Embedded brain-computer semantic recognition system and method

Abstract

The invention discloses an embedded brain-computer semantic recognition system and method. The system comprises an analog front-end module, a processing module and a display module, the analog front-end module comprises a power management module, a low-pass filtering module and an analog-to-digital conversion module; the processing module comprises a signal acquisition unit, a signal decoding unit, a storage unit and a logic control unit; the signal acquisition unit receives the electroencephalogram signal output by the analog-to-digital conversion module; the storage unit is used for storing the collected electroencephalogram data; the signal decoding unit performs feature extraction and recognition on the electroencephalogram signals; and the logic control unit is used for executing a corresponding task according to the identification result output by the signal decoding unit. According to the system, effective capture and preprocessing of the electroencephalogram signals are achieved through the analog front-end module, and the system adult cost is reduced. According to the invention, the signal acquisition and decoding processes are integrated in mobile processing, and the electroencephalogram data are directly transmitted by adopting an inter-process communication mechanism, so that the real-time performance and accuracy of signal processing are ensured.

Description

Technical Field

[0001] The present invention belongs to the technical field of brain-computer interfaces, and particularly relates to an embedded brain-computer semantic recognition system and method. Background Technique

[0002] A brain-computer interface (BCI) system is a new type of human-computer interaction system, which provides a new output channel for humans to interact with the surrounding environment without relying on muscles and peripheral nerves. This system has broad application prospects and important research value in the fields of neuroscience, medical rehabilitation, and human-machine automatic control.

[0003] The signal flow of a typical BCI system can be divided into: signal acquisition, preprocessing and feature extraction, recognition and classification, and application interface. First, signal acquisition obtains electroencephalogram (EEG) signals from the subject's brain through invasive or non-invasive methods and converts them into processable digital signals through a digitization process. Next, in the preprocessing and feature extraction stage, noise and artifacts are removed, and features related to current brain activity or specific tasks are extracted, laying a foundation for subsequent analysis. Then, the signals after feature extraction enter the recognition and classification stage, where classification algorithms analyze and process these features to generate control commands to drive external devices. Finally, the application interface feeds back the responses of external devices to the subject, enabling the subject to perceive the output of the system through visual, auditory, and other means, realizing the closed-loop operation of the BCI system.

[0004] Currently, in the signal acquisition stage, commercial EEG acquisition devices are generally used to acquire EEG signals, but there are several deficiencies. First, commercial devices are usually expensive, especially high-precision devices, and their popularity is limited, restricting the application of BCI technology to a wider population. Second, the hardware and software of commercial devices are often not publicly available, lacking sufficient openness and transparency, which limits the possibility for researchers and developers to customize, optimize, and improve the devices. In addition, commercial EEG devices have poor flexibility. Usually, they provide standardized solutions and cannot be adjusted individually according to specific application requirements, thus reducing their applicability in certain specific tasks.

[0005] The preprocessing, feature extraction, and recognition and classification stages of signals usually rely on a personal computer for remote computing. Although these computational processes are crucial for data analysis, such systems rely on a fixed hardware platform, leading to several significant problems. First, personal computer devices are usually large and bulky, lacking sufficient portability and flexibility, making it difficult to quickly deploy the system in actual use, especially in application scenarios that require dynamic adjustment, which limits the mobility of the system. Second, the high-performance computing hardware required by personal computers not only increases the cost of the device but also burdens the entire BCI system economically, restricting the popularity of the system. In addition, this architecture is prone to delays in the signal processing process, affecting real-time performance and making it difficult to meet the requirements of efficiently executing complex algorithms, thus affecting the response speed and user experience of the BCI system.

[0006] Current brain-computer interface systems are insufficient in terms of miniaturization, portability, and integration, making it difficult to popularize brain-computer interface technology in daily life. Summary of the Invention

[0007] Aiming at the deficiencies in the prior art, the purpose of the present invention is to provide an embedded brain-computer semantic recognition system and method, which combines the characteristics of low power consumption, high integration, and strong real-time performance, optimizes the electroencephalogram signal acquisition and decoding processes in the traditional BCI architecture, and improves the mobility, response speed, and application flexibility of the system. By integrating the entire electroencephalogram signal processing process into an embedded platform, this research can not only reduce the hardware dependence on the traditional PC platform but also effectively improve the real-time performance and convenience of the system, expanding the application potential of BCI technology in various actual application scenarios.

[0008] In a first aspect, the present invention provides an embedded brain-computer semantic recognition system, which includes an analog front-end module, a processing module, and a display module; the analog front-end module includes a power management module, a low-pass filter module, and an analog-to-digital conversion module; the processing module includes a signal acquisition unit, a signal decoding unit, a storage unit, and a logic control unit; the signal acquisition unit receives the electroencephalogram signal output by the analog-to-digital conversion module; the storage unit is used for the acquired electroencephalogram data; the signal decoding unit performs feature extraction and recognition on the electroencephalogram signal; the logic control unit is used to execute corresponding tasks according to the recognition result output by the signal decoding unit.

[0009] During the feature extraction process, the signal decoding unit decomposes the preprocessed electroencephalogram signal into multiple sub-band signals through multiple band-pass filters; performs correlation analysis on each sub-band signal with a reference signal respectively to obtain the correlation coefficient corresponding to each sub-band signal; during the recognition process, performs weighted averaging on the correlation coefficients obtained at different reference signal frequencies respectively to obtain the weighted correlation coefficient corresponding to different reference signal frequencies; takes the reference signal frequency corresponding to the highest weighted correlation coefficient as the target frequency.

[0010] The display module is used to provide visual stimuli for inducing electroencephalogram (EEG) signals to the user and to display the EEG signal recognition results.

[0011] Preferably, the signal acquisition unit integrates and serializes the EEG signals output by the analog-to-digital conversion module.

[0012] Preferably, before feature extraction of the EEG signals, the signal decoding unit reduces the sampling rate of the EEG signals and uses a notch filter to suppress power frequency interference.

[0013] Preferably, the low-pass filter module includes a differential capacitor connected between two differential signal lines and a common-mode capacitor connected between the signal line and the ground wire.

[0014] Preferably, the analog front-end module further includes a power management module; the power management module converts the voltage provided by the processing module and supplies power to the analog front-end module.

[0015] Preferably, the analog front-end module further includes a high-gain amplifier; the high-gain amplifier is used to amplify the filtered EEG analog signals; the amplified EEG analog signals are converted into EEG digital signals by the analog-to-digital conversion module.

[0016] Preferably, the analog-to-digital conversion module uses an integrated analog-to-digital conversion chip; the high-gain amplifier is built into the analog-to-digital conversion module.

[0017] In a second aspect, the present invention provides an embedded brain-computer semantic recognition method, which uses the aforementioned embedded brain-computer semantic recognition system; the embedded brain-computer semantic recognition method includes an EEG signal acquisition method, a signal decoding method, and a stimulation paradigm method; the EEG signal acquisition method is executed by the signal acquisition unit and is used to control the analog front-end module to read EEG signals, perform low-pass filtering and analog-to-digital conversion on the EEG signals; the signal decoding method is executed by the signal decoding unit to perform feature extraction and recognition on the EEG signals; the stimulation paradigm method is executed by the logic control unit and is used to control the display module to provide visual stimuli to the user.

[0018] Preferably, the steps of the EEG signal acquisition method are as follows:

[0019] Step 1: Initialize the communication and direct memory access parameters of the processing module.

[0020] Step 2: Set the working parameters of the analog-to-digital conversion module.

[0021] Step 3: After receiving the reading instruction provided by the processing module, read and encode the digital EEG data; during the encoding process, the EEG data is divided into multiple independent data frames; channel numbers, timestamps, and data integrity check information are appended to each data frame.

[0022] Step 4: Transmit the encoded EEG data to the signal decoding method for processing.

[0023] Preferably, the stimulation paradigm method is as follows: After receiving the instruction, the logic control unit starts the stimulation task; the stimulation task includes the task process of one trial or multiple trials; the task process of each trial includes a cue stage, a stimulation flash stage, and a rest stage that are executed in sequence; in the cue stage, the display module displays a preparation cue message; in the stimulation flash stage, the display module displays a flashing semantic instruction; the user fixes their gaze on the semantic instruction, triggering an EEG response signal.

[0024] The beneficial effects of the present invention are:

[0025] 1. The present invention realizes the effective capture and preprocessing of EEG signals through the analog front-end module, greatly reducing the overall cost of the system. This module replaces traditional expensive commercial EEG acquisition devices, not only maintaining high-precision data acquisition but also providing an economically viable option for a wide range of applications.

[0026] 2. Utilizing the excellent computing performance of the Raspberry Pi, the present invention successfully integrates the functions of real-time acquisition, decoding, and stimulation paradigm display of EEG signals. Compared with traditional bulky computer solutions, the Raspberry Pi, with its small and portable characteristics, significantly improves the portability and practicality of the system.

[0027] 3. The present invention integrates the signal acquisition and decoding processes within the Raspberry Pi and uses the inter-process communication (IPC) mechanism to directly transmit EEG data, effectively avoiding common data delay problems in wireless transmission and ensuring the real-time and accurate signal processing.

[0028] 4. By concentrating the signal processing flow of the brain-computer interface on a single platform, the Raspberry Pi, the present invention greatly improves the integration degree of the system, simplifies the system architecture, and enhances the stability and reliability of the system, laying a foundation for the development of a more compact and efficient brain-computer interface system. Description of the Drawings

[0029] Figure 1 It is a schematic diagram of the system structure of an embodiment of the present invention.

[0030] Figure 2 It is a circuit schematic diagram of the power management module in an embodiment of the present invention.

[0031] Figure 3This is the circuit schematic diagram of the low-pass filtering module in the embodiment of the present invention.

[0032] Figure 4 This is the circuit schematic diagram of the analog-to-digital conversion module in the embodiment of the present invention.

[0033] Figure 5 This is the flowchart of the signal acquisition method in the embodiment of the present invention.

[0034] Figure 6 This is the flowchart of the signal decoding method in the embodiment of the present invention.

[0035] Figure 7 This is the flowchart of the stimulation paradigm method in the embodiment of the present invention. Detailed implementation

[0036] The following further describes the present invention.

[0037] Embodiment

[0038] As Figure 1 shown, an embedded brain-computer semantic recognition system includes an analog front-end module (AFE), a processing module, and a display module.

[0039] The analog front-end module includes a power management module, a low-pass filtering module, and an analog-to-digital conversion module, and is used for preliminary preprocessing of electroencephalogram (EEG) signals; specifically, the analog front-end module completes tasks such as noise filtering, signal amplification, and analog-to-digital conversion to ensure high-quality transmission and processing of EEG signals.

[0040] EEG signals are easily affected by external electromagnetic interference and interference from other physiological signals of the human body (such as electromyogram signals), resulting in a decline in signal quality. Therefore, the analog front-end module designs a low-pass filter to remove these noises and improve the signal quality.

[0041] The amplitude of EEG signals is small and usually needs to be amplified to meet the requirements of subsequent processing circuits. The analog front-end module amplifies these signals to a suitable level for processing through a high-gain amplifier built into the analog-to-digital conversion module, enabling subsequent circuits and algorithms to effectively process these signals.

[0042] After noise filtering and signal amplification, the analog signal needs to be converted into a digital signal for further processing. The analog front-end module converts the preprocessed analog signal into a digital signal through the analog-to-digital conversion module (ADC).

[0043] In this embodiment, the processing module is built based on Raspberry Pi. As the core processing platform of the system, Raspberry Pi has the advantages of low power consumption, high computing performance and modular design. It undertakes multiple tasks such as signal acquisition, processing, storage and control, and is connected to each module in the system through rich hardware peripheral interfaces. Thanks to the rich GPIO interfaces of Raspberry Pi, the present invention demonstrates extremely high hardware compatibility and functional expandability. This enables it to be flexibly configured according to different application scenario requirements, such as connecting to an analog front end for signal acquisition or connecting to a display screen to realize visual stimulus presentation, greatly enriching the application scope.

[0044] The processing module includes a signal acquisition unit, a signal decoding unit, a storage unit and a logic control unit; specifically as follows:

[0045] 1. Signal acquisition unit: First, Raspberry Pi is connected to the analog front-end module through the SPI interface to receive the digitized EEG signals transmitted from this module. After being processed by Raspberry Pi, these signals are integrated, serialized, and then transmitted to the signal decoding unit after being packed according to a preset protocol to ensure data integrity and real-time performance. Raspberry Pi is also connected to the display screen through the HDMI interface to display the stimulation paradigm and experimental feedback information in real time.

[0046] 2. Signal decoding unit: In the signal decoding stage, Raspberry Pi performs digital filtering, feature extraction and classification recognition on the received EEG data, extracts the signal features related to the user's intention, and provides support for subsequent applications.

[0047] 3. Storage unit: In terms of data recording, the processing module saves the acquired EEG data to the local storage device to meet the needs of real-time analysis and long-term storage.

[0048] 4. Logic control unit: As the control center of the system, the logic control unit performs logic control according to the processed EEG signals, drives the display content of the display module, and realizes the diversified application of EEG signals.

[0049] The signal acquisition unit includes an EEG data stream server; the signal decoding unit includes an EEG data stream client and a stimulation response server; the logic control unit includes a stimulation and response client. The EEG data stream server sends EEG signals to the EEG data stream client for subsequent feature extraction and recognition classification. The stimulation response server communicates bidirectionally with the stimulation and response client; the stimulation response server sends the semantic recognition result to the stimulation and response client. The stimulation and response client sends the signals of the start and end of the stimulation to the stimulation response server. Thus, the inter-process communication mechanism is realized.

[0050] The display module uses a display screen for user interaction and feedback. During the construction of the dataset, the display module provides visual stimuli, specifically by presenting visual targets (semantic labels) that flash at different frequencies to induce specific electroencephalogram (EEG) signals, especially steady-state visual evoked potentials (SSVEPs). These flashing visual stimuli change at a predetermined frequency on the screen, activating the user's visual cortex through the change of visual stimuli, thereby generating corresponding EEG signals.

[0051] During the actual semantic recognition process, by analyzing the EEG signals recorded from the user's brain, detecting whether they contain frequency components corresponding to a certain semantic label during the dataset construction process, and inferring the user's intention, corresponding operations or commands are then executed.

[0052] As Figure 2 shown, the power management module includes a voltage inverter, an adjustable voltage regulator, and a high-precision voltage regulator TPS73225 with reverse current protection. The model of the voltage inverter is TPS60403; the model of the adjustable voltage regulator is TPS72325; the model of the high-precision voltage regulator is TPS73225. The power management module is supplied with 5V and 3.3V voltages provided by the processing module, and converts the 5V voltage into 2.5V and -2.5V; it is used to supply power to the analog circuits of the analog front-end module; while the 3.3V voltage supplies power to the digital circuit part of the analog front-end module. The above voltage management ensures the accurate acquisition of analog signals and the stability of digital signals during transmission. At the same time, in order to effectively isolate the interference of digital voltage on analog voltage, the power management module uses a 0-ohm resistor for single-point grounding to separate the analog ground and the digital ground, avoiding the coupling of noise in digital signals to analog signals and ensuring the accuracy of analog signals.

[0053] As a noise suppression scheme, a low-pass anti-aliasing filter is designed to effectively suppress high-frequency signals. In this filter, the low-pass filtering module selects a 4.7k-ohm resistor, a 47nF differential capacitor, and a 4.7nF common-mode capacitor to form a low-pass filter. The differential capacitor is connected between two differential signal lines, and the common-mode capacitor is connected between the signal line and the ground.

[0054] The designs of both the differential capacitor and the common-mode capacitor are aimed at improving the common-mode rejection performance of the circuit. When a common-mode interference is simultaneously applied to two signal lines, the differential capacitor can make the common-mode signal generate the same response on two paths, so that the signal is effectively suppressed at the receiving end; the common-mode capacitor is mainly used to reduce the coupling between two signal lines and the ground, suppress the noise introduced through the common-mode path, especially reduce the interference of the common-mode noise from the power supply on the signal.

[0055] As Figure 4As shown in the figure, the analog-to-digital conversion module uses an integrated ADC module. Compared with building with discrete components, the integrated ADC module not only has stronger anti-interference ability, but also has advantages such as low power consumption, functional modularity, and portability, providing new ideas for the design of bioelectric signal acquisition devices. Therefore, the analog-to-digital conversion module selects the ADS1299 integrated chip of Texas Instruments as the ADC module. This chip is a low-noise, 8-channel, 24-bit synchronous sampling analog-to-digital converter designed specifically for biopotential measurement, with a built-in programmable gain amplifier, and each channel can be independently configured. The sampling frequency is between 250Hz and 16kHz. In order to meet the requirements of multi-channel EEG signal acquisition in the experiment, this AFE module uses two ADS1299 chips, which can simultaneously acquire 16-channel EEG signals to capture the activity signals of multiple regions of the brain.

[0056] This embedded brain-computer semantic recognition system is used to execute the EEG semantic information recognition method.

[0057] This EEG semantic information recognition method includes an EEG signal acquisition method, a signal decoding method, and a stimulation paradigm method.

[0058] The EEG signal acquisition method is used to obtain EEG signals in real time from the analog front-end module and communicate with the analog-to-digital conversion module through the hardware SPI interface of the processing module.

[0059] At the same time, initialize the EEG data stream server through the TCP protocol, establish communication with the EEG signal decoding program, and be used to transmit the acquired EEG signal data in real time. When the client successfully connects to the server, the program sends a continuous read instruction to the analog-to-digital conversion module and starts waiting for the DRDY pin interrupt signal of the analog-to-digital conversion module. Once the DRDY interrupt is detected, the program reads the EEG signal data from the analog-to-digital conversion module through the SPI interface, decodes the original data, and converts the digital quantity read from the register into the corresponding voltage value, expressed in microvolts.

[0060] After decoding, the acquired signals are re-encoded to adapt to the transmission protocol and client data processing requirements. Specifically, the program groups the acquired EEG signal data into fixed numbers of sampling points (for example, 10 sampling points) in chronological order, and each group of data is packed into an independent data frame. Each data frame not only contains the numerical information of the signal, but also additional channel numbers, timestamps, and data integrity check information to ensure the timing consistency and transmission reliability of the data. The encoded data frames are transmitted to the EEG signal decoding program in real time through the TCP server to provide support for subsequent signal processing, feature extraction, and classification and recognition.

[0061] The signal decoding method is used to receive and process the electroencephalogram (EEG) signal data sent by the EEG signal acquisition method, analyze and classify to obtain the user's intention, and send the recognition and classification results to the stimulation paradigm method via TCP, while visually presenting the result feedback on the display module.

[0062] As Figure 6 shown, the steps of the signal decoding method are as follows:

[0063] Step 1: Initialize the stimulation response server and the EEG data stream client, and establish a connection with the EEG data stream server to receive real-time EEG signal data. Subsequently, wait for the connection of the stimulation and response client from the stimulation paradigm method to complete the initialization of communication. When the client successfully connects, enter the standby state and listen for the arrival of the stimulation start signal.

[0064] Step 2: When the stimulation start signal arrives, receive the EEG signal data from the EEG data stream server in real time and save it in the buffer according to the time series. During this process, the program continuously listens and waits for the stimulation end signal.

[0065] Step 3: After the stimulation end signal arrives, stop writing new data to the buffer, and preprocess and extract features from the EEG signals received during the trial.

[0066] In the preprocessing stage, first reduce the sampling rate of the original EEG signal from 1000 Hz to 250 Hz to reduce the amount of data and improve the subsequent processing efficiency. Subsequently, use a 50 Hz notch filter to effectively suppress the power frequency interference.

[0067] In the feature extraction and recognition classification stage, use the feature extraction and classification method. The feature extraction and classification method includes a feature extraction module and a recognition classification module. In the feature extraction module, use multiple band-pass filters to decompose the preprocessed EEG signal into multiple sub-band signals Subsequently, for each sub-band signal and the reference signal of each target frequency perform canonical correlation analysis, and calculate the correlation coefficient matrix between it and the reference signals of each frequency. Among them, for the frequency f k of each target stimulus, its correlation coefficient vector ρ k consists of N correlation coefficients:

[0068]

[0069] In the formula, ρ(x, y) represents the correlation coefficient between x and y; Y is the reference signal matrix, containing the reference signals of all target frequencies; W X and W Y are linear transformation matrices, which act on the sub-band signal and the reference signal matrix Y to extract the feature components with the maximum correlation in their respective linear spaces.

[0070] In the recognition and classification module, the N correlation coefficients of each reference signal frequency are weighted and averaged to obtain the weighted correlation coefficient between the EEG signal and the reference signals of each frequency

[0071]

[0072] Here, ω(n) represents the weight coefficient, and its expression is as follows:

[0073] ω(n) = n -a +b, n ∈ [1, 2,..., N]

[0074] In the formula, a and b are 1.25 and 0.25 respectively.

[0075] Finally, the frequency of the reference signal corresponding to the maximum weighted correlation coefficient is regarded as the target frequency f target :

[0076]

[0077] When the target frequency f target is determined, the signal decoding method sends the recognition and classification result to the stimulation paradigm method through TCP, and at the same time visually displays the recognition result on the display screen to prompt the user's target operation status.

[0078] The stimulation paradigm method is connected to the display module through the HDMI interface, presents visual stimuli on the display module to induce EEG responses, and interacts with the signal decoding method to complete signal synchronization, display of recognition and classification results, and user feedback. The specific steps of the stimulation paradigm method are as follows:

[0079] Step 1: Initialize the stimulation and response client, and try to connect to the stimulation response server through TCP to establish communication. After successful connection, enter the standby state and wait for the user to press the Space key to start the stimulation task. After the stimulation task starts, send a stimulation start signal to the stimulation response server and enter the task process of one trial.

[0080] The duration of each trial is 7 seconds, including a 2 - second prompt phase, a 3 - second stimulation flashing phase, and a 2 - second rest phase.

[0081] Step 2: Prompt phase: A red triangle will be displayed on the display screen below the target stimulus block to indicate the current focus target, and at the same time prompt the user to fixate on the target to prepare to enter the stimulation phase.

[0082] Step 3. Stimulus Blinking Phase: In this embodiment, the recognized semantic information is orientation information, including up, down, left, and right; there are 4 target stimulus blocks, which are respectively distributed at the four positions of the upper, lower, left, and right of the display screen, and their blinking frequencies are 8Hz, 9Hz, 10Hz, and 11Hz respectively. During the stimulus blinking process, each stimulus block blinks at a frequency modulated by a sine wave, and its stimulus intensity Stim( n， f i ) is determined by the following formula:

[0083]

[0084] In the formula, fi represents the stimulus frequency of the i-th target, sin() is used to generate a sine signal, R represents the screen refresh rate, and n is the frame index.

[0085] The user needs to focus on a specific target stimulus block according to the prompt to generate a clear electroencephalogram response signal through steady-state visual evoked potential (SSVEP).

[0086] Step 4. Rest Phase: After the stimulus blinking phase ends, the program enters a short rest phase while waiting for the signal decoding program to complete the analysis of the recognition and classification results. After the recognition and classification results are sent to the stimulus paradigm program through TCP, the stimulus paradigm program will display the result feedback on the screen to prompt the user of the operation status corresponding to the recognition and classification results.

[0087] After the trial is completed, the program returns to the initial state. The user can choose to enter the next trial or exit the program by pressing the Esc key. Throughout the process, the program ensures the synchronous communication between the stimulus paradigm and the electroencephalogram signal acquisition and decoding modules, providing support for the efficient execution of the trial task.

Claims

1. An embedded brain-computer semantic recognition system, comprising an analog front-end module, a processing module and a display module; characterized in that: The analog front-end module includes a power management module, a low-pass filter module and an analog-to-digital conversion module; the processing module includes a signal acquisition unit, a signal decoding unit, a storage unit and a logic control unit; the signal acquisition unit receives the EEG signal output by the analog-to-digital conversion module; the storage unit is used to store the collected EEG data; the signal decoding unit extracts and recognizes the features of the EEG signal; the logic control unit is used to perform corresponding tasks according to the recognition results output by the signal decoding unit; During the feature extraction process, the signal decoding unit decomposes the preprocessed EEG signal into multiple sub-band signals through multiple band-pass filters; performs correlation analysis on each sub-band signal and the reference signal respectively to obtain the correlation coefficient corresponding to each sub-band signal; During the recognition process, the correlation coefficients obtained for different reference signal frequencies are weighted averaged to obtain weighted correlation coefficients corresponding to different reference signal frequencies; the reference signal frequency corresponding to the highest weighted correlation coefficient is taken as the target frequency; The display module is used to provide visual stimulation for inducing EEG signals to the user and to display EEG signal recognition results.

2. The embedded brain-computer semantic recognition system according to claim 1, characterized in that: The signal acquisition unit integrates and serializes the EEG signals output by the analog-to-digital conversion module.

3. The embedded brain-computer semantic recognition system according to claim 1, characterized in that: The signal decoding unit reduces the sampling rate of the EEG signal before extracting the features of the EEG signal, and uses a notch filter to suppress power frequency interference.

4. The embedded brain-computer semantic recognition system according to claim 1, characterized in that: The low-pass filter module includes a differential capacitor connected between two differential signal lines, and a common-mode capacitor connected between the signal line and the ground line.

5. The embedded brain-computer semantic recognition system according to claim 1, characterized in that: The analog front-end module also includes a power management module; the power management module converts the voltage provided by the processing module to supply power to the analog front-end module.

6. The embedded brain-computer semantic recognition system according to claim 1, characterized in that: The analog front-end module also includes a high-gain amplifier; the high-gain amplifier is used to amplify the filtered EEG analog signal; the amplified EEG analog signal is converted into an EEG digital signal by the analog-to-digital conversion module.

7. The embedded brain-computer semantic recognition system according to claim 6, characterized in that: The analog-to-digital conversion module adopts an integrated analog-to-digital conversion chip; the high-gain amplifier is built in the analog-to-digital conversion module.

8. An embedded brain-computer semantic recognition method, characterized in that: An embedded brain-computer semantic recognition system as described in claim 1 is used; the embedded brain-computer semantic recognition method includes an EEG signal acquisition method, a signal decoding method and a stimulation paradigm method; the EEG signal acquisition method is executed by a signal acquisition unit, which is used to control an analog front-end module to read EEG signals and perform low-pass filtering and analog-to-digital conversion on the EEG signals; the signal decoding method is executed by a signal decoding unit, which performs feature extraction and recognition on the EEG signals; the stimulation paradigm method is executed by a logic control unit, which is used to control a display module to provide visual stimulation to a user.

9. The embedded brain-computer semantic recognition method according to claim 1, characterized in that: The steps of the EEG signal acquisition method are as follows: Step 1: Initialize the communication and direct memory access parameters of the processing module; Step 2: Setting the working parameters of the analog-to-digital conversion module; Step 3: After receiving the read instruction provided by the processing module, the digital EEG data is read and encoded; during the encoding process, the EEG data is divided into multiple independent data frames; Each data frame is appended with a channel number, a timestamp, and data integrity check information; Step 4: Transmit the encoded EEG data to the signal decoding method for processing.

10. The embedded brain-computer semantic recognition method according to claim 1, characterized in that: The stimulation paradigm method is as follows: the logic control unit starts the stimulation task after receiving the instruction; the stimulation task includes a task flow of one trial or multiple trials; the task flow of each trial includes a prompt stage, a stimulation flashing stage and a rest stage which are executed in sequence; in the prompt stage, the display module displays the preparation prompt information; In the stimulation flashing stage, the display module displays flashing semantic instructions; the user gazes at the semantic instructions, triggering an electroencephalogram response signal.

Images