014]

[0015] The local optimal position of the ith particle at the tth iteration is The global optimal position of the ith particle at the tth iteration is The position of the i-th particle at the t-th iteration is The search speed of the ith particle at the t+1th iteration is The search speed of the tth iteration of the i-th particle is The position of the i-th particle at the t+1th iteration is The random numbers between 0 and 1 are τ1 and τ2, the acceleration factors are λ1 and λ2, and the inertia weight factor is η;

[0016] Calculate the similarity between EEG signals:

[0017]

[0018] Among them The EEG signal is The mth EEG signal is x m , EEG signal and EEG signal x m The similarity is EEG signals The probability density of EEG signalx m The probability density of EEG signals and EEG signal x m The weight factor is

[0019] Cluster the EEG signals with the greatest similarity to obtain the initial clusters, and calculate the overall silhouette coefficient of the clusters:

[0020]

[0021] The average distance between the mth EEG signal and the EEG signal within the initial cluster is l(m), and the average distance between the mth EEG signal and the EEG signal point between the initial clusters is The number of EEG signals is

[0022] Calculate the penalty term:

[0023]

[0024] The number of EEG signals with negative contour coefficients is N ng , the average distance between the ιth EEG signal and the initial intra-cluster EEG signal is l(ι), and the average distance between the ιth EEG signal and the initial inter-cluster EEG signal is

[0025] Construct the fitness function, the expression is:

[0026]

[0027] The fitness function is The actual number of clusters is A, the predicted number of clusters is X, the convolution operation is *, and the cutoff distance is φ o , the indicative function is EEG signals and EEG signal x m The distance is No. The cluster centers are EEG signals Between cluster centers The distance is EEG signals and cluster center ε m The distance between The membership function is

[0028] The initial cluster is updated through continuous iteration of the particle population until the fitness function value is maximized, then the iteration is stopped and the updated initial cluster is output as pattern data.

[0029] Furthermore, the method of using the associated data to perform intentional correction on the pattern data to obtain corrected data includes:

[0030] The associated data are analyzed to obtain the potential behavior of the patient, and the potential behavior is used to correct the pattern data to obtain the corrected data.

[0031] Furthermore, the method of performing electroencephalogram signal recognition on the correction data to obtain recognition data includes:

[0032] The overview space and detail space of the corrected data are decomposed step by step, and the overview space is transformed by the first conjugate orthogonal mirror filter bank and the second conjugate orthogonal mirror filter bank respectively to obtain the transformed overview space and detail space, and the expression is:

[0033]

[0034] β[n]=(-1) 1-n h[1-n]

[0035] The jth overview space is E j , the j+1th detail space is C j+1 , the j+1th transformation overview space is E j+1 , the 1-nth first conjugate orthogonal mirror filter group is h[1-n], the first conjugate orthogonal mirror filter group is h[n], and the second conjugate orthogonal mirror filter group is β[n];

[0036] The function family is defined using equivalent scaling functions and equivalent wavelet functions:

[0037]

[0038] The equivalent scaling function is The equivalent wavelet function is The integer is R and the function family is

[0039] The multi-resolution decomposition algorithm is used to perform multi-resolution decomposition on the function family, and the expression is:

[0040]

[0041] The adjoint of the 1-2kth function family is h * [1-2k], the adjoint of the 1-2kth first conjugate orthogonal mirror filter bank is h * [1-2k];

[0042] The number of channels of EEG signals collected gives the reference signal corresponding to the SSVEP visual stimulation frequency:

[0043]

[0044] The bth reference signal is v b , the number of stimulation targets is b, and the stimulation frequency of the bth stimulation target is f b , the number of sampling points is N s ;

[0045] Calculate the correlation coefficient based on the EEG signal and the reference signal corresponding to the bth stimulation frequency:

[0046]

[0047] The horizontal weight matrix is ​​D w , the vertical weight matrix is ​​D v , the EEG signal is w, if K is the stimulation frequency data, then the target frequency is calculated and identified:

[0048]

[0049] The identification data is output according to the identified target frequency.

[0050] Furthermore, a method for constructing a patient intention brain wave recognition feedback model based on the recognition data includes:

[0051] The objective function is constructed based on the recognition data, and the expression is:

[0052]

[0053] The actual identification data is The predicted recognition data is Actual identification data and predictive recognition data The loss function is The objective function is

[0054] The patient intention brainwave recognition feedback model includes empirical mode decomposition, decision tree algorithm, and machine learning pattern recognition network;

[0055] Empirical mode decomposition decomposes the input EEG signal to obtain local amplitude, phase and frequency, and outputs the local amplitude, phase and frequency as the required features;

[0056] The decision tree algorithm recursively divides the demand characteristics into multiple regions, and builds a tree model according to the decision rules of the regions to perform category prediction and obtain classified data;

[0057] The machine learning pattern recognition network performs intent matching on the classified data to obtain matching results, and outputs the intent with matching results greater than 0.863 as feedback results.

[0058] Furthermore, the method for optimizing the patient intention brain wave recognition feedback model according to the feedback error includes:

[0059] Introducing low-discrepancy sequences to optimize the search subpopulation, the expression is:

[0060]

[0061] The initial position of the searcher in the dth dimension is The random number of the low discrepancy sequence in the range of 0 to 1 is Se , the maximum value of the search sub-position is Z max , search for the minimum value of the sub-position Z min ;

[0062] Randomly divide the searchers into four independent groups and calculate the fitness of the searchers:

[0063]

[0064] The fitness function of the first group is g1, the current number of iterations is t, and the maximum number of iterations is t. max , the fitness function of the second group is g2, the fitness function of the third group is g3, and the fitness function of the fourth group is g4;

[0065] Calculate the fitness of the four groups of searchers and sort them in descending order, and use the first four searchers as attackers, interceptors, drivers, and chasers in turn;

[0066] Calculate the correlation coefficient:

[0067]

[0068] The correlation coefficient is The shrinkage coefficient is a, and the random vector between 0 and 1 is r1;

[0069] Update the fitness of the searcher according to the current number of iterations. When the random number p is less than 0.5 and the absolute value of the correlation coefficient is less than 1, update the position of the searcher to obtain the first position. The expression is:

[0070]

[0071] The attacker's position is Z b1 , the position of the interceptor is Z b2 , the position of the driver is Z b3 , the position of the pursuer is Z b4 , the first position of the searcher in the t+1th iteration is Z1(t+1), the actual feedback is Q, and the predicted feedback is

[0072] When the random number p is less than 0.5 and the absolute value of the correlation coefficient is greater than or equal to 1, the search sub-position is updated to obtain the second position, and the expression is:

[0073]

[0074] The second position of the searcher in the t+1th iteration is Z2(t+1), the random vector between 0 and 1 is r2, and the position of the random searcher is Z random , the search subposition of the tth iteration is Z(t);

[0075] When the random number p is greater than or equal to 0.5, the third position is obtained by updating the search sub-position according to the spiral ascent, and the expression is:

[0076]

[0077] The third position of the searcher in the t+1th iteration is Z3(t+1), and the target position of the searcher is Z * (t), the random number between -1 and 1 is ρ, and the spiral control constant is λ;

[0078] Otherwise, continue to iterate until the maximum number of iterations is reached, then stop searching.

[0079] Secondly, the feedback system for patient intention brainwave recognition based on brain-computer interface includes:

[0080] Data acquisition module: used to collect EEG signals and associated data of the patient's brain-computer interface, and pre-process the EEG signals and associated data;

[0081] Data correction module: used for performing pattern classification on the EEG signal to obtain pattern data, and performing intention correction on the pattern data using the associated data to obtain correction data;

[0082] Identification and construction module: used for performing electroencephalogram signal recognition on the correction data to obtain recognition data, and constructing a patient intention electroencephalogram recognition feedback model according to the recognition data;

[0083] Optimization feedback module: used to optimize the patient intention brain wave recognition feedback model according to the feedback error, input the data to be fed back into the optimized patient intention brain wave recognition feedback model, and output the feedback result.

[0084] The beneficial effects of the present invention are:

[0085] The present invention is a feedback method and system for patient intention brain wave recognition based on brain-computer interface. Compared with the prior art, the present invention has the following technical effects:

[0086] The present invention can improve the feedback accuracy of brain-computer interface brain wave recognition of patient intention through preprocessing, pattern classification, intention correction, EEG signal recognition, model construction and model optimization steps, thereby improving the feedback accuracy of brain-computer interface brain wave recognition of patient intention, and optimizing the feedback of brain-computer interface brain wave recognition of patient intention, which can greatly save resources and improve work efficiency, and can realize intelligent feedback of brain-computer interface brain wave recognition of patient intention, and perform intention correction on the feedback of brain-computer interface brain wave recognition of patient intention in real time, which is of great significance to the feedback of brain-computer interface brain wave recognition of patient intention, and can adapt to the feedback of brain wave recognition of patient intention of different standards and the feedback requirements of different brain wave recognition of patient intention, and has a certain universality. BRIEF DESCRIPTION OF THE DRAWINGS

[0087] Figure 1 A flowchart of the steps of the feedback method for patient intention brain wave recognition based on brain-computer interface of the present invention;

[0088] Figure 2 This is a block diagram of the feedback system for patient intention brain wave recognition based on brain-computer interface of the present invention. DETAILED DESCRIPTION

[0089] The present invention is further described below by means of specific embodiments. The illustrative embodiments and descriptions of the present invention are used to explain the present invention but are not intended to limit the present invention.

[0090] The feedback method and system of the present invention for patient intention brain wave recognition based on brain-computer interface comprises the following steps:

[0091] like Figure 1 As shown, in this embodiment, the following steps are included:

[0092] Collecting EEG signals and associated data of the patient's brain-computer interface, and preprocessing the EEG signals and the associated data;

[0093] In actual assessments, the associated data include medical records, medication history, lifestyle habits, emotions, and fatigue levels;

[0094] The list of needs included “want to rest,” “want to remove the tube,” “want to talk,” “thirst,” “want to have a bowel movement,” “want to see family,” “feel cold,” “feel hot,” “want to turn over,” “feel pain,” and “have phlegm”;

[0095] A 51-year-old male patient XX was selected as the research subject. He was paralyzed due to spinal cord injury. The EEG signal showed that beta wave activity increased in the sensory and motor cortex areas, which may indicate that the patient was trying to move his body or turn over. Theta waves had slight fluctuations in the occipital lobe area, which may reflect the patient's discomfort with the current body posture. Alpha wave activity was relatively uniform throughout the cerebral cortex, but slightly decreased, which may indicate that the patient was tired from maintaining the same posture for a long time.

[0096] Performing pattern classification on the EEG signal to obtain pattern data, and performing intention correction on the pattern data using the associated data to obtain correction data;

[0097] Performing electroencephalogram signal recognition on the correction data to obtain recognition data, and constructing a patient intention electroencephalogram recognition feedback model according to the recognition data;

[0098] Optimizing the patient intention brain wave recognition feedback model according to the feedback error, inputting the data to be fed back into the optimized patient intention brain wave recognition feedback model, and outputting the feedback result;

[0099] In the actual evaluation, the feedback result is that the person wants to turn over.

[0100] In this embodiment, the method of performing pattern classification on the EEG signal to obtain pattern data includes:

[0101] Add particles as a search tool for EEG signals and update the search speed and position of the particles. The expression is:

[0102]

[0103] The local optimal position of the ith particle at the tth iteration is The global optimal position of the ith particle at the tth iteration is The position of the i-th particle at the t-th iteration is The search speed of the ith particle at the t+1th iteration is The search speed of the tth iteration of the i-th particle is The position of the i-th particle at the t+1th iteration is The random numbers between 0 and 1 are τ1 and τ2, the acceleration factors are λ1 and λ2, and the inertia weight factor is η;

[0104] Calculate the similarity between EEG signals:

[0105]

[0106] Among them The EEG signal is The mth EEG signal is x m , EEG signal and EEG signal x m The similarity is EEG signals The probability density of EEG signalx m The probability density of EEG signals and EEG signal x m The weight factor is

[0107] Cluster the EEG signals with the greatest similarity to obtain the initial clusters, and calculate the overall silhouette coefficient of the clusters:

[0108]

[0109] The average distance between the mth EEG signal and the EEG signal within the initial cluster is l(m), and the average distance between the mth EEG signal and the EEG signal point between the initial clusters is The number of EEG signals is

[0110] Calculate the penalty term:

[0111]

[0112] The number of EEG signals with negative contour coefficients is N ng , the average distance between the ιth EEG signal and the initial intra-cluster EEG signal is l(ι), and the average distance between the ιth EEG signal and the initial inter-cluster EEG signal is

[0113] Construct the fitness function, the expression is:

[0114]

[0115] The fitness function is The actual number of clusters is A, the predicted number of clusters is X, the convolution operation is *, and the cutoff distance is φ o , the indicative function is EEG signals and EEG signal x m The distance is No. The cluster centers are EEG signals Between cluster centers The distance is EEG signals and cluster center ε m The distance between The membership function is

[0116] The initial cluster is updated through continuous iteration of the particle population until the fitness function value is maximized, then the iteration is stopped and the updated initial cluster is output as pattern data.

[0117] In this embodiment, the method of using the associated data to perform intentional correction on the pattern data to obtain the corrected data includes:

[0118] The associated data are analyzed to obtain the potential behavior of the patient, and the potential behavior is used to correct the pattern data to obtain the corrected data.

[0119] In this embodiment, the method of performing EEG signal recognition on the correction data to obtain recognition data includes:

[0120] The overview space and detail space of the corrected data are decomposed step by step, and the overview space is transformed by the first conjugate orthogonal mirror filter bank and the second conjugate orthogonal mirror filter bank respectively to obtain the transformed overview space and detail space, and the expression is:

[0121]

[0122] β[n]=(-1) 1-n h[1-n]

[0123] The jth overview space is E j , the j+1th detail space is C j+1 , the j+1th transformation overview space is E j+1 , the 1-nth first conjugate orthogonal mirror filter group is h[1-n], the first conjugate orthogonal mirror filter group is h[n], and the second conjugate orthogonal mirror filter group is β[n];

[0124] The function family is defined using equivalent scaling functions and equivalent wavelet functions:

[0125]

[0126] The equivalent scaling function is The equivalent wavelet function is The integer is R and the function family is

[0127] The multi-resolution decomposition algorithm is used to perform multi-resolution decomposition on the function family, and the expression is:

[0128]

[0129] The adjoint of the 1-2kth function family is h * [1-2k], the adjoint of the 1-2kth first conjugate orthogonal mirror filter bank is h * [1-2k];

[0130] The number of channels of EEG signals collected gives the reference signal corresponding to the SSVEP visual stimulation frequency:

[0131]

[0132] The bth reference signal is v b , the number of stimulation targets is b, and the stimulation frequency of the bth stimulation target is f b , the number of sampling points is N s ;

[0133] Calculate the correlation coefficient based on the EEG signal and the reference signal corresponding to the bth stimulation frequency:

[0134]

[0135] The horizontal weight matrix is ​​D w , the vertical weight matrix is ​​D v , the EEG signal is w, if K is the stimulation frequency data, then the target frequency is calculated and identified:

[0136]

[0137] The identification data is output according to the identified target frequency.

[0138] In this embodiment, the method for constructing a patient intention brain wave recognition feedback model according to the recognition data includes:

[0139] The objective function is constructed based on the recognition data, and the expression is:

[0140]

[0141] The actual identification data is The predicted recognition data is Actual identification data and predictive recognition data The loss function is The objective function is

[0142] The patient intention brainwave recognition feedback model includes empirical mode decomposition, decision tree algorithm, and machine learning pattern recognition network;

[0143] Empirical mode decomposition decomposes the input EEG signal to obtain local amplitude, phase and frequency, and outputs the local amplitude, phase and frequency as the required features;

[0144] The decision tree algorithm recursively divides the demand characteristics into multiple regions, and builds a tree model according to the decision rules of the regions to perform category prediction and obtain classified data;

[0145] The machine learning pattern recognition network performs intent matching on the classified data to obtain matching results, and outputs the intent with matching results greater than 0.863 as feedback results.

[0146] In this embodiment, the method for optimizing the patient intention brain wave recognition feedback model according to the feedback error includes:

[0147] Introducing low-discrepancy sequences to optimize the search subpopulation, the expression is:

[0148]

[0149] The initial position of the searcher in the dth dimension is The random number of the low discrepancy sequence in the range of 0 to 1 is S e , the maximum value of the search sub-position is Z max , search for the minimum value of the sub-position Z min ;

[0150] Randomly divide the searchers into four independent groups and calculate the fitness of the searchers:

[0151]

[0152] The fitness function of the first group is g1, the current number of iterations is t, and the maximum number of iterations is t. max , the fitness function of the second group is g2, the fitness function of the third group is g3, and the fitness function of the fourth group is g4;

[0153] Calculate the fitness of the four groups of searchers and sort them in descending order, and use the first four searchers as attackers, interceptors, drivers, and chasers in turn;

[0154] Calculate the correlation coefficient:

[0155]

[0156] The correlation coefficient is The shrinkage coefficient is a, and the random vector between 0 and 1 is r1;

[0157] Update the fitness of the searcher according to the current number of iterations. When the random number p is less than 0.5 and the absolute value of the correlation coefficient is less than 1, update the position of the searcher to obtain the first position. The expression is:

[0158]

[0159] The attacker's position is Z b1 , the position of the interceptor is Z b2 , the position of the driver is Z b3 , the position of the pursuer is Z b4 , the first position of the searcher in the t+1th iteration is Z1(t+1), the actual feedback is Q, and the predicted feedback is

[0160] When the random number p is less than 0.5 and the absolute value of the correlation coefficient is greater than or equal to 1, the search sub-position is updated to obtain the second position, and the expression is:

[0161]

[0162] The second position of the searcher in the t+1th iteration is Z2(t+1), the random vector between 0 and 1 is r2, and the position of the random searcher is Z random , the search subposition of the tth iteration is Z(t);

[0163] When the random number p is greater than or equal to 0.5, the third position is obtained by updating the search sub-position according to the spiral ascent, and the expression is:

[0164]

[0165] The third position of the searcher in the t+1th iteration is Z3(t+1), and the target position of the searcher is Z * (t), the random number between -1 and 1 is ρ, and the spiral control constant is λ;

[0166] Otherwise, continue to iterate until the maximum number of iterations is reached, then stop searching.

[0167] Secondly, the feedback system for patient intention brainwave recognition based on brain-computer interface includes:

[0168] Data acquisition module: used to collect EEG signals and associated data of the patient's brain-computer interface, and pre-process the EEG signals and associated data;

[0169] Data correction module: used for performing pattern classification on the EEG signal to obtain pattern data, and performing intention correction on the pattern data using the associated data to obtain correction data;

[0170] Identification and construction module: used for performing electroencephalogram signal recognition on the correction data to obtain recognition data, and constructing a patient intention electroencephalogram recognition feedback model according to the recognition data;

[0171] Optimization feedback module: used to optimize the patient intention brain wave recognition feedback model according to the feedback error, input the data to be fed back into the optimized patient intention brain wave recognition feedback model, and output the feedback result.

[0172] In the actual evaluation, the feedback system for patient intention brainwave recognition based on brain-computer interface includes software and hardware parts. The hardware part includes a data acquisition module and a patient support and connection module. The hardware part includes a data correction module, an identification construction module, and an optimization feedback module.

[0173] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A feedback method for patient intention brain wave recognition based on brain-computer interface, characterized in that: The following steps are involved: Collecting EEG signals and associated data of the patient's brain-computer interface, and preprocessing the EEG signals and the associated data; Performing pattern classification on the EEG signal to obtain pattern data, and performing intention correction on the pattern data using the associated data to obtain correction data; Performing electroencephalogram signal recognition on the correction data to obtain recognition data, and constructing a patient intention electroencephalogram recognition feedback model according to the recognition data; The patient intention brain wave recognition feedback model is optimized according to the feedback error, the data to be fed back is input into the optimized patient intention brain wave recognition feedback model, and the feedback result is output.

2. The feedback method for patient intention brain wave recognition based on brain-computer interface according to claim 1 is characterized in that: The method for performing pattern classification on the EEG signal to obtain pattern data comprises: Add particles as a search tool for EEG signals and update the search speed and position of the particles. The expression is: The local optimal position of the ith particle at the tth iteration is The global optimal position of the ith particle at the tth iteration is The position of the i-th particle at the t-th iteration is The search speed of the ith particle at the t+1th iteration is The search speed of the tth iteration of the i-th particle is The position of the i-th particle at the t+1th iteration is The random numbers between 0 and 1 are τ1 and τ2, the acceleration factors are λ1 and λ2, and the inertia weight factor is η; Calculate the similarity between EEG signals: Among them The EEG signal is The mth EEG signal is x m , EEG signal and EEG signal x m The similarity is EEG signals The probability density of EEG signalx m The probability density of EEG signals and EEG signal x m The weight factor is Cluster the EEG signals with the greatest similarity to obtain the initial clusters, and calculate the overall silhouette coefficient of the clusters: The average distance between the mth EEG signal and the EEG signal within the initial cluster is l(m), and the average distance between the mth EEG signal and the EEG signal point between the initial clusters is The number of EEG signals is Calculate the penalty term: The number of EEG signals with negative contour coefficients is N ng , the average distance between the ιth EEG signal and the initial intra-cluster EEG signal is l(ι), and the average distance between the ιth EEG signal and the initial inter-cluster EEG signal is Construct the fitness function, the expression is: The fitness function is The actual number of clusters is A, the predicted number of clusters is X, the convolution operation is *, and the cutoff distance is φ o , the indicative function is EEG signals and EEG signal x m The distance is The distance between the centers of the cth cluster is ε c , EEG signal and cluster center ε c The distance is EEG signals and cluster center ε m The distance between The membership function is The initial cluster is updated through continuous iteration of the particle population until the fitness function value is maximized, then the iteration is stopped and the updated initial cluster is output as pattern data.

3. The feedback method for patient intention brain wave recognition based on brain-computer interface according to claim 1 is characterized in that: The method of using the associated data to perform intentional correction on the pattern data to obtain correction data comprises: The associated data are analyzed to obtain the potential behavior of the patient, and the potential behavior is used to correct the pattern data to obtain the corrected data.

4. The feedback method for patient intention brain wave recognition based on brain-computer interface according to claim 1 is characterized in that: The method of performing electroencephalogram signal recognition on the correction data to obtain recognition data comprises: The overview space and detail space of the corrected data are decomposed step by step, and the overview space is transformed by the first conjugate orthogonal mirror filter bank and the second conjugate orthogonal mirror filter bank respectively to obtain the transformed overview space and detail space, and the expression is: β[n]=(-1) 1-n h[1-n] The jth overview space is E j , the j+1th detail space is C j+1 , the j+1th transformation overview space is E j+1 , the 1-nth first conjugate orthogonal mirror filter group is h[1-n], the first conjugate orthogonal mirror filter group is h[n], and the second conjugate orthogonal mirror filter group is β[n]; The function family is defined using equivalent scaling functions and equivalent wavelet functions: The equivalent scaling function is The equivalent wavelet function is The integer is R and the function family is The multi-resolution decomposition algorithm is used to perform multi-resolution decomposition on the function family, and the expression is: The adjoint of the 1-2kth function family is h * [1-2k], the adjoint of the 1-2kth first conjugate orthogonal mirror filter bank is h * [1-2k]; The number of channels of EEG signals collected gives the reference signal corresponding to the SSVEP visual stimulation frequency: The bth reference signal is v b , the number of stimulation targets is b, and the stimulation frequency of the bth stimulation target is f b , the number of sampling points is N s ; Calculate the correlation coefficient based on the EEG signal and the reference signal corresponding to the bth stimulation frequency: The horizontal weight matrix is ​​D w , the vertical weight matrix is ​​D v , the EEG signal is w, if K is the stimulation frequency data, then the target frequency is calculated and identified: The identification data is output according to the identified target frequency.

5. The feedback method for patient intention brain wave recognition based on brain-computer interface according to claim 1 is characterized in that: The method for constructing a patient intention brain wave recognition feedback model according to the recognition data comprises: The objective function is constructed based on the recognition data, and the expression is: The actual identification data is Predicted recognition data is Actual identification data and predictive recognition data The loss function is The objective function is The patient intention brainwave recognition feedback model includes empirical mode decomposition, decision tree algorithm, and machine learning pattern recognition network; Empirical mode decomposition decomposes the input EEG signal to obtain local amplitude, phase and frequency, and outputs the local amplitude, phase and frequency as the required features; The decision tree algorithm recursively divides the demand characteristics into multiple regions, and builds a tree model according to the decision rules of the regions to perform category prediction and obtain classified data; The machine learning pattern recognition network performs intent matching on the classified data to obtain matching results, and outputs the intent with matching results greater than 0.863 as feedback results.

6. The feedback method for patient intention brain wave recognition based on brain-computer interface according to claim 1 is characterized in that: The method for optimizing the patient intention brain wave recognition feedback model according to the feedback error comprises: Introducing low-discrepancy sequences to optimize the search subpopulation, the expression is: The initial position of the searcher in the dth dimension is The random number of the low discrepancy sequence in the range of 0 to 1 is S e , the maximum value of the search sub-position is Z max , search for the minimum value of the sub-position Z min ; Randomly divide the searchers into four independent groups and calculate the fitness of the searchers: The fitness function of the first group is g1, the current number of iterations is t, and the maximum number of iterations is t. max , the fitness function of the second group is g2, the fitness function of the third group is g3, and the fitness function of the fourth group is g4; Calculate the fitness of the four groups of searchers and sort them in descending order, and use the first four searchers as attackers, interceptors, drivers, and chasers in turn; Calculate the correlation coefficient: The correlation coefficient is The shrinkage coefficient is a, and the random vector between 0 and 1 is r1; Update the fitness of the searcher according to the current number of iterations. When the random number p is less than 0.5 and the absolute value of the correlation coefficient is less than 1, update the position of the searcher to obtain the first position. The expression is: The attacker's position is Z b1 , the position of the interceptor is Z b2 , the position of the driver is Z b3 , the position of the pursuer is Z b4 , the first position of the searcher in the t+1th iteration is Z1(t+1), the actual feedback is Q, and the predicted feedback is When the random number p is less than 0.5 and the absolute value of the correlation coefficient is greater than or equal to 1, the search sub-position is updated to obtain the second position, and the expression is: The second position of the searcher in the t+1th iteration is Z2(t+1), the random vector between 0 and 1 is r2, and the position of the random searcher is Z random , the search subposition of the tth iteration is Z(t); When the random number p is greater than or equal to 0.5, the third position is obtained by updating the search sub-position according to the spiral ascent, and the expression is: The third position of the searcher in the t+1th iteration is Z3(t+1), and the target position of the searcher is Z * (t), the random number between -1 and 1 is ρ, and the spiral control constant is λ; Otherwise, continue to iterate until the maximum number of iterations is reached, then stop searching.

7. A feedback system for patient intention brain wave recognition based on brain-computer interface, used to execute any method of claims 1-6, characterized in that: include: Data acquisition module: used to collect EEG signals and associated data of the patient's brain-computer interface, and pre-process the EEG signals and associated data; Data correction module: used for performing pattern classification on the EEG signal to obtain pattern data, and performing intention correction on the pattern data using the associated data to obtain correction data; Identification and construction module: used for performing electroencephalogram signal recognition on the correction data to obtain recognition data, and constructing a patient intention electroencephalogram recognition feedback model according to the recognition data; Optimization feedback module: used to optimize the patient intention brain wave recognition feedback model according to the feedback error, input the data to be fed back into the optimized patient intention brain wave recognition feedback model, and output the feedback result.

Images