Motion Imagery Time Calibration Method and Device Based on Time-Domain Energy

Abstract

The present invention provides a motor imagery time calibration method and device based on time domain energy, comprising: obtaining a first EEG signal of a user performing a motor imagery task; converting the first EEG signal into a feature set through a first preprocessing method and a second preprocessing method; using a multiple-iteration method, finding the optimal feature template parameter set of the first EEG signal from the feature set; determining the time window optimization feature of the second EEG signal through the optimal feature template parameter set, identifying the time window optimization feature in combination with an optimal classification model, obtaining a classification result, and providing real-time feedback to the user. The present invention performs feature conversion on the EEG signal through two preprocessing methods, eliminates useless data, and improves the accuracy of the EEG signal features. In combination with a multiple-iteration method, it finds the optimal feature template parameter set, adaptively selects the optimal time window, optimizes the classification model, and improves the recognition accuracy of the classifier.

Description

Technical Field

[0001] The present invention relates to the field of brain-computer interface signal processing technology, and in particular to a motor imagery time calibration method and device based on time domain energy. Background Art

[0002] A brain-computer interface (BCI) is a human-computer interaction method that directly communicates with a computer or external device through the human brain. The BCI system controls external devices such as robotic arms and wheelchairs to help disabled people regain a certain degree of motor ability, or provides typing services through the BCI system to provide patients with language disorders with a way to communicate with the outside world. Motor imagery is a classic brain-computer interface paradigm. When a person prepares to move a limb, time-related synchronization / desynchronization occurs in the motor cortex of the brain. Based on this principle, the brain-computer interface system can decode the movement intentions of patients with post-stroke sequelae, control external devices to enable their disabled limbs to move, and thus promote neural directional remodeling. When the brain-computer interface is used for limb rehabilitation of stroke patients, the patient's brain recovery status can be obtained by detecting the brain electrical activity signal (EEG) at the human scalp.

[0003] In the application of existing technology, the brain-computer interface system requires the user to start imagining at a certain point in time, and assumes that the user strictly follows the prompts to imagine. However, the start time of the user's imagination is different each time. If the fixed time window motion imagination data processing method in the existing technology is used, the data within the time window cannot accurately restore the motion characteristics, thereby reducing the recognition accuracy of the BCI system. Summary of the Invention

[0004] According to a first aspect of the present invention, a method for motor imagery time calibration based on time domain energy is provided, comprising:

[0005] Step 1: obtaining a first EEG signal of a user performing a motor imagery task;

[0006] Step 2: converting the first EEG signal into a feature set through a first preprocessing method and a second preprocessing method;

[0007] Step 3: using a multiple iteration method, finding the optimal feature template parameter set of the first EEG signal from the feature set;

[0008] Step 4: Determine the time window optimization feature of the second EEG signal using the optimal feature template parameter set, identify the time window optimization feature in combination with the optimal classification model, obtain the classification result and provide real-time feedback to the user.

[0009] Furthermore, in step 2, the first preprocessing method includes:

[0010] Performing power frequency filtering, first bandpass filtering, baseline calibration, intercepting the EEG signal, classifying, and removing noise data on the first EEG signal to form a first feature set;

[0011] The second pretreatment method comprises:

[0012] After the first EEG signal undergoes a first preprocessing, the leads of the left and right motor areas of the brain are taken out, and a second band-pass filtering and weighted sliding average are performed. The time domain energy difference between the left and right motor areas of the brain is calculated to form a second feature set.

[0013] Furthermore, the step 3 specifically includes:

[0014] Step 3.1: The initial time window t1 is 0, the length is w1, the first classification accuracy is 0, the template change rate is 0, and the sub-features are intercepted from the second feature set with the initial time window, and the average is taken as the optimal feature template;

[0015] Step 3.2, traverse the second feature set, and use the first time window optimization method to determine the optimal sub-time window and optimal sub-feature of a single sample in the second feature set to obtain a third feature set;

[0016] Step 3.3, based on the optimal sub-time window of each sample in the third feature set, extract data from the corresponding samples in the first feature set to form the fourth feature set;

[0017] Step 3.4, obtain the second classification accuracy through cross-validation;

[0018] Step 3.5: updating the feature template according to the first classification accuracy, the second classification accuracy, the third feature set, and the template change rate;

[0019] Step 3.6, repeat steps 3.2 to 3.5 until the end condition is met and the optimal feature template parameter set is obtained.

[0020] Furthermore, in step 3.2, the first time window optimization method includes:

[0021] Step 3.21, for a single sample of the second feature set, extract several sub-features by sliding the time window;

[0022] Step 3.22, calculate the similarity between all sub-features and the feature template;

[0023] In step 3.23, the optimal sub-feature and the optimal time window are determined based on the similarity between the sub-feature and the feature template and the similarity threshold.

[0024] Furthermore, the step 3.23 includes:

[0025] If the maximum similarity between all sub-features in a single sample and the feature template is less than the similarity threshold, then the sub-feature corresponding to the initial time window among all sub-features of the sample is regarded as the optimal sub-feature, and the initial time window is regarded as the optimal sub-time window;

[0026] If the maximum similarity between all sub-features in a single sample and the feature template is greater than or equal to the similarity threshold, the sub-feature corresponding to the maximum similarity is taken as the optimal sub-feature, and the time window corresponding to the maximum similarity is taken as the optimal sub-time window.

[0027] Furthermore, in step 3.5, updating the feature template includes:

[0028] If the second classification accuracy is greater than or equal to the first classification accuracy, the optimal feature template is updated to a temporary feature template, and the first classification accuracy is updated to the second classification accuracy;

[0029] If the second classification accuracy is less than the first classification accuracy, the optimal feature template is subtracted by a multiple of the template change rate, where the multiple is R, and the template change rate is updated to the current optimal feature template minus the previous optimal feature template.

[0030] Furthermore, in step 3.6, the termination condition is satisfied, including:

[0031] When the second classification accuracy rate does not exceed the first classification accuracy rate for N1 consecutive rounds, or when the total number of iterations exceeds N2 rounds, the iteration is terminated;

[0032] N1 and N2 are adjusted based on experience, with N1 being smaller than N2.

[0033] Furthermore, the step 4 specifically includes:

[0034] Applying the first preprocessing method to the second EEG signal to obtain a fifth feature set;

[0035] Applying the second preprocessing method to the second EEG signal to obtain a sixth feature set;

[0036] According to the feature template and the similarity threshold, the optimal sub-time window of a single sample in the sixth feature set is found by using the first time window optimization method;

[0037] According to the optimal sub-time window, extracting features from samples corresponding to the fifth feature set as features after time window optimization;

[0038] Utilize the optimal classification model to identify the features after time window optimization, obtain the classification results and provide real-time feedback to the user.

[0039] Furthermore, before performing step 4, the method further includes:

[0040] Using the fourth feature set that achieves the maximum first classification accuracy, training a classifier to obtain the optimal classification model;

[0041] The user completes the same motor imagery task as in the data collection phase according to the prompts;

[0042] After the user completes a single motor imagery task, the system captures the data from 1 second before the prompt appears to 4 seconds after the prompt appears as the second EEG signal.

[0043] According to a second aspect of the present invention, a device for calibrating motor imagery time based on time-domain energy is provided. The device applies the process steps of the above-mentioned method for calibrating motor imagery time based on time-domain energy, including:

[0044] A signal acquisition module is used to obtain the EEG signal of the user performing the motor imagery task;

[0045] Signal calculation module, used to analyze EEG signals, obtain the optimal feature template parameter set, and classify EEG data;

[0046] The process control module is used for data collection in the online stage, outputs prompts to users, and controls the pace of data acquisition and processing;

[0047] The interactive module is used to output prompt information and feedback information to the user.

[0048] Compared with the existing technology, the beneficial effects achieved by the present invention are as follows: the present invention performs feature conversion on EEG signals through two preprocessing methods, eliminates useless data, improves the accuracy of EEG signal features, combines multiple iterations to find the optimal feature template parameter set, and determines the optimal sub-features and optimal time windows through similarity calculation and similarity threshold setting to achieve adaptive selection of the optimal time window, optimize the classification model, improve the recognition accuracy of the classifier, and enhance the accuracy of motion feature restoration of motor imagery data, thereby increasing the recognition accuracy of the brain-computer interface system.

[0049] It should be understood that all combinations of the foregoing concepts and the additional concepts described in more detail below, as long as such concepts are not mutually inconsistent, can be considered part of the inventive subject matter of this disclosure. In addition, all combinations of the claimed subject matter are considered part of the inventive subject matter of this disclosure.

[0050] The foregoing and other aspects, embodiments, and features of the present invention will be more fully understood from the following description in conjunction with the accompanying drawings. Other additional aspects of the present invention, such as features and / or beneficial effects of the exemplary embodiments, will become apparent from the following description or through practice of specific embodiments according to the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] The accompanying drawings are not intended to be drawn to scale. In the accompanying drawings, each identical or nearly identical component shown in various figures may be represented by the same reference numeral. For clarity, not every component is labeled in every figure. Embodiments of various aspects of the present invention will now be described by way of example and with reference to the accompanying drawings.

[0052] Figure 1 1 is a flow chart of a method for calibrating motor imagery time based on time domain energy according to the present invention;

[0053] Figure 2 This is a schematic diagram of the module distribution of the motion imagery time calibration device based on time domain energy shown in the present invention. DETAILED DESCRIPTION

[0054] In order to better understand the technical content of the present invention, specific embodiments are given and described below with reference to the accompanying drawings.

[0055] Various aspects of the present invention are described in this disclosure with reference to the accompanying drawings, in which a number of illustrative embodiments are shown. The embodiments of the present disclosure are not necessarily intended to include all aspects of the present invention. It should be understood that the various concepts and embodiments introduced above, as well as those described in more detail below, can be implemented in any of many ways, because the concepts and embodiments disclosed herein are not limited to any embodiment. In addition, some aspects of the present disclosure may be used alone or in any appropriate combination with other aspects disclosed herein.

[0056] In the application of existing technology, the brain-computer interface system requires the user to start imagining at a certain point in time, and assumes that the user strictly follows the prompts to imagine. However, the start time of the user's imagination is different each time. If the fixed time window motion imagination data processing method in the existing technology is used, the data within the time window cannot accurately restore the motion characteristics, thereby reducing the recognition accuracy of the BCI system.

[0057] To address the above issues, this embodiment proposes a motion imagery time calibration method based on time domain energy. According to the user's offline data, the optimal time window is adaptively selected to optimize the classification model, realize the calibration of motion imagery time, and improve the system recognition accuracy.

[0058] According to an embodiment of the present invention, Figure 1 The flowchart shown in FIG. 1 shows a method for time calibration of motor imagery based on time domain energy, comprising the following steps:

[0059] Step 1: obtaining a first EEG signal of a user performing a motor imagery task;

[0060] Step 2: The first EEG signal is converted into a feature set through a first preprocessing method and a second preprocessing method;

[0061] Step 3: Using a multiple iteration method, find the optimal feature template parameter set of the first EEG signal from the feature set;

[0062] Step 4: Determine the time window optimization feature of the second EEG signal through the optimal feature template parameter set, identify the time window optimization feature in combination with the optimal classification model, obtain the classification result and provide real-time feedback to the user.

[0063] It should be noted that the calibration method provided in this embodiment includes two parts: offline training and online application. The offline training part is used to obtain feature thresholds, feature templates and corresponding classification models corresponding to different motor imagery tasks. The online application part performs real-time calculations based on the feature thresholds, feature templates and classification models obtained in the offline training part to obtain classification results and provide real-time feedback to the user.

[0064] Specifically, this embodiment combines some preferred or optional examples to further describe in detail the implementation and / or effects of certain examples of the present invention.

[0065] 1. Offline Training

[0066] [Acquiring EEG signals]

[0067] Step 1.1: Detect the user's scalp brain electrical activity signal (EEG, the acquisition area must cover the brain motor area), obtain the user's brain signal when performing the specified motor imagery task, and use it as the first EEG signal.

[0068] As a preferred solution, the electrode layout is based on the international 10-20 standard, the acquisition leads must include C3 and C4, the sampling rate is greater than 200Hz, and the motor imagery task is to imagine the left hand and the right hand clenched. After receiving the prompt, the user begins to imagine the corresponding hand clenching. During data collection, for each motor imagery task, the user needs to perform it N times.

[0069]

Preprocessing, conversion feature set

[0070] Step 2.1, performing a first preprocessing on the first EEG signal to obtain a first feature set for classification;

[0071] The first pre-processing method includes:

[0072] The first EEG signal is subjected to power frequency filtering, first bandpass filtering, baseline calibration, EEG signal interception, classification, and noise data removal to form a first feature set.

[0073] As a preferred solution, the first pretreatment can be performed as follows:

[0074] For each motor imagery task in the first EEG signal, the EEG data of leads C3 and C4 were taken and subjected to 50 Hz notch filtering, 5-40 Hz band-pass filtering, and sliding filtering. The sliding filtering used a Savitzky-Golay filter with an order of 1 and a window length of 1 second.

[0075] 5 seconds of data were collected from 1 second before to 4 seconds after each motor imagery prompt and marked according to the motor imagery task.

[0076] The average value of the data before each motor imagery prompt was taken for baseline correction;

[0077] The C3 data baseline is subtracted from the C4 data baseline, and the two sets of data are merged to obtain the first feature set.

[0078] Step 2.2, performing a second preprocessing on the first EEG signal to obtain a second feature set for calculating the time window;

[0079] The second pre-processing method includes:

[0080] After the first EEG signal undergoes the first preprocessing, the leads of the left and right motor areas of the brain are taken out, and a second band-pass filtering and weighted sliding average are performed. The time domain energy difference between the left and right motor areas of the brain is calculated to form a second feature set.

[0081] As a preferred solution, the second pretreatment can be performed as follows:

[0082] For each sample of the same motor imagery task in the first feature set, the EEG data of the left cerebral motor area-related lead C3 and the right cerebral motor area-related lead C4 were taken;

[0083] Perform 50Hz notch filtering and 8-16Hz bandpass filtering;

[0084] Sliding filter, the sliding filter adopts Savitzky-Golay filter, the order is 1, and the window length is 1 second;

[0085] 5 seconds of data were collected from 1 second before to 4 seconds after each motor imagery prompt and marked according to the motor imagery task.

[0086] The average value of the data before each motor imagery prompt was taken for baseline correction;

[0087] The C3 data baseline is subtracted from the C4 data baseline, and the two sets of data are merged to obtain the second feature set.

[0088] Preferably, the first preprocessing method and the second preprocessing method provided in this embodiment perform feature set conversion on the collected first EEG signal for classification and time window calculation, thereby providing accurate and reliable data support for subsequent comparison of similarity thresholds and acquisition of feature templates.

[0089]

Get the optimal feature template parameter set

[0090] Step 3.1: The initial time window t1 is 0, the length is w1, the first classification accuracy is 0, the template change rate is 0, and the sub-features are intercepted from the second feature set with the initial time window, and the average is taken as the optimal feature template;

[0091] Step 3.2, traverse the second feature set, use the first time window optimization method to determine the optimal sub-time window and optimal sub-feature of a single sample in the second feature set, and obtain the third feature set;

[0092] Step 3.3, based on the optimal sub-time window of each sample in the third feature set, extract data from the corresponding samples in the first feature set to form the fourth feature set;

[0093] Step 3.4, obtain the second classification accuracy through cross-validation;

[0094] Step 3.5, updating the feature template according to the first classification accuracy, the second classification accuracy, the third feature set, and the template change rate;

[0095] Step 3.6, repeat steps 3.2 to 3.5 until the end condition is met and the optimal feature template parameter set is obtained.

[0096] It should be noted that the second feature set contains different groups, and the samples of each group correspond to a motor imagination task. The data of each sample from the t1 position with a length of w1 are taken, and the average within the group is taken as the optimal feature template for the group of data.

[0097] As a preferred solution, t1 is 0 seconds and w1 is 4 seconds.

[0098] It should be noted that the number of samples in the third feature set is consistent with the number of samples in the second feature set, and the third feature set contains the optimal sub-time window and optimal sub-feature data of each sample in the second feature set.

[0099] It should be noted that there is a one-to-one relationship between the samples of the first feature set, the second feature set, the third feature set, and the fourth feature set. According to the start time of the sub-time window of a single sample in the third feature set, the data of the corresponding sample of the first feature set is intercepted as the fourth feature set.

[0100] In step 3.2, the first time window optimization method includes:

[0101] Step 3.21, for a single sample of the second feature set, extract several sub-features through a sliding time window;

[0102] Step 3.22, calculate the similarity between all sub-features and the feature template;

[0103] In step 3.23, the optimal sub-feature and the optimal time window are determined based on the similarity between the sub-feature and the feature template and the similarity threshold.

[0104] As a preferred solution, the sliding time window starts from the first data point, with an interval of 100 milliseconds and a time window length of w1, and the sample is divided into several sub-features.

[0105] As a preferred solution, the similarity between the sub-feature and the current template is calculated using the Pearson correlation coefficient, and the result is mapped to the interval [0,1]. The relevant mathematical formula is as follows:

[0106]

[0107] Among them, X is a single sub-feature, Y is the current template, and the lengths of the two are the same. The value range of P is [0,1]. The larger the value, the higher the similarity between the sub-feature and the current template.

[0108] Further, step 3.23 includes:

[0109] If the maximum similarity between all sub-features in a single sample and the feature template is less than the similarity threshold, then the sub-feature corresponding to the initial time window among all sub-features of the sample is regarded as the optimal sub-feature, and the initial time window is regarded as the optimal sub-time window;

[0110] If the maximum similarity between all sub-features in a single sample and the feature template is greater than or equal to the similarity threshold, the sub-feature corresponding to the maximum similarity is taken as the optimal sub-feature, and the time window corresponding to the maximum similarity is taken as the optimal sub-time window.

[0111] In step 3.4, the second classification accuracy is obtained through cross-validation, including:

[0112] Step 3.41: Divide the samples of the fourth feature set into 10 parts, of which the first part is the test set and the remaining 9 parts are training sets;

[0113] Step 3.42: After the training data is subjected to CSP feature extraction, the SVM classifier is trained to obtain the CSP and SVM classification model (motor imagery EEG-SVM binary classification);

[0114] Step 3.43, use the test set obtained in step 3.41 to perform classification and save the classification results;

[0115] Step 3.44, change the test set, repeat steps 3.51 to 3.52 until all data are used as test sets, and only one is used as the test set;

[0116] In step 3.45, the classification results of all samples are counted to obtain the average classification accuracy, which is used as the second classification accuracy.

[0117] It should be noted that in step 3.43, after classification, it is possible to determine which motor imagery task the sample belongs to, and whether it is correct can be determined based on the data label.

[0118] As a preferred solution, the fourth feature set is cross-validated to obtain the second classification accuracy. The larger the second classification accuracy value is, the better the quality of the features contained in the fourth feature set is, and the corresponding feature template parameter set is more suitable.

[0119] In step 3.5, update the feature template, including:

[0120] Step 3.51, select the sub-features of all samples from the third feature set, calculate the average, and obtain a temporary feature template, where the length of the temporary feature template is w1;

[0121] Step 3.52: If the second classification accuracy is greater than or equal to the first classification accuracy, the optimal feature template is updated to the temporary feature template, and the first classification accuracy is updated to the second classification accuracy;

[0122] In step 3.53, if the second classification accuracy is less than the first classification accuracy, the optimal feature template is subtracted by a multiple of the template change rate, which is R. The template change rate is updated to the current optimal feature template minus the previous optimal feature template.

[0123] In step 3.6, the termination conditions are met, including:

[0124] When the second classification accuracy does not exceed the first classification accuracy for N1 consecutive rounds, or when the total number of iterations exceeds N2 rounds, the iteration terminates;

[0125] N1 and N2 are adjusted based on experience, with N1 being smaller than N2.

[0126] As a preferred solution, N1 is 10 and N2 is 50, taking into account both iteration time and optimization effect.

[0127] As a preferred solution, the optimal template parameter set includes the optimal feature template, the optimal similarity threshold, the optimal multiple R, and the optimal initial time window. The similarity threshold, the multiple R, the starting point t1 and the length w1 of the initial time window are traversed within a certain range, and the iterative process of step 3 is repeated. The first classification accuracy at the end of each time is recorded to find the optimal parameters.

[0128] It should be noted that this embodiment uses the optimal feature template corresponding to the maximum first classification accuracy as the optimal feature template; uses the similarity threshold corresponding to the maximum first classification accuracy as the optimal similarity threshold; uses the multiple R corresponding to the maximum first classification accuracy as the optimal multiple R; and uses the initial time window corresponding to the maximum first classification accuracy as the optimal initial time window.

[0129] Preferably, this embodiment finds the optimal feature template parameter set through multiple iterations, and determines the optimal sub-features and optimal time windows through similarity calculation and similarity threshold setting, so as to realize adaptive selection of the optimal time window, optimize the classification model, improve the recognition accuracy of the classifier, and enhance the accuracy of motion feature restoration of motor imagery data, thereby increasing the recognition accuracy of the brain-computer interface system.

[0130] 2. Online Application

[0131] Determine the time window optimization characteristics of new EEG signals

[0132] Step 4.1: Using the fourth feature set that achieves the maximum first classification accuracy, train the classifier to obtain the optimal classification model (i.e., the classification model obtained in step 3.4 above);

[0133] Step 4.2: The user completes the same motor imagery task as in the data collection phase according to the prompts;

[0134] Step 4.3: After the user completes the single motor imagery task, the system captures the data from 1 second before the prompt to 4 seconds after the prompt as the second EEG signal;

[0135] Step 4.4, applying the first preprocessing method to the second EEG signal to obtain a fifth feature set;

[0136] Step 4.5, applying a second preprocessing method to the second EEG signal to obtain a sixth feature set;

[0137] Step 4.6: Based on the feature template and similarity threshold, the optimal sub-time window of a single sample in the sixth feature set is found using the first time window optimization method;

[0138] Step 4.7: According to the optimal sub-time window, extract features from the corresponding samples of the fifth feature set as the features after time window optimization;

[0139] Step 4.8: Use the optimal classification model to identify the features after time window optimization and obtain the classification results;

[0140] Step 4.9: Feedback the classification results to the user in real time;

[0141] Step 4.10: Repeat steps 4.2 to 4.9 until the online application phase is completed.

[0142] As a preferred solution, the feedback can be implemented through sound and / or image.

[0143] Preferably, compared with the features extracted by the traditional fixed time window method, the method provided in this embodiment can optimize the features according to the time window extracted by the optimal template parameter set, improve the accuracy of the classifier, and thus improve the recognition accuracy of the brain-computer interface system.

[0144] The CSP feature extraction, SVM trainer classification and feature recognition algorithm of the aforementioned EEG signal can be performed using methods and means in the existing technology and will not be described in detail in this example.

[0145] Based on the above embodiments, other aspects disclosed in the embodiments of the present invention further provide a motion imagery time calibration device based on time domain energy, comprising:

[0146] A signal acquisition module is used to obtain the EEG signal of the user performing the motor imagery task;

[0147] Signal calculation module, used to analyze EEG signals, obtain the optimal feature template parameter set, and classify EEG data;

[0148] The process control module is used for data collection in the online stage, outputs prompts to users, and controls the pace of data acquisition and processing;

[0149] The interactive module is used to output prompt information and feedback information to the user.

[0150] Reference Figure 2 , the signal calculation module includes:

[0151] A signal preprocessing unit, used to convert EEG signals into corresponding feature sets;

[0152] A time window optimization unit, configured to determine an optimal sub-time window from a feature set according to a feature template parameter set;

[0153] Iterative optimization unit, used to find the optimal feature template parameter set that best suits the EEG data from the iteration;

[0154] Classification training unit, used to train the classifier;

[0155] The classification and recognition unit is used to identify the specific classification of the signal based on the trained classifier and feature data;

[0156] The time window segmentation unit is used to extract features from a single sample according to the time window;

[0157] The similarity calculation unit is used to process the features provided by the time window segmentation unit and calculate the similarity between two different data segments.

[0158] It should be noted that the apparatus provided in this embodiment further includes a memory for storing operable instructions. When these instructions are executed by one or more processors, the one or more processors are caused to perform operations. These operations include the process of the motion imagery time calibration method based on time domain energy of the aforementioned embodiment, especially Figure 1 The process of the method shown.

[0159] While the present invention has been disclosed above with reference to preferred embodiments, this is not intended to limit the present invention. Persons skilled in the art will readily appreciate that various modifications and variations can be made without departing from the spirit and scope of the present invention. Therefore, the scope of protection of the present invention shall be determined by the claims.

Claims

1. A method for motor imagery time calibration based on time domain energy, characterized in that: include: Step 1: obtaining a first EEG signal of a user performing a motor imagery task; Step 2: converting the first EEG signal into a feature set through a first preprocessing method and a second preprocessing method; Step 3: Using a multiple iteration method, find the optimal feature template parameter set of the first EEG signal from the feature set: Step 3.1: The initial time window t1 is 0, the length is w1, the first classification accuracy is 0, the template change rate is 0, and the sub-features are intercepted from the second feature set with the initial time window, and the average is taken as the optimal feature template; Step 3.2, traverse the second feature set, and use the first time window optimization method to determine the optimal sub-time window and optimal sub-feature of a single sample in the second feature set to obtain a third feature set; Step 3.3, based on the optimal sub-time window of each sample in the third feature set, extract data from the corresponding samples in the first feature set to form the fourth feature set; Step 3.4, obtain the second classification accuracy through cross-validation, which includes: Step 3.41, divide the samples of the fourth feature set into a test set and a training set; Step 3.42: After the training data is processed by CSP to extract features, the SVM classifier is trained to obtain the CSP and SVM classification models. Step 3.5, updating the feature template according to the first classification accuracy, the second classification accuracy, the third feature set, and the template change rate, including: If the second classification accuracy is greater than or equal to the first classification accuracy, the first classification accuracy is updated to the second classification accuracy; Step 3.6: Repeat steps 3.2 to 3.5 until the end condition is met and the optimal feature template parameter set is obtained; and the first classification accuracy at the end of each iteration is recorded; Step 4: Determine the time window optimization features of the second EEG signal through the optimal feature template parameter set, identify the time window optimization features in combination with the optimal classification model, obtain the classification results and provide real-time feedback to the user; use the fourth feature set that achieves the maximum first classification accuracy to train the classifier according to step 3.42 to obtain the optimal classification model; after the user completes a single motor imagery task, the system intercepts the data from 1 second before the prompt to 4 seconds after the prompt appears as the second EEG signal.

2. The method for motor imagery time calibration based on time domain energy according to claim 1, characterized in that: In step 2, the first pretreatment method includes: Performing power frequency filtering, first bandpass filtering, baseline calibration, intercepting the EEG signal, classifying, and removing noise data on the first EEG signal to form a first feature set; The second pretreatment method comprises: After the first EEG signal undergoes a first preprocessing, the leads of the left and right motor areas of the brain are taken out, and a second band-pass filtering and weighted sliding average are performed. The time domain energy difference between the left and right motor areas of the brain is calculated to form a second feature set.

3. The method for motor imagery time calibration based on time domain energy according to claim 1, characterized in that: In step 3.2, the first time window optimization method includes: Step 3.21, for a single sample of the second feature set, extract several sub-features by sliding the time window; Step 3.22, calculate the similarity between all sub-features and the feature template; In step 3.23, the optimal sub-feature and the optimal time window are determined based on the similarity between the sub-feature and the feature template and the similarity threshold.

4. The method for motor imagery time calibration based on time domain energy according to claim 3, characterized in that: The step 3.23 includes: If the maximum similarity between all sub-features in a single sample and the feature template is less than the similarity threshold, then the sub-feature corresponding to the initial time window among all sub-features of the sample is regarded as the optimal sub-feature, and the initial time window is regarded as the optimal sub-time window; If the maximum similarity between all sub-features in a single sample and the feature template is greater than or equal to the similarity threshold, the sub-feature corresponding to the maximum similarity is taken as the optimal sub-feature, and the time window corresponding to the maximum similarity is taken as the optimal sub-time window.

5. The method for motor imagery time calibration based on time domain energy according to claim 1, characterized in that: In step 3.5, updating the feature template further includes: If the second classification accuracy is greater than or equal to the first classification accuracy, updating the optimal feature template to a temporary feature template; If the second classification accuracy is less than the first classification accuracy, the optimal feature template is subtracted by a multiple of the template change rate, where the multiple is R, and the template change rate is updated to the current optimal feature template minus the previous optimal feature template.

6. The method for motor imagery time calibration based on time domain energy according to claim 1, characterized in that: In step 3.6, the termination condition is satisfied, including: When the second classification accuracy rate does not exceed the first classification accuracy rate for N1 consecutive rounds, or when the total number of iterations exceeds N2 rounds, the iteration is terminated; N1 and N2 are adjusted based on experience, with N1 being smaller than N2; In addition, in step 3.6, the optimal template parameter set includes the optimal feature template and the optimal similarity threshold. The optimal similarity threshold refers to repeating the iterative process of step 3, recording the first classification accuracy at the end of each time, and taking the similarity threshold corresponding to the maximum first classification accuracy as the optimal similarity threshold.

7. The method for motor imagery time calibration based on time domain energy according to any one of claims 1 to 6, characterized in that: The calibration method includes two parts: offline training and online application. The online application part performs real-time calculation based on the feature threshold, feature template, and classification model obtained in the offline training part. Steps 1-3 correspond to the offline training part, and step 4 corresponds to the online application part. Specifically, it also includes: Applying the first preprocessing method to the second EEG signal to obtain a fifth feature set; Applying the second preprocessing method to the second EEG signal to obtain a sixth feature set; According to the feature template and the similarity threshold, the optimal sub-time window of a single sample in the sixth feature set is found by using the first time window optimization method; According to the optimal sub-time window, extracting features from samples corresponding to the fifth feature set as features after time window optimization; Utilize the optimal classification model to identify the features after time window optimization, obtain the classification results and provide real-time feedback to the user.

8. A motor imagery time calibration device based on time domain energy, said device applying the process steps of the motor imagery time calibration method based on time domain energy according to any one of claims 1 to 7, characterized in that: include: A signal acquisition module is used to obtain the EEG signal of the user performing the motor imagery task; The signal calculation module is used to analyze EEG signals, obtain the optimal feature template parameter set, and classify EEG data, including: A signal preprocessing unit, used to convert EEG signals into corresponding feature sets; A time window optimization unit, configured to determine an optimal sub-time window from a feature set according to a feature template parameter set; Iterative optimization unit, used to find the optimal feature template parameter set that best suits the EEG data from the iteration; Classification training unit, used to train the classifier; The classification and recognition unit is used to identify the specific classification of the signal based on the trained classifier and feature data; The time window segmentation unit is used to extract features from a single sample according to the time window; The similarity calculation unit is used to process the features provided by the time window segmentation unit and calculate the similarity between two different data segments; The process control module is used for data collection in the online stage, outputs prompts to users, and controls the pace of data acquisition and processing; The interactive module is used to output prompt information and feedback information to the user.

Images