OpenViBE brain-computer interface decoding control method and system driven by efficient channel selection algorithm

Abstract

The present application relates to the field of brain-computer interface technology, and discloses an OpenViBE brain-computer interface decoding control method and system driven by an efficient channel selection algorithm. Specifically, based on pre-collected EEG data, the comprehensive importance score of each channel is calculated for each task, a channel-task importance matrix is ​​constructed, and then the optimal channel subset is selected from the channel-task importance matrix. The EEG signal processing model is optimized according to the optimal channel subset, and the optimized EEG signal processing model is deployed to the OpenViBE platform for real-time EEG analysis and processing. While ensuring accuracy, the amount of calculation can be significantly reduced, the demand for hardware resources is reduced, and the real-time performance of the model on portable devices is improved. In addition, the corresponding channel selection strategy can be flexibly selected according to the specific application scenario, and the optimal channel subset can be selected from the channel-task importance matrix according to the selected channel selection strategy, which significantly improves the versatility of the model.

Description

Technical Field

[0001] The present application relates to the field of brain-computer interface technology, and specifically to an OpenViBE brain-computer interface decoding control method and system driven by an efficient channel selection algorithm. Background Art

[0002] Brain-computer interface (BMI) technology offers a new avenue for motor rehabilitation for people with disabilities, particularly by utilizing motor imagery (MI) to control devices such as prosthetic limbs or wheelchairs. Compared to non-invasive electroencephalography (EEG), implantable BMIs are more suitable for long-term, stable rehabilitation training due to their higher signal quality and more convenient use. Convolutional neural networks (CNNs) have shown great potential in MI analysis, as their ability to automatically learn features avoids tedious manual feature engineering. However, high-performance CNN models typically need to run on remote servers with powerful computing power, which can lead to data transmission delays, affect system availability, and pose data privacy and security risks.

[0003] To address these issues, one feasible solution is to perform local data processing on a portable computing terminal (such as a laptop). This solution balances computing power and portability, allowing doctors to monitor and adjust parameters in real time. It is also easier to maintain and upgrade than embedded systems. However, this solution also faces a huge challenge: how to run a high-performance CNN model while ensuring real-time performance and decoding accuracy within the limited computing resources of a portable computing terminal? Existing brain-computer interface systems based on platforms such as OpenViBE often rely on complex algorithms and large amounts of channel data, which are difficult to achieve real-time processing on portable devices.

[0004] The key contribution of lightweight models such as EEGNet in resolving the conflict between real-time performance and accuracy lies in their improved parameter and computational efficiency. By adopting techniques such as depthwise separable convolution and designing a compact network architecture, the model parameters and computational complexity are significantly reduced, enabling fast inference on resource-constrained hardware platforms and meeting the needs of real-time BCI applications. However, models such as EEGNet do not explicitly consider channel redundancy and directly process all input channels. This results in inefficient computation and makes the model prone to overfitting to noise and redundant information, which interferes with feature learning, reduces generalization ability, and makes model results difficult to interpret. Summary of the Invention

[0005] This application provides an OpenViBE brain-computer interface decoding control method driven by an efficient channel selection algorithm to solve the problem in the existing technology that lightweight models do not explicitly consider channel redundancy when running on resource-constrained hardware platforms, resulting in low computational efficiency and easy overfitting to noise and redundant information, thereby interfering with feature learning, reducing generalization ability, and making model results difficult to interpret.

[0006] Correspondingly, this application also provides an OpenViBE brain-computer interface decoding control system driven by an efficient channel selection algorithm to ensure the implementation and application of the above method.

[0007] In order to solve the above technical problems, the present application discloses an OpenViBE brain-computer interface decoding control method driven by an efficient channel selection algorithm, the method comprising:

[0008] Based on the pre-collected EEG data, the comprehensive importance score of each channel is calculated for each task, and a channel-task importance matrix is ​​constructed;

[0009] Determine a channel selection strategy, and select the optimal channel subset from the channel-task importance matrix based on the selected channel selection strategy;

[0010] Optimize the EEG signal processing model based on the optimal channel subset;

[0011] The optimized EEG signal processing model is deployed to the OpenViBE platform for real-time EEG analysis and processing.

[0012] Preferably, based on the pre-collected EEG data, the comprehensive importance score of each channel is calculated for each task, and a channel-task importance matrix is ​​constructed, including:

[0013] Use all channel data of EEG data to train the EEG signal processing model and obtain the feature weights of each layer;

[0014] For each task, the weight gradient information is calculated according to the feature weights of each layer, the importance score of each channel is obtained, and the channel-task importance matrix is ​​constructed.

[0015] Preferably, for each task, the weight gradient information is calculated according to the feature weights of each layer to obtain the importance score of each channel, and a channel-task importance matrix is ​​constructed, including:

[0016] For the current task, the validation set data is used to calculate the loss function, and the gradient of the feature weights of each layer to the loss function is obtained through back propagation;

[0017] Based on the gradient and feature weight of each layer, the channel importance score of each channel in each layer is calculated;

[0018] Calculate the comprehensive importance score of each channel according to the channel importance score of each channel in each layer;

[0019] According to the comprehensive importance score of each channel obtained under each task, a channel-task importance matrix is ​​constructed.

[0020] Preferably, the channel selection strategy includes:

[0021] The global strategy selects one or more channels from all tasks in descending order of comprehensive importance scores;

[0022] Task-specific strategies select one or more channels for a specific task in descending order of comprehensive importance scores;

[0023] The hybrid strategy selects one or more channels shared by multiple tasks in descending order of comprehensive importance scores, and selects one or more channels for specific tasks.

[0024] Preferably, determining a channel selection strategy and selecting an optimal channel subset from a channel-task importance matrix according to the selected channel selection strategy includes:

[0025] Determine the channel selection strategy from global strategy, task-specific strategy and hybrid strategy;

[0026] Determine a channel selection scheme based on the selected channel selection strategy;

[0027] The optimal channel subset is obtained by selecting from the channel-task importance matrix according to the channel selection scheme.

[0028] Preferably, the channel selection scheme is any one of the following:

[0029] Solution 1: predetermine the number of target channels K, select the K channels with the highest importance scores based on the selected channel selection strategy, and generate the optimal channel subset;

[0030] Solution 2: Set an importance threshold and select all channels with importance scores higher than the importance threshold according to the selected channel selection strategy to generate the optimal channel subset;

[0031] Option three: determine the constraints, and on the premise of satisfying the constraints, select the optimal channel combination according to the selected channel selection strategy to generate the optimal channel subset.

[0032] Preferably, optimizing the EEG signal processing model according to the optimal channel subset includes:

[0033] In the EEG data, data related to the optimal channel subset is selected to train the EEG signal processing model.

[0034] Preferably, based on the pre-collected EEG data, the comprehensive importance score of each channel is calculated for each task, and before constructing the channel-task importance matrix, the method further includes:

[0035] Filter the EEG data and use a preset method to remove eye movement and myoelectric artifacts from the EEG data.

[0036] Preferably, before deploying the optimized EEG signal processing model to the OpenViBE platform for real-time EEG analysis and processing, the method further comprises:

[0037] Configure the relevant parameters of the OpenViBE platform, including sampling rate, channel data, and data format.

[0038] This application also discloses an OpenViBE brain-computer interface decoding control system driven by an efficient channel selection algorithm, the system comprising:

[0039] The channel importance calculation module is used to calculate the comprehensive importance score of each channel for each task based on the pre-collected EEG data and construct a channel-task importance matrix;

[0040] The channel selection module is used to determine the channel selection strategy and select the optimal channel subset from the channel-task importance matrix according to the selected channel selection strategy;

[0041] Model optimization module, used to optimize the EEG signal processing model based on the optimal channel subset;

[0042] The model deployment module is used to deploy the optimized EEG signal processing model to the OpenViBE platform for real-time EEG analysis and processing.

[0043] In this application, based on the pre-collected EEG data, the comprehensive importance score of each channel is calculated for each task, and a channel-task importance matrix is ​​constructed. Then, the optimal channel subset is selected from the channel-task importance matrix, and the EEG signal processing model is optimized according to the optimal channel subset. The optimized EEG signal processing model is deployed to the OpenViBE platform for real-time EEG analysis and processing. It can significantly reduce the amount of calculation while ensuring accuracy, reduce the demand for hardware resources, and improve the real-time performance of the model on portable devices. In addition, the corresponding channel selection strategy can be flexibly selected according to the specific application scenario, and the optimal channel subset can be selected from the channel-task importance matrix according to the selected channel selection strategy, which significantly improves the versatility of the model.

[0044] Additional aspects and advantages of the present application will be given in the following description, which will become apparent from the following description, or will be understood through practice of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] The above and / or additional aspects and advantages of the present application will become apparent and easily understood from the following description of the embodiments in conjunction with the accompanying drawings, in which:

[0046] Figure 1Flowchart of the OpenViBE brain-computer interface decoding control method driven by the efficient channel selection algorithm provided in the embodiment of the present application;

[0047] Figure 2 The overall flow chart of decoding control using the OpenViBE brain-computer interface provided in the embodiment of the present application;

[0048] Figure 3 A flowchart of real-time model optimization and deployment provided in an embodiment of the present application;

[0049] Figure 4 A schematic diagram of the structure of the OpenViBE brain-computer interface decoding control system driven by the efficient channel selection algorithm provided in an embodiment of the present application;

[0050] Figure 5 A schematic diagram of the structure of the complete system for OpenViBE brain-computer interface decoding control provided in an embodiment of the present application. DETAILED DESCRIPTION

[0051] The following describes embodiments of the present application in detail. Examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present application and are not to be construed as limiting the present application.

[0052] It will be understood by those skilled in the art that, unless expressly stated otherwise, the singular forms "a", "an", "said" and "the" used herein may also include the plural forms. It should be further understood that the term "comprising" used in the specification of this application refers to the presence of features, integers, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or combinations thereof. It should be understood that when we refer to an element as being "connected" or "coupled" to another element, it may be directly connected or coupled to the other element, or there may be intermediate elements. In addition, "connected" or "coupled" as used herein may include wireless connections or wireless couplings. The term "and / or" used herein includes all or any units and all combinations of one or more associated listed items.

[0053] Those skilled in the art will understand that, unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by those skilled in the art in the art to which this invention belongs. It should also be understood that terms such as those defined in common dictionaries should be understood to have meanings consistent with their meanings in the context of the prior art and will not be interpreted in an idealized or overly formal sense unless specifically defined as herein.

[0054] The solution provided in the embodiments of the present application can be executed by any electronic device, such as a terminal device or a server, wherein the server can be an independent physical server, a server cluster or a distributed system composed of multiple physical servers, or a cloud server that provides cloud computing services. The terminal can be a smart phone, a tablet computer, a laptop computer, a desktop computer, a smart speaker, a smart watch, etc., but is not limited to this. The terminal and the server can be directly or indirectly connected via wired or wireless communication, and this application does not limit this. With respect to the technical problems existing in the prior art, the OpenViBE brain-computer interface decoding control method and system driven by an efficient channel selection algorithm provided in this application is intended to solve at least one of the technical problems of the prior art.

[0055] The following specific embodiments describe in detail the technical solution of the present application and how the technical solution of the present application solves the above-mentioned technical problems. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments. The embodiments of the present application will be described below in conjunction with the accompanying drawings.

[0056] The embodiment of the present application provides an OpenViBE brain-computer interface decoding control method driven by an efficient channel selection algorithm. Figure 1 As shown in , the method may include the following steps:

[0057] Step 101 : Based on the pre-collected EEG data, the comprehensive importance score of each channel is calculated for each task, and a channel-task importance matrix is ​​constructed.

[0058] In the embodiments of the present application, an EEG acquisition device (e.g., an EEG cap) may be used to collect EEG data from the user's scalp. The choice of acquisition device depends on the specific needs of the research, such as the required number of channels, sampling rate, and anti-interference capability.

[0059] Based on the collected EEG equipment, by constructing a channel-task importance matrix to evaluate the comprehensive importance of each channel under each task, the key channels can be effectively identified and data redundancy can be reduced.

[0060] Step 102: determine a channel selection strategy, and select an optimal channel subset from the channel-task importance matrix according to the selected channel selection strategy.

[0061] Step 103: Optimize the EEG signal processing model according to the optimal channel subset.

[0062] In the embodiment of the present application, different channel selection strategies can be selected for different application scenarios, and then the optimal channel subset suitable for the application scenario can be selected. The EEG signal processing model is optimized by the optimal channel subset obtained under a specific strategy, which can be applied to different task requirements and improve the versatility of the model.

[0063] Step 104 : deploying the optimized EEG signal processing model to the OpenViBE platform for real-time EEG analysis and processing.

[0064] In an embodiment of the present application, based on the pre-collected EEG data, the comprehensive importance score of each channel is calculated for each task, and a channel-task importance matrix is ​​constructed. Then, the optimal channel subset is selected from the channel-task importance matrix, and the EEG signal processing model is optimized according to the optimal channel subset. The optimized EEG signal processing model is deployed to the OpenViBE platform for real-time EEG analysis and processing. While ensuring accuracy, the amount of calculation can be significantly reduced, the demand for hardware resources is reduced, and the real-time performance of the model on portable devices is improved. In addition, the corresponding channel selection strategy can be flexibly selected according to the specific application scenario, and the optimal channel subset can be selected from the channel-task importance matrix according to the selected channel selection strategy, which significantly improves the versatility of the model.

[0065] In an optional embodiment, based on pre-collected EEG data, the comprehensive importance score of each channel is calculated for each task, and before constructing the channel-task importance matrix, the method further includes:

[0066] The EEG data is filtered to remove eye movement and myoelectric artifacts in the EEG data using a preset method.

[0067] In this embodiment, EEG data is preprocessed to obtain more accurate data that is easier to analyze and process. By configuring and connecting to the built-in modules provided by the OpenViBE platform, filtering can be performed based on research objectives, using independent component analysis (ICA) or other methods to remove artifacts such as eye movements and myoelectricity. Ultimately, the data is segmented into epochs for subsequent real-time EEG analysis and processing.

[0068] In an optional embodiment, based on pre-collected EEG data, the comprehensive importance score of each channel is calculated for each task, and a channel-task importance matrix is ​​constructed, including:

[0069] Use all channel data of EEG data to train the EEG signal processing model and obtain the feature weights of each layer;

[0070] For each task, the weight gradient information is calculated according to the feature weights of each layer, the importance score of each channel is obtained, and the channel-task importance matrix is ​​constructed.

[0071] In the embodiment of the present application, the EEG signal processing model can be any CNN network with a spatial filter layer, such as EEGNet, or a typical EEG classification network architecture such as Shallow ConvNet.

[0072] In an optional embodiment, for each task, weight gradient information is calculated based on the feature weights of each layer to obtain the importance score of each channel, and a channel-task importance matrix is ​​constructed, including:

[0073] For the current task, the validation set data is used to calculate the loss function, and the gradient of the feature weights of each layer to the loss function is obtained through back propagation;

[0074] Based on the gradient and feature weight of each layer, the channel importance score of each channel in each layer is calculated;

[0075] Calculate the comprehensive importance score of each channel according to the channel importance score of each channel in each layer;

[0076] According to the comprehensive importance score of each channel obtained under each task, a channel-task importance matrix is ​​constructed.

[0077] In the embodiment of the present application, the above-mentioned pre-processed full-channel data can be used to train the EEG signal processing model. After the training is completed, the weight parameters of the model are saved to prepare for the subsequent channel selection. If necessary, the learning rate or the number of training rounds needs to be adjusted to ensure that the network can fully learn the spatial feature information in the full-channel data. The feature weights of each layer are extracted from the trained model, including the weight parameters of multiple levels such as the spatial convolution layer and the temporal convolution layer. Based on the feature weights of each layer, the weight gradient information is calculated by back propagation of the loss function to construct the channel-task importance matrix.

[0078] In the embodiment of the present application, first, for each task category t, the loss function L is calculated using the validation set data. Then, the weights of each layer are obtained by back propagation. W Gradient of the loss function ∇W For the channel i In the layer j The gradient importance score in can be expressed as:

[0079]

[0080] Afterwards, based on the weight gradient information obtained previously, the comprehensive importance score of each channel is calculated:

[0081]

[0082] This step comprehensively considers the contribution of multiple layers of features and balances the importance of different layers through the weight coefficient α. The higher the score, the greater the influence of the channel on the model output.

[0083] In the embodiment of the present application, channel importance analysis is performed for each task. Constructing a channel-task importance matrix M ,in M [ i , t ] indicates channel i Task t This allows us to identify the channels that are most critical to a specific task, as well as channels that are important across multiple tasks.

[0084] In an optional embodiment, the channel selection strategy includes:

[0085] The global strategy selects one or more channels from all tasks in descending order of comprehensive importance scores;

[0086] Task-specific strategies select one or more channels for a specific task in descending order of comprehensive importance scores;

[0087] The hybrid strategy selects one or more channels shared by multiple tasks in descending order of comprehensive importance scores, and selects one or more channels for specific tasks.

[0088] In the embodiments of the present application, any one of the above channel selection strategies can be flexibly selected according to the actual application scenario to implement optimization solutions for different application scenarios.

[0089] In an optional embodiment, determining a channel selection strategy and selecting an optimal channel subset from a channel-task importance matrix according to the selected channel selection strategy includes:

[0090] Determine the channel selection strategy from global strategy, task-specific strategy and hybrid strategy;

[0091] Determine a channel selection scheme based on the selected channel selection strategy;

[0092] The optimal channel subset is obtained by selecting from the channel-task importance matrix according to the channel selection scheme.

[0093] In an optional embodiment, the channel selection scheme is any one of the following:

[0094] Solution 1: Predetermine the target number of channels K, select the K channels with the highest importance scores based on the selected channel selection strategy, and generate the optimal channel subset. This approach is intuitive and easy to control the model scale.

[0095] Solution 2: Set an importance threshold and select all channels with importance scores higher than the threshold based on the selected channel selection strategy to generate the optimal channel subset. This approach can adaptively retain the truly important channels.

[0096] Solution 3: Determine the constraints. Under the premise of satisfying the constraints, select the optimal channel combination based on the selected channel selection strategy to generate the optimal channel subset. This approach considers actual constraints such as computing resources and device limitations, transforms channel selection into an optimization problem, and selects the optimal channel combination under the premise of satisfying the constraints. The specific implementation of this solution is as follows:

[0097] The first step is to build constraints. From the perspective of computing resource constraints, factors such as the hardware platform's memory and CPU computing power must be considered. For example, if the hardware platform has limited memory, the amount of data storage involved in the channel selection process can be used as a constraint. Assume that the memory size occupied by each channel data is m The upper limit of the memory that the hardware platform can provide for storing channel data is M max , then the constraints can be expressed as , here S Represents the set of channels to be selected, m i It is a channel i Data occupies memory. From the perspective of device limitations, it needs to be constructed according to the characteristics of the brain-computer interface device used. For example, the sampling frequency limit of the device, if the maximum sampling frequency of the device is f max , and different channel combinations may affect the overall data sampling frequency f , then the constraint can be set to f ≤ f max .

[0098] The second step is to optimize and select the best channel combination. First, define the objective function, with the balance between the channel importance score and the constraints satisfied as the goal. i The importance score is M [ i , t ] (t is the task), the objective function F ( S ) can be defined as ,in is the penalty term for violating the constraint, λ is the balance coefficient, which is used to adjust the weight between importance score and constraint satisfaction.

[0099] In the optimization algorithm selection, a greedy algorithm can be used, starting from the channel with the highest importance score, and adding channels to the candidate set one by one.S Each time a channel is added, it is checked whether all constraints are met. If not, the channel is abandoned and the next channel with a higher score is added until no more channels can be added. The candidate set at this time is S It is the optimal channel combination.

[0100] In an optional embodiment, optimizing the EEG signal processing model according to the optimal channel subset includes:

[0101] In the EEG data, data related to the optimal channel subset is selected to train the EEG signal processing model.

[0102] This can be achieved by freezing the parameters of the network's main layers and adjusting only the weights associated with the selected channels. Fine-tuning uses a small learning rate (typically 0.1 or 0.01 of the initial learning rate) to avoid disrupting the model's learned feature representations. Training is also done using the original training data, but retaining only the selected channels. This allows the model to adapt to the reduced channel configuration.

[0103] The embodiment of the present application uses the dual means of channel selection preprocessing and model optimization to significantly reduce the amount of calculation while ensuring accuracy, reduce the demand for hardware resources, and improve the real-time performance of the system on portable devices.

[0104] In an optional embodiment, after optimizing the EEG signal processing model according to the optimal channel subset, the method further includes:

[0105] The performance of the optimized EEG signal processing model was evaluated on the validation set.

[0106] Evaluation metrics include: classification accuracy, ensuring that channel optimization does not significantly degrade model performance; channel compression ratio, calculating the ratio of the number of retained channels to the original number of channels; inference time, verifying the real-time performance of the optimized model; and resource usage, evaluating the degree of reduction in memory usage and computational load.

[0107] In an optional embodiment, the optimized EEG signal processing model is deployed to the OpenViBE platform for real-time EEG analysis and processing, including:

[0108] The optimized model is deployed to the OpenViBE platform for real-time EEG analysis and processing.

[0109] Deploying the optimized model to the OpenViBE platform requires leveraging its Python scripting integration capabilities. This is primarily accomplished by creating a Python script that inherits from the OVBox class. Within the script, appropriate input and output interfaces must be defined to handle signal input and classification output, and the necessary initialization, processing, and termination functions must be implemented. During the initialization phase, the optimized model must be loaded, signal preprocessing parameters must be set, and a channel selection strategy corresponding to the optimized channel subset must be configured. During the signal processing phase, the script receives real-time EEG signals, performs necessary data format conversion, invokes the model for prediction, and outputs classification results. After scripting is complete, a Python Scripting Box is added to OpenViBE Designer, the script path is configured, and the corresponding signal input and output streams are connected. Finally, system debugging is performed to verify the data flow path, confirm real-time processing performance, optimize processing latency, and monitor resource usage to ensure the model runs stably and reliably on the OpenViBE platform.

[0110] In an optional embodiment, before deploying the optimized EEG signal processing model to the OpenViBE platform, the method further includes:

[0111] Configure the relevant parameters of the OpenViBE platform, including sampling rate, channel data, and data format.

[0112] Configure the above parameters to ensure that the OpenViBE platform can correctly receive and process EEG data. After the configuration is completed, it is necessary to test to ensure that the EEG data can be correctly transmitted from the EEG acquisition device to the OpenViBE platform.

[0113] In an optional embodiment, after deploying the optimized EEG signal processing model to the OpenViBE platform, the method further includes:

[0114] The classification results obtained through real-time EEG analysis and processing on the OpenViBE platform are converted into control instructions, and the control instructions are output to the specified device to instruct the specified device to perform the corresponding operation.

[0115] There are two main approaches to implementing peripheral control on the OpenViBE platform: one is to use a network protocol (TCP / IP or UDP) to convert classification results into control instructions in a Python script and send them via a network socket to a control program running on a local or remote device. The control program receives and parses these instructions and then performs the corresponding device operations. The other is to use the LSL (Lab Streaming Layer) interface to send control instructions as a data stream through OpenViBE's LSL output interface. An LSL client program is written on the receiving end to receive these instructions and control the designated device. To ensure real-time and reliable control, it is necessary to establish a stable communication mechanism, including error handling and status feedback. At the same time, an appropriate control instruction buffering strategy should be set to avoid misoperation caused by signal jitter. In addition, corresponding control logic and safety protection mechanisms should be designed to meet the specific requirements of different types of peripherals.

[0116] For example, in the embodiment of the present application, the OpenViBE platform is used to perform real-time EEG analysis and processing, and the specific implementation steps of peripheral control are as follows: Figure 2 As shown:

[0117] Step 1: Use EEG data acquisition equipment to collect EEG data;

[0118] Step 2: Drive the OpenViBE platform to receive EEG data. In the embodiment of the present application, the OpenViBE platform is configured to receive and process data from the EEG acquisition device.

[0119] Step 3: preprocessing the EEG data to remove eye movement and myoelectric artifacts in the EEG data;

[0120] Step 4: Real-time model optimization and deployment, i.e., optimizing the EEG signal processing model through steps 101 to 103 above, and in step 104, deploying the optimized EEG signal processing model on the OpenViBE platform to implement real-time EEG analysis and processing on the OpenViBE platform;

[0121] Step 5: Peripheral control, which converts the classification results obtained through real-time EEG analysis and processing on the OpenViBE platform into control instructions, and outputs the control instructions to the specified device to instruct the specified device to perform the corresponding operation.

[0122] More specifically, if Figure 3 As shown, the real-time optimization and deployment of the EEG signal processing model in step 4 may include a data preparation stage, a feature analysis stage, and a model optimization stage.

[0123] During the data preparation phase:

[0124] a. Use all channel data to train the model; after training, save the model weight parameters to prepare for subsequent channel selection.

[0125] b. Multi-layer feature weight extraction: Extract the feature weights of each layer from the trained model.

[0126] c. Calculate weight gradient information: For each task t, use the validation set data to calculate the loss function L, then obtain the gradient ∇W of the feature weight W of each layer to the loss function through backpropagation, and determine the gradient importance score of channel i in layer j.

[0127] In the feature analysis phase:

[0128] d. Calculate gradient-weighted channel importance: Based on the weight gradient information obtained in step c, calculate the comprehensive importance score of each channel.

[0129] e. Task-specific channel analysis: Perform channel importance analysis for each task and construct a channel-task importance matrix M.

[0130] f. Determine the channel selection strategy: Select the corresponding strategy from the global strategy, task-specific strategy, and hybrid strategy;

[0131] g. Select the optimal channel subset: Determine the final channel selection scheme based on the selected strategy, and select the optimal channel subset based on the final channel selection scheme;

[0132] During the model optimization phase:

[0133] h. Retrain the model using specific channels: Freeze the parameters of the main layers of the network and only adjust the weights related to the selected channels.

[0134] i. Evaluate model performance: Comprehensively evaluate the performance of the optimized model on the validation set.

[0135] j. Final model deployment: Deploy the optimized model to the OpenViBE platform.

[0136] Based on the above method, the embodiment of the present application can effectively solve the channel redundancy problem, significantly improve decoding accuracy, generalization ability and computing efficiency, and apply this method to build a more stable, more reliable and easier to use brain-computer interface system.

[0137] Based on the same principle as the method provided in the embodiment of the present application, the embodiment of the present application also provides an OpenViBE brain-computer interface decoding control system driven by an efficient channel selection algorithm, such as Figure 4 As shown, the system includes:

[0138] The channel importance calculation module 401 is used to calculate the comprehensive importance score of each channel for each task based on the pre-collected EEG data, and construct a channel-task importance matrix.

[0139] In the embodiments of the present application, an EEG acquisition device (e.g., an EEG cap) may be used to collect EEG data from the user's scalp. The choice of acquisition device depends on the specific needs of the research, such as the required number of channels, sampling rate, and anti-interference capability.

[0140] Based on the collected EEG equipment, by constructing a channel-task importance matrix to evaluate the comprehensive importance of each channel under each task, the key channels can be effectively identified and data redundancy can be reduced.

[0141] The channel selection module 402 is configured to determine a channel selection strategy and select an optimal channel subset from the channel-task importance matrix according to the selected channel selection strategy.

[0142] The model optimization module 403 is used to optimize the EEG signal processing model according to the optimal channel subset.

[0143] In the embodiment of the present application, different channel selection strategies can be selected for different application scenarios, and then the optimal channel subset suitable for the application scenario can be selected. The EEG signal processing model is optimized by the optimal channel subset obtained under a specific strategy, which can be applied to different task requirements and improve the versatility of the model.

[0144] The model deployment module 404 is used to deploy the optimized EEG signal processing model to the OpenViBE platform for real-time EEG analysis and processing.

[0145] In an embodiment of the present application, based on the pre-collected EEG data, the comprehensive importance score of each channel is calculated for each task, and a channel-task importance matrix is ​​constructed. Then, the optimal channel subset is selected from the channel-task importance matrix, and the EEG signal processing model is optimized according to the optimal channel subset. The optimized EEG signal processing model is deployed to the OpenViBE platform for real-time EEG analysis and processing. While ensuring accuracy, the amount of calculation can be significantly reduced, the demand for hardware resources is reduced, and the real-time performance of the model on portable devices is improved. In addition, the corresponding channel selection strategy can be flexibly selected according to the specific application scenario, and the optimal channel subset can be selected from the channel-task importance matrix according to the selected channel selection strategy, which significantly improves the versatility of the system.

[0146] In an optional embodiment, the channel importance calculation module 401 includes:

[0147] The first channel importance calculation submodule is used to train the EEG signal processing model using all channel data of the EEG data to obtain the feature weights of each layer;

[0148] The second channel importance calculation submodule is used to calculate the weight gradient information for each task based on the feature weights of each layer, obtain the importance score of each channel, and construct a channel-task importance matrix.

[0149] In an optional embodiment, the second channel importance calculation submodule includes:

[0150] The first channel importance calculation unit is used to calculate the loss function for the current task using the validation set data, and obtain the gradient of the feature weights of each layer to the loss function through back propagation;

[0151] The second channel importance calculation unit is used to calculate the channel importance score of each channel in each layer based on the gradient and feature weight of each layer;

[0152] A third channel importance calculation unit is used to calculate the comprehensive importance score of each channel according to the channel importance score of each channel in each layer;

[0153] The fourth channel importance calculation unit is used to construct a channel-task importance matrix based on the comprehensive importance score of each channel obtained under each task.

[0154] In an optional embodiment, the channel selection strategy includes:

[0155] The global strategy selects one or more channels from all tasks in descending order of comprehensive importance scores;

[0156] Task-specific strategy, which selects one or more channels for a specific task in descending order of comprehensive importance scores;

[0157] The hybrid strategy selects one or more channels shared by multiple tasks in descending order of comprehensive importance scores, and selects one or more channels for specific tasks.

[0158] In an optional embodiment, the channel selection module 402 includes:

[0159] a first channel selection submodule, configured to determine a channel selection strategy from a global strategy, a task-specific strategy, and a hybrid strategy;

[0160] A second channel selection submodule is used to determine a channel selection scheme based on the selected channel selection strategy;

[0161] The third channel selection submodule is used to select the optimal channel subset from the channel-task importance matrix according to the channel selection scheme.

[0162] In an optional embodiment, the channel selection scheme is any one of the following:

[0163] Solution 1: predetermine the number of target channels K, select the K channels with the highest importance scores based on the selected channel selection strategy, and generate the optimal channel subset;

[0164] Solution 2: Set an importance threshold and select all channels with importance scores higher than the importance threshold according to the selected channel selection strategy to generate the optimal channel subset;

[0165] Option three: determine the constraints, and on the premise of satisfying the constraints, select the optimal channel combination according to the selected channel selection strategy to generate the optimal channel subset.

[0166] In an optional embodiment, the model optimization module 403 includes:

[0167] The model optimization submodule is used to select data related to the optimal channel subset in the EEG data to train the EEG signal processing model.

[0168] In an optional embodiment, the system further includes:

[0169] The EEG preprocessing module is used to filter the EEG data and remove eye movement and myoelectric artifacts from the EEG data using a preset method.

[0170] In an optional embodiment, the system further includes:

[0171] The OpenViBE driver module is used to configure the relevant parameters of the OpenViBE platform, including sampling rate, channel data, and data format.

[0172] In an optional embodiment, the system further includes:

[0173] The peripheral control module is used to convert the classification results obtained through real-time EEG analysis and processing on the OpenViBE platform into control instructions, and output the control instructions to the specified device to instruct the specified device to perform the corresponding operation.

[0174] For example, in the embodiment of the present application, the OpenViBE platform is used to perform real-time EEG analysis and processing, and the specific system for peripheral control is as follows: Figure 5 As shown:

[0175] EEG acquisition device 501, used to collect EEG data;

[0176] OpenViBE driver module 502, used to drive the OpenViBE platform to receive EEG data; in the embodiment of the present application, the OpenViBE platform is configured to receive and process data from the EEG acquisition device;

[0177] An EEG preprocessing module 503 is used to preprocess the EEG data to remove eye movement and myoelectric artifacts in the EEG data;

[0178] A real-time model optimization and deployment module 504 is used to optimize the EEG signal processing model through the channel importance calculation module 401, the channel selection module 402, and the model optimization module, and deploy the optimized EEG signal processing model on the OpenViBE platform through the model deployment module 404 to implement real-time EEG analysis and processing on the OpenViBE platform;

[0179] The peripheral control module 505 is used to convert the classification results obtained through real-time EEG analysis and processing on the OpenViBE platform into control instructions, and output the control instructions to the designated device to instruct the designated device to perform corresponding operations.

[0180] The OpenViBE brain-computer interface decoding control system driven by the efficient channel selection algorithm provided in the embodiment of the present application can achieve Figures 1 to 4 To avoid repetition, the various processes implemented in the method embodiment will not be described again here.

[0181] The OpenViBE brain-computer interface decoding control system driven by the efficient channel selection algorithm of the embodiment of the present application can execute the OpenViBE brain-computer interface decoding control method driven by the efficient channel selection algorithm provided in the embodiment of the present application. The implementation principle is similar. The actions performed by each module and unit in the OpenViBE brain-computer interface decoding control system driven by the efficient channel selection algorithm in each embodiment of the present application correspond to the steps in the OpenViBE brain-computer interface decoding control method driven by the efficient channel selection algorithm in each embodiment of the present application. For the detailed functional description of each module of the OpenViBE brain-computer interface decoding control system driven by the efficient channel selection algorithm, please refer to the description in the corresponding OpenViBE brain-computer interface decoding control method driven by the efficient channel selection algorithm shown in the previous text, which will not be repeated here.

[0182] The above description is merely a preferred embodiment of the present application and an illustration of the technical principles employed. Those skilled in the art should understand that the scope of the disclosure herein is not limited to technical solutions formed by specific combinations of the aforementioned technical features. It also encompasses other technical solutions formed by any combination of the aforementioned technical features or their equivalents, without departing from the aforementioned disclosure. For example, a technical solution formed by replacing the aforementioned features with (but not limited to) technical features with similar functions disclosed in this application.

Claims

1. An OpenViBE brain-computer interface decoding control method driven by an efficient channel selection algorithm, characterized in that: The method comprises: Based on pre-collected EEG data, a comprehensive importance score of each channel is calculated for each task, and a channel-task importance matrix is ​​constructed, including: using all channel data of the EEG data to train an EEG signal processing model to obtain feature weights of each layer; For each task, weight gradient information is calculated based on the feature weights of each layer to obtain the importance score of each channel, and the channel-task importance matrix is ​​constructed, including: for the current task, using the validation set data to calculate the loss function, and obtaining the gradient of the feature weights of each layer to the loss function through back propagation; based on the gradients and feature weights of each layer, the channel importance score of each channel in each layer is calculated; based on the channel importance score of each channel in each layer, the comprehensive importance score of each channel is calculated; and based on the comprehensive importance score of each channel obtained under each task, the channel-task importance matrix is ​​constructed; Determining a channel selection strategy, and selecting an optimal channel subset from the channel-task importance matrix according to the selected channel selection strategy, wherein the channel selection strategy includes: a global strategy, selecting one or more channels from all tasks in descending order of comprehensive importance scores; a task-specific strategy, selecting one or more channels for a specific task in descending order of comprehensive importance scores; and a hybrid strategy, selecting one or more channels shared by multiple tasks and selecting one or more channels for a specific task in descending order of comprehensive importance scores; Optimizing the EEG signal processing model according to the optimal channel subset; The optimized EEG signal processing model is deployed to the OpenViBE platform for real-time EEG analysis and processing.

2. The OpenViBE brain-computer interface decoding control method driven by the efficient channel selection algorithm according to claim 1 is characterized in that: Determining a channel selection strategy and selecting an optimal channel subset from the channel-task importance matrix according to the selected channel selection strategy includes: determining a channel selection strategy from among the global strategy, the task-specific strategy, and the hybrid strategy; Determine a channel selection scheme based on the selected channel selection strategy; The optimal channel subset is obtained by selecting from the channel-task importance matrix according to the channel selection scheme.

3. The OpenViBE brain-computer interface decoding control method driven by the efficient channel selection algorithm according to claim 2 is characterized in that: The channel selection scheme is any one of the following: Solution 1: predetermine the target number of channels K, select the K channels with the highest importance scores according to the selected channel selection strategy, and generate the optimal channel subset; Solution 2: Set an importance threshold, select all channels with importance scores higher than the importance threshold according to the selected channel selection strategy, and generate the optimal channel subset; Solution three: determine the constraints, and on the premise that the constraints are met, select the optimal channel combination according to the selected channel selection strategy to generate the optimal channel subset.

4. The OpenViBE brain-computer interface decoding control method driven by the efficient channel selection algorithm according to claim 1 is characterized in that: Optimizing the EEG signal processing model according to the optimal channel subset includes: In the EEG data, data related to the optimal channel subset is selected to train the EEG signal processing model.

5. The OpenViBE brain-computer interface decoding control method driven by the efficient channel selection algorithm according to claim 1 is characterized in that: Before calculating the comprehensive importance score of each channel for each task based on the pre-collected EEG data and constructing the channel-task importance matrix, the method further includes: The EEG data is filtered, and eye movement and myoelectric artifacts in the EEG data are removed using a preset method.

6. The OpenViBE brain-computer interface decoding control method driven by the efficient channel selection algorithm according to claim 1 is characterized in that: Before deploying the optimized EEG signal processing model to the OpenViBE platform for real-time EEG analysis and processing, the method further includes: Configure the relevant parameters of the OpenViBE platform, including sampling rate, channel data, and data format.

7. An OpenViBE brain-computer interface decoding control system driven by an efficient channel selection algorithm, characterized in that: The system is implemented according to any one of claims 1 to 6, and the system comprises: The channel importance calculation module is used to calculate the comprehensive importance score of each channel for each task based on the pre-collected EEG data and construct a channel-task importance matrix; a channel selection module, configured to determine a channel selection strategy and select an optimal channel subset from the channel-task importance matrix according to the selected channel selection strategy; A model optimization module, configured to optimize the EEG signal processing model according to the optimal channel subset; The model application module is used to deploy the optimized EEG signal processing model to the OpenViBE platform for real-time EEG analysis and processing.

Images