Wearable eeg signal processing method and device based on feature fusion

Abstract

The application discloses a feature fusion-based wearable electroencephalogram signal processing method and device. The method comprises the following steps: synchronously latching the main channel electroencephalogram signal from a bioelectric electrode and the multi-modal auxiliary signal of each auxiliary sensor interface in a preset period; inputting the multi-modal auxiliary signal into a pre-trained interference identification model to output a structured interference quantization vector, wherein the interference quantization vector contains the confidence and intensity of different interference sources; processing the main channel electroencephalogram signal according to the confidence and intensity of different interference sources and a preset parameterized anti-interference algorithm library to obtain final electroencephalogram data; quantifying the quality reliability of the final electroencephalogram data, storing the ternary mapping relationship between the current period, the quality reliability and the final electroencephalogram data, and sending the ternary mapping relationship to a downstream application program. The application realizes adaptive matching of the processing strategy and the real-time interference scene, thereby effectively avoiding over-processing or under-processing and greatly improving the data robustness and accuracy of the downstream application.

Description

Technical Field

[0001] This application relates to the fields of brain-computer interface and neural engineering technology, and in particular to a wearable EEG signal processing method and device based on feature fusion. Background Technology

[0002] With the rapid development of brain-computer interfaces, the monitoring and analysis of electroencephalogram (EEG) signals has expanded from controlled laboratory environments to dynamic daily environments. It has become commonplace for users to wear head-mounted EEG devices for activities such as mobile office work, outdoor activities, driving, rehabilitation training, and even light exercise. In this dynamic environment, the EEG signal acquisition and processing system must have stable operating capabilities to provide a reliable data foundation for online analysis of brain states (such as attention, fatigue, and emotion).

[0003] In related technologies, the suppression of interference in EEG signals is mostly based on classic filtering techniques in the field of signal processing, such as bandpass filtering to remove power frequency interference and high-frequency noise, and adaptive filtering to suppress artifacts related to known reference signals.

[0004] However, in dynamic environments, signal degradation is often the result of the combined and coupled effects of multiple interference sources, such as motion artifacts, slow drift in skin impedance caused by sweat, and transient poor contact of devices. Existing suppression strategies targeting single, preset interference types cannot perform online diagnosis and quantification of the dominant interference mode in real-time processing, resulting in a lack of targeted processing.

[0005] Secondly, since the specific cause of the interference cannot be identified, existing technologies cannot dynamically adjust the processing algorithm and parameters according to the real-time situation. This one-size-fits-all approximation processing is prone to overprocessing or underprocessing in dynamic environments, making the effectiveness and fidelity of the final output signal unstable. Summary of the Invention

[0006] This application provides a wearable EEG signal processing method and apparatus based on feature fusion. To provide a basic understanding of some aspects of the disclosed embodiments, a brief summary is given below. This summary is not intended as a general commentary, nor is it intended to identify key / important components or describe the scope of protection of these embodiments. Its sole purpose is to present some concepts in a simple form as a prelude to the detailed description that follows.

[0007] In a first aspect, embodiments of this application provide a wearable EEG signal processing method based on feature fusion, applied to wearable EEG devices, the method comprising:

[0008] Synchronously latch the main channel EEG signal from the bioelectric electrodes and the multimodal auxiliary signals from each auxiliary sensor interface within a preset period. The multimodal auxiliary signals include at least the real-time contact impedance value of the electrodes, the instantaneous data of the inertial measurement unit, and the environmental data of the environmental sensor. The multimodal auxiliary signal is input into the pre-trained interference identification model, and the output is a structured interference quantization vector, which contains the confidence and intensity of different interference sources. Based on the confidence and intensity of different interference sources and the preset parameterized anti-interference algorithm library, the main channel EEG signal is processed to obtain the final EEG data; the preset parameterized anti-interference algorithm library stores multiple signal processing program modules that can be dynamically called. Each signal processing program module contains dynamically configurable adjustment parameters and each signal processing program module is used to suppress noise or eliminate artifacts for a certain type of interference source. The quality and reliability of the final EEG data are quantified, and the ternary mapping relationship between the current cycle, quality and reliability, and the final EEG data is stored and sent to downstream applications.

[0009] Secondly, embodiments of this application provide a wearable EEG signal processing device based on feature fusion, the device comprising: The multimodal signal acquisition module is used to synchronously latch the main channel EEG signal from the bioelectric electrodes and the multimodal auxiliary signals from each auxiliary sensor interface within a preset period. The multimodal auxiliary signals include at least the real-time contact impedance value of the electrodes, the instantaneous data of the inertial measurement unit, and the environmental data of the environmental sensor. The interference identification model processing module is used to input multimodal auxiliary signals into a pre-trained interference identification model and output a structured interference quantization vector, which contains the confidence and intensity of different interference sources. The EEG signal processing module is used to process the EEG signal of the main channel according to the confidence and intensity of different interference sources and the preset parameterized anti-interference algorithm library to obtain the final EEG data. The preset parameterized anti-interference algorithm library stores multiple signal processing program modules that can be dynamically called. Each signal processing program module contains dynamically configurable adjustment parameters and each signal processing program module is used to suppress noise or eliminate artifacts for a certain type of interference source. The data storage and distribution module is used to quantify the quality and reliability of the final EEG data, store the ternary mapping relationship between the current cycle, quality and reliability, and the final EEG data, and send it to the downstream application.

[0010] The technical solutions provided in this application embodiment may include the following beneficial effects: In this embodiment, on the one hand, after obtaining the interference quantization vector, signal processing program modules can be dynamically selected and combined from a pre-set parameterized anti-interference algorithm library. This process can dynamically adjust the execution order and specific parameters of the processing algorithm according to the confidence level, intensity, and coupling relationship of different interference sources, achieving adaptive matching between the processing strategy and the real-time interference scenario. This effectively avoids overprocessing or underprocessing, significantly improves the reliability of the output signal, and greatly enhances the data robustness and accuracy of downstream applications. On the other hand, by synchronously latching multimodal auxiliary signals and inputting them into a pre-trained interference identification model, the current interference can be diagnosed and quantified in a timely manner, obtaining interference quantization vectors containing the confidence level and intensity of different interference sources. This provides a precise list of interference causes for subsequent processing, thus solving the problem of traditional single, pre-set strategies lacking specificity, and making processing decisions based on evidence.

[0011] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and do not limit this application. Attached Figure Description

[0012] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0013] Figure 1 This is a schematic flowchart of a wearable EEG signal processing method based on feature fusion provided in an embodiment of this application; Figure 2 This is a schematic diagram of a wearable EEG device provided in an embodiment of this application; Figure 3 This is a model architecture diagram of a pre-trained interference identification model provided in an embodiment of this application; Figure 4 This is a schematic flowchart illustrating a dynamic algorithm combination list generation process provided in an embodiment of this application; Figure 5 This is a schematic diagram illustrating a parameter instantiation process provided in an embodiment of this application; Figure 6 This is a schematic block diagram illustrating a serial processing procedure for a main channel EEG signal provided in an embodiment of this application. Figure 7 This is a schematic block diagram of a wearable EEG signal processing procedure provided in an embodiment of this application; Figure 8 This is an interface diagram showing the parameter information of the main channel EEG signal and the multimodal auxiliary signals of each auxiliary sensor interface in the background, as provided in an embodiment of this application. Figure 9This is a flowchart illustrating an interference identification model training method provided in an embodiment of this application; Figure 10 This is a schematic diagram of the structure of a wearable EEG signal processing device based on feature fusion provided in an embodiment of this application; Figure 11 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0014] The following description and accompanying drawings fully illustrate specific embodiments of this application to enable those skilled in the art to practice them.

[0015] It should be understood that the described embodiments are merely some, not all, of the embodiments in this application. All other embodiments obtained by those skilled in the art based on the embodiments in this application without inventive effort are within the scope of protection of this application.

[0016] In the following description, when referring to the accompanying drawings, the same numbers in different drawings denote the same or similar elements unless otherwise indicated. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0017] In the description of this application, it should be understood that the terms "first," "second," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance. Those skilled in the art can understand the specific meaning of the above terms in this application based on the specific circumstances. Furthermore, in the description of this application, unless otherwise stated, "multiple" refers to two or more. "And / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship.

[0018] Currently, the suppression of interference in EEG signals is mostly based on classic filtering techniques in the field of signal processing, such as bandpass filtering to remove power frequency interference and high-frequency noise, and adaptive filtering to suppress artifacts related to known reference signals.

[0019] The inventors recognized that in dynamic environments, signal degradation is often the result of the combined and coupled effects of multiple interference sources, such as motion artifacts, slow drift in skin impedance caused by sweat, and momentary poor device contact. Existing suppression strategies targeting single, preset interference types cannot perform online diagnosis and quantification of the dominant interference pattern in real-time processing, resulting in a lack of targeted processing. Secondly, because the specific causes of interference cannot be identified, existing technologies struggle to dynamically adjust processing algorithms and parameters according to the real-time context. This one-size-fits-all approximation easily leads to overprocessing or underprocessing in dynamic environments, causing instability in the effectiveness and fidelity of the final output signal.

[0020] To address the existing technical problems, this application provides a wearable EEG signal processing method and device based on feature fusion to solve the aforementioned related technical issues. In the embodiments of this application, on the one hand, after obtaining the interference quantization vector, signal processing program modules can be dynamically selected and combined from a pre-set parameterized anti-interference algorithm library. This process can dynamically adjust the execution order and specific parameters of the processing algorithm according to the confidence level, intensity, and coupling relationship of different interference sources, achieving adaptive matching between the processing strategy and the real-time interference scenario. This effectively avoids overprocessing or underprocessing, significantly improves the reliability of the output signal, and greatly enhances the data robustness and accuracy of downstream applications. On the other hand, by synchronously latching multimodal auxiliary signals and inputting them into a pre-trained interference identification model, the current interference can be diagnosed and quantified in a timely manner, obtaining interference quantization vectors containing the confidence level and intensity of different interference sources. This provides a precise list of interference causes for subsequent processing, thus solving the problem of traditional single, pre-set strategies lacking specificity, making processing decisions more data-driven. The following describes the process in detail using exemplary embodiments.

[0021] The following will be combined with the appendix Figure 1 -Appendix Figure 9 This application provides a detailed description of the wearable EEG signal processing method based on feature fusion, as provided in the embodiments of this application. This method can be implemented using a computer program and can run on a wearable EEG signal processing device based on feature fusion and the von Neumann architecture. The computer program can be integrated into an application or run as a standalone utility application.

[0022] Please see Figure 1 This document provides a flowchart illustrating a wearable EEG signal processing method based on feature fusion, applicable to wearable EEG devices. Figure 1 As shown, the method in this application embodiment includes the following steps: S101, synchronously latches the main channel EEG signal from the bioelectric electrode and the multimodal auxiliary signals from each auxiliary sensor interface within a preset period. The multimodal auxiliary signals include at least the real-time contact impedance value of the electrode, the instantaneous data of the inertial measurement unit, and the environmental data of the environmental sensor. Synchronous latching involves simultaneously capturing and temporarily storing signal data from different sources under the same precise time reference (clock signal). This method ensures strict alignment of EEG signals and auxiliary sensor data on the time axis, avoiding data misalignment caused by acquisition time differences. The preset period is a pre-defined fixed time segment (e.g., every 100 milliseconds, every second, or every 256 sampling points). The wearable EEG device acquires data according to this fixed time segment. Bioelectric electrodes are sensors attached to the scalp to directly acquire weak brain potential changes (i.e., EEG signals). The main channel EEG signal is the core raw data acquired through the bioelectric electrodes, containing information about the user's brain activity, but also mixed with various noises. Multimodal auxiliary signals are data acquired by other types of sensors besides the EEG itself, used to help determine the current state or sources of interference (such as resistance, acceleration, temperature, etc.). The real-time electrode contact impedance value is a physical quantity that measures the quality of contact between the electrode and the scalp. Excessive impedance indicates loose electrodes, dried gel, or poor contact, which directly leads to increased signal noise or drift. Inertial measurement units (IMUs) typically contain chips for accelerometers and gyroscopes. They are used to detect the wearer's head movements (such as head shaking, nodding, and walking vibrations), and this data is often used to identify motion artifacts caused by movement. Environmental sensors are devices used to detect parameters of the physical environment around the wearer, such as temperature and humidity.

[0023] In some embodiments of this application, the intelligent EEG monitoring head-mounted device integrates a dry electrode EEG sensor, an impedance detection circuit, a six-axis IMU chip, and a temperature and humidity sensor. The system is set to a preset period of 128 milliseconds (corresponding to 128 data points at a sampling rate of 1000Hz). At this time, the main control chip first sends a latching pulse signal, and then the main channel reads the raw voltage waveform data collected by the bioelectric electrodes within the past 128 milliseconds as the main channel EEG signal. Auxiliary channel 1 reads the value measured by the impedance detection circuit at the current moment, for example, the impedance of the left frontal electrode is 45 kΩ, as the real-time contact impedance value of the electrode. Auxiliary channel 2 reads the three-axis acceleration and three-axis angular velocity of the IMU chip at the current moment, for example, detecting a rapid head nodding motion, with the Z-axis acceleration abruptly changing to 1.2g, as instantaneous data from the inertial measurement unit. Auxiliary channel 3 reads the values ​​from the temperature and humidity sensor, for example, the current ambient temperature is 25°C and the relative humidity is 60%, as environmental data from the environmental sensor. Finally, the above four sets of data (128-point EEG waveform, 45kΩ impedance value, 1.2g acceleration value, and 25℃ / 60% environmental value) were stamped with the same timestamp, packaged into a data package, and used as input for the pre-trained interference identification model.

[0024] For example, a user wears a wearable EEG device while on a treadmill for focus training. The user is running, their head rhythmically bobbing up and down with each stride, and due to sweating, the contact of the forehead electrodes becomes unstable. Within a preset period of 10 seconds to 10.128 seconds, the main channel EEG signal recorded a highly volatile waveform, which includes both the user's focus and... The wave also contained significant low-frequency fluctuations. At that instant, the system simultaneously latched the following multimodal auxiliary signals, including: the real-time contact impedance value of the electrodes: a sudden jump from the normal 20kΩ to 85kΩ (indicating poor contact due to sweating); instantaneous data from the inertial measurement unit: detecting a periodic impact with a frequency of 2Hz and an amplitude of 0.8g in the vertical direction (Y-axis) (corresponding to the foot vibrations of a user running); and environmental data from the environmental sensor: detecting regular changes in ambient light intensity (corresponding to the flickering of gym lights).

[0025] For example Figure 2 As shown, in the electrode array of the wearable EEG device, the left side is a mesh-like whole worn on the head, and the right side is first a biomass interface material used to penetrate hair and fit tightly to the skin to reduce contact resistance; secondly, breathable fabric electrodes, which adopt a micron-level (300μm) conductive fiber weaving structure to take into account both signal conduction and wearing comfort; and finally, a miniaturized integrated device, which highly integrates signal amplification and processing circuits on a flexible substrate, together constituting the brain-computer interface hardware of this application.

[0026] S102 inputs the multimodal auxiliary signal into the pre-trained interference identification model and outputs a structured interference quantization vector, which contains the confidence and intensity of different interference sources. The pre-trained interference identification model is an algorithm that has been trained using a large amount of historical data. This model can automatically determine the types and severity of interference present based on various input auxiliary signals. Different interference sources are specific causes of EEG signal distortion, such as motion artifacts caused by head movements, contact noise caused by loose electrodes, electromyographic interference caused by muscle tension, and power frequency interference caused by power fluctuations. Confidence level characterizes the model's belief in the existence of a certain interference; a higher value indicates that the model is more certain of the interference's presence. Intensity characterizes the severity or magnitude of the interference's impact on the EEG signal. For example, the more intense the movement, the greater the intensity value of the motion interference.

[0027] In one possible implementation, the system acquires multimodal auxiliary signals at the current moment, which are IMU data: X-axis acceleration. angular velocity along the Y-axis Impedance data: Right ear electrode impedance (Normal value) Environmental data: Low electromagnetic noise level. The system inputs the vector composed of the above values ​​into a pre-trained interference identification model. The model performs multiple calculations to analyze that high acceleration represents motion and high impedance represents poor contact. The model outputs a structured interference quantification vector, with the following data structure: [Interference source type, confidence level, intensity].

[0028] For example: Item 1: [Motion artifact, 0.95, 8.2] (meaning: 95% certainty that it is motion interference, intensity level 8.2) Item 2: [Contact noise, 0.88, 6.5] (meaning: 88% certainty that it is poor contact, intensity level 6.5) Item 3: [Power frequency interference, 0.05, 0.1] (meaning: almost no power frequency interference).

[0029] For example Figure 3 As shown, the pre-trained interference identification model includes an input layer, a feature fusion layer, a hidden layer, and an output layer.

[0030] In some embodiments of this application, the specific process of inputting multimodal auxiliary signals into a pre-trained interference identification model and outputting a structured interference quantization vector includes: the input layer normalizes and aligns the real-time contact impedance values ​​of the electrodes, the instantaneous data of the inertial measurement unit, and the environmental data of the environmental sensor, and outputs a first multidimensional feature vector; the feature fusion layer concatenates or weights the first multidimensional feature vector to generate a second multidimensional feature vector; the hidden layer performs nonlinear transformation and high-level feature extraction on the second multidimensional feature vector to learn and characterize the coupling relationship and independent patterns between different interference sources, and outputs high-level abstract features; the output layer maps the high-level abstract features to a structured interference quantization vector for output; the node structure of the output layer corresponds to a predefined set of interference source types, and each node outputs a multidimensional tuple, which includes at least the confidence and intensity of the corresponding interference source type, and the structured interference quantization vector is composed of multidimensional tuples corresponding to all interference source types.

[0031] S103, based on the confidence and intensity of different interference sources and the preset parameterized anti-interference algorithm library, the main channel EEG signal is processed to obtain the final EEG data; the preset parameterized anti-interference algorithm library stores multiple signal processing program modules that can be dynamically called. Each signal processing program module contains dynamically configurable adjustment parameters and each signal processing program module is used to suppress noise or eliminate artifacts for a type of interference source. In some embodiments of this application, the specific process of processing the main channel EEG signal to obtain the final EEG data based on the confidence and intensity of different interference sources and a preset parameterized anti-interference algorithm library includes: generating a dynamic algorithm combination list based on the confidence and intensity of different interference sources; the dynamic algorithm combination list includes module identifiers of multiple signal processing program modules and execution order information of each signal processing program module; dynamically calling each signal processing program module from the preset parameterized anti-interference algorithm library according to the module identifier; configuring dynamically configurable adjustment parameters of each signal processing program module to obtain multiple target signal processing program modules; and sequentially calling the corresponding target signal processing program modules according to the execution order information of each signal processing program module to serially process the main channel EEG signal to obtain the final EEG data.

[0032] The module identifier is a unique numeric or character label that indexes a specific signal processing program in the preset algorithm library, used to achieve precise location and dynamic loading of the algorithm module. Execution order information is metadata detailing the sequential call relationships during data stream processing, used to ensure the timing consistency and logical correctness of operations across modules in a serial processing flow. The signal processing program module is used to perform specific types of noise suppression or artifact removal on the input bioelectrical signals. Dynamically configurable adjustment parameters are variable parameters within the signal processing program module used to control algorithm behavior. The parameter values ​​can be adjusted in real time during system operation based on externally input interference quantification indicators to adapt to changing signal environments.

[0033] In this embodiment, an algorithm combination list is dynamically generated based on the confidence and intensity of the interference source, and the signal processing module is called from the algorithm library as needed and its adjustment parameters are configured accordingly. This achieves serialized processing of the EEG signal of the main channel. This method abandons the limitations of the traditional fixed processing flow and can flexibly adjust the processing strategy and parameter configuration according to the real-time interference characteristics. It avoids signal distortion caused by over-processing in interference-free scenarios and ensures the targeting and effectiveness of processing in complex interference environments. In the end, higher quality EEG data is obtained, and the adaptability and reliability of wearable EEG signal processing are improved.

[0034] In some embodiments of this application, the specific process of generating a dynamic algorithm combination list based on the confidence and intensity of different interference sources includes: comparing the confidence and intensity of different interference sources with the activation threshold corresponding to different interference sources to determine at least one target interference source that is greater than the activation threshold; obtaining the module identifier of the signal processing program module corresponding to each target interference source from the pre-constructed mapping relationship between interference sources and module identifiers of processing program modules; determining the execution order information of each signal processing program module according to the execution priority and logical dependency relationship between each signal processing program module in the preset parameterized anti-interference algorithm library; and using the module identifier of the signal processing program module corresponding to each target interference source and the execution order information of each signal processing program module as a dynamic algorithm combination list.

[0035] For example Figure 4 As shown, Figure 4This is a schematic flowchart illustrating the dynamic algorithm combination list generation process provided in this application. First, the confidence level and intensity of different interference sources are determined by combining a pre-trained interference identification model. Then, by comparing the confidence level and intensity with a preset activation threshold, target interference sources with values ​​greater than the threshold are selected, while interference sources with values ​​less than the threshold are skipped. Next, a mapping table is queried to obtain the signal processing program module identifier corresponding to the target interference source. Furthermore, based on the execution priority and logical dependency relationships between modules in the preset parameterized anti-interference algorithm library, the execution order information of each signal processing program module is determined. Finally, the module identifiers and execution order information are integrated into a dynamic algorithm combination list to guide the subsequent signal processing flow.

[0036] In one possible implementation, a smart EEG headset is monitoring a user's attention in a quiet indoor environment, but the user is chewing gum. The system reads the current interference data, obtaining: EMG interference (chewing): 95% confidence, intensity 7.0; EOG interference (blinking): 40% confidence, intensity 2.0; Power frequency interference (power supply): 99% confidence, intensity 1.5. The system reads preset activation thresholds, obtaining: EMG threshold: confidence > 80% and intensity > 5.0; EOG threshold: confidence > 60% and intensity > 3.0; Power frequency threshold: confidence > 90% and intensity > 1.0. At this point, the EMG interference (95% > 80%, 7.0 > 5.0) is identified as the target interference source. The EOG interference (40% < 60%) is ignored. The power frequency interference (99% > 90%, 1.5 > 1.0) is identified as the target interference source. The system queries the mapping table and finds that the target interference source, electromyography interference, corresponds to the module identifier EMG_Filter. The target interference source, power frequency interference, corresponds to the module identifier Notch_50Hz. The logical dependencies in the algorithm library are read: Notch_50Hz must be executed before EMG_Filter. Therefore, the first execution order is Notch_50Hz, and the second is EMG_Filter. The final output is a dynamic algorithm combination list: [{ID:Notch_50Hz,Order:1},{ID:EMG_Filter,Order:2}].

[0037] In some embodiments of this application, the specific process of configuring dynamically configurable adjustment parameters for each signal processing module to obtain multiple target signal processing modules includes: extracting the intensity corresponding to the target interference source for each signal processing module from the interference quantization vector; determining the target adjustment parameters associated with each signal processing module and the target interference source according to a preset parameter mapping relationship; wherein, the parameter mapping rule defines the mapping relationship between the interference source type, the signal processing module, and the dynamically configurable adjustment parameters; converting the intensity corresponding to the target interference source into a specific parameter value of the target adjustment parameter; and assigning the specific parameter value to the target adjustment parameter of the signal processing module to instantiate and configure each signal processing module to obtain multiple target signal processing modules.

[0038] The preset parameter mapping relationship is a predefined lookup table that establishes an association index between the interference source type, the signal processing program module identifier, and the dynamically configurable adjustment parameters, used to determine the specific parameter objects that need to be adjusted for a particular module in the current interference scenario.

[0039] For example Figure 5 As shown, Figure 5 This is a schematic diagram of a parameter instantiation process provided in this application. First, each target module is traversed to extract the intensity of the corresponding interference source. Then, by querying the preset parameter mapping rules, the parameter to be adjusted associated with the interference source type and the signal processing program module is determined. Next, the interference source intensity value is converted into the specific parameter value of the adjustment parameter. Finally, the signal processing program module is instantiated and configured based on the specific parameter value to generate the target signal processing program module. By iteratively checking whether there are any unconfigured modules, until all modules are configured, multiple target signal processing program modules are finally ready.

[0040] In one possible implementation, the system reads the intensity value of 0.85 (normalized value, range 0-1) corresponding to the target interference source, motion artifact, from the current interference quantization vector. The system queries a preset parameter mapping table, and the rules show that for motion artifact interference, the key parameter that the corresponding adaptive motion artifact removal module needs to adjust is the smoothing window width of the filter. Therefore, the target adjustment parameter is determined to be the smoothing window width. The system converts the intensity value 0.85 into a specific parameter value according to a preset conversion formula (e.g., window width = base value + intensity × coefficient). Assuming the base value is 10 and the coefficient is 20, the calculated specific parameter value is 10 + 0.85 × 20 = 27. The value 27 is assigned to the smoothing window width parameter of the adaptive motion artifact removal module. This module has been instantiated and configured, becoming a target signal processing program module, and its internal parameters have been precisely adjusted to handle motion interference with an intensity of 0.85.

[0041] In some embodiments of this application, the specific process of sequentially calling the corresponding target signal processing program modules based on the execution order information of each signal processing program module to serially process the main channel EEG signal and obtain the final EEG data includes: initializing a sliding window and using the main channel EEG signal as the initial input data stream of the sliding window; calling the target signal processing program module corresponding to each signal processing program module based on the execution order information of each signal processing program module; reading the current input data stream from the sliding window; wherein, for the first called target signal processing program module, the current input data stream is the main channel EEG signal; for other target signal processing program modules, the current input data stream is the intermediate processing data stream output by the previous target signal processing program module; inputting the current input data stream into the target signal processing program module to execute the interference source processing function of the target signal processing program according to the adjustment parameters configured in the target signal processing program module, thereby obtaining the intermediate processing data stream or the final processing result; writing the intermediate processing data stream to the sliding window as the current input data stream of the next target signal processing program module; or, when each target signal processing program module finishes execution, using the final processing result as the final EEG data.

[0042] The sliding window is a buffer mechanism used to extract fixed-length data segments from continuous time series data. It achieves continuous segmentation of the data stream by moving the starting position according to a preset step size, so as to maintain the real-time performance and continuity of the processing.

[0043] For example Figure 6 As shown, Figure 6 This application provides a schematic block diagram of a serial processing procedure for main channel EEG signals. The process begins by initializing a sliding window and using the main channel EEG signal as the initial input data stream. Subsequently, according to a preset execution order, target signal processing modules are called sequentially. The current input data stream is read from the sliding window (for the first module, it's the initial input data stream; for subsequent modules, it's the intermediate processing data stream output by the previous module). Each module is then driven to execute an interference source processing function based on its configured adjustment parameters to generate an intermediate processing data stream or the final processing result. If the current module is not the last, the intermediate processing data stream is written back to the sliding window for the next module to call and the loop continues until all modules have finished executing. Finally, the final processing result is output as the final EEG data.

[0044] In one possible implementation, the system initializes a sliding window, loading a 2-second main channel EEG signal (containing power frequency noise and motion noise) as the initial input data stream. According to the execution order, the first target signal processing module, such as a notch filter module, is called. The current input data stream, which is the raw main channel EEG signal, is read from the sliding window. This data stream is then input to the notch filter module, which executes an interference source processing function (50Hz notch filter algorithm) according to preset parameters, outputting an intermediate processed data stream (power frequency noise removed, but still containing motion noise). The system writes this intermediate processed data stream back to the sliding window, overwriting or updating the original raw data, as input for the next stage. The second target signal processing module, such as a motion artifact removal module, is called. The current input data stream, which is the intermediate processed data stream output from the previous stage, is read from the sliding window. This data stream is then input to the motion artifact removal module, which executes an interference source processing function (motion artifact correction algorithm), outputting the final processing result. Since all modules have completed execution, the system determines this final processing result as the final EEG data.

[0045] For example Figure 7 As shown, Figure 7 This is a schematic block diagram of a wearable EEG signal processing procedure provided in this application. First, multimodal signal acquisition is performed, and interference identification and quantization are performed on the acquired signals to generate an interference quantization vector. Then, the processing path is dynamically selected by determining whether the interference quantization vector is greater than a preset activation threshold: if it is greater than the activation threshold, dynamic algorithm combination is triggered, and then parameter configuration and serial processing are performed; if it is not greater than the activation threshold, pass-through output is performed. Finally, the signal after serial processing or the signal after pass-through output needs to be quality evaluated to obtain data sent to the downstream application.

[0046] Among them, the parameter information in the background includes the main channel EEG signal and the multimodal auxiliary signals of each auxiliary sensor interface, such as... Figure 8 As shown.

[0047] S104 quantifies the quality reliability of the final EEG data and stores the ternary mapping relationship between the current cycle, quality reliability, and final EEG data, and sends it to the downstream application.

[0048] In some embodiments of this application, the specific process of quantifying the quality reliability of the final EEG data includes: calculating at least one time-frequency domain feature index of the final EEG data, wherein the time-frequency domain feature index is used to characterize the signal quality of the final EEG data in the time domain or frequency domain; calculating a basic quality score based on at least one time-frequency domain feature index; calculating an interference environment correction factor based on the confidence and intensity of at least one target interference source; and fusing the basic quality score with the interference environment correction factor to obtain a scalar form of quality reliability.

[0049] Among them, time-frequency domain feature indicators are numerical parameters extracted from the final EEG data to quantify the statistical characteristics of the signal in the time or frequency dimensions. They objectively reflect the signal quality of the final EEG data, such as signal-to-noise ratio, power spectral density, root mean square value, or spectral entropy. The baseline quality score is calculated based on the time-frequency domain feature indicators using a preset mapping function or weighting algorithm, and is used to characterize the original quality level of the final EEG data under the current signal characteristics. The interference environment correction factor refers to a weighting coefficient or correction amount calculated based on the confidence and intensity of at least one target interference source using a preset attenuation function or logical rule, used to reflect the potential impact of the current interference environment on the quality of the final EEG data.

[0050] In one possible implementation, the quality of the processed final EEG data has been assessed, with the primary interference source identified as power line interference, with a confidence level of 0.85 and an intensity of 0.7 (after normalization). First, time-frequency domain characteristic indicators of the final EEG data are calculated, such as a signal-to-noise ratio (SNR) of 15 dB and a spectral flatness of 0.6. Based on a pre-defined mapping table, the SNR of 15 dB is mapped to a basic quality score. The score is based on a spectral flatness of 0.6 and a mapping-based quality score. The final baseline quality score is calculated using a weighted average method. ,For example Based on the confidence level of the target interference source (power frequency interference). and strength The interference environment correction factor is calculated using a preset product attenuation function. .For example, Finally, the basic quality score will be calculated. With interference environment correction factor Perform fusion. For example, use multiplicative fusion to obtain the scalar form of quality reliability. .

[0051] In some embodiments of this application, the current cycle and quality reliability in the ternary mapping relationship are sent to the downstream brain-computer interface decoder application through a preset inter-process communication interface or network protocol (such as TCP / IP, MQTT), providing a reliable data foundation for online analysis of brain states (such as attention, fatigue, and emotion).

[0052] In this embodiment, on the one hand, after obtaining the interference quantization vector, signal processing program modules can be dynamically selected and combined from a pre-set parameterized anti-interference algorithm library. This process can dynamically adjust the execution order and specific parameters of the processing algorithm according to the confidence level, intensity, and coupling relationship of different interference sources, achieving adaptive matching between the processing strategy and the real-time interference scenario. This effectively avoids overprocessing or underprocessing, significantly improves the reliability of the output signal, and greatly enhances the data robustness and accuracy of downstream applications. On the other hand, by synchronously latching multimodal auxiliary signals and inputting them into a pre-trained interference identification model, the current interference can be diagnosed and quantified in a timely manner, obtaining interference quantization vectors containing the confidence level and intensity of different interference sources. This provides a precise list of interference causes for subsequent processing, thus solving the problem of traditional single, pre-set strategies lacking specificity, and making processing decisions based on evidence.

[0053] Please see Figure 9 The following is a flowchart illustrating an interference identification model training method provided in this application embodiment. Figure 9 As shown, the method in this application embodiment includes the following steps: S201, acquire each multimodal auxiliary signal sample synchronously collected within the historical sampling period; S202, determine the structured interference quantization vector truth label corresponding to each multimodal auxiliary signal sample, the interference quantization vector truth label contains the confidence and intensity for multiple interference sources; S203, store the binary mapping relationship between each multimodal auxiliary signal sample and the structured interference quantization vector truth label corresponding to each multimodal auxiliary signal sample, and obtain each set of historical samples; S204, Create an interference identification model; In some embodiments of this application, the specific process of creating an interference identification model includes: using a machine learning algorithm to construct an input layer for receiving and preprocessing multimodal auxiliary signals, a feature fusion layer for fusing multimodal features, at least one hidden layer for performing nonlinear feature transformation, and an output layer for outputting a structured interference quantization vector; integrating the input layer, feature fusion layer, hidden layer, and output layer sequentially into a network structure; defining a composite loss function; the composite loss function includes a classification loss component for measuring the difference between the interference source confidence level output by the model and the ground truth label, and a regression loss component for measuring the difference between the interference source intensity output by the model and the ground truth label; integrating the composite loss function into the network structure to obtain the interference identification model.

[0054] The functional expression for the composite loss function is as follows: ; in, It is a composite loss function, representing the model parameters. The function is used to measure the total error between the model's overall output and the true label. The goal of training is to minimize the value of this function. This is the classification loss weight coefficient. It's a hyperparameter used to adjust the contribution of the classification task (interference source identification) to the total loss. By adjusting... The model can balance the two tasks of identifying the type of interference and quantifying the intensity of interference. This is the classification loss component. This component is specifically used to measure the difference between the error source confidence level in the model output and the ground truth label. It is the confidence vector of the interference source predicted by the model (e.g., the output probability distribution of the Softmax layer). These are the interference source category labels in the truth labels (e.g., one-hot encoded labels). This can be implemented using the cross-entropy loss function. This is the regression loss weighting coefficient. It is a hyperparameter used to adjust the contribution of the regression task (interference source intensity quantification) to the total loss. This is the regression loss component. This component is specifically used to measure the difference between the intensity of the disturbance source in the model output and the ground truth label. It is the predicted intensity value of the interference source (e.g., the output scalar of the regression layer). This is the true value of the interference source intensity in the truth label. It can be implemented using the Mean Squared Error (MSE) loss function or the Smooth L1 Loss.

[0055] S205, input the multimodal auxiliary signal samples from the target historical samples into the interference identification model to obtain the predicted structured interference quantization vector; the target historical samples are each group of historical samples; S206, calculate the loss value between the predicted structured interference quantization vector and the ground truth label in the target historical sample; when the loss value reaches the minimum, the pre-trained interference identification model is obtained.

[0056] In this embodiment, on the one hand, after obtaining the interference quantization vector, signal processing program modules can be dynamically selected and combined from a pre-set parameterized anti-interference algorithm library. This process can dynamically adjust the execution order and specific parameters of the processing algorithm according to the confidence level, intensity, and coupling relationship of different interference sources, achieving adaptive matching between the processing strategy and the real-time interference scenario. This effectively avoids overprocessing or underprocessing, significantly improves the reliability of the output signal, and greatly enhances the data robustness and accuracy of downstream applications. On the other hand, by synchronously latching multimodal auxiliary signals and inputting them into a pre-trained interference identification model, the current interference can be diagnosed and quantified in a timely manner, obtaining interference quantization vectors containing the confidence level and intensity of different interference sources. This provides a precise list of interference causes for subsequent processing, thus solving the problem of traditional single, pre-set strategies lacking specificity, and making processing decisions based on evidence.

[0057] The following are embodiments of the apparatus described in this application, which can be used to execute the embodiments of the method described in this application. For details not disclosed in the apparatus embodiments of this application, please refer to the embodiments of the method described in this application.

[0058] Please see Figure 10 This illustration shows a schematic diagram of a wearable EEG signal processing device based on feature fusion, provided in an exemplary embodiment of this application. This wearable EEG signal processing device based on feature fusion can be implemented as all or part of an electronic device through software, hardware, or a combination of both. The device 1 includes a multimodal signal acquisition module 10, an interference identification model processing module 20, an EEG signal processing module 30, and a data storage and distribution module 40.

[0059] The multimodal signal collection module 10 is used to synchronously latch the main channel EEG signal from the bioelectric electrodes and the multimodal auxiliary signals from each auxiliary sensor interface within a preset period. The multimodal auxiliary signals include at least the real-time contact impedance value of the electrodes, the instantaneous data of the inertial measurement unit, and the environmental data of the environmental sensor. The interference identification model processing module 20 is used to input multimodal auxiliary signals into a pre-trained interference identification model and output a structured interference quantization vector, which includes the confidence and intensity of different interference sources. The EEG signal processing module 30 is used to process the EEG signal of the main channel according to the confidence and intensity of different interference sources and the preset parameterized anti-interference algorithm library to obtain the final EEG data. The preset parameterized anti-interference algorithm library stores multiple signal processing program modules that can be dynamically called. Each signal processing program module contains dynamically configurable adjustment parameters and each signal processing program module is used to suppress noise or eliminate artifacts for a certain type of interference source. The data storage and distribution module 40 is used to quantify the quality and reliability of the final EEG data, store the ternary mapping relationship between the current cycle, quality and reliability, and the final EEG data, and send it to the downstream application.

[0060] It should be noted that the wearable EEG signal processing device based on feature fusion provided in the above embodiments is only illustrated by the division of the above functional modules when executing the wearable EEG signal processing method based on feature fusion. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above. In addition, the wearable EEG signal processing device based on feature fusion provided in the above embodiments and the wearable EEG signal processing method embodiments based on feature fusion belong to the same concept, and the implementation process is detailed in the method embodiments, which will not be repeated here.

[0061] The sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0062] In this embodiment, on the one hand, after obtaining the interference quantization vector, signal processing program modules can be dynamically selected and combined from a pre-set parameterized anti-interference algorithm library. This process can dynamically adjust the execution order and specific parameters of the processing algorithm according to the confidence level, intensity, and coupling relationship of different interference sources, achieving adaptive matching between the processing strategy and the real-time interference scenario. This effectively avoids overprocessing or underprocessing, significantly improves the reliability of the output signal, and greatly enhances the data robustness and accuracy of downstream applications. On the other hand, by synchronously latching multimodal auxiliary signals and inputting them into a pre-trained interference identification model, the current interference can be diagnosed and quantified in a timely manner, obtaining interference quantization vectors containing the confidence level and intensity of different interference sources. This provides a precise list of interference causes for subsequent processing, thus solving the problem of traditional single, pre-set strategies lacking specificity, and making processing decisions based on evidence.

[0063] This application also provides a computer-readable medium having program instructions stored thereon, which, when executed by a processor, implement the wearable EEG signal processing method based on feature fusion provided in the above-described method embodiments.

[0064] This application also provides a computer program product containing instructions that, when run on a computer, cause the computer to execute the wearable EEG signal processing method based on feature fusion of the above-described method embodiments.

[0065] Please see Figure 11 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Figure 11As shown, the electronic device 1000 may include: at least one processor 1001, at least one network interface 1004, a user interface 1003, a memory 1005, and at least one communication bus 1002.

[0066] The communication bus 1002 is used to realize the connection and communication between these components.

[0067] The user interface 1003 may include a display screen and a camera. Optionally, the user interface 1003 may also include a standard wired interface and a wireless interface.

[0068] The network interface 1004 may optionally include a standard wired interface or a wireless interface (such as a Wi-Fi interface).

[0069] The processor 1001 may include one or more processing cores. The processor 1001 connects to various parts within the electronic device 1000 using various interfaces and lines. It executes various functions and processes data by running or executing instructions, programs, code sets, or instruction sets stored in the memory 1005, and by calling data stored in the memory 1005. Optionally, the processor 1001 may be implemented using at least one hardware form selected from Digital Signal Processing (DSP), Field-Programmable Gate Array (FPGA), and Programmable Logic Array (PLA). The processor 1001 may integrate one or more of the following: a Central Processing Unit (CPU), a Graphics Processing Unit (GPU), and a modem. The CPU primarily handles the operating system, user interface, and applications; the GPU is responsible for rendering and drawing the content to be displayed on the screen; and the modem handles wireless communication. It is understood that the modem may also be implemented as a separate chip without being integrated into the processor 1001.

[0070] The memory 1005 may include random access memory (RAM) or read-only memory. Optionally, the memory 1005 may include a non-transitory computer-readable storage medium. The memory 1005 can be used to store instructions, programs, code, code sets, or instruction sets. The memory 1005 may include a program storage area and a data storage area, wherein the program storage area may store instructions for implementing an operating system, instructions for at least one function (such as touch function, sound playback function, image playback function, etc.), instructions for implementing the above-described method embodiments, etc.; the data storage area may store data involved in the above-described method embodiments, etc. Optionally, the memory 1005 may also be at least one storage system located remotely from the aforementioned processor 1001. Figure 11 As shown, the memory 1005, which serves as a computer storage medium, may include an operating system, a network communication module, a user interface module, and a wearable EEG signal processing application based on feature fusion.

[0071] exist Figure 11 In the illustrated electronic device 1000, the user interface 1003 is mainly used to provide an input interface for the user and to acquire user input data; while the processor 1001 can be used to call the wearable EEG signal processing application based on feature fusion stored in the memory 1005, and specifically perform the following operations: Synchronously latch the main channel EEG signal from the bioelectric electrodes and the multimodal auxiliary signals from each auxiliary sensor interface within a preset period. The multimodal auxiliary signals include at least the real-time contact impedance value of the electrodes, the instantaneous data of the inertial measurement unit, and the environmental data of the environmental sensor. The multimodal auxiliary signal is input into the pre-trained interference identification model, and the output is a structured interference quantization vector, which contains the confidence and intensity of different interference sources. Based on the confidence and intensity of different interference sources and the preset parameterized anti-interference algorithm library, the main channel EEG signal is processed to obtain the final EEG data; the preset parameterized anti-interference algorithm library stores multiple signal processing program modules that can be dynamically called. Each signal processing program module contains dynamically configurable adjustment parameters and each signal processing program module is used to suppress noise or eliminate artifacts for a certain type of interference source. The quality and reliability of the final EEG data are quantified, and the ternary mapping relationship between the current cycle, quality and reliability, and the final EEG data is stored and sent to downstream applications.

[0072] In one embodiment, when the processor 1001 processes the main channel EEG signal according to the confidence and intensity of different interference sources and a preset parameterized anti-interference algorithm library to obtain the final EEG data, it specifically performs the following operations: Based on the confidence level and intensity of different interference sources, a dynamic algorithm combination list is generated; the dynamic algorithm combination list includes the module identifiers of multiple signal processing program modules and the execution order information of each signal processing program module; Based on the module identifier, each signal processing program module is dynamically called from the preset parameterized anti-interference algorithm library; Configure the dynamically configurable adjustment parameters of each signal processing module to obtain multiple target signal processing modules; Based on the execution order information of each signal processing module, the corresponding target signal processing module is called sequentially to process the main channel EEG signal serially, thus obtaining the final EEG data.

[0073] In one embodiment, when the processor 1001 generates a dynamic algorithm combination list based on the confidence and intensity of different interference sources, it specifically performs the following operations: By comparing the confidence and intensity of different interference sources with the activation thresholds corresponding to different interference sources, at least one target interference source with an activation threshold greater than the activation threshold can be identified. Obtain the module identifier of the signal processing module corresponding to each target interference source from the pre-built mapping relationship between the module identifiers of the interference sources and the processing module; Based on the execution priority and logical dependency relationships between the signal processing program modules in the preset parameterized anti-interference algorithm library, the execution order information of each signal processing program module is determined; The module identifier of the signal processing program module corresponding to each target interference source and the execution order information of each signal processing program module are used as a dynamic algorithm combination list.

[0074] In one embodiment, when the processor 1001 executes dynamically configurable adjustment parameters to configure each signal processing module and obtains multiple target signal processing modules, it specifically performs the following operations: Extract the intensity of the target interference source corresponding to each signal processing module from the interference quantization vector; Based on the preset parameter mapping relationship, the target adjustment parameters associated with each signal processing program module and target interference source are determined; wherein, the parameter mapping rule defines the mapping relationship between interference source type, signal processing program module and dynamically configurable adjustment parameters; Convert the intensity corresponding to the target interference source into specific parameter values ​​of the target adjustment parameters; Specific parameter values ​​are assigned to the target adjustment parameters of the signal processing program module to instantiate and configure each signal processing program module, resulting in multiple target signal processing program modules.

[0075] In one embodiment, when the processor 1001 executes the execution order information of each signal processing program module, sequentially calls its corresponding target signal processing program module to serially process the main channel EEG signal to obtain the final EEG data, it specifically performs the following operations: Initialize the sliding window and use the main channel EEG signal as the initial input data stream for the sliding window; Based on the execution order information of each signal processing module, the target signal processing module corresponding to each signal processing module is called; The current input data stream is read from the sliding window; for the first target signal processing module called, the current input data stream is the main channel EEG signal; for other target signal processing modules called, the current input data stream is the intermediate processing data stream output by the previous target signal processing module. The current input data stream is input into the target signal processing program module, and the interference source processing function of the target signal processing program is executed on the current input data stream according to the adjustment parameters configured in the target signal processing program module, so as to obtain the intermediate processing data stream or the final processing result. The intermediate processing data stream is written to a sliding window to serve as the current input data stream for the next target signal processing module; or, upon completion of each target signal processing module, the final processing result is used as the final EEG data.

[0076] In one embodiment, when the processor 1001 performs the quantification of the quality reliability of the final EEG data, it specifically performs the following operations: Calculate at least one time-frequency domain feature index of the final EEG data, which is used to characterize the signal quality of the final EEG data in the time or frequency domain. Calculate the basic quality score based on at least one time-frequency domain characteristic index; Calculate the interference environment correction factor based on the confidence level and intensity of at least one target interference source; By fusing the basic quality score with the interference environment correction factor, a scalar form of quality reliability is obtained.

[0077] In one embodiment, when the processor 1001 executes the following operations when it inputs a multimodal auxiliary signal into a pre-trained interference identification model and outputs a structured interference quantization vector: The input layer normalizes and aligns the real-time contact impedance values ​​of the electrodes, the instantaneous data of the inertial measurement unit, and the environmental data of the environmental sensor with time, and outputs the first multi-dimensional feature vector. The feature fusion layer concatenates or weights the first multidimensional feature vector to generate the second multidimensional feature vector; The hidden layer performs nonlinear transformation and high-level feature extraction on the second multidimensional feature vector to learn and characterize the coupling relationship and independent patterns between different interference sources, and outputs high-level abstract features; The output layer maps high-level abstract features into structured interference quantization vectors for output. The node structure of the output layer corresponds to a predefined set of interference source types. Each node outputs a multidimensional tuple, which includes at least the confidence and intensity of the corresponding interference source type. The structured interference quantization vector is composed of multidimensional tuples corresponding to all interference source types.

[0078] In one embodiment, when the processor 1001 executes the generation of a pre-trained interference identification model, it specifically performs the following operations: Acquire each multimodal auxiliary signal sample synchronously acquired within the historical sampling period; Determine the structured ground truth label of the interference quantization vector corresponding to each multimodal auxiliary signal sample. The ground truth label of the interference quantization vector contains the confidence and intensity for multiple interference sources. Store the binary mapping relationship between each multimodal auxiliary signal sample and the structured interference quantization vector truth label corresponding to each multimodal auxiliary signal sample to obtain each set of historical samples; Create an interference identification model; The multimodal auxiliary signal samples in the target historical samples are input into the interference identification model to obtain the predicted structured interference quantization vector; the target historical samples are each group of historical samples; Calculate the loss value between the predicted structured interference quantization vector and the ground truth label in the target historical sample; when the loss value reaches its minimum, the pre-trained interference identification model is obtained.

[0079] In one embodiment, when the processor 1001 executes the creation of the interference identification model, it specifically performs the following operations: Using machine learning algorithms, an input layer is constructed for receiving and preprocessing multimodal auxiliary signals, a feature fusion layer for fusing multimodal features, at least one hidden layer for performing nonlinear feature transformation, and an output layer for outputting structured interference quantization vectors. The input layer, feature fusion layer, hidden layer, and output layer are sequentially integrated into a network structure. Define a composite loss function; the composite loss function includes a classification loss component that measures the difference between the confidence of the interference source in the model output and the true label, and a regression loss component that measures the difference between the intensity of the interference source in the model output and the true label. By integrating the composite loss function into the network structure, an interference identification model is obtained.

[0080] In this embodiment, on the one hand, after obtaining the interference quantization vector, signal processing program modules can be dynamically selected and combined from a pre-set parameterized anti-interference algorithm library. This process can dynamically adjust the execution order and specific parameters of the processing algorithm according to the confidence level, intensity, and coupling relationship of different interference sources, achieving adaptive matching between the processing strategy and the real-time interference scenario. This effectively avoids overprocessing or underprocessing, significantly improves the reliability of the output signal, and greatly enhances the data robustness and accuracy of downstream applications. On the other hand, by synchronously latching multimodal auxiliary signals and inputting them into a pre-trained interference identification model, the current interference can be diagnosed and quantified in a timely manner, obtaining interference quantization vectors containing the confidence level and intensity of different interference sources. This provides a precise list of interference causes for subsequent processing, thus solving the problem of traditional single, pre-set strategies lacking specificity, and making processing decisions based on evidence.

[0081] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program for wearable EEG signal processing based on feature fusion can be stored in a computer-readable storage medium. When executed, the program can include the processes of the embodiments of the above methods. The storage medium for the program for wearable EEG signal processing based on feature fusion can be a magnetic disk, optical disk, read-only memory, or random access memory, etc.

[0082] The above-disclosed embodiments are merely preferred embodiments of this application and should not be construed as limiting the scope of this application. Therefore, any equivalent variations made in accordance with the claims of this application shall still fall within the scope of this application.

Claims

1. A wearable EEG signal processing method based on feature fusion, characterized in that, Applied to wearable EEG devices, the method includes: The system synchronously latches the main channel EEG signals from the bioelectric electrodes and the multimodal auxiliary signals from each auxiliary sensor interface within a preset period. The multimodal auxiliary signals include at least the real-time contact impedance values ​​of the electrodes, the instantaneous data of the inertial measurement unit, and the environmental data of the environmental sensors, including temperature and humidity. The inertial measurement unit includes an accelerometer and a gyroscope. The multimodal auxiliary signal is input into a pre-trained interference identification model, which outputs a structured interference quantization vector. The interference quantization vector includes the confidence and intensity of different interference sources. The confidence is used to characterize the model's credibility with the existing interference sources. The intensity is used to characterize the severity or magnitude of the interference's effect on the EEG signal. Based on the confidence and intensity of different interference sources and a preset parameterized anti-interference algorithm library, the main channel EEG signal is processed to obtain the final EEG data. This includes: generating a dynamic algorithm combination list based on the confidence and intensity of the different interference sources; the dynamic algorithm combination list includes module identifiers of multiple signal processing program modules and execution order information for each signal processing program module; dynamically calling each signal processing program module from the preset parameterized anti-interference algorithm library based on the module identifiers; configuring dynamically configurable adjustment parameters for each signal processing program module to obtain multiple target signal processing program modules; sequentially calling the corresponding target signal processing program modules according to the execution order information of each signal processing program module to serially process the main channel EEG signal to obtain the final EEG data; the preset parameterized anti-interference algorithm library stores multiple dynamically callable signal processing program modules, each signal processing program module containing dynamically configurable adjustment parameters, and each signal processing program module is used for noise suppression or artifact elimination of a type of interference source. The quality reliability of the final EEG data is quantified, and a ternary mapping relationship between the current period, the quality reliability, and the final EEG data is stored and sent to the downstream application. Quantifying the quality reliability of the final EEG data includes: calculating at least one time-frequency domain feature index of the final EEG data, the time-frequency domain feature index being used to characterize the signal quality of the final EEG data in the time or frequency domain; calculating a basic quality score based on the at least one time-frequency domain feature index; calculating an interference environment correction factor based on the confidence and intensity of the at least one target interference source; and fusing the basic quality score with the interference environment correction factor to obtain a scalar form of quality reliability.

2. The method according to claim 1, characterized in that, The dynamic algorithm combination list generated based on the confidence and intensity of the different interference sources includes: By comparing the confidence and intensity of the different interference sources with the activation threshold corresponding to the different interference sources, at least one target interference source that is greater than the activation threshold is identified. Obtain the module identifier of the signal processing module corresponding to each target interference source from the pre-built mapping relationship between the module identifiers of the interference sources and the processing module; Based on the execution priority and logical dependency relationship between each signal processing program module in the preset parameterized anti-interference algorithm library, the execution order information of each signal processing program module is determined; The module identifier of the signal processing program module corresponding to each target interference source and the execution order information of each signal processing program module are used as a dynamic algorithm combination list.

3. The method according to claim 1, characterized in that, The configuration of dynamically configurable adjustment parameters for each signal processing module yields multiple target signal processing modules, including: From the interference quantization vector, extract the intensity of the target interference source corresponding to each signal processing module; Based on a preset parameter mapping relationship, target adjustment parameters associated with each signal processing program module and the target interference source are determined; wherein, the parameter mapping rule defines the mapping relationship between the interference source type, the signal processing program module and the dynamically configurable adjustment parameters; Convert the intensity corresponding to the target interference source into a specific parameter value of the target adjustment parameter; The specific parameter values ​​are assigned to the target adjustment parameters of the signal processing program module to instantiate and configure each signal processing program module, thereby obtaining multiple target signal processing program modules.

4. The method according to claim 1, characterized in that, The step involves sequentially calling the corresponding target signal processing module based on the execution order information of each signal processing module to serially process the main channel EEG signal and obtain the final EEG data, including: Initialize the sliding window and use the main channel EEG signal as the initial input data stream for the sliding window; Based on the execution order information of each signal processing module, the target signal processing module corresponding to each signal processing module is called; The current input data stream is read from the sliding window; wherein, for the first target signal processing program module to be called, the current input data stream is the main channel EEG signal; for other target signal processing program modules, the current input data stream is the intermediate processing data stream output by the previous target signal processing program module. The current input data stream is input into the target signal processing program module, so that the interference source processing function of the target signal processing program is executed on the current input data stream according to the adjustment parameters configured in the target signal processing program module, to obtain an intermediate processing data stream or a final processing result; The intermediate processing data stream is written to the sliding window as the current input data stream for the next target signal processing module; or, upon completion of each target signal processing module, the final processing result is used as the final EEG data.

5. The method according to any one of claims 1-4, characterized in that, The pre-trained interference identification model includes an input layer, a feature fusion layer, a hidden layer, and an output layer; The step of inputting the multimodal auxiliary signal into a pre-trained interference identification model and outputting a structured interference quantization vector includes: The input layer normalizes and aligns the real-time contact impedance value of the electrode, the instantaneous data of the inertial measurement unit, and the environmental data of the environmental sensor with time, and outputs a first multi-dimensional feature vector. The feature fusion layer concatenates or weights the first multidimensional feature vector to generate a second multidimensional feature vector. The hidden layer performs nonlinear transformation and high-level feature extraction on the second multidimensional feature vector to learn and characterize the coupling relationship and independent patterns between different interference sources, and outputs high-level abstract features; The output layer maps the high-level abstract features to the structured interference quantization vector for output; the node structure of the output layer corresponds to a predefined set of interference source types, and each node outputs a multidimensional tuple, which includes at least the confidence and intensity of the corresponding interference source type. The structured interference quantization vector is composed of the multidimensional tuples corresponding to all interference source types.

6. The method according to any one of claims 1-4, characterized in that, Generate a pre-trained interference identification model according to the following steps: Acquire each multimodal auxiliary signal sample synchronously acquired within the historical sampling period; Determine a structured ground truth label for the interference quantization vector corresponding to each multimodal auxiliary signal sample, wherein the ground truth label for the interference quantization vector includes confidence and intensity for multiple interference sources; Store the binary mapping relationship between each multimodal auxiliary signal sample and the structured interference quantization vector truth label corresponding to each multimodal auxiliary signal sample to obtain each set of historical samples; Create an interference identification model; The multimodal auxiliary signal samples from the target historical samples are input into the interference identification model to obtain the predicted structured interference quantization vector; the target historical samples are each group of historical samples; Calculate the loss value between the predicted structured interference quantization vector and the ground truth label in the target historical sample; when the loss value reaches its minimum, obtain the pre-trained interference identification model.

7. The method according to claim 6, characterized in that, The creation of the interference identification model includes: Using machine learning algorithms, an input layer is constructed for receiving and preprocessing multimodal auxiliary signals, a feature fusion layer for fusing multimodal features, at least one hidden layer for performing nonlinear feature transformation, and an output layer for outputting structured interference quantization vectors. The input layer, the feature fusion layer, the hidden layer, and the output layer are sequentially integrated into a network structure; Define a composite loss function; the composite loss function includes a classification loss component that measures the difference between the confidence of the interference source in the model output and the true label, and a regression loss component that measures the difference between the intensity of the interference source in the model output and the true label. The composite loss function is integrated into the network structure to obtain the interference identification model.

8. A wearable EEG signal processing device based on feature fusion, characterized in that, The device includes: A multimodal signal acquisition module is used to synchronously latch the main channel EEG signal from the bioelectric electrodes and the multimodal auxiliary signals from each auxiliary sensor interface within a preset period. The multimodal auxiliary signals include at least the real-time contact impedance value of the electrodes, the instantaneous data of the inertial measurement unit, and the environmental data of the environmental sensors, including temperature and humidity. The inertial measurement unit includes an accelerometer and a gyroscope. The interference identification model processing module is used to input the multimodal auxiliary signal into a pre-trained interference identification model and output a structured interference quantization vector. The interference quantization vector includes the confidence and intensity of different interference sources. The confidence is used to characterize the model's credibility to the existing interference sources. The intensity is used to characterize the severity or magnitude of the interference's effect on the EEG signal. An EEG signal processing module is used to process the main channel EEG signal according to the confidence and intensity of different interference sources and a preset parameterized anti-interference algorithm library to obtain final EEG data. This includes: generating a dynamic algorithm combination list based on the confidence and intensity of the different interference sources; the dynamic algorithm combination list includes module identifiers of multiple signal processing program modules and execution order information of each signal processing program module; dynamically calling each signal processing program module from the preset parameterized anti-interference algorithm library according to the module identifier; configuring dynamically configurable adjustment parameters of each signal processing program module to obtain multiple target signal processing program modules; and sequentially calling the corresponding target signal processing program modules according to the execution order information of each signal processing program module to serially process the main channel EEG signal to obtain final EEG data. The preset parameterized anti-interference algorithm library stores multiple dynamically callable signal processing program modules, each signal processing program module containing dynamically configurable adjustment parameters, and each signal processing program module is used for noise suppression or artifact elimination of a type of interference source. The data storage and distribution module is used to quantify the quality reliability of the final EEG data, store the ternary mapping relationship between the current period, the quality reliability, and the final EEG data, and send it to the downstream application. Quantifying the quality reliability of the final EEG data includes: calculating at least one time-frequency domain feature index of the final EEG data, the time-frequency domain feature index being used to characterize the signal quality of the final EEG data in the time or frequency domain; calculating a basic quality score based on the at least one time-frequency domain feature index; calculating an interference environment correction factor based on the confidence and intensity of the at least one target interference source; and fusing the basic quality score with the interference environment correction factor to obtain a scalar form of quality reliability.

Images