Wheelchair control system based on motor imagery electroencephalogram signals and surface electromyogram signals

Abstract

The application provides a wheelchair control system based on motor imagery electroencephalogram signals and surface electromyogram signals, and relates to the technical field of brain-computer control, comprising a data acquisition module, a system mode selection module, a signal processing module and a decision module; the data acquisition module is used for acquiring physiological signals of a wheelchair user and sending the physiological signals to the system mode selection module; the system mode selection module is used for selecting a processing mode of the physiological signals; the signal processing module is used for pre-processing the physiological signals and extracting signal features for identification; the signal processing module sends an identification result to the decision module; the decision module is used for fusing multiple instruction information and converting the instruction information into a final wheelchair control instruction, and outputting the control instruction to the wheelchair. The application can improve the operation accuracy of the control system, reduce the fatigue of the user when performing motor imagery behavior, improve the user experience and the application performance of the wheelchair control system, and expand the user range of the brain-controlled wheelchair.

Description

Technical Field

[0001] This invention relates to the field of brain-computer interface technology, and more particularly to a wheelchair control system based on motor imagery EEG signals and surface electromyography signals. Background Technology

[0002] Brain-Computer Interface (BCI) technology provides a method for communication and control that replaces natural methods by measuring and processing neural activity signals from the brain. It establishes direct communication between the brain and external devices without relying on typical neural communication channels such as peripheral nerves and muscles. BCI systems analyze and process electroencephalogram (EEG) signals to identify the brain's intentions and communicate or control external devices based on the final recognition results. Due to the technical characteristics and advantages of BCI systems, initial research aimed to assist paralyzed and disabled individuals. Using BCI technology to control robots or engage in human-computer interaction can improve the quality of life for the elderly, paralyzed, and disabled.

[0003] Currently, domestic and international scholars mainly focus on three types of EEG signals: Event-Related Potential (ERP), Motor Imagery (MI), and Steady-State Visual Evoke Potential (SSVEP). Visual stimulation-based BCI robot control systems, such as ERP-BCI and SSVEP-BCI, can increase user fatigue during actual operation. Although visual stimulation-based BCI systems have more distinct EEG signal characteristics and higher resolution, users need to continuously focus on the stimulus interface, resulting in a poor user experience. MI-BCI, on the other hand, induces brain neural behavior through imagined movement, without requiring external stimulation. Although motor imagery is a skill that requires learning and training and has limitations such as a limited number of brain modal recognition categories, it offers advantages such as spontaneous EEG activity, independence from external stimuli, and significant EEG evoked phenomena. From the perspective of practical application and human-computer interaction, MI-BCI is more suitable for robot control. Currently, most research in the field of motor imagery brain-computer interfaces, both domestically and internationally, focuses on binary classification problems such as left / right hand recognition and four-class classification problems such as left / right hand recognition, foot recognition, and tongue recognition. However, due to factors such as the nonlinearity and non-stationarity of EEG signals, large individual and age differences, and strong randomness, the accuracy of EEG signal decoding and recognition is relatively low. Summary of the Invention

[0004] To address the aforementioned problems of poor user experience, low recognition rate, and poor application performance in existing brain-computer interface wheelchair control systems, this invention provides a wheelchair control system based on motor imagery EEG signals and surface electromyography (EMG) signals. This invention primarily employs a hybrid brain-computer interface control system composed of EEG and EMG signals. Multi-sensor pattern recognition results are fused through decision-making to output control commands, improving the operational accuracy of the control system and reducing user fatigue when performing motor imagery actions, thereby enhancing the user experience and the application performance of the wheelchair control system.

[0005] The technical means employed in this invention are as follows:

[0006] A wheelchair control system based on motor imagery EEG signals and surface electromyography signals includes: a data acquisition module, a system mode selection module, a signal processing module, and a decision-making module;

[0007] The data acquisition module is used to collect physiological signals from wheelchair users and send them to the system mode selection module; the physiological signals include motor imagery electroencephalogram (EEG) signals and gesture surface electromyography (EMG) signals.

[0008] The system mode selection module is used to select the processing mode of physiological signals. The processing modes include training mode and working mode. In the training mode, a classifier model is generated based on the user's personal data. In the working mode, the system mode selection module sends the physiological signals to the signal processing module.

[0009] The signal processing module is used to preprocess physiological signals and extract signal features for identification. It uses a support vector machine algorithm to train and classify the signal features to obtain the identification results. The signal processing module then sends the identification results to the decision module.

[0010] The decision module is used to integrate multiple instruction information and transform them into the final wheelchair control instructions, and output the control instructions to the wheelchair.

[0011] Furthermore, the signal processing module includes an electroencephalogram (EEG) signal processing module and a surface electromyography (SEMG) signal processing module. The EEG signal processing module processes the EEG signals of motor imagery, and the SEMG signal processing module processes the SEMG signals of gestures.

[0012] Furthermore, the buttons for the training mode and the working mode are located on the operation interface, which also includes a basic parameter setting button, a directional motion status indicator button, a central circular motion potential energy indicator button, and a four-segment motion speed indicator button.

[0013] Furthermore, in the aforementioned working mode, the physiological signals collected by the data acquisition module are processed by the signal processing module and the decision fusion module before being sent to the wheelchair, and corresponding prompts are displayed on the control panel.

[0014] Furthermore, the EEG signal processing module processes the motor imagery EEG signals through the following steps:

[0015] Signal preprocessing: The motor imagery EEG signal is subjected to rereference and independent component analysis to remove electrooculography artifacts;

[0016] Spatial feature extraction: The preprocessed motor imagery EEG signal was subjected to three bandpass filters at 8-12Hz, 12-30Hz and 8-30Hz to obtain three signal data segments; spatial filters were constructed for the three signal data segments respectively through a common spatial mode and spatial feature extraction and fusion were performed.

[0017] Recognition and classification: Support vector machine is used as a classifier to obtain the recognition and classification results of EEG signals.

[0018] Furthermore, the surface electromyography signal processing module processes the gesture surface electromyography signals by including the following steps:

[0019] Signal preprocessing; performing 50Hz notch filtering on the signal data of each channel;

[0020] Time-domain feature extraction; the time-domain features include the signal's mean, root mean square, skewness, waveform factor, peak factor, impulse factor, and margin factor, and the feature values ​​of each channel are fused to form a feature vector;

[0021] Recognition and classification; support vector machines are used as classifiers to obtain the recognition and classification results of EEG signals.

[0022] Furthermore, the identification and classification by the signal processing module includes the following steps:

[0023] The Support Vector Machine (SVM) algorithm is used for training and classification to obtain recognition results. The kernel function used is the Gaussian kernel, and the formula is as follows:

[0024]

[0025] Where: x i Let x represent the i-th sample. j Let represent the j-th sample, σ be the Gaussian kernel bandwidth, and σ > 0;

[0026] The classifier decision mode is selected as OvR mode to handle the four-class classification problem, that is, four SVM classifiers are constructed and the final recognition result is output by voting.

[0027] Furthermore, the training mode includes the following steps:

[0028] Complete the training mode parameter settings in the basic parameters, and save the parameter settings after the settings are complete. The training mode parameter settings include the number of training sessions and the duration of a single training session.

[0029] Click the training mode button to start training data collection;

[0030] Users can select training task buttons. Clicking the central circular motion potential energy indicator light will start the training task data collection. After the collection is completed, the central circular motion potential energy indicator light will turn off. When all single training tasks are completed, all button indicator lights will turn off, and the next training cycle will begin.

[0031] Furthermore, in the training mode, the feature selection optimization module based on genetic algorithm is selected to improve model performance. In this case, the EEG signal processing module uses a filter bank co-space pattern method based on rhythm features, including:

[0032] The fitness function calculation and the pattern recognition accuracy of the feature matrix are mapped together as follows:

[0033] Fit(f(x)) = Accuracy i

[0034] Accuracy i The result of ten-fold cross-validation given the feature matrix input for the i-th feature combination;

[0035] Individual genotype encoding uses binary encoding;

[0036] Determine the relevant parameters of the genetic algorithm, including population size, crossover probability, and mutation probability;

[0037] The feature matrix and label vector obtained after processing the EEG dataset are read as a preparatory part of the feature selection optimization process;

[0038] Begin the feature selection optimization process based on genetic algorithms.

[0039] Furthermore, the feature selection optimization process based on the genetic algorithm includes the following steps:

[0040] Construct the initial population, generate an initial group of a individuals and calculate the fitness value of the genotype of each individual in the group, construct a roulette wheel and select the high-quality individuals to add to the initial population;

[0041] Once the pre-set population size b is met, in order to improve search efficiency and optimization effect, individuals within the population have a high probability of crossover, and partial mapping crossover is performed based on the crossover probability Pc.

[0042] Perform single-point mutation operations based on the mutation probability Pv on a single gene of each individual genotype in the population.

[0043] Calculate the fitness values ​​of all individuals in the new population, and record the genotype and fitness value of the best individual. Replace the old population with the new population and start the next iteration of evolution. After the iteration is completed, obtain the final feature selection optimization results.

[0044] Compared with the prior art, the present invention has the following advantages:

[0045] This invention discloses a wheelchair control system based on motor imagery EEG signals and surface electromyography (EMG) signals. The system employs a hybrid brain-computer interface (BCI) control system composed of EEG and EMG signals. Multi-sensor pattern recognition results are fused through decision-making to output control commands, improving the accuracy of the control system and reducing user fatigue during motor imagery activities, thereby enhancing the user experience and the overall performance of the wheelchair control system. This invention utilizes a traditional machine learning-based MI (Motion Imagery) EEG signal decoding method. Compared to deep learning methods, this method has a shorter model training time and can generate a customized classifier model based on the user's individual physiological signal data, thus mitigating individual and age-related differences to some extent. Attached Figure Description

[0046] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0047] Figure 1 This is the overall system framework diagram of the present invention.

[0048] Figure 2 This is a schematic diagram of the user interface of the present invention.

[0049] Figure 3 This is a decision mapping diagram for the present invention.

[0050] Figure 4 This is a flowchart of the system workflow under the decision fusion of the present invention.

[0051] Figure 5 This is a flowchart of the training mode of the present invention.

[0052] Figure 6This is a flowchart illustrating the common-space mode of the filter bank based on rhythmic features of the present invention.

[0053] Figure 7 This is a diagram illustrating the feature selection optimization process based on genetic algorithms in this invention. Detailed Implementation

[0054] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other. The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0055] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The following description of at least one exemplary embodiment is merely illustrative and is in no way intended to limit the present invention or its application or use. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0056] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0057] Unless otherwise specifically stated, the relative arrangement, numerical expressions, and values ​​of the components and steps described in these embodiments do not limit the scope of the invention. It should also be understood that, for ease of description, the dimensions of the various parts shown in the drawings are not drawn to actual scale. Techniques, methods, and devices known to those skilled in the art may not be discussed in detail, but where appropriate, such techniques, methods, and devices should be considered part of the specification. In all examples shown and discussed herein, any specific values ​​should be interpreted as merely exemplary and not as limitations. Therefore, other examples of exemplary embodiments may have different values. It should be noted that similar reference numerals and letters in the following figures denote similar items; therefore, once an item is defined in one figure, it need not be further discussed in subsequent figures.

[0058] This invention provides a wheelchair control system based on motor imagery and surface electromyography signals, such as... Figure 1 As shown, it includes:

[0059] The data acquisition module is used to collect four types of motor imagery EEG signal data (left hand, right hand, tongue and both feet) and gesture surface electromyography signal data (relaxed posture, open hand gesture, number 2 gesture and fist gesture);

[0060] The user interface module is used to select the system operating mode, set basic parameters, and display the current wheelchair operating status.

[0061] The system mode selection module is used to select the data processing mode for the collected data. The training mode can generate a classifier model customized according to the user's personal data, while the working mode can use the existing classifier model to carry out the normal operation of the system and send wheelchair control signals based on the input signal.

[0062] The EEG signal processing module preprocesses the signal, extracts the cosmic pattern features of the signal, uses the SVM algorithm for training and classification, and obtains the recognition results.

[0063] The surface electromyography signal processing module preprocesses the signal, extracts the temporal features of the signal, uses the SVM algorithm for training and classification, and obtains the recognition results.

[0064]

[0065] The SVM kernel function is chosen to be a Gaussian kernel, as shown in the formula above, where x i Let x represent the i-th sample. j Let represent the j-th sample, σ be the Gaussian kernel bandwidth, and σ > 0.

[0066] The classifier decision mode is selected as OvR mode (one pair of others) to address the problem in this invention, which is to handle the four-class classification problem, that is, to construct four SVM classifiers and output the final recognition result by voting.

[0067] The decision fusion control module, through the construction of the decision layer, fuses the modal results and transforms them into the final wheelchair control command output.

[0068] Furthermore, the data acquisition module includes the following components:

[0069] EEG signal data acquisition includes setting the amplifier parameters, lead electrode settings, and sampling frequency settings of the EEG signal acquisition device;

[0070] Surface electromyography (EMG) signal data acquisition, including parameter settings for the EMG signal acquisition device.

[0071] Furthermore, the system's user interface module includes basic parameter settings, wheelchair status display, and an operation interface:

[0072] The selection of working modes includes working mode and training mode.

[0073] Basic parameter settings, including basic parameter settings for EEG signal acquisition devices and surface electromyography signal acquisition devices, as well as relevant parameter settings for training modes;

[0074] The wheelchair status display and user interface are used for the daily operation of the wheelchair control system, such as... Figure 2 As shown, it includes three function buttons: training mode, working mode, and basic parameter settings; four triangular directional motion status indicator lights; a central circular motion potential energy indicator light; and a four-segment motion speed indicator light. In working mode, the indicator lights are inactive and only provide a prompting function. In training mode, the indicator lights are active and used to select the input of training data status labels, providing a prompting function.

[0075] Furthermore, the system mode selection module includes a training mode and a working mode. The system defaults to working mode upon startup. In this mode, the system operates normally, reading data from the acquisition device, processing it through the signal processing module and the decision fusion module, and then sending control commands to the wheelchair. Corresponding prompts are displayed on the control panel. In training mode, users can select the appropriate prompt button according to the training plan and parameter settings to start inputting training data. After all training data has been input, the system processes the dataset, trains the classifier model, and saves it. Specifically, as follows... Figure 5 As shown.

[0076] In training mode, you first need to complete the training mode parameter settings in the basic parameters, including the number of training sessions and the duration of a single training session (minimum duration is 10 seconds). After setting, save the parameter settings. Return to the main interface and click the training mode button to start training data collection. In each training session, you can choose the training task button (i.e., the triangular directional motion status indicator light and the four-segment motion speed indicator light). Clicking the central circular motion potential energy indicator light will start the training task data collection. After the collection is completed, the central circular motion potential energy indicator light will turn off. When all single training tasks are completed, all button indicator lights will turn off, and the next training cycle will begin.

[0077] In training mode, the feature selection optimization module based on a genetic algorithm can be enabled to improve model performance. When this function is enabled, the EEG signal processing module uses the rhythm feature-based filter bank common space pattern method (RF-FBCSP), and the algorithm flow is as follows: Figure 6 As shown.

[0078] The feature selection optimization process based on genetic algorithms is as follows: Figure 7As shown. The process of feature selection optimization is equivalent to solving a problem of maximizing an objective function. In a specific problem, the objective function result is the recognition accuracy of a certain feature matrix. Therefore, the calculation of the fitness function and the pattern recognition accuracy of the feature matrix are mapped together as follows:

[0079] Fit(f(x)) = Accuracy i

[0080] Accuracy i The result of ten-fold cross-validation is given as the input of the feature matrix of the i-th feature combination.

[0081] Due to the fitness function mapping relationship and the characteristics of the data feature matrix, binary encoding is used for individual genotype coding. This facilitates subsequent genetic operations and avoids the risk of causing a Hamming cliff when solving the feature selection optimization problem. The feature selection optimization process begins by determining relevant parameters of the genetic algorithm, including population size, crossover probability, and mutation probability. Next, the feature matrix and label vector obtained after processing the EEG dataset are read; this is the preparatory part of the feature selection optimization process. After the preparatory work is completed, the evolutionary process of the genetic algorithm begins, such as... Figure 7 As shown, the initial population is first constructed, generating an initial group of 'a' individuals and calculating the fitness value of each individual's genotype within the group. A roulette wheel is constructed, and high-performing individuals are selected and added to the initial population. Once the pre-set population size 'b' is met, to improve search efficiency and optimization, individuals within the population have a high probability of crossover. Partial mapping crossover is performed pairwise based on the crossover probability Pc. After the crossover operation, single-point mutation operations based on the mutation probability Pv are performed on the single genes of each individual's genotype within the population. After the crossover and mutation operations are completed, the fitness values ​​of all individuals in the new population are calculated, and the genotype and fitness value of the highest-performing individual are recorded. The new population replaces the old population, and the next iteration of evolution begins. After the iteration of evolution is completed, the final feature selection optimization result is obtained.

[0082] When using genetic algorithms for feature selection optimization, relevant control parameters also need to be set. To improve the computation speed and effect of feature selection optimization, the specific parameters are set as follows: initial group size is set to 5, population size is set to 400, mating probability is 60%, mutation probability is 2%, and number of iterations is 500.

[0083] Furthermore, the EEG signal processing module comprises three parts: signal preprocessing, spatial feature extraction, feature selection optimization, and recognition and classification. The multi-channel EEG signals undergo preprocessing to remove oculomotor artifacts using rereference and independent component analysis; the preprocessed signals are then subjected to three bandpass filters (8-12Hz, 12-30Hz, and 8-30Hz) to obtain three signal segments; spatial filters are constructed for each of the three signal segments using a common spatial pattern, and spatial feature extraction and fusion are performed, with the common spatial pattern's spatial filter bank constructed using a one-to-one (OvO) mode; a support vector machine is used as a classifier to obtain the recognition and classification results of the EEG signals.

[0084] Furthermore, the surface electromyography (EMG) signal processing module includes signal preprocessing, temporal feature extraction, and recognition and classification. Signal preprocessing involves applying a 50Hz notch filter to the signal data from each channel. Temporal features include the signal's mean, root mean square, skewness, waveform factor, peak factor, impulse factor, and margin factor; the feature values ​​from each channel are fused to form a feature vector. A support vector machine is used as a classifier to obtain the EEG signal recognition and classification results.

[0085] Furthermore, the decision-making module includes a decision layer and wheelchair control command output. The decision layer primarily constructs decision-making rules to fuse the recognition and classification results received from parallel channels with the contained command information, transforming them into final wheelchair control commands. This decision fusion improves the wheelchair's execution accuracy and application performance. The decision layer consists of decision mapping and decision fusion. Decision mapping is the mapping and association between the physiological signal recognition results of user behavior and control commands, such as... Figure 3 As shown, the EEG signals of the user when performing motor imagery are processed by the processing module to obtain the recognition results of the user's imagined specific limb movements, which are mapped to the wheelchair's movement mode selection. The specific mapping relationship is as follows: imagined tongue movement - forward mode, imagined two-leg movement - backward mode, imagined left hand movement - left turn mode, and imagined right hand movement - right turn mode. The surface electromyography signals of the user when performing hand gestures are processed by the processing module to obtain the recognition results of the user's hand gestures, which are mapped to the wheelchair's speed mode as follows: relaxed posture - wheelchair stopped, open hand gesture - wheelchair running at low speed, digit 2 gesture - wheelchair running at medium speed, and clenched fist gesture - wheelchair running at high speed.

[0086] Decision fusion refers to the integration of two types of control commands. Within the decision-making layer architecture, based on the recognition results of signal data from parallel channels obtained from the current wheelchair status, a judgment is made. The specific workflow of the decision-making layer is as follows: Figure 4As shown, after the system is powered on, the default movement mode is forward, and the default speed selection is stationary. The system reads the recognition results from the surface electromyography (EMG) signal processing module in real time and makes a fusion decision based on the recognition results and the wheelchair's current movement state. If the recognition result is a relaxed posture-stationary state, the wheelchair immediately stops moving. If the recognition result is one of low, medium, or high speed movements, and the wheelchair is currently stationary, the wheelchair executes the movement under the current movement mode and speed selection mode; otherwise, it maintains the current movement state. If the wheelchair remains stationary for more than 3 seconds, it first maintains the stationary state, resets timer 1, and counts counter 2. Next, counter 2 is used to determine whether a forced adjustment to forward movement mode is needed. If counter 2 counts less than 2, it begins reading and processing 2 seconds of EEG signals, changing the movement mode based on the recognition results. If counter 2 counts twice, it means that neither of the two movement mode switching results is the user's desired mode, and a forced adjustment to forward movement mode is made. After the movement mode is modified, the surface EMG signal processing recognition results are read again. If the gesture recognition result still remains stationary for 3 seconds, the next movement mode selection is initiated. By setting the forced adjustment trigger threshold of counter 2, the motion state is adjusted to the most commonly used forward state, thereby improving the performance of practical applications.

[0087] This invention utilizes ten 10-fold cross-validation based on the training dataset of the 4th Brain-Computer Interface Competition 2a. The results are shown in Table 1. The technical advantage of this invention compared to other algorithms lies in its lower model building and parameter calculation time and cost compared to large-scale algorithm models within the deep learning framework. It can also build customized models using individual user data, thus mitigating the problem of significant differences in physiological signal data among users to some extent.

[0088] Table 1. Results of tenfold cross-validation of EEG data related to motor imagery.

[0089]

[0090] Table 2 shows the classification results of various methods based on the 4th Brain-Computer Interface Competition 2a dataset. FBCSP belongs to traditional machine learning methods, while ATCNet and HDNN-TL belong to deep learning methods. Compared with FBCSP, the RF-FBCSP method studied in this paper has better validation results. The kappa value of HDNN-TL reflects the advantages of its deep learning algorithm, indicating good model performance, but the model is relatively complex. ATCNet's recognition accuracy is lower than the method studied in this paper when considering inter-subject differences, but its recognition accuracy is higher (85.38%) when not considering inter-subject differences. The large difference between these two cases may be due to the influence of dataset size and model capacity. In the case of small sample datasets, the RF-FBCSP method studied in this paper has certain advantages, meeting the need to generate personalized models using user-specific data and mitigating the impact of inter-subject differences to some extent.

[0091] Table 2. Classification results of various methods based on the 4th Brain-Computer Interface Competition 2a dataset.

[0092]

[0093] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A wheelchair control system based on motor imagery electroencephalogram signals and surface electromyogram signals, characterized in that, The system comprises a data acquisition module, a system mode selection module, a signal processing module and a decision module. The data acquisition module is used for collecting physiological signals of a wheelchair user and sending the physiological signals to the system mode selection module; the physiological signals include motor imagery electroencephalogram signals and gesture surface electromyogram signals. The system mode selection module is used for selecting a processing mode of the physiological signals; the processing mode includes a training mode and a working mode; in the training mode, a classifier model is generated according to personal data of the user; in the working mode, the system mode selection module sends the physiological signals to the signal processing module. The training mode comprises the following steps: Basic parameter setting in the training mode is completed; after the setting is completed, the parameter setting is saved; the training mode parameter setting includes the number of training times and the duration of single training; A training mode button is clicked to start training data acquisition; A training task button can be selected autonomously; when a central circular motion potential prompt light is clicked, training task data acquisition is started; after the acquisition is completed, the central circular motion potential prompt light is extinguished; when all single training tasks are completed, all button indicator lights are extinguished, and the next training period is started; In the training mode, a feature selection optimization function module based on a genetic algorithm is selected to improve the performance of the model; at this time, the electroencephalogram signal processing module uses a filter bank common spatial pattern method based on rhythm characteristics, which comprises the following steps: The calculation of the fitness function is mapped to the pattern recognition accuracy of the feature matrix, and the relationship is as follows: Binary encoding is used for individual genotype coding; wherein is the result of a ten-fold cross-validation for the feature matrix input of the i combination of features; Genetic algorithm related parameters are determined, including population size, crossover probability and mutation probability; The feature matrix and label vector obtained after processing of the electroencephalogram data set are read as a preparation part in the feature selection optimization process; The feature selection optimization process based on the genetic algorithm is started; The feature selection optimization process based on the genetic algorithm comprises the following steps: An initial population is constructed, an initial group of a individuals is generated, and the fitness value of each individual genotype in the group is calculated; a roulette wheel is constructed, and high-quality individuals are selected to join the initial population; When the preset population size b is met, the search efficiency and optimization effect are improved by making the individuals in the population have a high probability of crossover; based on the crossover probability Pc, partial mapping crossover is performed in pairs; Single-point mutation operation is performed on the single-point gene of each individual genotype in the population based on the mutation probability Pv; The fitness values of all individuals in the new population are calculated, and the genotype and fitness value of the best individual are recorded; the new population replaces the old population, and the next iteration evolution is started; after the iteration evolution is completed, the final feature selection optimization result is obtained; The buttons of the training mode and the working mode are arranged on an operation interface; the operation interface is further provided with a basic parameter setting button, a directional motion state prompt light button, a central circular motion potential prompt light button and a four-section motion speed prompt light button. ​ The signal processing module is used for pre-processing the physiological signals and extracting signal features for recognition, training and classifying the signal features using a support vector machine algorithm, and obtaining a recognition result; the signal processing module sends the recognition result to the decision module; The signal processing module includes an electroencephalogram signal processing module and a surface electromyogram signal processing module, the electroencephalogram signal processing module processes motor imagery electroencephalogram signals, and the surface electromyogram signal processing module processes gesture surface electromyogram signals; The electroencephalogram signal processing module processes motor imagery electroencephalogram signals, including the following steps: Signal pre-processing: the motor imagery electroencephalogram signals are subjected to re-reference and independent component analysis to remove eye movement artifacts; Spatial feature extraction: the motor imagery electroencephalogram signals after signal pre-processing are subjected to 8-12 Hz, 12-30 Hz and 8-30 Hz three-band pass filtering to obtain three segments of signal data; the three segments of signal data are respectively constructed into spatial filters by common spatial pattern and subjected to spatial feature extraction fusion; Recognition and classification: a support vector machine is used as a classifier to obtain a recognition and classification result of the electroencephalogram signals; The surface electromyogram signal processing module processes gesture surface electromyogram signals, including the following steps: Signal pre-processing: the signal data of each channel is subjected to 50 Hz notch filtering processing; Time domain feature extraction: the time domain features include mean, root mean square, skewness, waveform factor, peak factor, pulse factor and margin factor of the signals, and the feature values of each channel are fused to form a feature vector; Recognition and classification: a support vector machine is used as a classifier to obtain a recognition and classification result of the surface electromyogram signals; The recognition and classification of the signal processing module includes the following steps: SVM algorithm is used for training and classification to obtain a recognition result, wherein the selected kernel function is Gaussian kernel, and the formula is as follows: wherein: denotes the ith sample, denotes the jth sample, is a Gaussian kernel bandwidth, and ; The classifier decision mode is selected as OvR mode to process four classification problems, that is, four SVM classifiers are constructed, and the final recognition result is output by voting; The decision module is used to fuse and convert multiple instruction information into a final wheelchair control instruction, and output the control instruction to the wheelchair.

2. The wheelchair control system based on motor imagery electroencephalogram signals and surface electromyogram signals according to claim 1, characterized in that, In the working mode, the physiological signals collected by the data acquisition module are subjected to signal processing by the signal processing module and decision fusion by the decision fusion module, and control instructions are sent to the wheelchair, and corresponding prompts are displayed on the control panel.

Images