A method and system for motion intention recognition for brain-computer interface

Abstract

A motion intention recognition method and system for brain-computer interface relate to the technical field of signal processing and recognition, through collecting functional near-infrared spectroscopy signals of the frontal lobe brain area of a user under a paradigm task, obtaining original light intensity data, preprocessing the light intensity data, and converting the preprocessed data into a hemodynamic response signal; the hemodynamic response signal is subjected to multiple feature extraction respectively, and the extracted multiple features are combined into a feature matrix; the feature matrix is subjected to statistical feature optimization to generate a feature selection mask, and the preferred feature set is determined based on the feature selection mask; support vector machine is used as a classifier, and the preferred feature set is input into the classifier as a training set for model training, the present disclosure realizes redundant feature removal and high-contribution feature recognition based on feature statistical optimization, considers the information mining depth and operation efficiency of the blood oxygen concentration signal, and realizes real-time presentation of the classification result.

Description

Technical Field

[0001] This disclosure relates to the fields of signal processing and recognition technology, and more specifically, to a method and system for recognizing motor intentions for brain-computer interfaces. Background Technology

[0002] Functional near-infrared spectroscopy (fNIRS) is a method for calculating brain activity by measuring the scattering and absorption of light emitted by an optical system as it passes through brain tissue. Compared to other brain imaging techniques, fNIRS offers significant advantages such as high ecovalidity, environmental robustness, and low deployment costs. In recent years, fNIRS-based brain-computer interface (BCI) systems have demonstrated great practical potential in acquiring and recognizing neural signals in everyday situations, attracting widespread attention from researchers.

[0003] fNIRS signal analysis technology, based on a modified Lambert-Beer law, converts light intensity signals into blood oxygen concentration signals. When a subject uses a specific brain region, metabolic stress causes capillary dilation in the corresponding area, leading to higher hemoglobin concentrations. The resulting light scattering and absorption effects are directly reflected in the measured light intensity data to match the activation patterns of the subject's brain regions. After obtaining hemodynamic activation data, the analytical method used determines the accuracy of stimulus recognition.

[0004] However, the related technologies have problems such as difficulty in balancing the depth of information mining of blood oxygen concentration signals with computational efficiency, feature selection being greatly affected by the specificity of brain activation time, and poor real-time performance. Summary of the Invention

[0005] In view of the many shortcomings of the existing technology, this disclosure provides a method and system for motor intention recognition in brain-computer interfaces.

[0006] The first aspect of this disclosure provides a method for recognizing motor intentions in a brain-computer interface, comprising the following steps: acquiring raw light intensity data by collecting functional near-infrared spectral signals from the frontal lobe of a user under a paradigm task; preprocessing the light intensity data and converting the preprocessed data into hemodynamic response signals; extracting multiple features from the hemodynamic response signals and combining the extracted features into a feature matrix, wherein the feature extraction includes time-domain feature extraction, frequency-domain feature extraction, network feature extraction, and curve fitting parameter extraction; performing statistical feature optimization on the feature matrix to generate a feature selection mask, and determining a preferred feature set based on the feature selection mask, wherein the statistical feature optimization includes dataset stratification, normality test, difference significance test, and feature recurrence rate test; using a support vector machine as a classifier, inputting the preferred feature set as a training set into the classifier for model training, and obtaining a trained model; and based on the trained model, performing task classification on the preferred feature set to be recognized to obtain classification results, and outputting the motor intentions corresponding to the classification results to the user interface.

[0007] Furthermore, in some embodiments, before acquiring the functional near-infrared spectral signal of the frontal lobe region of the user under the paradigm task, the method further includes: setting parameters for the paradigm prompts on the real-time classification interface according to the experimental paradigm design to guide the user to perform the paradigm task, wherein the parameter settings include classification task settings, paradigm trial settings, baseline time settings, task time settings, and resting time settings.

[0008] In some embodiments, preprocessing the light intensity data and converting the preprocessed data into a hemodynamic response signal includes: extracting the light intensity data before the end of each trial to obtain partial light intensity data; filtering the partial light intensity data to remove physiological noise, baseline drift, and spike noise to obtain filtered data; and converting the filtered data into a blood oxygen concentration signal using a modified Lambert-Beer law, wherein the blood oxygen concentration signal includes oxyhemoglobin concentration data and deoxyhemoglobin concentration data.

[0009] In some embodiments, the hemodynamic response signal is subjected to multiple feature extractions, including: time-domain feature extraction to obtain the time information of the hemodynamic response signal, wherein the time information includes mean, variance, slope, peak value, peak time, kurtosis, quadratic coefficient, and approximate entropy; frequency-domain feature extraction to obtain the frequency information of the hemodynamic response signal, wherein the frequency information includes centroid frequency, mean square frequency, and frequency variance; network feature extraction to obtain the spatial information of the hemodynamic response signal, wherein the spatial information includes node importance, average path length, clustering coefficient, node clustering coefficient, average degree, and degree distribution; and curve fitting parameters of the hemodynamic response signal are extracted using a deep learning network.

[0010] In some embodiments, the deep learning network includes a multilayer perceptron and a long short-term memory network. The curve fitting parameters for extracting hemodynamic response signals using the deep learning network include: simultaneously using a multilayer perceptron and a long short-term memory network to fit the response curve of each channel of the hemodynamic response signal, and using the trained network parameters as features to form a feature matrix. The multilayer perceptron network for fitting the response curve includes an input layer, a hidden layer with 10 neurons, and an output layer. The long short-term memory network for fitting the response curve includes a sequence input layer, a long short-term memory layer, a modified linear unit activation layer, a fully connected layer, and a regression layer.

[0011] In some embodiments, the preferred generation of a feature selection mask for statistical features of the feature matrix includes: using the feature matrix obtained from valid trials as samples, performing stratified sampling on the dataset including all samples according to a set rule to obtain stratified data subsets, combining each data subset with the same samples according to the sample label and classification task to obtain a sample set; performing a normality test on the sample set to determine whether the channel feature distribution of the sample set conforms to a normal distribution, and generating a first feature selection mask to indicate the normality test result; performing a difference significance test on the sample set to determine whether the channel feature distribution under different labels in the sample set conforms to the difference significance, and generating a second feature selection mask to indicate the difference significance result; performing a feature recurrence rate test on the sample set, statistically analyzing the recurrence rate of the channel features, and generating a third feature selection mask based on the recurrence rate and a configured recurrence rate threshold to indicate the recurrence rate test result.

[0012] In some embodiments, performing a normality test on the sample set to determine whether the channel feature distribution of the sample set conforms to a normal distribution and generating a first feature selection mask to indicate the normality test result includes: grouping the sample set according to different labels for each layer of the sample set; calculating the p-value for each group of samples independently using the Jarque-Bera test method; if the p-value meets the configured first floating threshold, determining that the channel feature distribution within the current sample group conforms to a normal distribution, and setting the first feature selection mask to 1; or, if the p-value does not meet the configured first floating threshold, determining that the channel feature distribution within the current sample group does not conform to a normal distribution, and setting the first feature selection mask to 0.

[0013] In some embodiments, performing a significance test on the sample set to determine whether the channel feature distribution under different labels in the sample set meets the significance of difference, and generating a second feature selection mask to indicate the significance of difference results includes: for each stratified sample set, grouping the sample set according to different classification tasks; calculating the p-value for each group of samples using the t-test method and the analysis of variance method; if the p-value meets the configured second floating threshold, determining that the channel feature distribution within the current sample group meets the significance of difference, and the second feature selection mask is 1; or, if the p-value does not meet the configured second floating threshold, determining that the channel feature distribution within the current sample group does not meet the significance of difference, and the second feature selection mask is 0.

[0014] In some embodiments, a feature recurrence rate test is performed on the sample set, the recurrence rate of channel features is counted, and a third feature selection mask is generated based on the recurrence rate and a configured recurrence rate threshold to indicate the recurrence rate test result. This includes: counting the number of channel features in all strata that simultaneously conform to a normal distribution and significant differences, and combining the number of all strata to obtain the recurrence rate; if the recurrence rate meets the configured recurrence rate threshold, the channel feature is determined to pass the recurrence rate test, and the third feature selection mask is 1; or, if the recurrence rate does not meet the configured recurrence rate threshold, the channel feature is determined to fail the recurrence rate test, and the third feature selection mask is 0.

[0015] A second aspect of this disclosure provides a motor intention recognition device for a brain-computer interface, comprising: a data acquisition module for acquiring raw light intensity data by acquiring functional near-infrared spectral signals from the frontal lobe of a user under a paradigm task, preprocessing the light intensity data, and converting the preprocessed data into a hemodynamic response signal; and a feature extraction module for extracting various features from the hemodynamic response signal and combining the extracted features into a feature matrix, wherein the feature extraction includes time-domain feature extraction, frequency-domain feature extraction, network feature extraction, and curve fitting parameter extraction; and feature... The statistical feature selection module performs statistical feature selection on the feature matrix to generate a feature selection mask. The mask is then used to determine the preferred feature set. Statistical feature selection includes dataset stratification, normality testing, difference significance testing, and feature recurrence rate testing. The training module uses a support vector machine as a classifier, inputting the preferred feature set as the training set to train the model. The classification module, based on the trained model, performs task classification on the preferred feature set to be identified, obtains the classification results, and outputs the corresponding motion intent to the user interface.

[0016] The third aspect of this disclosure provides a real-time classification interface for motor intention recognition in a brain-computer interface, including a paradigm setting module, a data reading module, a recognition module, and a classification result presentation module. The paradigm setting module is used to set relevant parameters for paradigm prompt presentation. The data reading module is used to read signals collected in real time by the fNIRS device to obtain raw data. The recognition module is used to preprocess, extract features, and optimize features from the collected raw data before inputting it into the classifier to obtain classification results. The classification result presentation module is used to present the classification results in a visual form in real time.

[0017] A fourth aspect of this disclosure provides an electronic device comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the aforementioned first aspect.

[0018] A fifth aspect of this disclosure provides a non-transitory computer-readable storage medium storing computer instructions, wherein the computer instructions are used to cause a computer to perform the aforementioned first aspect.

[0019] A sixth aspect of this disclosure provides a computer program product including a computer program that, when executed by a processor, implements the first aspect described above.

[0020] Due to the adoption of the above technical solutions, the beneficial effects of this disclosure are as follows: This disclosure calculates the hemodynamic activation signal under the paradigm task based on frontal lobe fNIRS data, combines multiple feature extraction methods and deep learning to obtain a feature matrix to fully express the data information, and achieves redundant feature removal and high contribution feature identification through statistical optimization methods, taking into account both the information mining depth and computational efficiency of blood oxygen concentration signals. Based on the above methods, model training and real-time classification interface deployment are completed, taking into account both information mining depth and computational efficiency, solving the problems of feature selection being greatly affected by the specificity of brain activation time and poor real-time performance, and achieving high-precision motion decision recognition. Attached Figure Description

[0021] The above and other objects, features, and advantages of this disclosure will become clearer from the following description of embodiments of the present disclosure with reference to the accompanying drawings, in which:

[0022] Figure 1 A flowchart illustrating a motion intention recognition method for a brain-computer interface according to an embodiment of the present disclosure is shown schematically.

[0023] Figure 2 The illustration schematically shows a paradigmatic experimental design of a motor intention recognition method for a brain-computer interface according to an embodiment of the present disclosure;

[0024] Figure 3 The flowchart illustrating a preprocessing method for a motion intention recognition method for a brain-computer interface according to an embodiment of the present disclosure is shown.

[0025] Figure 4 A flowchart illustrating a feature extraction method for a brain-computer interface motion intention recognition method according to an embodiment of the present disclosure is shown.

[0026] Figure 5 A flowchart illustrating a feature statistical optimization method for a motor intention recognition method for a brain-computer interface according to an embodiment of the present disclosure is shown.

[0027] Figure 6 A schematic diagram of a motion intention recognition system for a brain-computer interface according to an embodiment of the present disclosure is shown.

[0028] Figure 7 A schematic diagram of an electronic device according to an embodiment of the present disclosure is shown. Detailed Implementation Plan

[0029] The exemplary embodiments of this disclosure are described below with reference to the accompanying drawings, including various details of the embodiments to aid understanding, and should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of this disclosure. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.

[0030] To make the objectives, technical solutions, and advantages of the embodiments of this disclosure clearer, the relevant technologies of the technical solutions of this disclosure will first be described.

[0031] In the fNIRS-BCI study on motor intention recognition, it was found that the deep learning-based feature extraction method generates more features, resulting in longer computation time and decreased recognition accuracy. In particular, it is difficult to achieve high-precision recognition of complex motor intentions, especially for measurements of the frontal lobe brain region, making it unsuitable for application in everyday situations.

[0032] To identify different motion patterns, a reliable decoding algorithm is needed to extract effective information from fNIRS signals. The quality of the measured data depends primarily on the performance of the acquisition equipment and the acquisition techniques. To minimize noise during the measurement phase, a suitable filtering algorithm should be selected to address the disturbances that accompany the measurement. Meanwhile, in the BCI system, using all features as input results in significant redundancy. Non-task-related features lead to longer computation time and reduced training performance of the classification model. When neural networks are used for feature extraction, a large number of features are extracted, making the disadvantages of retaining redundant features even more pronounced.

[0033] Given that existing fNIRS-BCI systems have limitations in terms of classification accuracy and the complexity of classification tasks, they cannot meet the requirements of brain-computer interface applications in everyday scenarios due to problems such as difficulty in balancing information mining depth and computational efficiency, feature selection being greatly affected by the specificity of brain activation time, and poor real-time performance.

[0034] Therefore, in order to solve the problems in related technologies, this invention provides a method for motor intention recognition in brain-computer interfaces. This method optimizes the effective feature extraction of fNIRS signals based on hemodynamic response signals measured in the frontal lobe brain region. It combines multiple feature extraction methods and deep learning to obtain a feature matrix to fully express data information. Statistical optimization methods are used to remove redundant features and identify high-contribution features. The method also completes model training and deployment on a real-time classification interface, balancing information mining depth and computational efficiency. This solves the problems of feature selection being greatly affected by the specificity of brain activation time and poor real-time performance.

[0035] The following description, with reference to the accompanying drawings, describes a method, system, real-time classification interface, and electronic device for recognizing motor intentions using a brain-computer interface.

[0036] Figure 1 A flowchart illustrating a motion intention recognition method for a brain-computer interface according to an embodiment of the present disclosure is shown schematically.

[0037] like Figure 1 As shown, the method includes the following steps 101-105.

[0038] Step 101: Obtain raw light intensity data by collecting functional near-infrared spectral signals of the frontal lobe brain region of the user under the paradigm task, preprocess the light intensity data, and convert the preprocessed data into hemodynamic response signals.

[0039] In some embodiments, before acquiring the functional near-infrared spectral signal of the frontal lobe brain region of the user under the paradigm task, the method further includes: setting parameters for the paradigm prompts on the real-time classification interface according to the experimental paradigm design to guide the user to perform the paradigm task, wherein the parameter settings include classification task settings, paradigm trial settings, baseline time settings, task time settings, and resting time settings.

[0040] In some embodiments, the classification task setting is used to deploy the algorithm model according to the selected classification task and change the image presentation of the classification result presentation module to adapt to the selected task; the paradigm trial setting is used to configure the number of cycles of task prompts and rest prompts on the paradigm prompt screen; the baseline time setting is used to configure the duration of the baseline rest state before the paradigm starts; the task time setting is used to set the duration of the task prompt presentation on the paradigm prompt screen; and the rest time setting is used to set the duration of the rest prompt presentation on the paradigm prompt screen.

[0041] In some embodiments, the paradigm is designed to perform tasks on five types of movements: resting, left-hand grasping, right-hand grasping, left-hand finger tapping, and right-hand finger tapping.

[0042] In one embodiment of this disclosure, such as Figure 2 As shown, a single paradigm experiment begins with a 30-second baseline time, followed by 20 task blocks in five cycles of the rest-left-hand grasp-rest-right-hand grasp sequence (Trial). The duration of each rest and task block can be 5 seconds, 10 seconds, or 15 seconds, with a total experimental time of 230 seconds. When the task duration is 15 seconds, the paradigm experiment consists of 12 task blocks in three cycles of the rest-left-hand grasp-rest-right-hand grasp sequence.

[0043] Furthermore, the motor execution in the paradigm experiment can be a tapping action. A single paradigm experiment starts with a 30-second baseline block, followed by 20 task blocks in five cycles of rest-left-hand tapping-rest-right-hand tapping sequence. The duration of each rest and task block can be 5 seconds, 10 seconds, or 15 seconds, with a total experimental time of 230 seconds. When the task duration is 15 seconds, the paradigm experiment consists of 12 task blocks in three cycles of rest-left-hand tapping-rest-right-hand tapping sequence.

[0044] It should be noted that, in order to ensure the standardization of the paradigm execution, the subjects maintained a comfortable posture and screen distance while seated and completed the task according to the prompts presented on the screen. A complete grasping action was set as follows: starting from a natural state, each finger first extends outward, pauses slightly after reaching the maximum abduction state, then grasps inward with force, holds the fist for a short period of time, and finally returns to the natural state; the finger tapping action was performed at a frequency of 3Hz.

[0045] In one embodiment of this disclosure, a total of 500 data samples were collected from three subjects in a single measurement cycle, including 100 sets each of resting, left-hand grasping, right-hand grasping, left-hand finger tapping, and right-hand finger tapping.

[0046] In some embodiments, the fNIRS signal acquisition device is used to achieve parallel acquisition of all measurement channels based on a high-precision digital phase-locked loop algorithm. It is equipped with two near-infrared light wavelengths of 775nm and 855nm, and is equipped with 10 light sources and 4 detectors. The sampling frequency is set to 2Hz.

[0047] In some embodiments of this disclosure, according to the paradigm task, all data before the end time of each sequence is extracted, this part of the data is filtered, and before feature extraction, the light intensity signal is converted into a blood oxygen concentration signal based on the modified Lambert-Beer law, and then only the data during the task is retained according to the tag.

[0048] like Figure 3 As shown, the light intensity data is preprocessed and the preprocessed data is converted into a hemodynamic response signal, including the following steps 201-203.

[0049] Step 201: Extract the light intensity data before the end of each trial to obtain partial light intensity data.

[0050] Step 202: Filter a portion of the light intensity data to remove physiological noise, baseline drift, and spike noise, and obtain the filtered data.

[0051] In some embodiments, the filtering method uses an 8th-order 0.45Hz Butterworth low-pass filter to remove physiological noise such as heartbeat and breathing. For each channel of data, a moving standard deviation of 5 and a Sobel filter are used to identify the segment to be processed. For the segment to be processed, cubic spline interpolation and a Sacitzky-Golay filter are used to remove baseline drift and spike noise, respectively.

[0052] Step 203: The filtered data is converted into a blood oxygen concentration signal using the modified Lambert-Beer law. The blood oxygen concentration signal includes oxyhemoglobin (HbO) concentration data and deoxyhemoglobin (HbR) concentration data.

[0053] In some embodiments, the modified Lambert-Beer law calculation formula is as follows:

[0054] ;

[0055] ;

[0056] Where L is the total average path length. This refers to the intensity attenuation at a wavelength of 775nm. This refers to the intensity attenuation at a wavelength of 855nm. The extinction coefficient of deoxyhemoglobin for near-infrared light at a wavelength of 775 nm. The extinction coefficient of deoxyhemoglobin for near-infrared light at a wavelength of 855nm. The extinction coefficient of oxyhemoglobin for near-infrared light at a wavelength of 775 nm. The extinction coefficient of oxyhemoglobin for near-infrared light at a wavelength of 855nm. This refers to the concentration of oxyhemoglobin. This refers to the concentration of deoxyhemoglobin.

[0057] In some embodiments, oxyhemoglobin concentration data is input into step 102 for feature extraction.

[0058] Step 102: Perform various feature extractions on the hemodynamic response signal and combine the extracted features into a feature matrix. The feature extraction includes time-domain feature extraction, frequency-domain feature extraction, network feature extraction, and curve fitting parameter extraction.

[0059] In some embodiments of this disclosure, the preprocessed data is converted into a hemodynamic response signal and input to a feature extraction module. The feature extraction module includes a time-domain feature extraction module, a frequency-domain feature extraction module, a network feature extraction module, and a deep learning module. The time-domain feature extraction module is used to extract the time information of the hemodynamic response, the frequency-domain feature extraction module is used to extract the frequency information of the hemodynamic response, the network feature extraction module is used to extract the spatial information of the hemodynamic response, and the deep learning module is used to extract the fitting parameters of the hemodynamic response curve.

[0060] In some embodiments, such as Figure 4 As shown, the hemodynamic response signal is subjected to various feature extractions, including the following steps 301-304.

[0061] Step 301: Extract time-domain features from the hemodynamic response signal to obtain the time information of the hemodynamic response signal. The time information includes mean, variance, slope, peak value, peak time, kurtosis, quadratic coefficient, and approximate entropy.

[0062] In some embodiments, by formula Calculate the mean value of the hemodynamic response signal within the task response window, where, N represents the mean, and N represents the number of data points in the window. This represents the HbO / HbR concentration value at the q-th time point within the window.

[0063] In some embodiments, by formula Calculate the variance of the hemodynamic response signal within the task response window, where, Represents variance.

[0064] In some embodiments, by formula Calculate the peak value of the hemodynamic response signal within the task response window, where E[Z] 4 The ] operator represents the expectation (mean) value, where EX represents the mean u of sample points X, and DX represents the variance of X. , Z represents the standard deviation of sample point X. This indicates that a standardized value is calculated for each sample point.

[0065] In some embodiments, a quadratic polynomial is fitted to each channel, and the polynomial fitting is solved by QR decomposition to extract the slope and quadratic coefficients.

[0066] Specifically, suppose there are m channels that need to be fitted, and the signal of each channel can be represented as an nth-order polynomial over time x:

[0067] ;

[0068] in, For the polynomial of the th The coefficient of the order term, is the coefficient of the highest-order term, and y is the observed signal value for each channel.

[0069] Furthermore, the system of equations is constructed as follows:

[0070] ;

[0071] in, A matrix composed of vectors of variables of various orders; , where n is the highest order of the variable; , where m is the number of channels to be fitted;

[0072] Then, perform QR decomposition on matrix A according to the following formula to solve for the vectors.

[0073] ;

[0074] in, Let be a coefficient vector, and have It is a constant. The value represents the slope we are looking for. This represents the coefficient of the quadratic term. It is an orthogonal matrix. Denotes the transpose of the orthogonal matrix Q. It is a non-singular upper triangular matrix.

[0075] Step 302: Extract frequency domain features from the hemodynamic response signal to obtain frequency information of the hemodynamic response signal, including centroid frequency, mean square frequency, and frequency variance.

[0076] In some embodiments, the centroid frequency is obtained using the following operational expression:

[0077] ;

[0078] in, Represents the center of gravity frequency. The power spectral density (PSD) can be calculated using the Welch method.

[0079] In some embodiments, the mean square frequency is obtained using the following computational expression:

[0080] ;

[0081] In some embodiments, the frequency variance is calculated based on the centroid frequency and PSD, specifically using the following calculation expression:

[0082] ;

[0083] Step 303: Extract network features from the hemodynamic response signal to obtain spatial information of the hemodynamic response. The spatial information includes node importance, average path length, clustering coefficient, node clustering coefficient, average degree, and degree distribution.

[0084] In some embodiments, the Pearson correlation coefficient between each channel is first calculated to construct an adjacency matrix. The edges between nodes are constructed using thresholds of 0.3, 0.4, 0.5, 0.6, 0.7, and 0.8, respectively. The number of rows in the network feature matrix represents the selected threshold, and the number of columns represents the extracted network features. For each threshold, the node importance, average path length, clustering coefficient, node clustering coefficient, average degree, and degree distribution characteristics are calculated in the adjacency matrix.

[0085] In some embodiments, node importance includes degree centrality, betweenness centrality, and eigenvector centrality. Degree centrality refers to the sum of the edge weights connecting the node, betweenness centrality refers to the frequency of the node's occurrence in all shortest paths, and eigenvector centrality refers to the weighted sum of the importance of the node's neighbors.

[0086] In some embodiments, the average path length is calculated using the following expression:

[0087] ;

[0088] in, This represents the average path length. Let z be the shortest path length from node z to j. This represents the maximum number of edges.

[0089] In some embodiments, the clustering coefficient is obtained using the following operational expression:

[0090] ;

[0091] in, Representing nodes, j and k are nodes that belong to. The set of adjacent nodes , , For nodes Adjacent points , The actual number of edges between them For nodes The number of adjacent nodes.

[0092] Step 304: Use a deep learning network to extract curve fitting parameters for the hemodynamic response signal.

[0093] In embodiments of this disclosure, the core objective of Fitting Response Curves is to establish a mathematical relationship between brain activity and hemodynamic signals (such as changes in HbO / HbR concentration).

[0094] In some embodiments of this disclosure, step 304 includes: simultaneously using a multilayer perceptron (MLP) and a long short-term memory (LSTM) network to fit the response curve of each channel to the hemodynamic response signal, and using the trained network parameters as features to form a feature matrix.

[0095] The MLP network for fitting the response curve includes an input layer, a hidden layer with 10 neurons, and an output layer. The LSTM network for fitting the response curve includes a sequence input layer, an LSTM layer, a modified linear unit activation layer, a fully connected layer, and a regression layer.

[0096] Step 103: Perform statistical feature optimization on the feature matrix, generate a feature selection mask, and determine the preferred feature set based on the feature selection mask. Statistical feature optimization includes dataset stratification, normality test, difference significance test, and feature recurrence rate test.

[0097] In some embodiments of this disclosure, the features extracted by each feature extraction module are combined into a feature matrix and input into a statistical feature optimization module. The statistical feature optimization module includes a dataset stratification module, a normality test module, a difference significance test module, and a feature recurrence rate test module. The dataset stratification module is used to divide the sample set for statistical testing and to assign labels to each group of samples. The normality test module is used to test the normality level of the sample distribution of the sample set. The difference significance test module is used to test the difference significance of the sample distribution under different labels in the sample set. The feature recurrence rate test module is used to test the number of times the same feature is repeatedly selected under different strata.

[0098] In some embodiments, such as Figure 5 As shown, the statistical feature optimization of the feature matrix and the generation of the feature selection mask include the following steps 401-404.

[0099] Step 401: Using the feature matrix obtained from the effective trials as samples, perform stratified sampling on the dataset including all samples according to the set rules to obtain stratified data subsets. Based on the sample labels and classification tasks, combine each data subset with the same samples to obtain a sample set.

[0100] In the embodiments of this disclosure, the dataset stratification module first determines that every 5 samples constitute a stratified sampling based on the number of samples in the total input dataset, thereby obtaining a stratified dataset. Each dataset with the same sample data is then combined according to the label and classification task and input into the subsequent normality test module, difference significance test module, and feature recurrence rate test module.

[0101] In some embodiments, the degree of difference between each corresponding element of the two sets of samples is calculated. The features with significant differences retained in this way are used as the initial feature matrix 1. After increasing the number of samples drawn, the same method is used to form the initial feature matrix 2. This operation is continued until more samples are drawn, resulting in feature matrix n. After obtaining all the initial feature matrices, the number of times each feature appears in the initial feature matrices is used to determine whether it is a preferred feature input.

[0102] In some embodiments, the size of the temporal feature matrix is ​​(7, 10), meaning that 7 temporal features are extracted from each of the 10 channels; the size of the frequency domain feature matrix is ​​(4, 10), meaning that 4 frequency domain features are extracted from each of the 10 channels; the size of the network feature matrix is ​​(6, 24), meaning that 24 network features are extracted from each of the 6 threshold selections for building the network from the adjacency matrix; the size of the MLP network parameter feature matrix is ​​(31, 10), meaning that 31 features are extracted from each of the 10 channels; and the size of the LSTM network parameter feature matrix is ​​(571, 10), meaning that 571 features are extracted from each of the 10 channels.

[0103] Step 402: Perform a normality test on the sample set to determine whether the channel feature distribution of the sample set conforms to a normal distribution, and generate a first feature selection mask to indicate the normality test result.

[0104] In some embodiments, performing a normality test on the sample set to determine whether the channel feature distribution of the sample set conforms to a normal distribution and generating a first feature selection mask to indicate the normality test result includes: grouping the sample set according to different labels for each layer of the sample set; calculating the p-value for each group of samples independently using the Jarque-Bera test method; if the p-value meets the configured first floating threshold, determining that the channel feature distribution within the current sample group conforms to a normal distribution, and setting the first feature selection mask to 1; or, if the p-value does not meet the configured first floating threshold, determining that the channel feature distribution within the current sample group does not conform to a normal distribution, and setting the first feature selection mask to 0.

[0105] Specifically, firstly, based on different labels, the normality of the feature sample distribution is calculated, and the Jarque-Bera (JB) test is used to calculate the p-value. Then, based on the floating threshold settings, it is determined whether the feature distribution on the corresponding channel conforms to a normal distribution. The formula for calculating the JB test is as follows:

[0106] ;

[0107] Where N represents the sample size, i.e., the number of data points within the window, S represents the skewness value, and K represents the kurtosis value. The formulas for calculating the skewness value S and the kurtosis value K are as follows:

[0108] ;

[0109] ;

[0110] in, This is the sample mean.

[0111] Furthermore, a first feature selection mask is generated. The result of the normality test is indicated by 0 and 1, where 1 indicates that the corresponding location feature is selected and 0 indicates that the corresponding location feature is removed.

[0112] Step 403: Perform a significance test on the sample set to determine whether the channel feature distribution under different labels in the sample set meets the significance test, and generate a second feature selection mask to indicate the significance test result.

[0113] In some embodiments, performing a significance test on the sample set to determine whether the channel feature distribution under different labels in the sample set meets the significance of difference, and generating a second feature selection mask to indicate the significance of difference results includes: for each stratified sample set, grouping the sample set according to different classification tasks; calculating the p-value for each group of samples using the t-test method and the analysis of variance method; if the p-value meets the configured second floating threshold, determining that the channel feature distribution within the current sample group meets the significance of difference, and the second feature selection mask is 1; or, if the p-value does not meet the configured second floating threshold, determining that the channel feature distribution within the current sample group does not meet the significance of difference, and the second feature selection mask is 0.

[0114] Specifically, depending on the classification task, t-tests and ANOVA are used to calculate p-values, and then the significance of differences in features across channels is configured according to the floating threshold. The independent samples t-test compares whether there is a significant difference between the means of two independent samples X and Y. The formula for the statistic is as follows:

[0115] ;

[0116] in, Represents the sample mean of X. Represents the sample mean of Y. (X) represents the sample standard deviation of X. (Y) represents the standard deviation of the Y sample, n1 represents the sample size of X, and n2 represents the sample size of Y.

[0117] The samples are divided into groups 1-j based on different classification tasks. The formula for calculating the statistics using the analysis of variance method is as follows:

[0118] First, calculate the sample mean and sample variance:

[0119] ;

[0120] ;

[0121] The variance between groups is calculated by the following formula:

[0122] ;

[0123] in, This represents the sample mean for the j-th classification task. This represents the sample size for the j-th classification task. This represents the value of the i-th sample point in the j-th classification task. The variance between groups, This represents the sum of squares between groups (SSB). This represents the total sample mean for all groups. For SSB, the degrees of freedom are the total sample mean. The calculation formula is as follows:

[0124] ;

[0125] in, This represents the sum of the sizes of each sample. In this case, the within-group variance (MSE) can be calculated as:

[0126] ;

[0127] in, The sum of squared errors (SSE) is the sum of squared errors. For SSE, the degrees of freedom.

[0128] Furthermore, a second feature selection mask is generated. The result of the significance test indicates the difference, consisting of 0 and 1. 1 indicates that the corresponding positional feature is selected, and 0 indicates that the corresponding positional feature is removed.

[0129] Step 404: Perform feature recurrence rate test on the sample set, count the recurrence rate of channel features, and generate a third feature selection mask based on the recurrence rate and the configured recurrence rate threshold to indicate the recurrence rate test result.

[0130] In the embodiments of this disclosure, the number of features that simultaneously meet the channel characteristics and conform to the normal distribution and have significant differences under each layered sampling is sequentially superimposed to obtain the number of channel features that meet the configured normality requirements and significant differences under different layered conditions. Then, the feature selection mask is output by the configured recurrence rate threshold.

[0131] In some embodiments, a feature recurrence rate test is performed on the sample set, the recurrence rate of channel features is counted, and a third feature selection mask is generated based on the recurrence rate and a configured recurrence rate threshold to indicate the recurrence rate test result. This includes: counting the number of channel features in all strata that simultaneously conform to a normal distribution and significant differences, and combining the number of all strata to obtain the recurrence rate; if the recurrence rate meets the configured recurrence rate threshold, the channel feature is determined to pass the recurrence rate test, and the third feature selection mask is 1; or, if the recurrence rate does not meet the configured recurrence rate threshold, the channel feature is determined to fail the recurrence rate test, and the third feature selection mask is 0.

[0132] In embodiments of this disclosure, the preferred output of the statistical features is:

[0133] ;

[0134] in, Let be the input feature matrix of the i-th sample, and isnan denotes the removal of 0 elements from the matrix. For all input features of the i-th sample, , , These are the results of the normality test, the significance test, and the recurrence rate test, respectively, consisting of 0 and 1. 1 indicates that the corresponding positional feature is selected, while 0 indicates that the corresponding positional feature is removed.

[0135] Step 104: Use a support vector machine as a classifier, input the selected feature set as the training set into the classifier to train the model, and obtain the trained model.

[0136] According to embodiments of this disclosure, a support vector machine (SVM) is used as a classifier. A preferred feature set is input as a training set into the classifier to train the model parameters. Newly input data is used as a test set to verify the model's recognition accuracy. Alternatively, 70% of the total dataset is used as the training set and 30% as the validation set. The training set is input into the model to train the model parameters, and the validation set data is input into the trained model to verify the accuracy. Recall rate ,Fraction and accuracy The calculation formulas for evaluating the model training performance are as follows:

[0137] ;

[0138] ;

[0139] ;

[0140] ;

[0141] Where TP (True Positive) represents the number of samples correctly predicted as positive by the model, FP (False Positive) represents the number of samples incorrectly predicted as positive by the model but actually as negative, FN (False Negative) represents the positive samples missed by the model, and TN (True Negative) represents the negative samples correctly excluded by the model. In the calculation formula, Represents accuracy R represents recall rate .

[0142] According to embodiments of this disclosure, the maximum classification accuracy is 91% for a left-hand grasp-right-hand grasp binary task, 91% for a left-hand tapping-right-hand tapping binary task, 93% for a resting-grasping-tapping triadic task, and 65% for a resting-left-hand grasping-right-hand grasping-left-hand tapping-right-hand tapping quintuple task.

[0143] Step 105: Based on the trained model, perform task classification on the preferred feature set to be identified to obtain the classification results, and output the motion intent corresponding to the classification results to the user interface.

[0144] Figure 6 A motion intention recognition system 500 for a brain-computer interface according to an embodiment of the present disclosure is illustrated schematically, such as... Figure 6 As shown, the system includes:

[0145] The data acquisition module 510 is used to acquire raw light intensity data by acquiring functional near-infrared spectral signals from the frontal lobe brain region of the user under the paradigm task, preprocess the light intensity data, and convert the preprocessed data into hemodynamic response signals.

[0146] The feature extraction module 520 is used to extract various features from the hemodynamic response signal and combine the extracted features into a feature matrix. The feature extraction includes time-domain feature extraction, frequency-domain feature extraction, network feature extraction, and curve fitting parameter extraction.

[0147] The feature statistical optimization module 530 is used to perform statistical feature optimization on the feature matrix to generate a feature selection mask. The preferred feature set is determined through the feature selection mask. The statistical feature optimization includes dataset stratification, normality test, difference significance test, and feature recurrence rate test.

[0148] Training module 540: The training module is used to use a support vector machine as a classifier, and inputs the preferred feature set as the training set into the classifier to train the model and obtain the trained model.

[0149] The classification module 550 is used to classify the preferred feature set to be identified based on the trained model, obtain the classification result, and output the corresponding motion intention to the user interface.

[0150] In some embodiments, the system further includes a parameter setting module, which is used to set parameters for paradigm prompts on the real-time classification interface based on the experimental paradigm design, so as to guide the user to perform the paradigm task. The parameter settings include classification task settings, paradigm trial settings, baseline time settings, task time settings, and rest time settings.

[0151] In some embodiments, the data acquisition module 510 is specifically used to: extract light intensity data before the end of each trial to obtain partial light intensity data; filter the partial light intensity data to remove physiological noise, baseline drift, and spike noise from the partial light intensity data to obtain filtered data; and convert the filtered data into a blood oxygen concentration signal using a modified Lambert-Beer law, wherein the blood oxygen concentration signal includes oxyhemoglobin concentration data and deoxyhemoglobin concentration data.

[0152] In some embodiments, the feature extraction module 520 is specifically used for: performing time-domain feature extraction on the hemodynamic response signal to obtain time information of the hemodynamic response signal, wherein the time information includes mean, variance, slope, peak value, peak time, kurtosis, quadratic coefficient, and approximate entropy; performing frequency-domain feature extraction on the hemodynamic response signal to obtain frequency information of the hemodynamic response signal, wherein the frequency information includes centroid frequency, mean square frequency, and frequency variance; performing network feature extraction on the hemodynamic response signal to obtain spatial information of the hemodynamic response, wherein the spatial information includes node importance, average path length, clustering coefficient, node clustering coefficient, average degree, and degree distribution; and using a deep learning network to extract curve fitting parameters of the hemodynamic response signal.

[0153] In some embodiments, the deep learning network includes a multilayer perceptron and a long short-term memory network. The curve fitting parameters for extracting hemodynamic response signals using the deep learning network include: simultaneously using a multilayer perceptron and a long short-term memory network to fit the response curve of each channel of the hemodynamic response signal, and using the trained network parameters as features to form a feature matrix. The multilayer perceptron network for fitting the response curve includes an input layer, a hidden layer with 10 neurons, and an output layer. The long short-term memory network for fitting the response curve includes a sequence input layer, a long short-term memory layer, a modified linear unit activation layer, a fully connected layer, and a regression layer.

[0154] In some embodiments, the feature statistics optimization module 530 is specifically used to: take the feature matrix obtained from the effective trials as samples, perform stratified sampling on the data set including all samples according to the set rules to obtain stratified data subsets, combine each data subset with the same samples according to the sample label and classification task to obtain a sample set; perform a normality test on the sample set to determine whether the channel feature distribution of the sample set conforms to a normal distribution, and generate a first feature selection mask to indicate the normality test result; perform a difference significance test on the sample set to determine whether the channel feature distribution under different labels in the sample set conforms to the difference significance, and generate a second feature selection mask to indicate the difference significance result; perform a feature recurrence rate test on the sample set, count the recurrence rate of the channel features, and generate a third feature selection mask according to the recurrence rate and the configured recurrence rate threshold to indicate the recurrence rate test result.

[0155] In some embodiments, performing a normality test on the sample set to determine whether the channel feature distribution of the sample set conforms to a normal distribution and generating a first feature selection mask to indicate the normality test result includes: grouping the sample set according to different labels for each layer of the sample set; calculating the p-value for each group of samples independently using the Jarque-Bera test method; if the p-value meets the configured first floating threshold, determining that the channel feature distribution within the current sample group conforms to a normal distribution, and setting the first feature selection mask to 1; or, if the p-value does not meet the configured first floating threshold, determining that the channel feature distribution within the current sample group does not conform to a normal distribution, and setting the first feature selection mask to 0.

[0156] In some embodiments, performing a significance test on the sample set to determine whether the channel feature distribution under different labels in the sample set meets the significance of difference, and generating a second feature selection mask to indicate the significance of difference results includes: for each stratified sample set, grouping the sample set according to different classification tasks; calculating the p-value for each group of samples using the t-test method and the analysis of variance method; if the p-value meets the configured second floating threshold, determining that the channel feature distribution within the current sample group meets the significance of difference, and the second feature selection mask is 1; or, if the p-value does not meet the configured second floating threshold, determining that the channel feature distribution within the current sample group does not meet the significance of difference, and the second feature selection mask is 0.

[0157] In some embodiments, a feature recurrence rate test is performed on the sample set, the recurrence rate of channel features is counted, and a third feature selection mask is generated based on the recurrence rate and a configured recurrence rate threshold to indicate the recurrence rate test result. This includes: counting the number of channel features in all strata that simultaneously conform to a normal distribution and significant differences, and combining the number of all strata to obtain the recurrence rate; if the recurrence rate meets the configured recurrence rate threshold, the channel feature is determined to pass the recurrence rate test, and the third feature selection mask is 1; or, if the recurrence rate does not meet the configured recurrence rate threshold, the channel feature is determined to fail the recurrence rate test, and the third feature selection mask is 0.

[0158] According to embodiments of this disclosure, a real-time interface for motion intent is provided, including a paradigm setting module, a data reading module, a recognition module, and a classification result presentation module. The paradigm setting module is used to set relevant parameters for paradigm prompt presentation. The data reading module is used to read signals collected in real time by an fNIRS device. The recognition module is used to preprocess, extract, and optimize the features of the collected raw data before inputting it into a classifier to obtain classification results. The classification result presentation module is used to present the classification results in a visual form in real time.

[0159] According to embodiments of this disclosure, the paradigm setting module includes classification task setting, paradigm trial setting, baseline time setting, task time setting, and rest time setting. The classification task setting is used to deploy the algorithm model according to the selected classification task and change the image presentation of the classification result presentation module to adapt to the selected task. The paradigm trial setting is used to configure the number of cycles for the task prompts and rest prompts on the paradigm prompt screen. The baseline time setting is used to configure the duration of the baseline rest state before the paradigm starts. The task time setting is used to set the duration of the task prompt presentation on the paradigm prompt screen. The rest time setting is used to set the duration of the rest prompt presentation on the paradigm prompt screen.

[0160] According to embodiments of this disclosure, the data reading module reads data in real time from the classification interface based on the storage path when the fNIRS device collects the data, and inputs it into the recognition module.

[0161] According to embodiments of this disclosure, the recognition module first segments the input reading data based on the number of completed paradigm tasks presented, retains the data before the end of the last resting presentation, inputs it into the preprocessing module, and after feature extraction and feature selection, inputs it into the trained support vector machine model, and obtains the output classification result, which is then input into the classification result presentation module.

[0162] According to embodiments of this disclosure, the classification result presentation module reads the classification results according to the set classification task, selects and displays the corresponding hand action images that match the classification results in real time on the interface, and finally saves the data generated during the process and the classification results through MATLAB.

[0163] The technological advancements achieved by this invention due to the adoption of the above technical solutions are as follows:

[0164] This disclosure provides a method, device, and real-time classification interface for motor intention recognition in brain-computer interfaces. It offers a feature-based statistical optimization-based fNIRS-BCI motor intention recognition scheme to improve the performance of hand motor intention recognition. This addresses the limitations of related technologies, such as low accuracy in complex motor classification tasks using fNIRS in the frontal lobe region and the difficulty in balancing sufficient feature extraction with computational efficiency. The method boasts high reliability, can optimize classification results for multiple individuals, and achieves generalization performance, enabling the training of an overall model applicable to all individuals. It provides comprehensive feature extraction methods and high-reliability and validity feature statistical optimization methods, satisfying the optimization of classifier training results after feature optimization while maintaining high interpretability, supporting a more intuitive understanding of the relationship between statistical parameters and classifier training results. The designed real-time classification interface, utilizing the trained classification model, can display the subject's action execution intention in real time, providing an important tool and methodological reference for practical brain-computer interface applications focused on motor decision recognition.

[0165] First, the fNIRS acquisition device was used to collect brain activation data of the frontal lobe region of the subject according to the paradigm requirements. After preprocessing to identify and filter non-task-related noise, the conversion between light intensity signal and blood oxygen concentration signal was realized using the modified Lambert-Beer law.

[0166] Then, the network parameters, after being fitted and trained by the time feature extraction module, frequency feature extraction module, network feature extraction module, and deep learning methods such as MLP and LSTM networks, are combined into a feature matrix. Each element of the combined feature matrix undergoes normality test, difference significance test, and recurrence rate test. By adjusting the significance levels of the normality test, difference significance test, and feature recurrence frequency, a feature selection mask matrix is ​​constructed. Based on the mask matrix, the retained features and redundant features are selected and removed, and the classifier is trained by the support vector machine model.

[0167] Finally, the trained model is deployed to the real-time classification interface. After configuring the paradigm parameters, the test subject begins to complete the paradigm task according to the prompts presented on the screen. The interface reads the real-time collected data and uses the same preprocessing and feature extraction to output the classification results displayed on the interface.

[0168] Figure 7 A schematic block diagram of an example electronic device 600 that can be used to implement embodiments of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0169] like Figure 7 As shown, device 600 includes a computing unit 601, which can perform various appropriate actions and processes based on a computer program stored in ROM (Read-Only Memory) 602 or a computer program loaded from storage unit 608 into RAM (Random Access Memory) 603. RAM 603 may also store various programs and data required for the operation of device 600. The computing unit 601, ROM 602, and RAM 603 are interconnected via bus 604. I / O (Input / Output) interface 605 is also connected to bus 604.

[0170] Multiple components in device 600 are connected to I / O interface 605, including: input unit 606, such as keyboard, mouse, etc.; output unit 607, such as various types of monitors, speakers, etc.; storage unit 608, such as disk, optical disk, etc.; and communication unit 609, such as network card, modem, wireless transceiver, etc. Communication unit 609 allows device 600 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0171] The computing unit 601 can be various general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 601 include, but are not limited to, CPUs (Central Processing Units), GPUs (Graphics Processing Units), various special-purpose AI (Artificial Intelligence) computing chips, various computing units running machine learning model algorithms, DSPs (Digital Signal Processors), and any suitable processor, controller, microcontroller, etc. The computing unit 601 performs the various methods and processes described above, such as a motion intention recognition method for a brain-computer interface. For example, in some embodiments, the motion intention recognition method for a brain-computer interface can be implemented as a computer software program tangibly contained in a machine-readable medium, such as storage unit 608. In some embodiments, part or all of the computer program can be loaded and / or installed on device 600 via ROM 602 and / or communication unit 609. When the computer program is loaded into RAM 603 and executed by the computing unit 601, one or more steps of the methods described above can be performed. Alternatively, in other embodiments, the computing unit 601 may be configured by any other suitable means (e.g., by means of firmware) to perform the aforementioned motion intention recognition method for brain-computer interfaces.

[0172] Various implementations of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, FPGAs (Field Programmable Gate Arrays), ASICs (Application-Specific Integrated Circuits), ASSPs (Application-Specific Standard Products), SOCs (System-on-Chips), CPLDs (Complex Programmable Logic Devices), computer hardware, firmware, software, and / or combinations thereof. These various implementations may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0173] The program code used to implement the methods of this disclosure may be written in any combination of one or more programming languages. This program code may be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing apparatus, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code may be executed entirely on a machine, partially on a machine, as a standalone software package partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0174] In the context of this disclosure, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, RAM, ROM, EPROM (Electrically Programmable Read-Only Memory) or flash memory, optical fiber, CD-ROM (Compact Disc Read-Only Memory), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0175] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device for displaying information to the user (e.g., a CRT (Cathode-Ray Tube) or LCD (Liquid Crystal Display) monitor); and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0176] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as data servers), or middleware components (e.g., application servers), or frontend components (e.g., user computers with graphical user interfaces or web browsers through which users can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., communication networks). Examples of communication networks include LANs (Local Area Networks), WANs (Wide Area Networks), the Internet, and blockchain networks.

[0177] Computer systems can include clients and servers. Clients and servers are generally geographically separated and typically interact via communication networks. The client-server relationship is created by computer programs running on the respective computers and having a client-server relationship with each other. A server can be a cloud server, also known as a cloud computing server or cloud host, a hosting product within the cloud computing service system that addresses the shortcomings of traditional physical hosts and VPS (Virtual Private Server) services, such as high management difficulty and weak business scalability. Servers can also be servers for distributed systems or servers incorporating blockchain technology.

[0178] It's important to note that artificial intelligence (AI) is the study of enabling computers to simulate certain human thought processes and intelligent behaviors (such as learning, reasoning, thinking, and planning). It encompasses both hardware and software technologies. AI hardware technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, and big data processing. AI software technologies primarily include computer vision, natural language processing, machine learning / deep learning, big data processing, and knowledge graph technologies.

[0179] It should be understood that the various forms of processes shown above can be used to rearrange, add, or delete steps. For example, the steps described in this disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this disclosure can be achieved, and this is not limited herein.

[0180] The specific embodiments described above do not constitute a limitation on the scope of protection of this disclosure. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this disclosure should be included within the scope of protection of this disclosure.

Claims

1. A method for recognizing motor intentions in a brain-computer interface, characterized in that, The method includes: By collecting functional near-infrared spectral signals from the frontal lobe of the user under the paradigm task, raw light intensity data is obtained, the light intensity data is preprocessed, and the preprocessed data is converted into hemodynamic response signals. Multiple features are extracted from the hemodynamic response signal, and the extracted features are combined into a feature matrix. The feature extraction includes time-domain feature extraction, frequency-domain feature extraction, network feature extraction, and curve fitting parameter extraction. Statistical feature optimization is performed on the feature matrix to generate a feature selection mask. Based on the feature selection mask, a preferred feature set is determined. The statistical feature optimization includes dataset stratification, normality test, difference significance test, and feature recurrence rate test. A support vector machine is used as a classifier, and the preferred feature set is input into the classifier as a training set to train the model and obtain a trained model. Based on the trained model, the preferred feature set to be identified is classified to obtain the classification result, and the motion intention corresponding to the classification result is output to the real-time classification interface. The process of generating a feature selection mask by optimizing the statistical features of the feature matrix includes: The feature matrix obtained from the effective trials is used as a sample. The dataset including all samples is stratified according to the set rules to obtain stratified data subsets. Based on the label of the sample and the classification task, each data subset with the same sample is combined to obtain a sample set. Perform a normality test on the sample set to determine whether the channel feature distribution of the sample set conforms to a normal distribution, and generate a first feature selection mask to indicate the normality test result; Perform a difference significance test on the sample set to determine whether the channel feature distribution under different labels in the sample set meets the difference significance, and generate a second feature selection mask to indicate the difference significance result; The sample set is subjected to a feature recurrence rate test, the recurrence rate of channel features is statistically analyzed, and a third feature selection mask is generated based on the recurrence rate and the configured recurrence rate threshold to indicate the recurrence rate test result.

2. The method for motor intention recognition for brain-computer interfaces according to claim 1, characterized in that, in, Before acquiring the functional near-infrared spectral signals of the frontal lobe brain region of the user under the paradigm task, the method further includes: Based on the experimental paradigm design, parameters are set for the paradigm prompts on the real-time classification interface to guide users to perform the paradigm task. The parameter settings include classification task settings, paradigm trial settings, baseline time settings, task time settings, and rest time settings.

3. The method for motor intention recognition for brain-computer interfaces according to claim 2, characterized in that, in, Preprocessing the light intensity data and converting the preprocessed data into hemodynamic response signals includes: Extract the light intensity data before the end of each trial to obtain partial light intensity data; The partial light intensity data is filtered to remove physiological noise, baseline drift, and spike noise, resulting in filtered data. The filtered data is converted into a blood oxygen concentration signal using a modified Lambert-Beer law, wherein the blood oxygen concentration signal includes oxyhemoglobin concentration data and deoxyhemoglobin concentration data.

4. The method for motor intention recognition for brain-computer interfaces according to claim 1, characterized in that, in, The extraction of multiple features from the hemodynamic response signal includes: The hemodynamic response signal is subjected to time-domain feature extraction to obtain the time information of the hemodynamic response signal, wherein the time information includes mean, variance, slope, peak value, peak time, kurtosis, quadratic coefficient and approximate entropy; Frequency domain feature extraction is performed on the hemodynamic response signal to obtain the frequency information of the hemodynamic response signal, wherein the frequency information includes the centroid frequency, mean square frequency, and frequency variance; Network feature extraction is performed on the hemodynamic response signal to obtain the spatial information of the hemodynamic response. The spatial information includes node importance, average path length, clustering coefficient, node clustering coefficient, average degree, and degree distribution. The curve fitting parameters of the hemodynamic response signal are extracted using a deep learning network.

5. The method for motor intention recognition for brain-computer interfaces according to claim 4, characterized in that, in, The deep learning network includes a multilayer perceptron and a long short-term memory network, and the curve fitting parameters for extracting the hemodynamic response signal using the deep learning network include: The response curve of each channel is fitted using both the multilayer perceptron and the long short-term memory network simultaneously for the hemodynamic response signal, and the trained network parameters are used as features to form a feature matrix. The network for fitting the response curve of the multilayer perceptron includes an input layer, a hidden layer with 10 neurons, and an output layer. The network for fitting the response curve of the long short-term memory network includes a sequence input layer, a long short-term memory layer, a modified linear unit activation layer, a fully connected layer, and a regression layer.

6. The method for motor intention recognition for brain-computer interfaces according to claim 1, characterized in that, in, Perform a normality test on the sample set to determine whether the channel feature distribution of the sample set conforms to a normal distribution, and generate a first feature selection mask to indicate the normality test result, including: For each layer of the sample set, the sample set is grouped according to the different labels; The p-value was calculated independently for each sample group using the Jarque-Bera test. If the p-value satisfies the configured first floating threshold, it is determined that the channel feature distribution within the current sample group conforms to a normal distribution, and the first feature selection mask is set to 1, or... If the p value does not meet the configured first floating threshold, it is determined that the channel feature distribution in the current sample group does not conform to the normal distribution, and the first feature selection mask is 0.

7. The method for motor intention recognition for brain-computer interfaces according to claim 1, characterized in that, in, Perform a significance test on the sample set to determine whether the channel feature distribution under different labels in the sample set meets the significance test, and generate a second feature selection mask to indicate the significance test result, including: For each stratified sample set, the sample set is grouped according to the different classification tasks; For each sample group, the p-value was calculated using both the t-test and analysis of variance methods. If the p-value satisfies the configured second floating threshold, it is determined that the channel feature distribution within the current sample group meets the significant difference, and the second feature selection mask is set to 1, or... If the p-value does not meet the configured second floating threshold, it is determined that the channel feature distribution within the current sample group does not meet the significance of difference, and the second feature selection mask is 0.

8. The method for motor intention recognition for brain-computer interfaces according to claim 1, characterized in that, in, The sample set is subjected to a feature recurrence rate test, the recurrence rate of channel features is statistically analyzed, and a third feature selection mask is generated based on the recurrence rate and a configured recurrence rate threshold to indicate the recurrence rate test results, including: The number of channels in all strata that simultaneously conform to a normal distribution and show significant differences is counted, and the recurrence rate is obtained by combining the number of all strata. If the reproducibility rate meets the configured reproducibility rate threshold, the channel feature is determined to have passed the reproducibility rate test, and the third feature is selected with a mask of 1, or... If the reproducibility rate does not meet the configured reproducibility rate threshold, the channel feature is determined to fail the reproducibility rate test, and the third feature is selected with a mask of 0.

9. A motion intention recognition system for brain-computer interfaces, characterized in that, include: The data acquisition module is used to acquire raw light intensity data by acquiring functional near-infrared spectral signals of the frontal lobe brain region of the user under the paradigm task, preprocess the light intensity data, and convert the preprocessed data into hemodynamic response signals. The feature extraction module is used to extract various features from the hemodynamic response signal and combine the extracted features into a feature matrix. The feature extraction includes time-domain feature extraction, frequency-domain feature extraction, network feature extraction, and curve fitting parameter extraction. The feature statistical optimization module is used to perform statistical feature optimization on the feature matrix to generate a feature selection mask, and to determine the preferred feature set through the feature selection mask. The statistical feature optimization includes dataset stratification, normality test, difference significance test, and feature recurrence rate test. The training module is used to employ a support vector machine as a classifier, inputting the preferred feature set as a training set into the classifier to train the model and obtain a trained model. The classification module is used to perform task classification on the preferred feature set to be identified based on the trained model, obtain the classification result, and output the motion intention corresponding to the classification result to the user interface.

Images