A brain-controlled target control method and device based on ECoG signals

Abstract

The application discloses a brain control target control method and device based on an ECoG signal, and acquires real-time electroencephalogram signals collected through predetermined channels in ECoG electrodes, wherein the predetermined channels are selected according to the control contribution of the electroencephalogram signals collected by each channel to the brain control target; the real-time electroencephalogram signals are input to a preset coordinate decoding model and a speed decoding model to obtain decoded coordinates and decoded speed of the brain control target, wherein the coordinate decoding model and the speed decoding model are trained by the electroencephalogram signals of the predetermined channels and the corresponding coordinates and speed of the brain control target; and the running track of the brain control target is generated according to the decoded coordinates and the decoded speed, so that the control strategy optimization of the brain control target is completed according to the ECoG signal, and the accurate execution of the brain control instruction is ensured.

Description

Technical Field

[0001] This application relates to the field of medical technology assistance technology, specifically to a brain-controlled target control method and device based on ECoG signals. Background Technology

[0002] Brain-computer interface (BCI) technology is a technology that directly connects the brain to an external target device, enabling direct control of the device through the brain. With the development of brain-controlled technology, various applications have been realized, such as brain-controlled sentence spelling, control of robotic arms, and control of wheelchairs.

[0003] Most existing technologies rely on decoding and analyzing electroencephalogram (EEG) signals to achieve target control, while research on brain-controlled technologies based on electrocorticogram (ECoG) signals is relatively limited. Furthermore, current methods for optimizing control strategies in brain-controlled applications based on algorithmic decoding results often require additional signals such as blinking, facial electromyography (EMG) signals, or radar for comprehensive decision-making and strategy optimization. While this approach improves the accuracy of brain-controlled commands to some extent, it increases system complexity and reduces user experience.

[0004] Therefore, how to accurately execute brain control commands based on ECoG signals is a technical problem that needs to be solved. Summary of the Invention

[0005] This application provides a brain-controlled target control method and device based on ECoG signals, which can accurately execute brain-controlled commands and precisely control brain-controlled targets based on ECoG signals.

[0006] In a first aspect, embodiments of this application provide brain-controlled target control based on ECoG signals, wherein the brain-controlled target control method based on ECoG signals includes:

[0007] Real-time EEG signals acquired through predetermined channels in ECoG electrodes are obtained, wherein the predetermined channels are selected based on the contribution of the EEG signals acquired by each channel to the control of the brain-controlled target.

[0008] The real-time EEG signal is input to a preset coordinate decoding model and velocity decoding model to obtain the decoded coordinates and decoding velocity of the brain-controlled target. The coordinate decoding model and the velocity decoding model are trained by the EEG signal of the predetermined channel with the coordinates and velocity corresponding to the brain-controlled target.

[0009] The trajectory of the mind-controlled target is generated based on the decoding coordinates and the decoding speed.

[0010] In conjunction with the first aspect, in one implementation, before acquiring real-time EEG signals acquired through predetermined channels in the ECoG electrodes, the predetermined channels are selected based on the contribution of the EEG signals acquired through each channel to the control target of the brain-controlled object, including:

[0011] The EEG signals collected from each channel of the ECoG electrode and the real coordinates corresponding to the brain-controlled target are acquired simultaneously, and the acquired EEG signals are denoised.

[0012] The EEG signals collected from each channel are cut into multiple window segments according to a preset step size, and the real speed corresponding to the brain control target is calculated based on the real coordinates of each window segment.

[0013] Each window segment is converted from a time-domain signal to a time-frequency graph, and the time-frequency graphs in each channel are divided to obtain the first training set and the first test set for each channel.

[0014] The time-frequency graphs in the first training set of each channel are used as input, and the real velocities in each direction are used as labels to train the deep learning model, thereby obtaining the velocity prediction models for each direction of each channel.

[0015] The time-frequency graphs from the first test set of each channel are input into the velocity prediction model for each direction to obtain the predicted velocity for each direction.

[0016] Based on the first correlation coefficient between the predicted velocity and the actual velocity in each direction of each channel, the control contribution coefficient of each channel to the brain control target is determined.

[0017] The top q channels with the largest control contribution coefficients are identified as the predetermined channels;

[0018] Each direction includes the x-direction and y-direction in the coordinate system, or the x-direction, y-direction and z-direction.

[0019] In one implementation, training the coordinate decoding model and the velocity decoding model before inputting the real-time EEG signal to the preset coordinate decoding model and velocity decoding model includes:

[0020] The window segment of the EEG signal acquired by the predetermined channel is converted from a time-domain signal to a frequency-domain signal;

[0021] Based on the preset sampling frequency and extraction frequency, each window segment is decomposed into frequency bands, and the average energy amplitude of each frequency band is calculated to extract the features of the window segment, thereby obtaining the feature data of each window segment.

[0022] The feature data of each window segment is used as input, and the real coordinates of each direction are used as labels. Multiple coordinate decoding models for each direction are trained separately.

[0023] The feature data of each window segment is used as input, and the real velocity in each direction is used as a label to train multiple velocity decoding models for each direction.

[0024] In one implementation, after acquiring real-time EEG signals collected through a predetermined channel in the ECoG electrode, and before inputting the real-time EEG signals into a preset coordinate decoding model and velocity decoding model to obtain the decoded coordinates and decoding velocity of the brain-controlled target, the process includes:

[0025] The real-time EEG signal is subjected to noise reduction processing;

[0026] The denoised real-time EEG signal is cropped into real-time window segments according to the preset step size. The real-time window segments are converted from time-domain signals to frequency-domain signals, and feature extraction is performed on the real-time window segments to obtain feature data of each real-time window segment.

[0027] In one implementation, the real-time EEG signal is input to a preset coordinate decoding model and a velocity decoding model to obtain the decoded coordinates and decoding velocity of the brain-controlled target, including:

[0028] The feature data of u consecutive real-time window segments are input into multiple coordinate decoding models and multiple velocity decoding models in each direction to obtain multiple decoded coordinates and multiple decoded velocities in each direction of each real-time window segment in the u consecutive real-time window segments.

[0029] The final decoding coordinates for each direction are obtained by optimizing the calculation based on multiple decoding coordinates for each real-time window segment.

[0030] The final decoding speed for each direction is obtained by optimizing the calculation based on the decoding speed of each real-time window segment in multiple directions.

[0031] In one embodiment, after obtaining the decoding coordinates and decoding speed of the mind-controlled target, and before generating the trajectory of the mind-controlled target based on the decoding coordinates and decoding speed, the method further includes:

[0032] Differential calculation is performed on the final decoded coordinates of u real-time window segments to obtain u-1 reverse velocities;

[0033] Calculate the second correlation coefficient between the final decoding speed in the x-direction and the reverse decoding speed in the x-direction, and the third correlation coefficient between the final decoding speed in the y-direction and the reverse decoding speed in the y-direction.

[0034] If the second correlation coefficient is greater than or equal to the sum of the third correlation coefficient and the first reference coefficient, and is greater than the first reference coefficient, then the average of the final decoding speed and the reverse speed in each direction is used as the correction speed in each direction, resulting in u-1 correction speeds.

[0035] Otherwise, the minimum of the final decoding speed and the reverse speed in each direction is taken as the correction speed for each direction, resulting in u-1 correction speeds.

[0036] In one embodiment, after obtaining the decoding coordinates and decoding speed of the mind-controlled target, and before generating the trajectory of the mind-controlled target based on the decoding coordinates and decoding speed, the method further includes:

[0037] Perform inverse difference calculations on u-1 corrected velocities to obtain the corresponding u inverse coordinates;

[0038] Calculate the fourth correlation coefficient between the final decoded coordinates in the x-direction and the reversed coordinates in the x-direction, and the fifth correlation coefficient between the final decoded coordinates in the y-direction and the reversed coordinates in the y-direction.

[0039] If the fourth correlation coefficient is greater than or equal to the sum of the fifth correlation coefficient and the second reference coefficient, and is greater than the second reference coefficient, then the mean of the final decoded coordinates and the reversed coordinates in each direction is used as the corrected coordinates in each direction to obtain u corrected coordinates;

[0040] Otherwise, the smallest of the final decoded coordinates and the reversed coordinates of each direction of the real-time window segment is used as the corrected coordinate for each direction to obtain the corrected coordinates of each real-time window segment, thus obtaining u corrected coordinates.

[0041] In one implementation, after determining the corrected coordinates and corrected velocity for each direction of each real-time window segment, the method further includes:

[0042] Based on the three sigma principle, abnormal correction coordinates in each direction are eliminated in each preset smoothing step size;

[0043] Based on the three sigma principle, abnormal correction speeds in each direction of each smoothing step size are eliminated;

[0044] The smoothing step size is greater than 2*u real-time window segments.

[0045] In one implementation, generating the trajectory of the brain-controlled target based on the decoding coordinates and the decoding speed includes:

[0046] The motion intention curve of the brain-controlled target is obtained by fitting the corrected coordinates after removing outliers;

[0047] Calculate the slope of the tangent at each corrected coordinate in the motion intention curve, and determine the angle between the slope of the tangent at each corrected coordinate and the corrected velocity direction corresponding to the corrected coordinate.

[0048] If the included angle is less than or equal to a preset included angle threshold, then the average value of all target correction coordinates between the first target correction coordinate corresponding to the included angle and the second target correction coordinate k correction coordinates away from the first target correction distance is taken as the decision coordinate, and the average value of the correction speeds corresponding to all target correction coordinates is taken as the decision speed.

[0049] If the included angle is greater than the included angle threshold, then the third target correction coordinate corresponding to the included angle and the fourth target correction coordinate that is k correction coordinates away from the third target correction coordinate are connected to form a target straight line. The distances between each target correction coordinate and the target straight line between the third target correction coordinate and the fourth target correction coordinate are calculated. The target correction coordinate with the largest distance is taken as the center coordinate. The t target correction coordinates before and after the center coordinate are taken as decision coordinates. The correction speed corresponding to the decision coordinate is taken as the decision speed.

[0050] Connect the decision coordinates to generate the trajectory of the brain-controlled target after a current interval of k coordinates;

[0051] The radius of curvature of each decision coordinate in the running trajectory is determined. When the radius of curvature is less than or equal to a preset radius threshold, the corresponding decision velocity direction is corrected to point to the center of curvature of the corresponding decision coordinate.

[0052] Secondly, embodiments of this application provide a brain-controlled target control device based on ECoG signals, the brain-controlled target control device based on ECoG signals comprising:

[0053] The acquisition module is used to acquire real-time EEG signals collected through predetermined channels in the ECoG electrodes, wherein the predetermined channels are selected according to the control contribution of the EEG signals collected by each channel to the brain-controlled target.

[0054] A decoding module is used to input the real-time EEG signal to a preset coordinate decoding model and velocity decoding model to obtain the decoded coordinates and decoding velocity of the brain-controlled target, wherein the coordinate decoding model and the velocity decoding model are trained by the EEG signal of the predetermined channel with the coordinates and velocity corresponding to the brain-controlled target;

[0055] A generation module is used to generate the trajectory of the brain-controlled target based on the decoding coordinates and the decoding speed.

[0056] This application provides a brain-controlled target control method and apparatus based on ECoG signals. It acquires real-time EEG signals collected through predetermined channels in ECoG electrodes, wherein the predetermined channels are selected based on the contribution of the EEG signals collected from each channel to the control of the brain-controlled target. The real-time EEG signals are input to preset coordinate decoding models and velocity decoding models to obtain the decoded coordinates and decoding velocity of the brain-controlled target. The coordinate decoding model and the velocity decoding model are trained using the EEG signals from the predetermined channels and the coordinates and velocities corresponding to the brain-controlled target. The running trajectory of the brain-controlled target is generated based on the decoded coordinates and decoding velocity, thus achieving optimization of the control strategy for the brain-controlled target based on ECoG signals and ensuring the accurate execution of brain-controlled commands. Attached Figure Description

[0057] Figure 1 This is a flowchart illustrating an embodiment of the brain-controlled target control method based on ECoG signals according to this application.

[0058] Figure 2 This is a schematic diagram of the functional modules of an embodiment of the brain-controlled target control device based on ECoG signals of this application. Detailed Implementation

[0059] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present application.

[0060] To make the objectives, technical solutions, and advantages of this application clearer, the embodiments of this application will be described in further detail below with reference to the accompanying drawings.

[0061] In a first aspect, embodiments of this application provide a brain-controlled target control method based on ECoG signals.

[0062] In one embodiment, reference is made to Figure 1 , Figure 1 This is a flowchart illustrating the first embodiment of the brain-controlled target control method based on ECoG signals according to this application. Figure 1 As shown, brain-controlled target control methods based on ECoG signals include:

[0063] Step S101: Acquire real-time EEG signals collected through predetermined channels in the ECoG electrodes, wherein the predetermined channels are selected based on the contribution of the EEG signals collected by each channel to the control of the brain-controlled target.

[0064] Step S102: Input the real-time EEG signal into the preset coordinate decoding model and velocity decoding model to obtain the decoding coordinates and decoding velocity of the brain-controlled target. The coordinate decoding model and the velocity decoding model are obtained by training the EEG signal of the predetermined channel with the coordinates and velocity corresponding to the brain-controlled target.

[0065] Step S103: Generate the running trajectory of the brain-controlled target based on the decoding coordinates and the decoding speed.

[0066] It is worth noting that the predetermined channels in the ECoG electrodes, as well as the training coordinate decoding model and the training velocity decoding model, selected in this embodiment can be completed offline in advance before executing the brain-controlled target control method based on ECoG signals.

[0067] The ECoG signal-based brain-controlled target control method provided in this embodiment can control various brain-controlled targets, such as brain-controlled wheelchairs and brain-controlled robotic arms. The following description uses a brain-controlled wheelchair as an example to illustrate the ECoG signal-based brain-controlled target control method of this application.

[0068] Specifically, before acquiring the real-time EEG signals collected through the predetermined channels in the ECoG electrodes in step S101, the predetermined channels are selected based on the contribution of the EEG signals collected by each channel to the control of the brain-controlled target, including:

[0069] Step S201: Synchronously acquire the EEG signals collected from each channel of the ECoG electrode and the real coordinates corresponding to the brain-controlled target, and perform noise reduction processing on the acquired EEG signals.

[0070] It's worth noting that the channels of an ECoG electrode refer to the multiple independent sensors on the ECoG electrode used to record electroencephalogram (EEG) signals. The true coordinates of a brain-controlled target are its spatial position coordinates, and these true coordinates include multiple directions. For example, the true coordinates of a brain-controlled wheelchair are (x, y), including the true coordinates in the x and y directions; the coordinates of a brain-controlled robotic arm are (x, y, z), consisting of the true coordinates in the x, y, and z directions.

[0071] In this embodiment, the methods for denoising EEG signals include Independent Component Analysis (ICA), wavelet decomposition, EMD, or machine learning, and the specific denoising method can be selected according to the specific task requirements.

[0072] Step S202: According to the preset step size S, the EEG signals collected by each channel are cut into multiple window segments, and the real speed corresponding to the brain control target is calculated according to the real coordinates corresponding to each window segment.

[0073] It is worth noting that in this embodiment, the preset step size S can be 600 points, 800 points, or 2000 points, etc., and the value of the preset step size S can be selected according to the actual situation. After the EEG signals of each channel are clipped into window segments, each channel includes multiple window segments (i.e., EEG signal segments), and each window segment corresponds to multiple coordinates of the brain-controlled target.

[0074] By averaging the true coordinates of all x-directions and all y-directions within the same window, the true coordinates of that window segment can be obtained. For each channel, the corresponding real coordinates are assigned according to the time sequence of the window segment. The true velocity (v) of the brain-controlled target is obtained by performing first-order difference processing for each window segment. x ,v y Among them, the actual speed (v) of the mind-controlled target. x ,v y This includes the actual velocity v in the x-direction. x and the true velocity v in the y direction y .

[0075] S203. Convert each window segment from a time-domain signal into a time-frequency graph, and divide the time-frequency graph in each channel to obtain the first training set and the first test set for each channel.

[0076] Specifically, the time-domain signals of each window segment can be converted into time-frequency graphs using methods such as Fourier transform, short-time Fourier transform, or wavelet transform. After partitioning, the first training set for each channel includes multiple time-frequency graphs for model training and corresponding real velocities, while the first test set includes multiple time-frequency graphs for model testing and corresponding real velocities.

[0077] S204. Using the time-frequency graphs from the first training set of each channel as input, and the actual velocities in each direction as labels, train the deep learning model to obtain the velocity prediction models for each direction of each channel.

[0078] As an example, for each channel, M deep learning models can be set up to predict the velocity in the x-direction and M deep learning models to predict the velocity in the y-direction. In this embodiment, the ECoG electrode includes K channels. Taking the k-th channel as an example, the time-frequency plot from the first test set is used as input, and the corresponding true velocity v in the x-direction is... x Using these as labels, Q deep learning models for predicting velocity in the x-direction are trained to obtain Q x-direction velocity prediction models for the k-th channel, i.e., Q k-models. vxUsing the time-frequency graph from the first test set as input, the corresponding true velocity v in the y-direction is... x As labels, Q deep learning models for predicting y-direction velocity are trained to obtain Q y-direction velocity prediction models for the k-th channel, i.e., Q k-models. vy Where k represents the k-th channel, the ECoG electrode has a total of K channels, and the model vx The velocity prediction model represents the velocity in the x-direction. vy This represents the velocity prediction model in the y-direction. In this embodiment, the deep learning model can be a CNN model.

[0079] Q can be set to any value as needed. A value greater than or equal to 1 indicates that multiple models are trained to select the best one for each direction.

[0080] S205. Input the time-frequency graphs from the first test set of each channel into the velocity prediction model for each direction to obtain the predicted velocity for each direction.

[0081] Taking the k-th channel as an example, multiple time-frequency maps from the first test set are input into the velocity prediction model k-model in the x-direction. vx k-model for velocity prediction in the y-direction vy The predicted velocity Tk-list in the x-direction is obtained from the output. vxi Predicted velocity Tk-list in the y-direction vyi , where i represents the i-th time-frequency plot in the first test set.

[0082] S206. Based on the first correlation coefficient between the predicted velocity in each direction of each channel and the corresponding actual velocity in each direction, determine the control contribution coefficient of each channel to the brain-controlled target.

[0083] It is worth noting that, taking the k-th channel as an example, the predicted velocity Tk-list in the x-direction is calculated. vxi k-list of true velocities in the x-direction corresponding to the test set vxi The first correlation coefficient kD between them vxi Calculate the predicted velocity Tk-list in the y-direction vyi k-list of true velocities in the y-direction corresponding to the test set vyi The first correlation coefficient kD between them vyi and will The corresponding value is used as the control contribution coefficient of the k-th channel to the brain control target. This process can be repeated to calculate the control contribution coefficient of each channel to the brain control target.

[0084] S207. Determine the top q channels with the largest control contribution coefficients as the predetermined channels.

[0085] Exemplary control contribution coefficients for K channels Sort the channels from largest to smallest and select the top q channels as the reserved channels, or select the top q channels from the total number of channels, whichever is the first preset percentage. q can be a value based on requirements, such as 4, 8, or 10, and the preset percentage can be a value based on requirements, such as the top 10% or 20%.

[0086] It is worth noting that after selecting the predetermined channels, the channel number of each predetermined channel is determined. If the ECoG electrode has only one CMOS circuit, the EEG signals collected by this q predetermined channels are retained for normal transmission, and the other channels are turned off. If the ECoG electrodes share a CMOS circuit in a single row / column or multiple rows / columns, the CMOS circuit containing the q predetermined channels is retained, and the other CMOS circuits are turned off.

[0087] Further, before inputting the real-time EEG signal into the preset coordinate decoding model and velocity decoding model in step S102, training the coordinate decoding model and the velocity decoding model includes the following steps:

[0088] Step S301: Convert a window segment of the EEG signal acquired through the predetermined channel from a time-domain signal to a frequency-domain signal. This conversion can be performed based on Fourier transform.

[0089] Step S302: Decompose each window segment into frequency bands according to the preset sampling frequency and extraction frequency, and calculate the average energy amplitude of each frequency band to extract the features of the window segment, thereby obtaining the feature data of each window segment.

[0090] As an example, n frequency ranges of interest are extracted from the frequency domain signal, such as n = 3, 5, or 10. The signal sampling frequency is PHZ, and the extraction frequency is from... Select n points from the range, and the frequency range of each point is... in It can be set based on experience, such as wait.

[0091] Based on the amplitudes corresponding to n frequency bands, the average energy amplitude of each frequency band is calculated and stored in a list format. Thus, the data corresponding to the same window segment of the original data consists of:

[0092] [[L 11 ,L 12 …L 1S ],[L 21 ,L 22 …L 2S ],…L[LK1 ,L K2 …L KS ]]

[0093] become:

[0094] [W 11 W 12 …W 1n W 21 W 22 …W 2n W K1 W K2 …W Kn ]

[0095] This expands the data length at any given time from K channels to n*K, where K is the number of channels and S is the preset step size for dividing the window segments. The EEG signal of each predetermined channel can be decomposed into n frequency bands, such as 0-4Hz, 11-30Hz, etc., through frequency domain decomposition.

[0096] Step S303: Take the feature data of each window segment as input, take the real coordinates of each direction as labels, and train the multiple coordinate decoding models for each direction.

[0097] As an example, the coordinate decoding model for each direction includes a coordinate decoding model for the x-direction and a coordinate decoding model for the y-direction. By combining multiple coordinate decoding models for the x-direction and multiple coordinate decoding models for the y-direction, the accuracy of the decoded coordinates can be improved. The models can be at least two of the following: random forest, decision tree, backpropagation neural network, and SVM.

[0098] During model training, the feature data of each window segment is used as input, and the corresponding true coordinates in the x-direction are used as labels to train the multi-coordinate decoding model in the x-direction; the feature data of each window segment is used as input, and the corresponding true coordinates in the y-direction are used as labels to train the multi-coordinate decoding model in the y-direction.

[0099] Step S304: Take the feature data of each window segment as input, take the real velocity in each direction as label, and train the multiple velocity decoding models in each direction.

[0100] Similarly, the velocity decoding model for each direction includes a velocity decoding model in the x-direction and a velocity decoding model in the y-direction. By combining multiple velocity decoding models in the x-direction and multiple velocity decoding models in the y-direction, the accuracy of the decoding speed can be improved. The model type can be at least two of the following: random forest, decision tree, backpropagation neural network, and SVM.

[0101] During model training, the feature data of each window segment is used as input, and the corresponding true velocity in the x-direction is used as the label to train multiple velocity decoding models in the x-direction; the feature data of each window segment is used as input, and the corresponding true velocity in the y-direction is used as the label to train multiple velocity decoding models in the y-direction.

[0102] After completing the offline selection of the predetermined channel and training of the coordinate decoding model and velocity decoding model, online decoding can be started.

[0103] Specifically, after acquiring the real-time EEG signal through a predetermined channel in the ECoG electrode in step S101, before inputting the real-time EEG signal into a preset coordinate decoding model and velocity decoding model in step S102 to obtain the decoding coordinates and decoding velocity of the brain-controlled target, the process includes:

[0104] The real-time EEG signal acquired in step S101 is subjected to noise reduction processing. The noise-reduced real-time EEG signal is then clipped into real-time window segments according to the preset step size S. These real-time window segments are converted from time-domain signals to frequency-domain signals, and feature extraction is performed on each real-time window segment to obtain feature data for each segment. The specific methods for noise reduction, signal conversion, and feature extraction can be the same as described above and will not be repeated here.

[0105] Further, step S102 inputs the real-time EEG signal to a preset coordinate decoding model and velocity decoding model to obtain the decoding coordinates and decoding velocity of the brain-controlled target, specifically including:

[0106] Step S1021: Input the feature data of u consecutive real-time window segments into the multiple coordinate decoding model and the multiple velocity decoding model in each direction to obtain multiple decoded coordinates and multiple decoded velocities in each direction of each real-time window segment in the u consecutive real-time window segments.

[0107] When there are H types of multiple coordinate decoding models and multiple velocity decoding models in each direction, then for each real-time window segment, we can obtain H x-direction decoding coordinates, H y-direction decoding coordinates, H x-direction decoding velocities, and H y-direction decoding velocities.

[0108] Step S1022: Optimize the calculation based on the multiple decoding coordinates of each direction of each real-time window segment to obtain the final decoding coordinates of each direction.

[0109] As an example, for each real-time window segment, the H x-direction decoding coordinates are input into a preset decision model and the particle swarm optimization algorithm or whale optimization algorithm is used. The pre-x predicted by the optimization algorithm is used as the final decoding coordinate in the x-direction of the real-time window segment. Similarly, the final decoding coordinate pre-y in the y-direction of the real-time window segment can be obtained. Different optimization algorithms can be selected for the x and y directions.

[0110] During online decoding, the coordinate decoding model is loaded sequentially. Decoding u consecutive real-time window segments will predict u final decoded coordinates (x, y).

[0111] Step S1023: Optimize the calculation based on the multiple decoding speeds of each direction of each real-time window segment to obtain the final decoding speed for each direction.

[0112] It is worth noting that the exemplary method in step S1022 can be used to load the velocity decoding model to predict u final velocity coordinates (v) for decoding u consecutive real-time window segments. x ,v y ).

[0113] In one embodiment, after obtaining the decoding coordinates and decoding speed of the mind-controlled target in step S102, and before generating the running trajectory of the mind-controlled target based on the decoding coordinates and decoding speed in step S103, the method further includes:

[0114] Step S401: Perform differential calculation on the final decoded coordinates of u real-time window segments to obtain u-1 reverse velocity.

[0115] As an example, the first-order difference is performed on the first u final decoded coordinates (x, y) calculated in step S1022 to obtain the u-1 backward velocity calculated from the coordinates. (The reverse speed corresponding to the first final decoded coordinate is discarded).

[0116] Step S402: Calculate the second correlation coefficient between the final decoding speed in the x direction and the reverse decoding speed in the x direction, and the third correlation coefficient between the final decoding speed in the y direction and the reverse decoding speed in the y direction.

[0117] Exemplary calculation of the final decoding speed v in the x-direction x Reverse velocity in the x direction The second correlation coefficient D between them uvx Calculate the final decoding speed v in the y-direction. y Reverse velocity in the y direction The third correlation coefficient D between them uvy .

[0118] Step S403: If the second correlation coefficient is greater than or equal to the sum of the third correlation coefficient and the first reference coefficient, and is greater than the first reference coefficient, then the average of the final decoding speed and the reverse speed in each direction is used as the correction speed in each direction, resulting in u-1 correction speeds.

[0119] As an example, in this embodiment, the first reference coefficient is Δ, if D uvx ≥ΔandD uvy If ≥Δ, then the corrected u-1 velocities are

[0120] Step S404: Otherwise, take the minimum of the final decoding speed and the reverse speed in each direction as the correction speed for each direction, and obtain u-1 correction speeds.

[0121] Exemplary, if D is not satisfied uvx ≥ΔandD uvy If ≥Δ, then the corrected velocities of u-1 are: The initial velocity of the first u-th correction is set to 0, resulting in the corrected velocities for the first consecutive u-window segments. Starting from the second consecutive u-window segment, the velocity at the first velocity position is equal to the last corrected velocity of the previous consecutive u-window segment.

[0122] Furthermore, after obtaining the decoding coordinates and decoding speed of the mind-controlled target in step S102, and before generating the running trajectory of the mind-controlled target based on the decoding coordinates and decoding speed in step S103, the method further includes:

[0123] Step S501: Perform reverse differential calculation on u-1 corrected velocities to obtain the corresponding u reverse coordinates.

[0124] Specifically, the final decoded coordinates of the starting position are taken, and the velocity corrected in step S404 is subjected to an inverse first-order difference to obtain the corresponding inverse coordinates (x). u ,y u ).

[0125] Step S502: Calculate the fourth correlation coefficient between the final decoded coordinates in the x-direction and the reversed coordinates in the x-direction, and the fifth correlation coefficient between the final decoded coordinates in the y-direction and the reversed coordinates in the y-direction.

[0126] As an example, calculate the final decoded coordinate x in the x-direction and the inverse coordinate x in the x-direction. u The fourth correlation coefficient D between them ux Calculate the final decoded coordinates y in the y-direction and the inverse coordinates y in the y-direction. u The fifth correlation coefficient D between them uy .

[0127] Step S503: If the fourth correlation coefficient is greater than or equal to the sum of the fifth correlation coefficient and the second reference coefficient, and is greater than the second reference coefficient, then the mean of the final decoded coordinates and the reversed coordinates in each direction is used as the corrected coordinates in each direction to obtain u corrected coordinates.

[0128] As an example, in this embodiment, the second reference coefficient is Γ, if D ux ≥ΓandD uy If ≥Γ, then the corrected u coordinates are

[0129] Step S504: Otherwise, take the smallest of the final decoded coordinates and the reversed coordinates of each direction of the real-time window segment as the corrected coordinates of each direction to obtain the corrected coordinates of each real-time window segment, so as to obtain u corrected coordinates.

[0130] Exemplary, if D is not satisfied ux ≥ΓandD uy If ≥Γ, then the corrected u coordinates are (min(x,x)). u ),min(y,y u )).

[0131] It is worth noting that Δ and Γ can be obtained from experimental tests or based on experience, such as Γ taking values ​​of 0.85 or 0.90, etc., without specific restrictions here.

[0132] Furthermore, after determining the corrected coordinates and corrected speed for each direction of each real-time window segment, the process also includes:

[0133] Abnormal correction coordinates in each direction are removed based on the three Sigma principle; abnormal correction velocities in each direction are removed based on the three Sigma principle; wherein, the smoothing step size is greater than 2*u real-time window segments.

[0134] As an example, a smoothing step size of L is selected, where L ≥ 2*u. The x-coordinate of each corrected coordinate within the smoothing step size L is calculated to determine if it is a singular value. i ∈|x i -mean(x i∈L )|≥3(x i∈L Then determine x. i If x is a singular value, remove it. i At the same time, remove y i That is, the coordinates (x) were removed. i y i At this point, of the original L coordinate pairs, L1 remain, i.e., (x L1 ,y L1 ).

[0135] Next, determine (x) L ,y L For each corrected coordinate, is the y-direction coordinate a singular value? If y j ∈|y j -mean(y j∈L1 )|≥3·std(y j∈L1 ), then determine y j Remove this y for singular values j At the same time, remove x j That is, the coordinates (x) were removed. j y j ) to obtain (x L2 ,y L2 At this point, only L2 coordinate pairs remain.

[0136] Based on the anomaly correction coordinates removed within the step size L, synchronous removal is performed. The anomaly correction speed in the middle is obtained

[0137] Further, step S103, generating the trajectory of the mind-controlled target based on the decoded coordinates and the decoded speed, includes:

[0138] Step S1031: Correct the coordinates (x) after removing outliers. L2 ,y L2 The motion intention curve of the brain-controlled target is obtained by fitting the points in the graph.

[0139] Step S1032: Calculate the slope of the tangent line at each corrected coordinate in the motion intention curve. And determine the slope of the tangent line for each corrected coordinate. Corrected velocity direction corresponding to the corrected coordinates The included angle θ L2_i .in, Depend on get.

[0140] Step S1033: If the included angle is less than or equal to a preset included angle threshold a, i.e., θ L2_i If ≤α, then the average of all target correction coordinates between the first target correction coordinate corresponding to the included angle and the second target correction coordinate which is r correction coordinates away from the first target correction coordinate is taken as the decision coordinate, and the average of the correction velocities corresponding to all target correction coordinates is taken as the decision velocity. Here, α can be any value, for example, 20, 25, or 30.

[0141] Step S1034: If the included angle is greater than the included angle threshold, i.e. θ L2_iIf the angle is greater than α, then connect the corrected coordinates of the third target corresponding to the included angle with the corrected coordinates of the fourth target, which are g corrected coordinates away from the corrected coordinates of the third target, to form the target straight line l. κ The system calculates the distances between each target correction coordinate and the target straight line between the third target correction coordinate and the fourth target correction coordinate. The target correction coordinate with the largest distance is taken as the center coordinate, and the t target correction coordinates before and after the center coordinate are taken as decision coordinates. The correction speed corresponding to the decision coordinate is taken as the decision speed. In this embodiment, t is set to 5.

[0142] Step S1035: Connect the decision coordinates obtained in S1033 and S1034 to generate the running trajectory of the brain-controlled target after a current interval of r coordinates.

[0143] Step S1036: Determine the radius of curvature of each decision coordinate in the running trajectory. When the radius of curvature is less than or equal to a preset radius threshold, correct the corresponding decision speed direction to point to the center of curvature of the corresponding decision coordinate.

[0144] As an example, determine the radius of curvature r at each decision coordinate position in the trajectory of step S1035. κi When r κi When ≤β, the velocity direction at this decision coordinate position is forcibly corrected to point towards the center of the curvature circle, where β can be given based on actual experience or obtained by experimental methods; and the velocity magnitude is forcibly corrected to the range of [v1,v2], where v1 is to ensure the lower limit of the velocity to prevent it from being too slow, and v2 is to ensure the upper limit of the velocity to avoid the risk of excessive centrifugal force during turning.

[0145] In another application scenario, taking a brain-controlled robotic arm as an example, when controlling the wrist of the robotic arm, taking the left hand as an example, the origin of the coordinate system is the left shoulder axis (0,0,0), the coronal axis is the x-axis, the sagittal axis is the y-axis, and the vertical axis is the z-axis.

[0146] Step S601: Channel Contribution Ranking and Determining Pre-defined Channels. Pre-defined channels are selected based on the contribution of EEG signals acquired from each channel to the control of the brain-controlled robotic arm. Specifically, this includes:

[0147] Step S6011, referring to step S202, the true spatial coordinates of the wrist are (x, y, z), and the true velocity is (v). x ,v y ,v z ).

[0148] Step S6012, referring to steps S203 to S204, with (v x ,v y ,v zA regression model is trained using the labels to obtain the k-model. vx k-model vy k-model vz .

[0149] Step S6013, referring to steps S205 to S206, obtain kD vxi kD vyi kD vzi And and

[0150] Step S6014, referring to step S207, yields q predetermined channels.

[0151] Step S602, Feature Extraction. Refer to steps S301 to S302 to obtain the feature list.

[0152] Step S603: Feature fitting, training the coordinate decoding model and the velocity decoding model. Referring to steps S303 to S304, obtain pre-x, pre-y, pre-z, and pre-v. x ,pre-v y ,pre-v z .

[0153] Online decoding and control phase:

[0154] Step S604, refer to step S207 to control the start and stop of the control channel.

[0155] Step S605: Real-time EEG signal denoising, window segmentation, conversion, and feature extraction.

[0156] Step S606: Decoding coordinates and decoding speed correction, and outlier removal. Refer to steps S401-S404 and S501-S504 to obtain the corrected (x) L2 ,y L2 ,z L2 )and

[0157] Step S607: Delay decision-making to determine the motion path.

[0158] Step S6071: The length from the left shoulder axis to the left wrist is r. Remove (x) L2 ,y L2 ,z L2 )middle The point, to obtain a new coordinate pair (x r1 ,y r1 ,z r1 After removing the velocity pairs corresponding to those times, new velocity pairs are obtained.

[0159] Step S6072: When the arm extends backward, the extreme value of the y-axis corresponds to the spatial angle θ of the arm. Discard coordinate pairs that violate the θ→y correspondence to obtain new coordinate pairs (x...). r2 ,y r2 ,z r2 ) and new

[0160] Step S6073 performs the following processing on the xy plane, xz plane, and yz plane respectively, taking the xy plane as an example:

[0161] (1) Referring to steps S1031-S1032, obtain the slope of the tangent at each coordinate point in the xy plane. and the direction of velocity at this coordinate.

[0162] (2) Referring to steps S1033-S1034, obtain the coordinate points retained in the xy plane and their corresponding velocity pairs;

[0163] (3) Repeat (1) and (2) to obtain the coordinate points and their corresponding velocity pairs retained in the xz plane and yz plane.

[0164] (4) Combine the (x,y), (x,z), and (y,z) values ​​of the three planes obtained in (3) to form (x,y,z), and retain only the values ​​in the planes that are ... The coordinate pairs that have appeared in the previous coordinate pairs are used to form new coordinate pairs (x, y, y). r3 ,y r3 ,z r3 ) and their corresponding velocity pairs

[0165] Secondly, embodiments of this application also provide a brain-controlled target control device based on ECoG signals.

[0166] In one embodiment, reference is made to Figure 2 , Figure 2 This is a functional module diagram of an embodiment of the brain-controlled target control device based on ECoG signals according to this application. Figure 2 As shown, the brain-controlled target control device based on ECoG signals includes:

[0167] The acquisition module is used to acquire real-time EEG signals collected through predetermined channels in the ECoG electrodes, wherein the predetermined channels are selected according to the control contribution of the EEG signals collected by each channel to the brain-controlled target.

[0168] A decoding module is used to input the real-time EEG signal to a preset coordinate decoding model and velocity decoding model to obtain the decoded coordinates and decoding velocity of the brain-controlled target, wherein the coordinate decoding model and the velocity decoding model are trained by the EEG signal of the predetermined channel with the coordinates and velocity corresponding to the brain-controlled target;

[0169] A generation module is used to generate the trajectory of the brain-controlled target based on the decoding coordinates and the decoding speed.

[0170] Furthermore, in one embodiment, the device further includes a selection module, which is used to:

[0171] The EEG signals collected from each channel of the ECoG electrode and the real coordinates corresponding to the brain-controlled target are acquired simultaneously, and the acquired EEG signals are denoised.

[0172] The EEG signals collected from each channel are cut into multiple window segments according to a preset step size, and the real speed corresponding to the brain control target is calculated based on the real coordinates of each window segment.

[0173] Each window segment is converted from a time-domain signal to a time-frequency graph, and the time-frequency graphs in each channel are divided to obtain the first training set and the first test set for each channel.

[0174] The time-frequency graphs in the first training set of each channel are used as input, and the real velocities in each direction are used as labels to train the deep learning model, thereby obtaining the velocity prediction models for each direction of each channel.

[0175] The time-frequency graphs from the first test set of each channel are input into the velocity prediction model for each direction to obtain the predicted velocity for each direction.

[0176] Based on the first correlation coefficient between the predicted velocity and the actual velocity in each direction of each channel, the control contribution coefficient of each channel to the brain control target is determined.

[0177] The top q channels with the largest control contribution coefficients are identified as the predetermined channels;

[0178] Each direction includes the x-direction and y-direction in the coordinate system, or the x-direction, y-direction and z-direction.

[0179] Furthermore, in one embodiment, the device further includes a training module, which is used for:

[0180] The window segment of the EEG signal acquired by the predetermined channel is converted from a time-domain signal to a frequency-domain signal;

[0181] Based on the preset sampling frequency and extraction frequency, each window segment is decomposed into frequency bands, and the average energy amplitude of each frequency band is calculated to extract the features of the window segment, thereby obtaining the feature data of each window segment.

[0182] The feature data of each window segment is used as input, and the real coordinates of each direction are used as labels. Multiple coordinate decoding models for each direction are trained separately.

[0183] The feature data of each window segment is used as input, and the real velocity in each direction is used as a label to train multiple velocity decoding models for each direction.

[0184] Furthermore, in one embodiment, the device further includes a processing module for:

[0185] The real-time EEG signal is subjected to noise reduction processing;

[0186] The denoised real-time EEG signal is cropped into real-time window segments according to the preset step size. The real-time window segments are converted from time-domain signals to frequency-domain signals, and feature extraction is performed on the real-time window segments to obtain feature data of each real-time window segment.

[0187] Furthermore, in one embodiment, the decoding module is also used for:

[0188] The feature data of u consecutive real-time window segments are input into multiple coordinate decoding models and multiple velocity decoding models in each direction to obtain multiple decoded coordinates and multiple decoded velocities in each direction of each real-time window segment in the u consecutive real-time window segments.

[0189] The final decoding coordinates for each direction are obtained by optimizing the calculation based on multiple decoding coordinates for each real-time window segment.

[0190] The final decoding speed for each direction is obtained by optimizing the calculation based on the decoding speed of each real-time window segment in multiple directions.

[0191] Furthermore, in one embodiment, the device further includes a correction module, which is used to:

[0192] Differential calculation is performed on the final decoded coordinates of u real-time window segments to obtain u-1 reverse velocities;

[0193] Calculate the second correlation coefficient between the final decoding speed in the x-direction and the reverse decoding speed in the x-direction, and the third correlation coefficient between the final decoding speed in the y-direction and the reverse decoding speed in the y-direction.

[0194] If the second correlation coefficient is greater than or equal to the sum of the third correlation coefficient and the first reference coefficient, and is greater than the first reference coefficient, then the average of the final decoding speed and the reverse speed in each direction is used as the correction speed in each direction, resulting in u-1 correction speeds.

[0195] Otherwise, the minimum of the final decoding speed and the reverse speed in each direction is taken as the correction speed for each direction, resulting in u-1 correction speeds.

[0196] Furthermore, in one embodiment, the correction module is also used for:

[0197] Perform inverse difference calculations on u-1 corrected velocities to obtain the corresponding u inverse coordinates;

[0198] Calculate the fourth correlation coefficient between the final decoded coordinates in the x-direction and the reversed coordinates in the x-direction, and the fifth correlation coefficient between the final decoded coordinates in the y-direction and the reversed coordinates in the y-direction.

[0199] If the fourth correlation coefficient is greater than or equal to the sum of the fifth correlation coefficient and the second reference coefficient, and is greater than the second reference coefficient, then the mean of the final decoded coordinates and the reversed coordinates in each direction is used as the corrected coordinates in each direction to obtain u corrected coordinates;

[0200] Otherwise, the smallest of the final decoded coordinates and the reversed coordinates of each direction of the real-time window segment is used as the corrected coordinate for each direction to obtain the corrected coordinates of each real-time window segment, thus obtaining u corrected coordinates.

[0201] Furthermore, in one embodiment, the device further includes a rejection module, which is used to:

[0202] Based on the three sigma principle, abnormal correction coordinates in each direction are eliminated in each preset smoothing step size;

[0203] Based on the three sigma principle, abnormal correction speeds in each direction of each smoothing step size are eliminated;

[0204] The smoothing step size is greater than 2*u real-time window segments.

[0205] Furthermore, in one embodiment, the generation module is further configured to:

[0206] The motion intention curve of the brain-controlled target is obtained by fitting the corrected coordinates after removing outliers;

[0207] Calculate the slope of the tangent at each corrected coordinate in the motion intention curve, and determine the angle between the slope of the tangent at each corrected coordinate and the corrected velocity direction corresponding to the corrected coordinate.

[0208] If the included angle is less than or equal to a preset included angle threshold, then the average value of all target correction coordinates between the first target correction coordinate corresponding to the included angle and the second target correction coordinate k correction coordinates away from the first target correction distance is taken as the decision coordinate, and the average value of the correction speeds corresponding to all target correction coordinates is taken as the decision speed.

[0209] If the included angle is greater than the included angle threshold, then the third target correction coordinate corresponding to the included angle and the fourth target correction coordinate that is k correction coordinates away from the third target correction coordinate are connected to form a target straight line. The distances between each target correction coordinate and the target straight line between the third target correction coordinate and the fourth target correction coordinate are calculated. The target correction coordinate with the largest distance is taken as the center coordinate. The t target correction coordinates before and after the center coordinate are taken as decision coordinates. The correction speed corresponding to the decision coordinate is taken as the decision speed.

[0210] Connect the decision coordinates to generate the trajectory of the brain-controlled target after a current interval of k coordinates;

[0211] The radius of curvature of each decision coordinate in the running trajectory is determined. When the radius of curvature is less than or equal to a preset radius threshold, the corresponding decision velocity direction is corrected to point to the center of curvature of the corresponding decision coordinate.

[0212] The functions of each module in the ECoG signal-based brain control target control device correspond to the steps in the ECoG signal-based brain control target control method embodiment, and their functions and implementation processes will not be described in detail here.

[0213] It should be noted that the sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0214] The terms "comprising" and "having," and any variations thereof, in the specification, claims, and accompanying drawings of this application are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or apparatus that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to such process, method, product, or apparatus. The terms "first," "second," and "third," etc., are used to distinguish different objects, etc., and do not indicate a sequence, nor do they limit "first," "second," and "third" to different types.

[0215] In some processes described in the embodiments of this application, multiple operations or steps are included in a specific order. However, it should be understood that these operations or steps may not be executed in the order they appear in the embodiments of this application, or they may be executed in parallel. The sequence number of the operation is only used to distinguish different operations, and the sequence number itself does not represent any execution order. In addition, these processes may include more or fewer operations, and these operations or steps may be executed sequentially or in parallel, and these operations or steps may be combined.

[0216] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method.

[0217] The above are merely preferred embodiments of this application and do not limit the patent scope of this application. Any equivalent structural or procedural transformations made using the content of this application's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of this application.

Claims

1. A brain-controlled target control method based on ECoG signals, characterized in that, The brain-controlled target control method based on ECoG signals includes: Real-time EEG signals acquired through predetermined channels in ECoG electrodes are obtained, wherein the predetermined channels are selected based on the contribution of the EEG signals acquired by each channel to the control of the brain-controlled target. The real-time EEG signal is input to a preset coordinate decoding model and velocity decoding model to obtain the decoded coordinates and decoding velocity of the brain-controlled target. The coordinate decoding model and the velocity decoding model are trained by the EEG signal of the predetermined channel with the coordinates and velocity corresponding to the brain-controlled target. The trajectory of the mind-controlled target is generated based on the decoding coordinates and the decoding speed; Before acquiring real-time EEG signals through predetermined channels in the ECoG electrodes, the predetermined channels are selected based on the contribution of the EEG signals acquired through each channel to the control of the brain-controlled target, including: The EEG signals collected from each channel of the ECoG electrode and the real coordinates corresponding to the brain-controlled target are acquired simultaneously, and the acquired EEG signals are denoised. The EEG signals collected from each channel are cut into multiple window segments according to a preset step size, and the real speed corresponding to the brain control target is calculated based on the real coordinates of each window segment. Each window segment is converted from a time-domain signal to a time-frequency graph, and the time-frequency graphs in each channel are divided to obtain the first training set and the first test set for each channel. The time-frequency graphs in the first training set of each channel are used as input, and the real velocities in each direction are used as labels to train the deep learning model, thereby obtaining the velocity prediction models for each direction of each channel. The time-frequency graphs from the first test set of each channel are input into the velocity prediction model for each direction to obtain the predicted velocity for each direction. Based on the first correlation coefficient between the predicted velocity and the actual velocity in each direction of each channel, the control contribution coefficient of each channel to the brain control target is determined. The top q channels with the largest control contribution coefficients are identified as the predetermined channels; Each direction includes the x-direction and y-direction in the coordinate system, or the x-direction, y-direction and z-direction.

2. The brain-controlled target control method based on ECoG signals as described in claim 1, characterized in that, Before inputting the real-time EEG signal into the preset coordinate decoding model and velocity decoding model, training the coordinate decoding model and the velocity decoding model includes: The window segment of the EEG signal acquired by the predetermined channel is converted from a time-domain signal to a frequency-domain signal; Based on the preset sampling frequency and extraction frequency, each window segment is decomposed into frequency bands, and the average energy amplitude of each frequency band is calculated to extract the features of the window segment, thereby obtaining the feature data of each window segment. The feature data of each window segment is used as input, and the real coordinates of each direction are used as labels. Multiple coordinate decoding models for each direction are trained separately. The feature data of each window segment is used as input, and the real velocity in each direction is used as a label to train multiple velocity decoding models for each direction.

3. The brain-controlled target control method based on ECoG signals as described in claim 2, characterized in that, After acquiring real-time EEG signals through predetermined channels in ECoG electrodes, and before inputting the real-time EEG signals into preset coordinate decoding models and velocity decoding models to obtain the decoded coordinates and decoding velocity of the brain-controlled target, the process includes: The real-time EEG signal is subjected to noise reduction processing; The denoised real-time EEG signal is cropped into real-time window segments according to the preset step size. The real-time window segments are converted from time-domain signals to frequency-domain signals, and feature extraction is performed on the real-time window segments to obtain feature data of each real-time window segment.

4. The brain-controlled target control method based on ECoG signals as described in claim 3, characterized in that, The real-time EEG signal is input into a preset coordinate decoding model and velocity decoding model to obtain the decoded coordinates and decoding velocity of the brain-controlled target, including: The feature data of u consecutive real-time window segments are input into multiple coordinate decoding models and multiple velocity decoding models in each direction to obtain multiple decoded coordinates and multiple decoded velocities in each direction of each real-time window segment in the u consecutive real-time window segments. The final decoding coordinates for each direction are obtained by optimizing the calculation based on multiple decoding coordinates for each real-time window segment. The final decoding speed for each direction is obtained by optimizing the calculation based on the decoding speed of each real-time window segment in multiple directions.

5. The brain-controlled target control method based on ECoG signals as described in claim 4, characterized in that, After obtaining the decoding coordinates and decoding speed of the mind-controlled target, and before generating the trajectory of the mind-controlled target based on the decoding coordinates and decoding speed, the method further includes: Differential calculation is performed on the final decoded coordinates of u real-time window segments to obtain u-1 reverse velocities; Calculate the second correlation coefficient between the final decoding speed in the x-direction and the reverse decoding speed in the x-direction, and the third correlation coefficient between the final decoding speed in the y-direction and the reverse decoding speed in the y-direction. If the second correlation coefficient is greater than or equal to the sum of the third correlation coefficient and the first reference coefficient, and is greater than the first reference coefficient, then the average of the final decoding speed and the reverse speed in each direction is used as the correction speed in each direction, resulting in u-1 correction speeds. Otherwise, the minimum of the final decoding speed and the reverse speed in each direction is taken as the correction speed for each direction, resulting in u-1 correction speeds.

6. The brain-controlled target control method using ECoG signals as described in claim 5, characterized in that, After obtaining the decoding coordinates and decoding speed of the mind-controlled target, and before generating the trajectory of the mind-controlled target based on the decoding coordinates and decoding speed, the method further includes: Perform inverse difference calculations on u-1 corrected velocities to obtain the corresponding u inverse coordinates; Calculate the fourth correlation coefficient between the final decoded coordinates in the x-direction and the reversed coordinates in the x-direction, and the fifth correlation coefficient between the final decoded coordinates in the y-direction and the reversed coordinates in the y-direction. If the fourth correlation coefficient is greater than or equal to the sum of the fifth correlation coefficient and the second reference coefficient, and is greater than the second reference coefficient, then the mean of the final decoded coordinates and the reversed coordinates in each direction is used as the corrected coordinates in each direction to obtain u corrected coordinates; Otherwise, the smallest of the final decoded coordinates and the reversed coordinates of each direction of the real-time window segment is used as the corrected coordinate for each direction to obtain the corrected coordinates of each real-time window segment, thus obtaining u corrected coordinates.

7. The brain-controlled target control method using ECoG signals as described in claim 6, characterized in that, After determining the corrected coordinates and corrected velocity for each direction of each real-time window segment, the following is also included: Based on the three sigma principle, abnormal correction coordinates in each direction are eliminated in each preset smoothing step size; Based on the three sigma principle, abnormal correction speeds in each direction of each smoothing step size are eliminated; Wherein, the smoothing step size is greater than 2 A real-time window fragment.

8. The brain-controlled target control method using ECoG signals as described in claim 7, characterized in that, The trajectory of the mind-controlled target is generated based on the decoding coordinates and the decoding speed, including: The motion intention curve of the brain-controlled target is obtained by fitting the corrected coordinates after removing outliers; Calculate the slope of the tangent at each corrected coordinate in the motion intention curve, and determine the angle between the slope of the tangent at each corrected coordinate and the corrected velocity direction corresponding to the corrected coordinate. If the included angle is less than or equal to a preset included angle threshold, then the average value of all target correction coordinates between the first target correction coordinate corresponding to the included angle and the second target correction coordinate k correction coordinates away from the first target correction distance is taken as the decision coordinate, and the average value of the correction speeds corresponding to all target correction coordinates is taken as the decision speed. If the included angle is greater than the included angle threshold, then the third target correction coordinate corresponding to the included angle and the fourth target correction coordinate that is k correction coordinates away from the third target correction coordinate are connected to form a target straight line. The distances between each target correction coordinate and the target straight line between the third target correction coordinate and the fourth target correction coordinate are calculated. The target correction coordinate with the largest distance is taken as the center coordinate. The t target correction coordinates before and after the center coordinate are taken as decision coordinates. The correction speed corresponding to the decision coordinate is taken as the decision speed. Connect the decision coordinates to generate the trajectory of the brain-controlled target after a current interval of k coordinates; The radius of curvature of each decision coordinate in the running trajectory is determined. When the radius of curvature is less than or equal to a preset radius threshold, the corresponding decision velocity direction is corrected to point to the center of curvature of the corresponding decision coordinate.

9. A brain-controlled target control device based on ECoG signals, characterized in that, The brain-controlled target control device based on ECoG signals includes: The acquisition module is used to acquire real-time EEG signals collected through predetermined channels in the ECoG electrodes, wherein the predetermined channels are selected according to the control contribution of the EEG signals collected by each channel to the brain-controlled target. A decoding module is used to input the real-time EEG signal to a preset coordinate decoding model and velocity decoding model to obtain the decoded coordinates and decoding velocity of the brain-controlled target, wherein the coordinate decoding model and the velocity decoding model are trained by the EEG signal of the predetermined channel with the coordinates and velocity corresponding to the brain-controlled target; A generation module is used to generate the trajectory of the brain-controlled target based on the decoding coordinates and the decoding speed; The device also includes a selection module, which is used for: The EEG signals collected from each channel of the ECoG electrode and the real coordinates corresponding to the brain-controlled target are acquired simultaneously, and the acquired EEG signals are denoised. The EEG signals collected from each channel are cut into multiple window segments according to a preset step size, and the real speed corresponding to the brain control target is calculated based on the real coordinates of each window segment. Each window segment is converted from a time-domain signal to a time-frequency graph, and the time-frequency graphs in each channel are divided to obtain the first training set and the first test set for each channel. The time-frequency graphs in the first training set of each channel are used as input, and the real velocities in each direction are used as labels to train the deep learning model, thereby obtaining the velocity prediction models for each direction of each channel. The time-frequency graphs from the first test set of each channel are input into the velocity prediction model for each direction to obtain the predicted velocity for each direction. Based on the first correlation coefficient between the predicted velocity and the actual velocity in each direction of each channel, the control contribution coefficient of each channel to the brain control target is determined. The top q channels with the largest control contribution coefficients are identified as the predetermined channels; Each direction includes the x-direction and y-direction in the coordinate system, or the x-direction, y-direction and z-direction.

Images