Lower limb movement brain-computer interface signal acquisition method based on target guidance

Abstract

The invention relates to the technical field of brain-computer interfaces, and discloses a lower limb movement brain-computer interface signal acquisition method based on target guidance. According to the method, when a target-oriented lower limb motor imagery task is executed, multi-channel electroencephalogram signals and task event marks are recorded synchronously, sensory motor rhythm frequency band signals are extracted through band-pass filtering, and electro-oculogram and myoelectricity artifacts are removed through independent component analysis. And training a spatial filter by using a common spatial pattern algorithm, and carrying out dimension reduction processing on the signal to obtain an optimal source signal. And by calculating the power value of the optimal source signal in the sensory movement rhythm frequency band, constructing an electroencephalogram feature vector representing the lower limb movement intention. According to the method, the signal purity and the recognition accuracy are improved, and the anti-interference capability of a brain-computer interface system is enhanced.

Description

Technical Field

[0001] This invention relates to the field of brain-computer interface technology and discloses a goal-oriented method for acquiring lower limb movement brain-computer interface signals. Background Technology

[0002] Traditional brain-computer interfaces often suffer from signal quality degradation and reduced recognition accuracy in lower limb motor intention recognition due to artifacts from electrooculography (EOG) and electromyography (EMG).

[0003] Existing methods often rely on single feature extraction, neglecting the impact of task orientation on neural activity patterns and making it difficult to capture dynamic movement intentions. Furthermore, insufficient signal stability in low signal-to-noise ratio environments limits the effectiveness of brain-computer interface systems in real-world scenarios.

[0004] Although the co-space pattern algorithm has been used for signal dimensionality reduction, traditional training methods have not fully considered the differentiated characteristics of the two types of tasks (such as walking and imagining stepping), resulting in limited source signal discrimination and failing to meet the requirements of high-precision motion decoding. Summary of the Invention

[0005] To address the aforementioned technical shortcomings, the purpose of this invention is to provide a goal-oriented method for acquiring lower limb motor brain-computer interface signals, thereby solving the problems of inaccurate acquisition and weak anti-interference ability of existing lower limb motor EEG signals.

[0006] To solve the above-mentioned technical problems, the present invention adopts the following technical solution: In a first aspect, the present invention provides a goal-oriented method for acquiring lower limb motor brain-computer interface signals, comprising: When performing a goal-oriented lower limb motor imagery task, the EEG signals acquired by the multi-channel EEG acquisition system and the task event markers generated by the goal-oriented task system are recorded simultaneously. The lower limb motor imagery task is walking imagery or stepping imagery. The task event markers include at least the task start, task type and feedback information trigger point. The EEG signal was bandpass filtered to extract the sensorimotor rhythm frequency band signal, and then independent component analysis was used to remove artifacts from electrooculography and electromyography to obtain the processed EEG signal. Based on the task start point in the task event marker, the EEG signal after a preset duration is extracted as the analysis data segment. The preset duration is set based on the execution time window of the goal-oriented task. The analyzed data segment is input into a spatial filter to perform signal dimensionality reduction and obtain the optimal source signal. The spatial filter is trained using a co-spatial pattern algorithm based on historical lower limb motor imagery task EEG data. The power value of the optimal source signal in the sensorimotor rhythm frequency band is calculated and used as the EEG feature vector representing the intention to move the lower limbs.

[0007] Preferably, in one possible implementation of the first aspect, bandpass filtering is performed on the electroencephalogram (EEG) signal to extract the sensorimotor rhythm frequency band signal, including: A zero-phase digital filter was used to filter the EEG signal, and the sensorimotor rhythm frequency band was set to a wide band of 8Hz to 30Hz. The filtered signal contains mu rhythm and beta rhythm components, which are used for artifact removal processing.

[0008] Preferably, in one possible implementation of the first aspect, the removal of electrooculography and electromyography artifacts using independent component analysis includes: Independent component analysis was performed on the sensorimotor rhythm frequency band signal after bandpass filtering, decomposing it into multiple statistically independent components; Identification is performed based on the matching degree between the spatiotemporal characteristics of each component and typical artifact patterns. The spatiotemporal characteristics include the brain topography distribution of the component and the kurtosis and skewness of the time series. After setting the components related to electrooculography (EOG) or electromyography (EMG) artifacts to zero, the EEG signal after artifact removal is reconstructed through inverse transformation.

[0009] Preferably, in one possible implementation of the first aspect, the identification based on the matching degree between the spatiotemporal characteristics of each component and typical artifact patterns includes: The correlation between the brain topography maps of each component and the weighted templates of the prefrontal cortex and periocular region is calculated. If the correlation coefficient exceeds the first threshold, it is determined to be an electrooculography artifact component. The kurtosis value of each component's time series is compared with a second threshold. If the kurtosis value exceeds the second threshold, it is determined to be an electromyographic artifact component.

[0010] Preferably, in one possible implementation of the first aspect, the spatial filter is trained using a co-spatial pattern algorithm based on historical lower limb motor imagery task EEG data, and includes: Multiple sets of EEG data segments belonging to the two categories of walking imagination and stepping imagination were extracted from historical lower limb motor imagination task EEG data. The covariance matrix of the EEG data segments for the two types of tasks was calculated separately, and the covariance matrix was normalized. Solve the generalized eigenvalue problem of the normalized mixture covariance matrix by arranging the eigenvectors to form a spatial projection matrix.

[0011] Preferably, in one possible implementation of the first aspect, the normalization process includes: The covariance matrix is ​​corrected using a regularization parameter calculated based on the trace of the covariance matrix, thereby improving the numerical stability of the spatial filter under low signal-to-noise ratio conditions.

[0012] Preferably, in one possible implementation of the first aspect, the process of obtaining the optimal source signal includes: The analyzed data segment is projected onto the spatial projection matrix obtained by the common space mode algorithm to obtain a set of source signals; The source signals are sorted according to the magnitude of the feature values ​​associated with the differences between the two types of tasks; The preset quantity source signals, which are ranked first and last, are selected together as the optimal source signal.

[0013] Preferably, in one possible implementation of the first aspect, the selection of the preset quantity source signals that are first and last in the sorting includes: Select the top m source signals with the largest eigenvalues ​​and the bottom n source signals with the smallest eigenvalues, where m and n are preset positive integers, and the sum of m and n accounts for a preset proportion of the number of original EEG signal channels.

[0014] Preferably, in one possible implementation of the first aspect, the power value calculation includes: For each optimal source signal, a sliding time window is used to calculate its signal variance as a power estimate. The calculated power estimates of each optimal source signal are combined sequentially into a multidimensional vector; This multidimensional vector is used as an EEG feature vector representing the intention to move the lower limbs.

[0015] Preferably, in one possible implementation of the first aspect, the step of calculating the signal variance as a power estimate using a sliding time window includes: The length of the sliding time window is matched with the preset duration of the analyzed data segment; After removing the mean from the signal data within each time window, calculate its variance, and take the average of the variance values ​​within all windows as the final power estimate of the optimal source signal.

[0016] The beneficial effects of this invention are as follows: by synchronously recording task event markers and multi-channel EEG signals, this invention enhances the time alignment accuracy of data, and by combining bandpass filtering and independent component analysis, it effectively removes artifacts from electrooculography and electromyography, thereby improving signal purity.

[0017] A co-space mode algorithm is used to train the spatial filter, and historical data is used to optimize the extraction of source signals, thereby enhancing the expressive power of the differences between the two types of tasks.

[0018] By calculating the optimal source signal power value through a sliding time window, a multi-dimensional feature vector is constructed to achieve efficient representation of lower limb movement intention.

[0019] This method maintains numerical stability even under low signal-to-noise ratio conditions, improves the recognition accuracy and anti-interference ability of brain-computer interface systems, and provides reliable technical support for rehabilitation engineering and human-computer interaction. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 This application provides a flowchart of a goal-oriented brain-computer interface signal acquisition method for lower limb movement. Detailed Implementation

[0022] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] Example 1: As Figure 1 As shown, this invention provides a goal-oriented method for acquiring lower limb motor brain-computer interface signals, comprising: When performing a goal-oriented lower limb motor imagery task, EEG signals acquired by a multi-channel EEG acquisition system and task event markers generated by a goal-oriented task system are recorded simultaneously. The lower limb motor imagery task is either walking imagery or stepping imagery. The task event markers include at least the task start, task type, and feedback information trigger point.

[0024] In this embodiment, the goal-oriented lower limb motor imagery task is implemented through an experimental platform consisting of a multi-channel EEG acquisition system and a goal-oriented task system. Subjects are placed in an electromagnetically shielded room, maintaining a sitting or standing posture, with their eyes focused on virtual environmental feedback displayed on a screen in front of them. The lower limb motor imagery task is defined as subjects actively imagining walking or stepping based on visual or auditory cues. Walking imagery requires subjects to simulate the intention of alternating steps with both legs, while stepping imagery focuses on the mental simulation of raising a leg in place.

[0025] The multi-channel EEG acquisition system uses a 64-lead electrode cap extended from the international 10-20 system. The reference electrodes are connected to the bilateral mastoid processes, and the ground electrode is placed on the forehead. The EEG signals are amplified by a differential amplifier and digitized using a 24-bit analog-to-digital converter to capture neural oscillatory activity.

[0026] The goal-oriented task system generates task event markers and synchronizes them with the EEG acquisition system via a parallel port. Task event markers include the task start point, task type identifier, and feedback trigger point. The task start point is triggered by a signal sent by the system at each start, marking the initiation time of the visualization task. The task type identifier distinguishes between walking visualization and stepping visualization, each corresponding to a different numerical code. The feedback trigger point is generated within a preset time window after task execution to trigger visual feedback in the virtual environment, such as virtual character movement or progress bar updates, to enhance the subject's sense of participation and goal orientation.

[0027] The synchronous recording mechanism is implemented through hardware timestamps. The EEG acquisition system and the task system use the same clock source to ensure that the event markers are time-aligned with the EEG signals. The event markers are embedded in the EEG data stream in the form of digital pulses, facilitating precise segmentation during subsequent offline analysis. In addition, the system records behavioral data during task execution, such as the subject's response time and accuracy, to help verify the validity of the EEG signals.

[0028] Bandpass filtering was performed on the EEG signal to extract the sensorimotor rhythm frequency band signal. Then, independent component analysis was used to remove artifacts from electrooculography and electromyography to obtain the processed EEG signal.

[0029] In this embodiment, bandpass filtering is implemented using a zero-phase digital filter. The filter is designed in bandpass mode, with a passband frequency range of 8Hz to 30Hz. This band covers key neural oscillatory activities in the sensorimotor cortex, including mu rhythms and beta rhythms. Mu rhythms occur in the range of 8Hz to 13Hz and are associated with motor inhibition states; beta rhythms are distributed in the range of 13Hz to 30Hz, reflecting neural synchronization activities after motor execution. During the filtering process, the original EEG signal is first forward-filtered, followed by inverse filtering to eliminate group delay, ensuring that the signal waveform is not distorted. The filtered signal retains the complete time-frequency characteristics of the sensorimotor rhythm.

[0030] Next, independent component analysis (ICA) was performed on the bandpass-filtered sensorimotor rhythm frequency band signal to separate statistically independent source components. ICA decomposed the multichannel EEG signal into several independent components by maximizing the non-Gaussianity criterion. Each component represents a potential source signal, including neural activity and various artifacts. The decomposition process was implemented using a fixed-point algorithm, which converged to a stable solution through iterative optimization. After decomposition, a set of independent components was obtained, each corresponding to a spatiotemporal pattern, including its brain topography distribution and time-series characteristics.

[0031] To identify EEG and EMG artifacts, the spatiotemporal characteristics of each component and their matching degree with typical artifact patterns were analyzed. Spatiotemporal characteristics included the brain topography distribution of the component and the kurtosis and skewness of the time series. For EEG artifact identification, the correlation between the brain topography of each component and weighted templates of the prefrontal cortex and periocular regions was calculated. The weighted templates were constructed based on prior knowledge, emphasizing high-weight regions of prefrontal electrodes (e.g., FP1, FP2, Fpz) and periocular electrodes (e.g., F7, F8). The correlation coefficient was calculated; if the correlation coefficient exceeded a first threshold, the component was determined to be an EEG artifact component. The first threshold was determined through historical data verification to balance recognition sensitivity and specificity; in this embodiment, it was set to 0.7.

[0032] For EMG artifact identification, the kurtosis values ​​of each component's time series are analyzed. Kurtosis characterizes the peaks in the signal distribution; EMG artifacts manifest as high-frequency bursts of activity, resulting in high kurtosis values ​​in the time series. The kurtosis of each component's time series is calculated and compared with a second threshold. If the kurtosis value exceeds the second threshold, it is determined to be an EMG artifact component. The second threshold is set based on the statistical distribution of a large amount of normal EEG data to ensure the capture of abnormal high-frequency activities; in this embodiment, it is set to 5. In addition, the skewness of the time series is also used as an auxiliary criterion; a skewness value deviating significantly from zero indicates an asymmetrical signal distribution, which can further confirm the nature of the artifact.

[0033] After identifying components related to electrooculography (EOG) or electromyography (EMG) artifacts, the signal values ​​of these components are zeroed out, effectively removing artifact contributions from the mixed signal. Subsequently, the EEG signal is reconstructed using inverse transformation of independent component analysis. The inverse transformation multiplies the zeroed-out component matrix with the mixed matrix to obtain the artifact-free multichannel EEG signal. This reconstructed signal preserves the neural activity components of sensorimotor rhythms while reducing interference from eye movements and muscle activity, thus improving signal quality. Finally, the processed EEG signal is obtained.

[0034] Based on the task start point in the task event marker, the EEG signal after the start point is extracted as the analysis data segment for a preset duration. The preset duration is based on the execution time window setting of the goal-oriented task.

[0035] In this embodiment, the task start point in the task event marker serves as the reference time point for EEG signal interception. This start point is marked by the goal-oriented task system at each start via a hardware synchronization mechanism, aligned with the timestamp of the multi-channel EEG acquisition system. The interception operation uses the start point as the zero point in time, extending a preset duration of continuous EEG signals as the analysis data segment. The preset duration is set based on the execution time window of the goal-oriented task, which defines the complete task execution cycle from the start of the task to the feedback information trigger point. This time window is determined through behavioral analysis of experimental data and covers the key neural response stages of the lower limb motor imagery task.

[0036] The length of the execution time window is set according to the task type, and this value is derived from the statistical results of pre-experiments. In this embodiment, a uniform 2000ms is used to capture the typical activation patterns of the sensorimotor cortex during walking and stepping imagery. During the extraction process, digital signal processing algorithms are used to extract analysis data segments from the EEG signals recorded throughout the entire time period. Each data segment corresponds to an independent task. Before extracting the data segments, the temporal synchronization between the event markers and the EEG signals is verified to ensure no temporal deviation. The length of the analysis data segments is fixed to avoid variability introduced by individual differences in task execution duration.

[0037] The analyzed data segments are input into a spatial filter to perform signal dimensionality reduction and obtain the optimal source signal. The spatial filter is trained using a co-spatial pattern algorithm based on historical lower limb motor imagery task EEG data.

[0038] In this embodiment, the spatial filter is trained based on historical lower limb motor imagery task EEG data and implemented using a shared spatial pattern algorithm. The historical data originates from EEG signals collected from multiple subjects in previous experiments. These signals were recorded during the execution of gait and stepping imagery tasks and underwent preprocessing. Multiple sets of EEG data segments belonging to both gait and stepping imagery tasks are extracted from the historical data. Each data segment is truncated to a preset duration based on the task start point, ensuring consistency with the time window of the currently analyzed data segment. The extraction of data segments considers the synchronicity of task execution, avoiding noise introduced due to timing deviations, while ensuring a balance in data volume between the two types of tasks, thus improving the robustness of the algorithm training.

[0039] For each task, the covariance matrix of its EEG data segments is calculated separately. The covariance matrix represents the linear correlation between multi-channel EEG signals and reflects the spatial distribution pattern of task-induced neural activity. During the calculation, the mean of each data segment is removed to eliminate the influence of DC bias, and then the covariance matrix is ​​solved using the standard formula. To improve the numerical stability of the matrix, especially under low signal-to-noise ratio conditions, the covariance matrix is ​​normalized. Normalization is achieved using a regularization parameter, which is calculated based on the trace of the covariance matrix. Specifically, the regularization parameter is determined by the product of the inverse of the trace and a small constant used to prevent matrix singularity. The normalized covariance matrix is ​​then corrected by linear combination to make its eigenvalue distribution more stable and reduce the risk of overfitting.

[0040] Next, we solve the generalized eigenvalue problem for the normalized mixture covariance matrix of the two tasks. The mixture covariance matrix is ​​obtained by weighted averaging of the covariance matrices of the walking and stepping imagery tasks, with the weight ratios set based on the number of data segments for each task. The generalized eigenvalue problem aims to find a set of eigenvectors that maximizes the variance ratio of the EEG signals for the two tasks. A numerical optimization algorithm is used to ensure computational efficiency and accuracy. The obtained eigenvectors are arranged in descending order of their corresponding eigenvalues. The eigenvalues ​​reflect the variance discrimination in that direction; a larger eigenvalue indicates a more significant contribution of the source signal to the task differences. The arranged eigenvectors are combined to form a spatial projection matrix, which is an orthogonal transformation matrix used to project the original multi-channel EEG signals into a new source signal space.

[0041] After the spatial projection matrix is ​​trained, it is applied to the current data segment being analyzed. The data segment is input to the spatial filter, which projects the data segment onto the spatial projection matrix through matrix multiplication, resulting in a set of source signals. The projection process achieves signal dimensionality reduction, converting high-dimensional multi-channel EEG signals into low-dimensional source signals, with each source signal representing an independent source of neural activity.

[0042] After obtaining the source signals, they are sorted according to the magnitude of the feature values ​​related to the differences between the two tasks. The feature values ​​were calculated during the training phase, and the sorting is done in descending order of feature values. The source signal with the largest feature value corresponds to the direction that maximizes the variance of the walking imagery task, and the source signal with the smallest feature value corresponds to the direction that maximizes the variance of the stepping imagery task. After sorting, a predetermined number of source signals, both the first and last in the sorted list, are selected as the optimal source signals. Specifically, the first m source signals with the largest feature values ​​and the last n source signals with the smallest feature values ​​are selected, where m and n are predetermined positive integers. The values ​​of m and n are determined through cross-validation to ensure that their sum accounts for a predetermined proportion of the original EEG signal channel count. This proportion is set to 20% to 30% to balance information preservation and dimensionality reduction. For example, if the original number of channels is 64, the sum of m and n can be 13, where m is 6 and n is 7, ensuring that the optimal source signals cover the most discriminative neural activity patterns for both tasks.

[0043] The selection of optimal source signals is based on the theoretical foundation of the co-space pattern algorithm, which states that source signals with large eigenvalues ​​carry strong discriminative information related to task execution, while source signals with small eigenvalues ​​may contain task-irrelevant noise or complementary information. By combining the source signals from both ends, EEG features of lower limb movement intentions are captured, enhancing the performance of subsequent classification or regression models.

[0044] The power value of the optimal source signal in the sensorimotor rhythm frequency band is calculated and used as the EEG feature vector representing the intention to move the lower limbs.

[0045] In this embodiment, the power value of the optimal source signal within the sensorimotor rhythm frequency band is calculated to construct an EEG feature vector representing the intention to move the lower limbs. Power estimation is performed for each optimal source signal, and the results are ultimately integrated into a multi-dimensional feature vector for subsequent pattern recognition or control tasks. The optimal source signal originates from the output of the spatial filter and already contains key neural oscillation information from the sensorimotor cortex. Power calculation quantifies its energy distribution, thereby capturing changes in neural activity induced by the motor imagery task.

[0046] For each optimal source signal, a sliding time window is applied for segmented processing. The time window starts from the signal's starting point and slides in a non-overlapping manner, covering the entire analysis data segment. The signal data within each time window is first subjected to mean removal processing to eliminate the effects of DC components and low-frequency drift. Mean removal is achieved by calculating the arithmetic mean of the signal within that window and subtracting this mean from each data point, ensuring that the signal is in a zero-mean state, so that the variance calculation only reflects the energy of the AC component.

[0047] After removing the mean, the variance of the signal within each time window is calculated. Specifically, for a time window containing N data points, the variance is calculated as the sum of the squares of the data points divided by N. This value represents the average power of the signal within that time window.

[0048] Subsequently, the arithmetic mean of the variances calculated for all time windows is used as the final power estimate of the optimal source signal for that path. The averaging process smooths out time fluctuations, improves the robustness of the power estimate, and reduces the impact of random noise. The final power estimate reflects the overall level of signal power throughout the entire task execution and is correlated with the intensity of neural activity related to lower limb motor intention.

[0049] After power estimation of all optimal source signals is completed, the power estimates of each channel are combined into a multidimensional vector in a predetermined order, based on the arrangement of the source signals in the spatial filter.

[0050] Ultimately, this multidimensional vector, serving as an EEG feature vector representing lower limb motor intention, is input into downstream classifiers or regression models for task type recognition or continuous control. The construction of the feature vector fully leverages the power characteristics of sensorimotor rhythms, achieving efficient decoding of lower limb motor imagery through variance estimation and vectorization, providing stable and interpretable input features for brain-computer interface systems.

[0051] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A goal-oriented method for acquiring lower limb motor brain-computer interface signals, characterized in that, include: When performing a goal-oriented lower limb motor imagery task, the EEG signals acquired by the multi-channel EEG acquisition system and the task event markers generated by the goal-oriented task system are recorded simultaneously. The lower limb motor imagery task is walking imagery or stepping imagery. The task event markers include at least the task start, task type, and feedback information trigger point. The EEG signal was bandpass filtered to extract the sensorimotor rhythm frequency band signal, and then independent component analysis was used to remove artifacts from electrooculography and electromyography to obtain the processed EEG signal. Based on the task start point in the task event marker, the EEG signal after a preset duration is extracted as the analysis data segment. The preset duration is set based on the execution time window of the goal-oriented task. The analyzed data segment is input into a spatial filter to perform signal dimensionality reduction and obtain the optimal source signal. The spatial filter is trained using a co-spatial pattern algorithm based on historical lower limb motor imagery task EEG data. The power value of the optimal source signal in the sensorimotor rhythm frequency band is calculated and used as the EEG feature vector representing the intention to move the lower limbs.

2. The method for acquiring lower limb motor brain-computer interface signals based on goal guidance according to claim 1, characterized in that, Bandpass filtering is performed on the electroencephalogram (EEG) signal to extract the sensorimotor rhythm frequency band signal, including: A zero-phase digital filter was used to filter the EEG signal, and the sensorimotor rhythm frequency band was set to a wide band of 8Hz to 30Hz. The filtered signal contains mu rhythm and beta rhythm components, which are used for artifact removal processing.

3. The method for acquiring lower limb motor brain-computer interface signals based on goal guidance according to claim 2, characterized in that, The removal of electrooculography and electromyography artifacts using independent component analysis includes: Independent component analysis was performed on the sensorimotor rhythm frequency band signal after bandpass filtering, decomposing it into multiple statistically independent components; Identification is performed based on the matching degree between the spatiotemporal characteristics of each component and typical artifact patterns. The spatiotemporal characteristics include the brain topography distribution of the component and the kurtosis and skewness of the time series. After setting the components related to electrooculography (EOG) or electromyography (EMG) artifacts to zero, the EEG signal after artifact removal is reconstructed through inverse transformation.

4. The method for acquiring lower limb motor brain-computer interface signals based on goal guidance according to claim 3, characterized in that, The identification based on the matching degree between the spatiotemporal characteristics of each component and typical artifact patterns includes: The correlation between the brain topography maps of each component and the weighted templates of the prefrontal cortex and periocular region is calculated. If the correlation coefficient exceeds the first threshold, it is determined to be an electrooculography artifact component. The kurtosis value of each component's time series is compared with a second threshold. If the kurtosis value exceeds the second threshold, it is determined to be an electromyographic artifact component.

5. The method for acquiring lower limb motor brain-computer interface signals based on goal guidance according to claim 1, characterized in that, The spatial filter is trained using a co-spatial pattern algorithm based on historical lower limb motor imagery task EEG data, and includes: Multiple sets of EEG data segments belonging to the two categories of walking imagination and stepping imagination were extracted from historical lower limb motor imagination task EEG data. The covariance matrix of the EEG data segments for the two types of tasks was calculated separately, and the covariance matrix was normalized. Solve the generalized eigenvalue problem of the normalized mixture covariance matrix by arranging the eigenvectors to form a spatial projection matrix.

6. The method for acquiring lower limb motor brain-computer interface signals based on goal guidance according to claim 5, characterized in that, The normalization process includes: The covariance matrix is ​​corrected using a regularization parameter calculated based on the trace of the covariance matrix, thereby improving the numerical stability of the spatial filter under low signal-to-noise ratio conditions.

7. The method for acquiring lower limb motor brain-computer interface signals based on goal guidance according to claim 6, characterized in that, The process of obtaining the optimal source signal includes: The analyzed data segment is projected onto the spatial projection matrix obtained by the common space mode algorithm to obtain a set of source signals; The source signals are sorted according to the magnitude of the feature values ​​associated with the differences between the two types of tasks; The preset quantity source signals, which are ranked first and last, are selected together as the optimal source signal.

8. The method for acquiring lower limb motor brain-computer interface signals based on goal guidance according to claim 7, characterized in that, The preset quantity source signals for selecting the first and last sorting points include: Select the top m source signals with the largest eigenvalues ​​and the bottom n source signals with the smallest eigenvalues, where m and n are preset positive integers, and the sum of m and n accounts for a preset proportion of the number of original EEG signal channels.

9. A method for acquiring lower limb motor brain-computer interface signals based on goal orientation according to claim 1, characterized in that, The power value calculation includes: For each optimal source signal, a sliding time window is used to calculate its signal variance as a power estimate. The calculated power estimates of each optimal source signal are combined sequentially into a multidimensional vector; This multidimensional vector is used as an EEG feature vector representing the intention to move the lower limbs.

10. A method for acquiring lower limb motor brain-computer interface signals based on goal guidance according to claim 9, characterized in that, The step of using a sliding time window to calculate the signal variance as a power estimate includes: The length of the sliding time window is matched with the preset duration of the analyzed data segment; After removing the mean from the signal data within each time window, calculate its variance, and take the average of the variance values ​​within all windows as the final power estimate of the optimal source signal.

Images