Asynchronous electroencephalogram signal acquisition system and method

Abstract

The invention discloses an asynchronous electroencephalogram signal acquisition system and method, and belongs to the technical field of biomedical engineering and neuroscience, and the system comprises a front-end asynchronously-triggered electroencephalogram signal acquisition circuit which is used for achieving asynchronous acquisition of electroencephalogram signals and improving the quality of the electroencephalogram signals so as to facilitate subsequent analysis and processing; the embedded parallel logic control module is used for realizing signal processing and data flow management; an electroencephalogram asynchronous physiological calculation chip is arranged in the embedded parallel logic control module and used for improving the speed and accuracy of data processing and physiological calculation of emotions, and emotion classification is achieved. By the adoption of the asynchronous electroencephalogram signal collecting system and method, the electroencephalogram signals can be efficiently collected and processed, and emotional state analysis is facilitated.

Description

Technical Field

[0001] The present invention relates to the technical fields of biomedical engineering and neuroscience, and in particular to an asynchronous electroencephalogram (EEG) signal acquisition system and method. Background Art

[0002] Electroencephalogram (EEG) is a technology that records the potential changes generated by the activities of brain neurons by arranging electrodes on the scalp. It is a key tool for monitoring and analyzing the functional state of the brain. The signals captured by EEG technology can reflect the activities of the brain in different emotional and cognitive states, and the monitoring of emotional functions is particularly important. By carefully analyzing the changes in EEG signals, electroencephalogram patterns related to specific emotional states can be identified, providing a scientific basis for the diagnosis and evaluation of abnormal emotional functions. Changes in emotional states are often accompanied by changes in the activity intensity of specific brain regions, which are manifested as fluctuations in different frequencies (such as alpha waves, beta waves, and theta waves) and amplitudes in the electroencephalogram.

[0003] However, most current EEG signal systems adopt synchronous acquisition methods. Synchronous acquisition means collecting data simultaneously from multiple EEG electrodes at a fixed frequency under the control of the same clock. This method has obvious limitations: it cannot flexibly respond to the rapid changes in EEG activities and usually generates a large amount of redundant data. This not only increases the burden of data storage and processing but also may cause the tracking of key diagnostic information to be missed during the analysis process. Secondly, the continuous use of a high sampling rate will greatly increase the energy consumption of the system, which is disadvantageous for portable or long-term monitoring devices.

[0004] On the contrary, asynchronous acquisition technology does not perform at fixed time intervals but triggers sampling according to the actual changes in EEG signals. The core advantage of this method is that it can automatically adjust the sampling timing according to the activities and importance of the data sources, thus significantly reducing the storage and processing of invalid data and improving the relevance and accuracy of data acquisition. In addition, this strategy greatly improves the energy efficiency of the system, enabling it to maintain low-power operation during long-term monitoring.

[0005] In the development process of EEG technology, traditional electroencephalogram signal processing systems often use microprocessors or microcontrollers as the core of data processing. Such systems rely on the central processing unit (CPU) to process data sequentially and face major problems including slow processing speed, limited parallel processing ability, and relatively high power consumption. These limitations are particularly prominent in EEG monitoring applications that require processing large amounts of data streams. Especially during multi-channel data acquisition, the CPU may have difficulty processing the data of all channels in real time, resulting in delays or data loss. In contrast, field-programmable gate arrays (FPGAs), as a highly flexible digital logic device, have the following several significant advantages. First is the high parallel processing ability. FPGAs can be configured to process multiple data streams simultaneously. For example, it can process multiple-channel EEG signals from different brain regions in parallel. This feature makes FPGAs particularly suitable for multi-lead EEG signal devices, effectively improving data processing speed and accuracy. Second is the low-power design. The design of FPGAs allows unused logic blocks to be turned off in the inactive state, significantly reducing the overall power consumption. This feature is particularly crucial for mobile or portable medical devices that require long-term battery power supply, ensuring that the device can operate continuously without frequent power source replacement. Then there is programmability and customizability. Different from traditional hard-coded logic chips, FPGAs allow hardware logic to be configured as needed, providing great flexibility for EEG signal processing. Developers can quickly reconfigure FPGAs according to specific monitoring requirements or algorithm updates, and adapt to new application scenarios without replacing hardware. And there is rapid prototyping. In the initial stage of the development of physiological computing chips, FPGAs can quickly implement and test new processing algorithms and chip designs. This flexibility and cost efficiency make FPGAs play an important role in the initial research and development of physiological computing chips.

[0006] In the field of electroencephalogram (EEG) signal processing and emotion state decoding, wearable devices have become increasingly important due to their continuous monitoring ability and comfort. However, traditional wearable devices have some technical problems, especially in data processing and power consumption management. Currently, most wearable devices collect EEG signals, then digitize the data and transmit it to the cloud for processing. Although this method can utilize powerful cloud computing resources, it also comes with several key problems: First, continuous data transmission significantly increases the power consumption of the device. Especially when the device operates relying on batteries or energy harvesters, frequent data sending will accelerate energy consumption and limit the usage time of the device. Second, after the sensor collects data, the data needs to be transmitted to the cloud for processing, and the processed results are then sent back to the device. This round-trip delay will cause unacceptable latency in applications that require immediate response, affecting the user experience and the practicality of the device. In contrast, dedicated physiological computing chips have high integration, low power consumption, and sufficient computing power to achieve local processing and decision-making in the sensing node. This is more suitable for wearable devices in terms of data volume transmission, system latency, and power consumption.

[0007] In the diagnosis and analysis of EEG-based abnormal emotion functions, traditional algorithms mainly include support vector machines (SVMs), artificial neural networks (ANNs), and deep learning models such as convolutional neural networks (CNNs). These methods have achieved certain success in emotion state analysis, but they usually require a large amount of computing resources and generally rely on high-performance CPUs and GPUs. On portable devices, the implementation of these algorithms on portable devices may be limited by both power consumption and processing capabilities. Compared with traditional algorithms, algorithms based on spiking neural networks (SNNs) provide a processing method that is closer to the biological nervous system. The working principle of SNNs is to process information by simulating the spiking activities of neurons, which is particularly suitable for processing EEG signals. The main advantages of SNNs include: First, low power consumption. SNNs can operate with very low energy consumption because it only consumes energy when neurons are activated, which makes it particularly suitable for power-constrained mobile and portable devices; Second, high timeliness. Due to its event-driven characteristics, SNNs are suitable for real-time or near-real-time data processing and can quickly respond to emotion changes in EEG signals; and hardware compatibility. SNNs can be directly implemented on hardware, and by designing dedicated neuromorphic physiological computing chips, the processing speed and efficiency can be further improved. Summary of the Invention

[0008] The object of the present invention is to provide an asynchronous EEG signal acquisition system and method that can efficiently acquire and process EEG signals for emotion state analysis.

[0009] To achieve the above object, the present invention provides an asynchronous EEG signal acquisition system, including:

[0010] A front-end asynchronous trigger electroencephalogram (EEG) signal acquisition circuit is used to achieve asynchronous acquisition of EEG signals and improve the quality of EEG signals for subsequent analysis and processing.

[0011] An embedded parallel logic control module is used to implement signal processing and data stream management. An EEG asynchronous physiological computing chip is provided in the embedded parallel logic control module to improve the speed and accuracy of data processing and physiological computing of emotions, and to achieve emotion classification.

[0012] Preferably, the EEG signal acquisition circuit includes EEG signal sensing electrodes, an EEG signal preprocessing circuit, and an asynchronous ADC circuit. The EEG signal sensing electrodes are flexible dry electrodes of graphene / liquid metal to improve the signal reception quality and the comfort of the subject. The EEG signal preprocessing circuit includes a low-noise amplifier, a low-pass filter, a 50 Hz power frequency notch filter, and a band-pass filter connected in sequence. The asynchronous ADC circuit includes a digital-to-analog converter, a multiplexer, a comparator, digital logic control, a counter, and digital logic gates, and is connected according to signal logic control.

[0013] Preferably, the embedded parallel logic control module uses an FPGA as the core and has the configuration and logic control of the asynchronous ADC circuit built-in.

[0014] Preferably, the EEG asynchronous physiological computing chip includes a data cache module, a pulse coding module, and a pulse neural network classifier.

[0015] Preferably, the data cache module adopts a direct mapped cache architecture and a high-speed static random access memory. The pulse coding module uses a rate coding-based strategy and combines dynamic adjustment of parameters. The pulse neural network classifier includes an input layer, a linear transformation layer, and a Leaky Integrate-and-Fire neuron layer, and implements learning and optimization strategies.

[0016] The present invention also provides an asynchronous EEG signal acquisition method, including the following steps:

[0017] S1. Acquire EEG signals and perform analysis and processing.

[0018] S2. After the processed EEG data is input, it is encoded by the pulse coding module to generate pulses.

[0019] S3. Input the generated pulses into the pulse neural network classifier for analysis and processing, and finally output the emotion classification category.

[0020] Preferably, in S2, the pulse coding module introduces a parameter para th , for the parameter parath Optimize it by dynamically adjusting the pulse threshold of each EEG processing channel. The pulse threshold of each EEG processing channel is calculated by the following formula:

[0021]

[0022] where N is the number of samples, l is the mean of the signal change rate in the samples, r is the standard deviation of the signal change rate in the samples, and V th (k) is the variable threshold of the k-th input channel;

[0023] On each EEG processing channel, when the change rate of the signal exceeds the corresponding variable threshold, a positive pulse is generated; when the change rate of the signal exceeds the variable threshold in the negative direction, a negative pulse is generated.

[0024] Preferably, the steps for optimizing the parameters are specifically as follows: First, calculate the statistical description features from the pulse sequences of each EEG processing channel. After feature selection, calculate the median of the pulse emission rate and the time difference between pulses; Second, use an SVM classifier to evaluate the feature performance under different settings, adopt the Bayesian optimization method to find the optimal value, evaluate the performance of the feature set, and adjust according to the obtained classification accuracy; Use the leave-one-out cross-validation method to ensure the generalization ability of the evaluation process, that is, leave one sample as the test set each time, and use all the remaining samples for training.

[0025] Preferably, in S3, after the generated pulses are input into the input layer of the pulse neural network classifier, in the first linear transformation layer, the input data is mapped to a 128-dimensional hidden space for linear transformation of the data. The output data is processed by the Leaky Integrate and Fire neuron layer to simulate the dynamic behavior of biological neurons, including the integration and leakage of membrane potential, and the firing process when reaching the threshold; The output data is input into the second linear transformation layer again to map the 128-dimensional features to the dimension corresponding to the number of output categories. Finally, the output data is input into the second Leaky Integrate and Fire neuron layer again to further integrate the features and directly use them for the classification task; The entire network is trained by minimizing the cross-entropy loss, and the classification performance is optimized by updating the weights, and finally the emotion classification category is output.

[0026] Therefore, the beneficial effects of the present invention adopting the above asynchronous EEG signal acquisition system and method are as follows:

[0027] (1) It can improve the signal reception quality and the comfort of the subject: By adopting flexible dry electrodes, the present invention not only enhances the signal reception effect but also improves the wearing comfort, which is suitable for long-term monitoring of EEG signals.

[0028] (2) It can optimize data acquisition efficiency and storage: By combining precise signal preprocessing and level-crossing asynchronous ADC, the present invention ensures the accuracy and reliability of data. It samples only when the signal passes through a preset level threshold, effectively reducing the acquisition of invalid data and decreasing the data storage and transmission requirements. In addition, this mechanism provides higher time resolution when the signal changes rapidly, which helps to accurately record fast-changing events in EEG signals, such as spikes and fluctuations, and optimizes the performance of portable EEG devices for long-term monitoring.

[0029] (3) It can classify mood disorders based on spiking neural networks: The spiking neural network classification method adopted by the present invention can be directly deployed on hardware, especially on dedicated chips designed for small portable devices, effectively supporting efficient and accurate classification of emotional states.

[0030] (4) It has the technical advantages of asynchronous acquisition and physiological calculation: The asynchronous acquisition mechanism allows capturing asynchronous signals from different brain regions, improving the accuracy of monitoring the activities of different brain regions and contributing to in-depth neuroscience research. Its event-triggered asynchronous mechanism responds in real time to specific neurophysiological signals and accurately records the temporal relationship of neural activities. In addition, due to not relying on a synchronous clock, the asynchronous ADC and dedicated chips reduce power consumption, extend the battery life of the device, and improve the efficiency of data processing and the anti-interference ability of the system, enhancing the overall system stability and reliability.

[0031] The technical solution of the present invention will be further described in detail below through the accompanying drawings and embodiments. Description of the Drawings

[0032] Figure 1 It is a schematic diagram of the composition of an asynchronous EEG signal acquisition system of the present invention;

[0033] Figure 2 It is a schematic diagram of the EEG signal preprocessing circuit of an asynchronous EEG signal acquisition system of the present invention;

[0034] Figure 3 It is a schematic diagram of the asynchronous ADC circuit of an asynchronous EEG signal acquisition system of the present invention;

[0035] Figure 4 It is a schematic diagram of the flow of an asynchronous EEG signal acquisition method of the present invention;

[0036] Figure 5 It is the optimization strategy diagram of parameter para th of the present invention. Detailed Embodiments

[0037] The technical solution of the present invention will be further described below through the accompanying drawings and embodiments.

[0038] Unless otherwise defined, the technical terms or scientific terms used in this invention shall have the ordinary meanings as understood by those of ordinary skill in the field to which this invention pertains. The "first", "second" and similar terms used in this invention do not denote any order, quantity or importance, but are only used to distinguish different components. Words such as "comprising" or "including" mean that the elements or items appearing before this word cover the elements or items listed after this word and their equivalents, without excluding other elements or items. Words such as "connected" or "coupled" are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. "Upper", "lower", "left", "right", etc. are only used to indicate relative position relationships, and when the absolute position of the object being described changes, the relative position relationship may also change accordingly.

[0039] Embodiment 1

[0040] As Figure 1 shown, this embodiment provides an asynchronous electroencephalogram (EEG) signal acquisition system, including:

[0041] A front-end asynchronous trigger EEG signal acquisition circuit, which is used to realize the asynchronous acquisition of EEG signals and improve the quality of EEG signals, so as to facilitate subsequent analysis and processing.

[0042] An embedded parallel logic control module, which is used to realize signal processing and data stream management. An EEG asynchronous physiological calculation chip is arranged in the embedded parallel logic control module to improve the speed and accuracy of data processing and the physiological calculation of emotions, and realize emotion classification.

[0043] Among them, the EEG signal acquisition circuit includes an EEG signal sensing electrode, an EEG signal preprocessing circuit and an asynchronous ADC circuit. The EEG signal sensing electrode is a flexible dry electrode of graphene / liquid metal to improve the signal reception quality and the comfort of the subject.

[0044] As Figure 2As shown, the EEG signal preprocessing circuit includes a low-noise amplifier, a low-pass filter, a 50Hz power frequency notch filter, and a band-pass filter connected in sequence. The main purpose of this circuit is to improve the quality of EEG signals for subsequent analysis and processing. The circuit first receives the original EEG signal, which contains the electrical activity information collected from the human brain. The weak original EEG signal received is amplified by a low-noise amplifier with a gain of 10 to 50 times. The amplified signal first passes through a low-pass filter to remove the high-frequency noise in the signal and retain the frequency components more critical for subsequent analysis. Subsequently, the signal passes through a 50Hz power frequency filter, which is specifically used to eliminate the interference generated by the power supply line. This step is particularly crucial when using EEG devices in an industrial environment because power frequency noise is a common interference source. Finally, the signal enters the band-pass filter. This step sets a specific frequency range that only allows the frequencies of theta waves (4 - 7Hz) and alpha waves (8 - 12Hz) to pass. The EEG signals in these two frequency ranges are very important in analyzing EEG activities because they are each associated with different brain functional states.

[0045] As Figure 3 shown, the asynchronous ADC circuit is mainly a Level-Crossing ADC circuit designed based on the asynchronous sampling principle, including a digital-to-analog converter, a multiplexer, a comparator, digital logic control, a counter, and digital logic gates, and connected according to signal logic control.

[0046] The sampling moment of this asynchronous ADC circuit is triggered by the change of the input signal itself, rather than an external clock. Specifically, it dynamically adjusts the sampling frequency by monitoring the crossing between the analog signal and a preset voltage threshold, thereby achieving efficient and accurate analog-to-digital conversion. The main components of the circuit and their functions are as follows: 1-bit DAC: This DAC is used to set the high (V H ) and low (V L ) voltage thresholds, which are adjusted according to the feedback control signal. These thresholds crucially determine the triggering timing of the ADC sampling process and are essential for the sensitive response to signal changes. Compared with traditional multi-bit DACs, the simplified 1-bit DAC significantly reduces power consumption and system complexity, making it particularly suitable for biomedical devices with limited power supply and long-term operation requirements. Multiplexer (MUX): The MUX is controlled by the logical output of the lower comparator (CMPD). CMPD compares the V ON output by the DAC with the reference voltage, where V M is the average value of V H and V L . According to the comparison result, the MUX dynamically selects to output V H or V L , in cooperation with V ONSent to the upper comparator CMPU. This ensures that the comparison window of the upper comparator is correctly set to capture the level crossing of the signal. Comparators (CMPD and CMPU): CMPD is responsible for monitoring the relationship between the DAC output V ON and the reference voltage V M , monitoring the direction change and threshold crossing of the signal. CMPU switches the comparison window according to the MUX output control signal. When the signal crosses the set threshold, it triggers data acquisition and CLC digital logic control. RLC and CLC logic: According to the characteristics of the EEG signal, two types of level crossing triggering methods are defined: continuous level crossing (CLC) and repetitive level crossing (RLC). The RLC logic controls the MUX to output rising or falling signals to the counter and generates the RLC crossing pulse C R . The CLC logic controls the DAC logic to generate the pulse C C triggered by the CLC crossing. These pulses form the change signal through a logical OR gate. The CLC trigger updates the entire system, while the RLC trigger only refreshes the counter. Up / Down Counter (U / DCounter): This counter records the number and direction of level crossing events, and this data is used to generate the digital representation of the analog signal.

[0047] The embedded parallel logic control module uses the FPGA as the core and has the configuration and logic control of the asynchronous ADC circuit built-in, which can achieve efficient signal processing and data flow management.

[0048] An EEG asynchronous physiological computing chip developed under the 40nm process technology is set in the embedded parallel logic control module, which is used to improve the speed and accuracy of data processing and physiological computing of emotions, and realize emotion classification. The chip includes a data cache module, a pulse coding module, and a pulse neural network classifier. The data cache module adopts a direct mapped cache architecture with a capacity of 2MB and a high-speed static random access memory (SRAM), supporting a 32-bit wide data interface. In the SRAM architecture, each read and write operation directly locates to a fixed storage location, making data access faster and reducing processing delay.

[0049] The pulse coding module uses a rate coding-based strategy, combined with the dynamic adjustment parameter para th , to effectively adjust the pulse thresholds of each channel to adapt to the changing signal intensity and improve the accuracy and efficiency of signal coding.

[0050] The pulse neural network classifier integrates a multi-layer neural network, including an input layer, multiple linear transformation layers, and a Leaky Integrate-and-Fire (LIF) neuron layer, and implements learning and optimization strategies. The classifier determines the final emotion category by comparing the pre-activation scores at the output layer of the network, achieving accurate emotion classification.

[0051] The system is used to effectively collect brain response signals under various external stimulus interventions, such as exercise intervention, neuroelectrical stimulation intervention, music intervention, etc. When external stimuli cause neural activity responses in specific brain regions, the electrical signals of these brain regions are collected by electroencephalogram (EEG) signal sensing electrodes and transmitted to the EEG signal preprocessing circuit. This circuit performs gain amplification by 10 to 50 times and denoising filtering on the weak original neural electrical signals. Multiple independent LC-ADCs in the asynchronous ADC circuit perform parallel asynchronous acquisition of the preprocessed EEG signals. Each LC-ADC unit dynamically adjusts its sampling threshold and frequency according to the actual level change of the signal, and triggers sampling only when a key change in the level occurs. This sampling mechanism significantly improves the time efficiency of data acquisition and reduces unnecessary data storage. The processed signals are transmitted in real time to the embedded parallel logic control module composed of an FPGA through the Serial Peripheral Interface (SPI). The embedded parallel logic control module is responsible for synchronously receiving, caching, and further analyzing these signals, including performing pulse coding, pulse feature extraction, and emotion classification through an SNN classifier. Finally, the result of emotion classification is output to the host application software through the Universal Asynchronous Receiver / Transmitter (UART) interface. The application software is responsible for visualizing and displaying the emotion classification results so that users can intuitively understand and analyze the data.

[0052] As Figure 4 shown, this embodiment also provides an asynchronous EEG signal acquisition method, including the following steps:

[0053] S1. Collect EEG signals and perform analysis and processing.

[0054] S2. After the processed EEG data is input, it is encoded by the pulse coding module to generate pulses. The pulse coding module introduces the parameter para th , optimizes the parameter para th , and dynamically adjusts the pulse threshold of each EEG processing channel. The pulse threshold of each EEG processing channel is calculated by the following formula:

[0055]

[0056] where N is the number of samples, l is the mean value of the signal change rate in the samples, r is the standard deviation of the signal change rate in the samples, and V th (k) is the variable threshold of the k-th input channel.

[0057] On each EEG processing channel, when the change rate of the signal exceeds the corresponding variable threshold, a positive pulse is generated; when the change rate of the signal exceeds the variable threshold in the negative direction, a negative pulse is generated.

[0058] For the parameter para thThe steps for optimization are as follows Figure 5 shown in the figure. Specifically: First, calculate the statistical description features from the pulse sequences of each EEG processing channel. After feature selection, calculate the pulse emission rate and the median of the time difference between pulses. Second, use an SVM classifier to evaluate the feature performance under different settings, adopt the Bayesian optimization method to find the optimal values, evaluate the performance of the feature set, and adjust according to the obtained classification accuracy. Use the leave-one-out cross-validation method to ensure the generalization ability of the evaluation process, that is, leave one sample as the test set each time, and use all the remaining samples for training.

[0059] S3. Input the generated pulses into a pulse neural network classifier (including an input layer, multiple linear transformation layers, and a Leaky Integrate-and-Fire (LIF) neuron layer) for analysis and processing, and finally output the emotion classification category.

[0060] After the generated pulses are input from the input layer of the pulse neural network classifier, in the first linear transformation layer, the input data is mapped to a 128-dimensional hidden space for linear transformation of the data. The output data is processed by the Leaky Integrate and Fire neuron layer to simulate the dynamic behavior of biological neurons, including the integration and leakage of the membrane potential, and the firing process when the threshold is reached. The output data is input to the second linear transformation layer again to map the 128-dimensional features to the dimension corresponding to the number of output categories. Finally, the output data is input to the second Leaky Integrate and Fire neuron layer again to further integrate the features and directly use them for the classification task. The entire network is trained by minimizing the cross-entropy loss, and the classification performance is optimized by updating the weights, and finally the emotion classification category is output.

[0061] Therefore, by adopting the above asynchronous EEG signal acquisition system and method, the present invention can efficiently acquire and process EEG signals for emotion state analysis.

[0062] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that they can still modify or equivalently replace the technical solutions of the present invention, and these modifications or equivalent replacements cannot make the modified technical solutions deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. An asynchronous EEG signal acquisition system, characterized in that: include: The front-end asynchronously triggered EEG signal acquisition circuit is used to realize asynchronous acquisition of EEG signals and improve the quality of EEG signals for subsequent analysis and processing; Embedded parallel logic control module for signal processing and data flow management; The embedded parallel logic control module is provided with an EEG asynchronous physiological computing chip for improving the speed and accuracy of data processing and physiological computing of emotions, thereby realizing emotion classification.

2. The asynchronous EEG signal acquisition system and method according to claim 1, characterized in that: The EEG signal acquisition circuit includes an EEG signal sensing electrode, an EEG signal preprocessing circuit and an asynchronous ADC circuit; the EEG signal sensing electrode is a flexible dry electrode of graphene / liquid metal to improve the signal reception quality and the comfort of the subject; the EEG signal preprocessing circuit includes a low-noise amplifier, a low-pass filter, a 50Hz industrial frequency notch filter and a band-pass filter connected in sequence; the asynchronous ADC circuit includes a digital-to-analog converter, a multiplexer, a comparator, a digital logic control, a counter, a digital logic gate, and is connected according to signal logic control.

3. The asynchronous EEG signal acquisition system and method according to claim 2, characterized in that: The embedded parallel logic control module uses FPGA as a core and has built-in configuration and logic control for the asynchronous ADC circuit.

4. The asynchronous EEG signal acquisition system and method according to claim 1, characterized in that: The EEG asynchronous physiological computing chip includes a data cache module, a pulse encoding module and a pulse neural network classifier.

5. The asynchronous EEG signal acquisition system and method according to claim 4, characterized in that: The data cache module adopts a direct-mapped cache architecture and a high-speed static random access memory, and the pulse encoding module uses a rate-based coding strategy combined with a dynamic adjustment parameter para th ; The pulse neural network classifier includes an input layer, a linear transformation layer and a Leaky Integrate-and-Fire neuron layer, and implements a learning and optimization strategy.

6. An asynchronous EEG signal acquisition method, characterized in that: The following steps are involved: S1, collect EEG signals and analyze and process them; S2, after the processed EEG data is input, it is encoded by the pulse encoding module to generate pulses; S3. Input the generated pulses into the pulse neural network classifier for analysis and processing, and finally output the emotion classification category.

7. The asynchronous EEG signal acquisition method according to claim 6, characterized in that: In S2, the pulse encoding module introduces the parameter para th , for the parameter para th Optimize and dynamically adjust the pulse threshold of each EEG processing channel. The pulse threshold of each EEG processing channel is calculated by the following formula: Where N is the number of samples, l is the mean of the signal change rate in the samples, r is the standard deviation of the signal change rate in the samples, and V th (k) is the variable threshold of the kth input channel; On each EEG processing channel, a positive pulse is generated when the rate of change of the signal exceeds the corresponding variable threshold; a negative pulse is generated when the rate of change of the signal exceeds the variable threshold in the negative direction.

8. The asynchronous EEG signal acquisition method according to claim 7, characterized in that: For parameter para th The optimization steps are as follows: first, the statistical descriptive features are calculated from the pulse train of each EEG processing channel. After feature selection, the median of the pulse firing rate and the time difference between pulses are calculated; second, the SVM classifier is used to evaluate the feature performance under different settings, and the Bayesian optimization method is used to find the optimal value. The performance of the feature set is evaluated and adjusted according to the obtained classification accuracy; the leave-one-out cross-validation method is used to ensure the generalization ability of the evaluation process, that is, one sample is left as the test set each time, and all other samples are used for training.

9. The asynchronous EEG signal acquisition method according to claim 6, characterized in that: In S3, after the generated pulses are input into the input layer of the pulse neural network classifier, the input data is mapped to a 128-dimensional latent space in the first linear transformation layer, and the linear transformation of the data is performed. The output data is processed by the LeakyIntegrate and Fire neuron layer to simulate the dynamic behavior of biological neurons, including the integration and leakage of membrane potential, and the discharge process when the threshold is reached; the output data is input into the second linear transformation layer again to map the 128-dimensional features to the dimensions corresponding to the number of output categories, and finally, the output data is input into the second Leaky Integrate and Fire neuron layer again to further integrate the features and directly use them for classification tasks; the entire network is trained by minimizing the cross entropy loss, and the classification performance is optimized by updating the weights, and finally the emotion classification category is output.

Images