Brain-controlled nursing system and method based on EEG signal recognition

Abstract

The present disclosure relates to the technical field of electroencephalogram (EEG) signal processing and identification, and proposes a brain-controlled nursing system and method based on electroencephalogram (EEG) signal identification, including: obtaining the electroencephalogram (EEG) signal during motor imagery and preprocessing it; extracting time-frequency features and channel-space features from the preprocessed EEG signal and constructing a feature set; setting a self-attention module after the input layer of the residual graph convolutional network to obtain an improved residual graph convolutional network; fusing the features in the feature set based on the trained improved residual graph convolutional network, and then classifying them through a fully connected layer to obtain a motor imagery intention discrimination result, and outputting a control instruction based on the discrimination result to control the robotic arm to move. The present disclosure adopts a nursing system and method based on an improved residual graph convolutional network and a brain-controlled robotic arm, which improves the classification and recognition accuracy of motor imagery of EEG signals and can better realize the nursing of patients with movement disorders.

Description

Technical Field

[0001] The present disclosure relates to the technical field related to classification and recognition of motor imagery of EEG signals, and in particular to a brain-controlled nursing system and method based on EEG signal recognition. Background Art

[0002] The statements in this section merely provide background information related to the present disclosure and do not necessarily constitute prior art.

[0003] BCI system is the abbreviation of Brain-Computer Interface System. It is a technical system that aims to achieve direct interaction between the human brain and the computer by directly reading and analyzing human brain activities. By associating EEG signals with specific intentions or instructions, the BCI system can achieve functions such as controlling external devices and performing specific tasks. For example, people with movement disorders can use the BCI system to control the movement of prostheses or wheelchairs. The BCI system has broad application prospects in the medical field. For example, it can be used to help paralyzed patients recover their motor functions, improve the communication ability of aphasic patients, and even treat mental illnesses such as depression. BCI technology can help patients with severe paralysis due to stroke to take care of them. Due to the patient's motor disorder, accompanying personnel are required to take care of the patient, which greatly increases the cost of labor. The BCI system is used to manipulate the robotic arm to fully induce and promote brain plasticity, so that patients with severe paralysis due to stroke can better manipulate the robotic arm to assist patients in picking up items, achieving the function of machine care.

[0004] Electroencephalogram (EEG) is a graph obtained by amplifying and recording the spontaneous biopotential of the brain from the scalp through sophisticated electronic instruments. It can observe the functional reorganization of the cerebral cortex, show brain electrical activity, and functional network connections. Its time resolution is at the millisecond level, and it can be used as a brain signal in BCI. Motor imagery (MI) is a dynamic experience in which the user considers the mental imagery of movement without activating any muscles or peripheral nerves. The motor imagery brain-computer interface (MI-BCI) is a system that converts brain signals generated during this imagination process into actionable sequences. Preliminary studies have shown that stroke patients can operate the MI-BCI system based on electroencephalogram (EEG) as effectively as healthy subjects.

[0005] The inventors found in their research that the MI-BCI system based on electroencephalogram (EEG) needs to accurately identify the motor imagery of EEG signals in order to provide accurate human-brain interaction. Currently, the BCI technology has a low accuracy rate in EEG signal recognition in the paralyzed patient care system, and thus cannot accurately execute the human brain's intentions. Summary of the invention

[0006] In order to solve the above problems, the present invention proposes a brain-controlled nursing system and method based on EEG signal recognition, which adopts an improved residual graph convolutional network to improve the classification and recognition accuracy of motor imagery of EEG signals, thereby converting the classified and recognized labels into instructions for controlling the movement of the robotic arm, and better realizing the care of patients with movement disorders.

[0007] In order to achieve the above objectives, the present disclosure adopts the following technical solutions:

[0008] One or more embodiments provide a brain control method based on EEG signal recognition, comprising the following steps:

[0009] Obtain the EEG signals during motor imagery and perform preprocessing;

[0010] Extract time-frequency features and channel-space features from the preprocessed EEG signals;

[0011] Combine the time-frequency features with the channel space features in the adjacency matrix to construct a feature set;

[0012] A self-attention module is set after the input layer of the residual graph convolutional network to obtain an improved residual graph convolutional network. The features in the feature set are fused based on the trained improved residual graph convolutional network, and then classified through the fully connected layer to obtain the motion imagination intention discrimination result. According to the discrimination result, the control command is output to control the robotic arm to move.

[0013] One or more embodiments provide a brain-controlled nursing system based on EEG signal recognition, including: an EEG acquisition device and a portable EEG signal processor;

[0014] The EEG acquisition equipment is used to collect EEG signals during motor imagery;

[0015] The portable EEG signal processor is configured as the above-mentioned brain control method based on EEG signal recognition.

[0016] An electronic device includes a memory and a processor, and computer instructions stored in the memory and executed on the processor. When the computer instructions are executed by the processor, the steps of the above-mentioned brain control method based on electroencephalogram signal recognition are completed.

[0017] Compared with the prior art, the present invention has the following beneficial effects:

[0018] The present invention extracts the time-frequency and spatial features of EEG signals and fuses them to obtain comprehensive feature information. At the same time, the recognition network is improved to capture useful information in EEG signals and improve recognition accuracy. The improved residual graph convolutional network is used to improve the classification and recognition accuracy of motor imagery of EEG signals, thereby converting the classification and recognition labels into instructions for controlling the movement of the robotic arm, and better realizing the care of patients with movement disorders.

[0019] The advantages of the present disclosure and the advantages of additional aspects will be described in detail in the following specific embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] The accompanying drawings constituting a part of the present disclosure are used to provide a further understanding of the present disclosure. The exemplary embodiments of the present disclosure and the description thereof are used to explain the present disclosure but do not constitute a limitation of the present disclosure.

[0021] Figure 1 is a flow chart of the brain control method of Embodiment 1 of the present disclosure;

[0022] Figure 2 is an adjacency matrix graph generated by calculating the Pearson correlation coefficient in Example 1 of the present disclosure;

[0023] Figure 3 is a graph showing the accuracy of classification and recognition in Example 1 of the present disclosure;

[0024] Figure 4 is a residual block network structure diagram of Embodiment 1 of the present disclosure;

[0025] Figure 5 is an overall module diagram of the nursing system of Embodiment 2 of the present disclosure;

[0026] Figure 6 is a hardware system diagram of a nursing system according to Embodiment 2 of the present disclosure; DETAILED DESCRIPTION

[0027] The present disclosure is further described below in conjunction with the accompanying drawings and embodiments.

[0028] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present disclosure. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present disclosure belongs.

[0029] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit exemplary embodiments according to the present disclosure. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprising" and / or "including" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof. It should be noted that, in the absence of conflict, the various embodiments in the present disclosure and the features in the embodiments can be combined with each other. The embodiments will be described in detail below in conjunction with the accompanying drawings.

[0030] The present invention is used for patients who are severely paralyzed due to stroke and have completely lost their ability to act independently. In view of the high cost of hospital care, the inconvenience of communication between hospitalized patients and their families, and the low recognition accuracy in the prior art, a brain-controlled nursing system and method based on EEG signal recognition is proposed. The improved residual graph convolutional network is combined with the motor imagery (MI) BCI system based on electroencephalogram (EEG) to realize feature extraction and classification recognition of EEG signals of patients with severe paralysis due to stroke. The classified and recognized labels are converted into instructions for controlling the movement of the mechanical arm, thereby controlling the movement of the mechanical arm to help patients with severe paralysis due to stroke to pick up items. The nursing of patients with severe paralysis due to stroke is better realized. The system disclosed in the present invention is suitable for home environments and has the advantages of simple and safe operation, high cost performance and high accuracy of EEG signal recognition. Some severely paralyzed patients can receive care at home. The nursing system can install monitoring equipment at home to monitor the patient's vital signs in real time, assist the patient in daily life activities, and send the patient and the mechanical arm movement status to the remote guardian through a remote sending device. The following is an explanation with specific embodiments.

[0031] Example 1

[0032] In the technical solutions disclosed in one or more embodiments, Figures 1 to 4 As shown, a brain control method based on EEG signal recognition includes the following steps:

[0033] Step 1, obtaining the EEG signal during motor imagery and preprocessing it;

[0034] Step 2: Use the MST algorithm to extract the time-frequency features of the preprocessed EEG signal; use the Pearson correlation coefficient to calculate the spatial features and generate the adjacency matrix between channels;

[0035] Step 3: Combine the time-frequency features with the channel space features in the adjacency matrix to construct a feature set;

[0036] Step 4: Set a self-attention module after the input layer of the residual graph convolutional network to obtain an improved residual graph convolutional network. Based on the trained improved residual graph convolutional network, fuse the features in the feature set, and then classify them through the fully connected layer to obtain the motion imagination intention discrimination result. According to the discrimination result, output the control command to control the robotic arm to move.

[0037] In this embodiment, the time-frequency features and spatial features of the EEG signal are extracted, and feature fusion is performed to obtain comprehensive feature information. At the same time, the recognition network is improved to capture useful information in the EEG signal and improve the accuracy of recognition.

[0038] In step 1, specifically, the EEG signal of a stroke patient with severe paralysis during motor imagery can be collected; a Neuroscan 64-channel EEG signal acquisition system with a sampling frequency of 1000 Hz can be used to collect EEG signal data of a stroke patient with severe paralysis.

[0039] Among them, the electroencephalogram (EEG) signal represents the position of each electrode in the acquisition device, which represents a node in the brain network, and the number of electrodes is the number of nodes N. The frontal lobe is the most developed lobe in the brain. It includes the motor area, premotor area and prefrontal lobe. It is the main area of ​​the brain for performing MI tasks. In order to reduce the interference of irrelevant information, 21 electrodes can be selected for EEG acquisition;

[0040] The corresponding electrodes for EEG signal acquisition of Neuroscan 64-channel EEG signal acquisition system are: “Fp1”, “Fp2”, “F7”, “F3”, “Fz”, “F4”, “F8”, “T7”, “Cz”, “C4”, “T8”, “P7”, “P3”, “Pz”, “P4”, “P8”, “O1”, “O2” and “Oz”.

[0041] Optionally, preprocessing may include downsampling, bandpass filtering, and artifact removal of the EEG signal;

[0042] The collected data is filtered and downsampled. Filtering can not only remove noise such as artifacts, but also select a specific frequency band to observe the signal.

[0043] The data preprocessing process is as follows:

[0044] Step 11: Downsample the EEG signal from 1000 Hz to 100 Hz to reduce the amount of calculation;

[0045] Step 12, band-pass filter the EEG signal and divide it into α wave (8-13) Hz and β wave (13-30) Hz; segment the EEG data into 6 s data segments;

[0046] Since the patients who performed motor imagery only performed the MI task within the corresponding 6-s task period, the EEG data were segmented into 6-s segments for further analysis.

[0047] Step 13: Perform artifact removal; optionally, ±70 μv can be set as the artifact threshold.

[0048] In order to better analyze the time-frequency characteristics of EEG data, the MST algorithm is used to perform time-frequency processing on the EEG data in step 2. The MST algorithm introduces an adaptive parameter improvement algorithm based on the ST algorithm, and continuously optimizes the window width to calculate more accurate high-frequency energy.

[0049] In this embodiment, MST can be defined as:

[0050]

[0051] Where x(t) is the EEG signal, τ is the translation factor, g(τ-t,f) is the Gaussian window function used to control the position of the Gaussian function on the t time axis, f is the frequency, and j is the imaginary unit, which is defined as follows:

[0052]

[0053] Among them, σ1(f) is the standard deviation of the Gaussian function, which is defined as follows:

[0054]

[0055] MST can obtain the optimal energy concentration in the time-frequency domain by continuously adjusting the two parameters p and q. MST is used to extract the time-frequency features of EEG data in the appropriate frequency band, referred to as PSD features. PSD is defined as:

[0056]

[0057] Where E represents the mean square error of MST.

[0058] In order to better analyze the relationship between EEG signal channels, in step 2, the Pearson coefficient is used to calculate the connectivity between EEG signal channels. The Pearson correlation coefficient is calculated for the EEG signals of 21 channels, and the absolute value of the Pearson correlation coefficient r(x1, x2) is taken. The absolute value Pearson correlation coefficient is used to generate a 21*21 adjacency matrix as the spatial feature. The Pearson correlation coefficient r(x1, x2) is defined as follows:

[0059]

[0060] Where N is the number of time points, x1 and x2 are the channel data to be analyzed, and Represents the average value of the observed channel. The result r∈[-1,1], where -1 indicates a perfect negative correlation and 1 indicates a perfect positive correlation. A value of zero indicates no linear signal relationship.

[0061] In step 3, time-frequency features and spatial features are extracted from the EEG signal data to construct a feature set.

[0062] In step 4, the method for fusing the time-frequency features and spatial features in the feature set is:

[0063] The improved residual graph convolutional neural network is used to fuse time-frequency features and spatial features. The time-frequency features are used as node features, and the spatial structure features are used as the structure of the graph. The residual graph convolutional neural network is used to learn the propagation and fusion of time-frequency features in the spatial structure. In this way, the graph structure learning ability of GCN can be fully utilized to effectively fuse time-frequency features and spatial features to obtain a more comprehensive and accurate feature representation.

[0064] In this embodiment, the time, space and spectrum features of the EEG signal are extracted and fused to ensure that the effective features of the data can be better learned during the classification and recognition process of the research model. These features can help us better understand the changes in time, frequency and space of the EEG signal, so as to more accurately identify and classify different EEG signal patterns.

[0065] In order to enable the model to better process input data flexibly and capture important parts of the input data, the model of this embodiment introduces an attention mechanism to dynamically allocate attention weights during the processing process to improve model performance.

[0066] Specifically, the residual graph convolutional network is improved, an attention mechanism module is added to the residual graph convolutional network, and a self-attention module is set after the input layer of the residual graph convolutional network. The improved residual graph convolutional network architecture proposed includes four residual blocks and a fully connected layer. Each residual block includes three graph convolutional layers, three Relu activation layers and a pooling layer, and a self-attention layer is added after each convolutional layer to dynamically obtain the weight relationship between nodes. The residual block structure is as follows Figure 4 As shown;

[0067] The self-attention module is configured to calculate the weight of the input data and multiply the input data by the weight coefficient. Specifically: for the input feature vector X, the self-attention module will pass w Q 、w V 、w K The three trainable weight matrices map out three vectors Q, V, and K of dimension L to calculate the attention weight of self-attention. The weight calculation formula is as follows:

[0068]

[0069] First, dot products are used on Q and K to evaluate the correlation between different tags, and the results are passed through the softmax function to obtain a weighted matrix, namely the attention score; then, the attention score is dot-producted with V.

[0070] In the formula, T represents the transpose of the matrix, d K Represents the dimension of the vector matrix K. Q, V, and K represent Query, Value, and Key, respectively, and are obtained by linearly transforming the same input feature vector X. The specific transformation formula is:

[0071] Q=XW Q

[0072] V=XW V

[0073] K=XW K

[0074] Finally, the calculated weight is used as the coefficient of the corresponding input data;

[0075] In this embodiment, residual graph convolution with an attention mechanism is used to train data and algorithm models, and the feature-extracted EEG signals are used to identify the intention of motor imagination through a fully connected layer, and labels are output; the labels are converted into robotic arm control instructions, and the instructions control the robotic arm to move.

[0076] Furthermore, in this embodiment, a collaborative robotic arm is set as a motion control module, which is used to classify and identify EEG signals in real time and convert them into motion instructions. The motion instructions are related to the motion state of the robotic arm. Specific instructions control the robotic arm to produce different motion results, thereby helping severely paralyzed stroke patients to pick up objects.

[0077] Further technical solutions include the following steps to train the improved residual graph convolutional network:

[0078] Step S1, obtaining an EEG signal data set, including EEG signals and corresponding motor imagery intention classification labels, and performing preprocessing;

[0079] Optionally, the ratio of the training set to the test set is 9:1, and finally the classification accuracy of the training set and the test set is obtained to generate an offline model;

[0080] Step S2: extracting time-frequency features from the preprocessed EEG signal using the MST algorithm; calculating spatial features using the Pearson correlation coefficient to generate an adjacency matrix between channels;

[0081] Step S3, combining the time-frequency features with the channel space features in the adjacency matrix to obtain a feature set;

[0082] Step S4, set a self-attention module after the input layer of the residual graph convolutional network, construct an improved residual graph convolutional network, fuse the feature set, classify the motion imagery intention through the fully connected layer, obtain the classification label, calculate the classification loss for training, and obtain the trained residual graph convolutional network.

[0083] The preprocessing method, feature extraction method and fusion method in the training process are the same as those in the above steps 1 to 3 and will not be repeated here.

[0084] After training, the training weights and algorithm model are migrated to a portable EEG processor for real-time analysis of the EEG signal intentions of severely paralyzed patients.

[0085] Furthermore, it also includes system monitoring to detect abnormal information of the system, including collision of robot arm movements and physiological state detection of patients.

[0086] The system monitoring function can be realized by using an embedded system detection device connected to the collaborative robot arm through an SPI serial communication bus;

[0087] Furthermore, it also includes remote monitoring, querying abnormal information through the terminal device of the remote guardian, and making corresponding remote control processing.

[0088] When the system monitoring module detects a collision, it can issue a command to automatically reset the robot arm. At the same time, all the operation information of the robot arm is stored in the memory, which is convenient for subsequent viewing and maintenance of the equipment.

[0089] The remote monitoring APP can receive system alarm information, the patient's physiological status and the movement information of the collaborative robotic arm, and can remotely operate the operation of home care equipment.

[0090] Furthermore, human-computer interaction is achieved by setting up a human-computer interaction module.

[0091] In this embodiment, the EEG signals of severely paralyzed stroke patients during motor imagery are collected, and the EEG signals are downsampled, bandpass filtered, and artifacts are removed; the time-frequency features of the EEG signals processed above are extracted using the MST algorithm, and the adjacency matrix between channels is generated using the Pearson correlation coefficient, and the channel space information features and time-frequency features in the adjacency matrix are combined together; the residual graph convolution with the introduction of the attention mechanism is used to train the data and the algorithm model, and the motor imagery intention is judged on the feature-extracted EEG signals through the fully connected layer, and a label is output; the label is converted into a robot arm control instruction, and the instruction controls the robot arm to move; the EEG signal recognition accuracy is high through the above recognition process; this embodiment is suitable for home environments, and has the advantages of simple and safe operation, high cost performance, etc.

[0092] To illustrate the technical effect of the solution in this embodiment, a simulation experiment was conducted, and the accuracy of classification and recognition is shown in the figure below. Figure 3 As shown, the accuracy rate can reach 95%, which can meet the accuracy requirement of motor imagery intention discrimination.

[0093] Example 2

[0094] Based on Example 1, this embodiment provides a brain-controlled nursing system based on EEG signal recognition, such as Figures 5 and 6 As shown, it includes: an EEG acquisition device and a portable EEG signal processor;

[0095] The EEG acquisition equipment is used to collect EEG signals during motor imagery;

[0096] The portable EEG signal processor is configured to execute the brain control method based on EEG signal recognition described in Example 1.

[0097] Optionally, the EEG acquisition device is NeuroScan 64-256-channel EEG system, including:

[0098] 1) 64-electrode EEG cap, used to collect EEG signals from the electrodes on the patient's head;

[0099] 2) Display screen, used as a stimulus presentation display terminal for playing the acquisition paradigm;

[0100] 3) Recording terminal, used to record the collected EEG information;

[0101] 4) Signal amplifier, used for signal amplification processing;

[0102] 5) Multi-parameter synchronizer, used for synchronous marking and event marking of received stimuli;

[0103] 6) Smart synchronization device, which wirelessly connects the wireless EEG / ERP amplifier to the computer via Wi-Fi;

[0104] 7) Transmission line, used to transmit signals;

[0105] The method of data collection is as follows: specifically, a patient, such as a severely paralyzed patient, sits in front of a display screen. Before each motor imagery (MI), left and right prompts and arrow patterns of the motor imagery will appear on the screen. The patient is prompted to imagine left-hand movement or right-hand movement. The duration of each motor imagery and the rest time during the motor imagery can be set, and multiple sets of motor imagery tests can be performed.

[0106] Each motor imagery can last for 6 seconds, and the rest time between each motor imagery can be 4 seconds. Each random motor imagery is performed 10 times for each hand. Patients need to perform 4-6 sets of motor imagery tests, with a 90-second rest period between each set of motor imagery tests.

[0107] Furthermore, a motion control module is provided for converting the results of the EEG signals into motion instructions, and transmitting the instructions to the cooperative robotic arm so that the robotic arm performs corresponding actions according to the instructions.

[0108] A control signal processing module is set in the portable EEG signal processor. The control signal processing module and the motion control module are connected via a WiFi network. The control signal processing module can integrate a wireless communication module (Wi-Fi) to achieve real-time data transmission. This will allow users to collect data without connecting to the data processing module, improving the portability and user experience of the device.

[0109] Furthermore, a human-computer interaction module is provided to realize human-computer interaction through a provided display, which is used for signal transmission prompts, model processing progress prompts and classification result presentations.

[0110] Optionally, the display may be an LED screen display.

[0111] This embodiment adds a human-computer interaction module and uses an LED screen as a human-computer interaction interface to increase the operability of the device.

[0112] Optionally, a system monitoring module is also included to detect abnormal information of the system, including collision of the robot arm movement and physiological status detection of the patient;

[0113] In some embodiments, a remote monitoring terminal is also provided, and a remote monitoring APP is provided to enable the patient's guardian to implement remote monitoring through the remote monitoring terminal.

[0114] In this embodiment, through the collaborative interactive control of the control signal processing module, motion control module, system monitoring module and remote monitoring APP, it is possible to synchronously control the robotic arm using EEG signals, monitor system alarm information, the patient's physiological state and the motion information of the collaborative robotic arm, and remotely operate the operation of the nursing system. The nursing system helps severely paralyzed stroke patients pick up items, alleviates the shortage of nursing resources, and has important clinical significance.

[0115] Example 3

[0116] Based on Example 1, this embodiment further provides an electronic device, including a memory and a processor, and computer instructions stored in the memory and running on the processor. When the computer instructions are executed by the processor, the steps of the brain control method based on EEG signal recognition described in Example 1 are completed.

[0117] The computer instructions, i.e. the decoded EEG information, are executed by the processor. The processor may be a central processing unit (CPU), and the memory may include a read-only memory (ROM) and a random access memory (RAM), which is used to store information.

[0118] Connect the LED screen to the above electronic equipment, and connect the electronic equipment to the collaborative robot arm. Design the interface layout on the LED screen, including signal transmission prompts, model processing progress prompts, and classification result display areas. According to the resolution and display capacity of the LED screen, rationally plan the arrangement and size of the interface elements. Write the control logic to display the corresponding information on the LED screen. Refresh the content on the LED screen according to the changes in the prompt information to provide real-time human-computer interaction feedback. During the processing, display the corresponding progress information on the LED screen interface according to the progress of the processing. Display the percentage of the processing progress. User interaction is required, and buttons are added to implement user control operations. The functions implemented include turning the power on and off, clearing the results, and starting the recognition program. At the same time, the operation information is stored in the memory to facilitate subsequent viewing and maintenance of the equipment.

[0119] The above description is only a preferred embodiment of the present disclosure and is not intended to limit the present disclosure. For those skilled in the art, the present disclosure may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present disclosure shall be included in the protection scope of the present disclosure.

[0120] Although the above describes the specific implementation methods of the present disclosure in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present disclosure. Technical personnel in the relevant field should understand that on the basis of the technical solution of the present disclosure, various modifications or variations that can be made by those skilled in the art without creative work are still within the scope of protection of the present disclosure.

Claims

1. A brain control method based on EEG signal recognition, characterized in that: The steps include: Step 1, obtaining the EEG signal during motor imagery and preprocessing it; The data preprocessing process is as follows: Step 11: downsample the EEG signal, reducing the sampling rate from 1000 Hz to 100 Hz; Step 12: bandpass filter the EEG signal to divide it into α wave and β wave; segment the EEG data into 6s data segments; Step 13: Remove artifacts; Step 2: Extract time-frequency features and channel-space features from the preprocessed EEG signal; The MST algorithm is used to extract the time-frequency features of the preprocessed EEG signals; The MST algorithm is an algorithm that introduces adaptive parameter improvement based on the ST algorithm, and continuously optimizes the window width to calculate more accurate high-frequency energy; MST is: in, x(t) It's the EEG signal. is the translation factor, is the Gaussian window function, which is used to control the Gaussian function in The position on the time axis, is the frequency, is the imaginary unit and is defined as follows: in, is the standard deviation of the Gaussian function and is defined as follows: MST obtains the optimal energy concentration in the time-frequency domain by continuously adjusting the two parameters p and q. MST is used to extract the time-frequency features of EEG data in the appropriate frequency band as PSD features. PSD is defined as: in, represents the mean square error of MST; The Pearson coefficient is used to calculate the strength of connectivity between EEG signal channels, and the Pearson correlation coefficient is calculated for EEG signals of different channels. The absolute value of the Pearson correlation coefficient is taken, and the adjacency matrix generated by the absolute value of the Pearson correlation coefficient is used as the channel space feature; Step 3: Combine the time-frequency features with the channel space features in the adjacency matrix to construct a feature set; Step 4: a self-attention module is set after the input layer of the residual graph convolutional network to obtain an improved residual graph convolutional network; based on the trained improved residual graph convolutional network, the features in the feature set are fused, and then classified through the fully connected layer to obtain the result of motion imagination intention discrimination, and the control command is output according to the discrimination result to control the robot arm to move; The steps for training the improved residual graph convolutional network include: Obtain an EEG signal dataset, including EEG signals and corresponding motor imagery intention classification labels, for preprocessing; The preprocessed EEG signal is used to extract time-frequency features using the MST algorithm; the spatial features are calculated using the Pearson correlation coefficient to generate an adjacency matrix between channels; Combine the time-frequency features with the channel space features in the adjacency matrix to obtain a feature set; A self-attention module is set after the input layer of the residual graph convolutional network to construct an improved residual graph convolutional network, the features in the feature set are fused, the motor imagery intention is classified through the fully connected layer to obtain the classification label, the classification loss is calculated for training, and the improved residual graph convolutional network after training is obtained; The preprocessing method and feature extraction method in the training process are the same as those in steps 1 to 3 above; The method for fusing the time-frequency features in the feature set and the channel space features in the adjacency matrix is: An improved residual graph convolutional network is used to fuse time-frequency features and spatial features. The time-frequency features are used as node features, and the spatial structure features are used as the structure of the graph. The improved residual graph convolutional network is used to learn the propagation and fusion of time-frequency features in the spatial structure. The improved residual graph convolutional network includes multiple residual blocks and a fully connected layer. Each residual block includes multiple graph convolutional layers and Relu activation layers, as well as a pooling layer. A self-attention layer is added after each graph convolutional layer to dynamically obtain the weight relationship between nodes.

2. The brain control method based on EEG signal recognition as claimed in claim 1, characterized in that: The self-attention module is configured to be used for the input feature vector ,pass , , The three trainable weight matrices map to three vectors of dimension L , and , and calculate the attention weight of self-attention. The weight calculation formula is as follows: First, in and The dot product is used to evaluate the correlation between different tags, and the result is passed through the softmax function to obtain the weighted matrix of attention scores; then, the attention scores are combined with Perform dot product calculations; In the formula, , T represents the transpose of the matrix, d K Represents the dimension of the vector matrix K; , , Representing Query, Value and Key respectively, they are input through the same feature vector The specific transformation formula is: 。 3. A brain-controlled nursing system based on EEG signal recognition, characterized in that: include: EEG acquisition equipment and portable EEG signal processor; The EEG acquisition equipment is used to collect EEG signals during motor imagery; The portable EEG signal processor is configured to execute the brain control method based on EEG signal recognition as described in any one of claims 1-2.

4. The brain-controlled nursing system based on EEG signal recognition as claimed in claim 3, characterized in that: include: The brain-controlled nursing system is also provided with a motion control module, which is used to convert the results of the EEG signal into motion instructions, and transmit the instructions to the cooperative robot arm, so that the robot arm performs corresponding actions according to the instructions; Alternatively, the brain-controlled nursing system is further provided with a motion control module for converting the result of the EEG signal into a motion command, and transmitting the command to the cooperative robot arm so that the robot arm performs corresponding actions according to the command; Alternatively, it also includes a system monitoring module for detecting abnormal information of the system, including collision of the robot arm movement and physiological state detection of the patient; Alternatively, a remote monitoring terminal is used to set up a remote monitoring APP, and a patient guardian can install the remote monitoring terminal to achieve remote monitoring.

5. An electronic device, characterized in that: It comprises a memory and a processor, and computer instructions stored in the memory and executed on the processor. When the computer instructions are executed by the processor, the steps of the brain control method based on EEG signal recognition as described in any one of claims 1 to 2 are completed.

Images