Dynamic noise reduction method, device, equipment and storage medium for brain-computer AI headset

Abstract

The present application discloses a dynamic noise reduction method, device, equipment and storage medium for brain-computer AI headphones, and the method includes: obtaining signal data collected by the brain-computer AI headphones and the corresponding noise segment and noise structure; obtaining the noise level corresponding to the noise structure, constructing a suppression chain according to the noise level, performing noise reduction decomposition in the suppression chain, determining a processing node according to the processing result corresponding to the noise reduction decomposition, and determining a scheduling plan according to the processing node; performing scene analysis on the scheduling plan, obtaining environmental characteristics, constructing a response chain and corresponding adaptation parameters; performing signal analysis on the feature data according to the adaptation parameters, constructing a protection chain, performing signal reorganization to generate protection data, determining a control position in the protection data, generating a control chain according to the control position, obtaining corresponding control parameters for noise reduction mapping, and obtaining corresponding mapping results; determining an output area according to the mapping results, completing noise reduction of the signal data, and outputting a pure signal.

Description

Technical Field

[0001] The present application relates to the field of brain-computer interface technology, and in particular to a dynamic noise reduction method, device, equipment and storage medium for brain-computer AI headphones. Background Art

[0002] With the application of brain-computer interface technology in daily scenarios, the anti-interference ability of brain-computer AI headsets has received more and more attention. Early noise reduction methods mainly used fixed parameter filtering technology, such as bandpass filtering, notch filtering, etc. Although certain specific noises can be suppressed in a stable environment, it is difficult to adapt to complex and changeable actual usage scenarios. In recent years, although adaptive noise reduction technology has made significant progress in the field of acoustics, there are still many problems when these methods are directly applied to EEG signals. Especially when dealing with layered noise, existing algorithms lack systematic noise feature extraction and classification methods.

[0003] In practical applications, the noise problems faced by brain-computer signals are mainly manifested in three aspects: first, instantaneous noise at the signal acquisition level, such as power frequency interference, equipment electromagnetic radiation, etc., which need to be identified and quickly suppressed in real time; second, short-term noise at the feature extraction level, such as waveform distortion caused by environmental vibration, human activity, etc., which affects the accuracy of feature extraction; finally, long-term noise at the state assessment level, such as baseline drift, temperature effect, etc., which leads to unstable state recognition results. Existing noise reduction algorithms have obvious deficiencies in dealing with these layered noises: lack of a complete noise feature map construction method, making it difficult to accurately locate noise; no systematic noise reduction parameter scheduling mechanism has been established, which cannot adapt to the needs of different scenarios; limited ability to protect signal features and track states, which easily leads to the loss of effective information.

[0004] At the same time, how to ensure the overall performance of the noise reduction system in a complex and changeable usage environment is also a challenge. The existing technology lacks a complete signal protection mechanism, making it difficult to maintain key features during the noise reduction process; the parameter mapping and optimization methods are not systematic enough to achieve environmental adaptation; the data integration and output strategies are not perfect, affecting the quality of the final signal. These problems seriously restrict the application effect of brain-computer headsets in actual scenarios.

[0005] Therefore, a method is urgently needed to solve at least one of the above problems. Summary of the invention

[0006] In a first aspect, an embodiment of the present application provides a dynamic noise reduction method for a brain-computer AI headset, comprising:

[0007] Obtain signal data collected by the brain-computer AI headset and the corresponding noise segment, obtain feature data corresponding to the noise segment, and obtain the noise structure corresponding to the feature data;

[0008] Obtain the noise level corresponding to the noise structure, build a suppression chain according to the noise level, perform noise reduction decomposition in the suppression chain, determine the processing node according to the processing result corresponding to the noise reduction decomposition, generate a process chain according to the processing node, and determine the scheduling plan in the process chain;

[0009] Perform scenario analysis on the scheduling plan, obtain environmental characteristics, build a response chain based on the environmental characteristics, and obtain the adaptation parameters corresponding to the response chain;

[0010] Perform signal analysis on the characteristic data according to the adaptation parameters, obtain the characteristic area, build a protection chain according to the characteristic area, reorganize the signal according to the protection chain to generate protection data, determine the control position in the protection data, generate a control chain according to the control position, and obtain control parameters corresponding to the control chain;

[0011] Perform noise reduction mapping on the control parameters, obtain the noise reduction interval and convert the noise reduction interval into a mapping chain, and obtain the mapping result corresponding to the mapping chain;

[0012] The output area is determined according to the mapping result, and the output chain is constructed according to the output area. The signal data is denoised in the output chain to output a pure signal.

[0013] In a second aspect, the present application further provides a dynamic noise reduction device, comprising:

[0014] A structure acquisition module, used to acquire signal data collected by the brain-computer AI headset and the corresponding noise segment, acquire feature data corresponding to the noise segment, and acquire the noise structure corresponding to the feature data;

[0015] A level acquisition module, used for acquiring a noise level corresponding to the noise structure, constructing a suppression chain according to the noise level, performing noise reduction decomposition in the suppression chain, determining a processing node according to a processing result corresponding to the noise reduction decomposition, generating a process chain according to the processing node, and determining a scheduling scheme in the process chain;

[0016] A scenario analysis module, used to perform scenario analysis on the scheduling scheme, obtain environmental features, build a response chain according to the environmental features, and obtain adaptation parameters corresponding to the response chain;

[0017] a parameter acquisition module, configured to perform signal analysis on the characteristic data according to the adaptation parameter, obtain a characteristic area, construct a protection chain according to the characteristic area, perform signal reorganization according to the protection chain to generate protection data, determine a control position in the protection data, generate a control chain according to the control position, and obtain a control parameter corresponding to the control chain;

[0018] A noise reduction mapping module, used to perform noise reduction mapping on the control parameter, obtain a noise reduction interval and convert the noise reduction interval into a mapping chain, and obtain a mapping result corresponding to the mapping chain;

[0019] A clean output module is used to determine an output area according to the mapping result, build an output chain according to the output area, complete noise reduction on the signal data in the output chain, and output a clean signal.

[0020] In a third aspect, the present application also provides a computer device, comprising a processor and a memory, wherein the memory is used to store a computer program, and when the computer program is executed by the processor, the dynamic noise reduction method of the brain-computer AI headset as described in the first aspect is implemented.

[0021] In a fourth aspect, the present application also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the dynamic noise reduction method for brain-computer AI headphones as described in the first aspect.

[0022] Compared with the prior art, this application has at least the following beneficial effects:

[0023] 1. This invention proposes a hierarchical noise reduction strategy, which realizes accurate identification and effective suppression of noise through the innovative combination of feature extraction, distribution analysis and suppression chain. Compared with the traditional single noise reduction method, this strategy can simultaneously handle three types of noise: instantaneous, short-term and long-term. It not only solves the problem of rapid suppression of sudden noise such as power frequency interference and motion artifacts, but also effectively handles continuous interference such as environmental vibration and equipment radiation, significantly improving the reliability of brain-computer headsets in complex environments.

[0024] 2. The present invention has established a complete feature protection mechanism, which innovatively combines data scheduling, scene analysis and parameter control. This mechanism can not only adaptively match different usage scenarios such as office, study, and commuting, but also protect the characteristics of key EEG bands such as α, β, and θ during the noise reduction process, solving the problem that traditional noise reduction methods easily cause effective signal distortion. Through dynamic parameter adjustment and feature area protection, the method achieves the optimal balance between noise reduction and feature protection, ensuring the accuracy of brain-computer interaction.

[0025] 3. The present invention implements a method-based signal optimization scheme, and establishes a complete signal processing chain through innovative noise reduction mapping and data integration technology. This scheme overcomes the difficulty of parameter optimization of traditional methods, realizes intelligent regulation of the noise reduction process, can automatically optimize processing parameters according to environmental changes, and ensures the continuity and stability of the output signal through multi-level data integration. While maintaining a good noise reduction effect, the method significantly improves the real-time and reliability of signal processing, and improves the practicality of brain-computer headphones.

[0026] It should be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 This is a flow chart of a dynamic noise reduction method for a brain-computer AI headset according to an embodiment of the present application;

[0028] Figure 2 This is a schematic diagram of the structure of a dynamic noise reduction device shown in an embodiment of the present application;

[0029] Figure 3 A schematic diagram of the structure of a computer device shown in an embodiment of the present application. DETAILED DESCRIPTION

[0030] In the following description, specific details such as specific system structures, technologies, etc. are provided for the purpose of illustration rather than limitation, so as to provide a thorough understanding of the embodiments of the present application. However, it should be clear to those skilled in the art that the present application may also be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to prevent unnecessary details from obstructing the description of the present application.

[0031] It should be understood that when used in the present specification and the appended claims, the term "comprising" indicates the presence of described features, wholes, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, wholes, steps, operations, elements, components and / or combinations thereof.

[0032] It should also be understood that the term “and / or” used in the specification and appended claims refers to any and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0033] As used in the specification and appended claims of this application, the term "if" can be interpreted as "when" or "uponce" or "in response to determining" or "in response to detecting", depending on the context. Similarly, the phrase "if it is determined" or "if [described condition or event] is detected" can be interpreted as meaning "uponce it is determined" or "in response to determining" or "uponce [described condition or event] is detected" or "in response to detecting [described condition or event]", depending on the context.

[0034] In addition, in the description of the present application specification and the appended claims, the terms "first", "second", "third", etc. are only used to distinguish the descriptions and cannot be understood as indicating or implying relative importance.

[0035] References to "one embodiment" or "some embodiments" etc. described in the specification of this application mean that one or more embodiments of the present application include specific features, structures or characteristics described in conjunction with the embodiment. Therefore, the statements "in one embodiment", "in some embodiments", "in some other embodiments", "in some other embodiments", etc. that appear in different places in this specification do not necessarily refer to the same embodiment, but mean "one or more but not all embodiments", unless otherwise specifically emphasized in other ways. The terms "including", "comprising", "having" and their variations all mean "including but not limited to", unless otherwise specifically emphasized in other ways.

[0036] The technical solution of the embodiment of the present application is introduced below.

[0037] With the application of brain-computer interface technology in daily scenarios, the anti-interference ability of brain-computer AI headsets has received more and more attention. Early noise reduction methods mainly used fixed parameter filtering technology, such as bandpass filtering, notch filtering, etc. Although certain specific noises can be suppressed in a stable environment, it is difficult to adapt to complex and changeable actual usage scenarios. In recent years, although adaptive noise reduction technology has made significant progress in the field of acoustics, there are still many problems when these methods are directly applied to EEG signals. Especially when dealing with layered noise, existing algorithms lack methodical noise feature extraction and classification methods.

[0038] In practical applications, the noise problems faced by brain-computer signals are mainly manifested in three aspects: first, instantaneous noise at the signal acquisition level, such as power frequency interference, equipment electromagnetic radiation, etc., which need to be identified and quickly suppressed in real time; second, short-term noise at the feature extraction level, such as waveform distortion caused by environmental vibration, human activity, etc., which affects the accuracy of feature extraction; finally, long-term noise at the state assessment level, such as baseline drift, temperature effect, etc., which leads to unstable state recognition results. Existing noise reduction algorithms have obvious deficiencies in dealing with these layered noises: lack of a complete noise feature map construction method, making it difficult to accurately locate noise; no noise reduction parameter scheduling mechanism has been established, and it cannot adapt to the needs of different scenarios; limited ability to protect signal features and track states, which easily leads to the loss of effective information.

[0039] At the same time, how to ensure the overall performance of the noise reduction method in a complex and changeable usage environment is also a challenge. The existing technology lacks a complete signal protection mechanism, making it difficult to maintain key features during the noise reduction process; the parameter mapping and optimization methods are not methodical enough to achieve environmental adaptation; the data integration and output strategies are not perfect, affecting the quality of the final signal. These problems seriously restrict the application effect of brain-computer headsets in actual scenarios.

[0040] To solve the above problems, please refer to Figure 1 , Figure 1 The following is a flowchart of a dynamic noise reduction method for a brain-computer AI headset provided in an embodiment of the present application. The dynamic noise reduction method for a brain-computer AI headset in an embodiment of the present application can be applied to computer devices, including but not limited to smartphones, laptops, tablet computers, desktop computers, physical servers, cloud servers and other devices. Figure 1 As shown, the dynamic noise reduction method of the brain-computer AI headset of this embodiment includes steps S101 to S106, which are described in detail as follows:

[0041] Step S101, obtain the signal data and corresponding noise segments collected by the brain-computer AI headset, obtain the characteristic data corresponding to the noise segment, and obtain the noise structure corresponding to the characteristic data.

[0042] Specifically, the signal data of the brain-computer headset is collected and the noise segment is identified. This process first needs to be carried out in the actual application scenario where the user wears the brain-computer headset. The method adopts a multi-channel signal acquisition scheme, including eight EEG acquisition channels and two reference electrodes located in different areas of the brain, to continuously collect EEG signals. The sampling period of each channel is set to 2 milliseconds, the sampling rate is 500 Hz, and the sampling accuracy is 24 bits. The collected signal is amplified 1000 times by the preamplifier and then digitized by the analog-to-digital converter. The method ensures that the electrode-skin contact impedance is always kept below 5 kilo-ohms through real-time monitoring, thereby ensuring the signal acquisition quality. In the signal preprocessing stage, the method uses a 50 Hz notch filter to preliminarily eliminate power frequency interference, and a 0.1 to 100 Hz bandpass filter is used to preprocess the original signal. In daily application scenarios, when users use brain-computer headsets on buses, vehicle bumps can cause electrode displacement, resulting in obvious motion artifacts, manifested as signal baseline drift and instantaneous spikes; when users work in a cafe, the surrounding customers walking and talking will generate environmental vibrations, causing the electrodes to shake slightly; in an office environment, computer monitors and air-conditioning equipment will generate continuous electromagnetic interference. The method monitors the time-frequency characteristics of the signal. When it is detected that the rate of change of the signal amplitude between adjacent sampling points exceeds the preset threshold, or the high-frequency energy is abnormally concentrated, or the signal variance undergoes a significant mutation within a specific time window, these time periods are marked as specific noise segments.

[0043] In some embodiments, obtaining feature data corresponding to a noise segment includes: extracting features from the noise segment to obtain a feature set corresponding to the noise segment; generating a noise spectrum according to the feature set; and extracting feature data from the noise spectrum.

[0044] The noise segment is feature extracted. The method uses a 300 millisecond analysis window to continuously analyze the marked noise segment. Each analysis window is preprocessed with a Hanning window function, and 50% overlap is maintained between adjacent windows to ensure the continuity of signal analysis. In each window, the method simultaneously calculates features in the time domain and frequency domain. The time domain features include the root mean square value, peak-to-peak value and waveform factor of the signal, which are used to characterize the basic fluctuation characteristics of the signal; the frequency domain features calculate the spectrum energy distribution in the range of 0-100 Hz by fast Fourier transforming the signal. In practical applications, for example, when a user walks in a shopping mall, the method can identify periodic motion artifacts mainly of 1-2 Hz, and superimposed with 50 Hz environmental power frequency interference; when the user is in an elevator, transient impact signals generated by elevator start-up and braking will be detected. For these different types of noise, the method uses a wavelet decomposition method, selects the db4 wavelet basis function for 5-layer decomposition, and obtains the time-frequency features corresponding to each frequency band of δ, θ, α, β and γ respectively. Through this series of feature extraction processes, the method finally obtains a feature set including time domain, frequency domain and time-frequency domain features.

[0045] The method first constructs a multidimensional feature space by generating a noise spectrum and outputting feature data through the feature set. The space represents the temporal characteristics of the noise with the time axis, the frequency axis reflects the spectral distribution of the signal, and the amplitude axis describes the change of signal strength. The method maps the parameters in the feature set to this feature space, and uses the density clustering algorithm to identify the main mode of feature distribution. For different types of noise, the method establishes a special feature template library, including motion artifact template M001, power frequency interference template M002, myoelectric interference template M003, etc. In the template matching process, the method uses a dynamic time warping algorithm to calculate the similarity between the current feature and the template, while considering the change trend of the feature in the time and frequency dimensions. For example, when a user is active in a subway station, the mixed noise detected by the method will contain the features of multiple templates at the same time. At this time, the main noise components are extracted by the principal component analysis method, and the contribution weights of each component are calculated. For each type of noise identified, the method updates its distribution characteristics in the feature space in real time, including the distribution center, discrete degree and change trend. Based on these analysis results, the method generates a real-time updated noise spectrum, which not only contains the spatial distribution characteristics of the noise, but also records the temporal evolution of the noise type. Finally, the method extracts and encodes the key information of the noise spectrum and outputs a structured feature data, including the coding identification of the noise type, the numerical value of the feature parameters of each dimension, the timestamp of occurrence, the confidence of recognition, and the description of the spatial distribution.

[0046] In some embodiments, obtaining the noise structure corresponding to the feature data includes: performing distribution analysis on the feature data to obtain the noise distribution corresponding to the feature data; mapping the noise distribution into a statistical domain to obtain a feature expression; and generating the noise structure according to the feature expression.

[0047] The output feature data is analyzed for distribution and the noise distribution is located. The method first processes the received feature data packet. The noise type code contained in the data packet is used to identify the noise type, the feature parameter value reflects the noise characteristics, the timestamp records the noise occurrence sequence, and the confidence level indicates the recognition reliability. For the feature data of each noise type, the method constructs a probability density function and analyzes its distribution law in the feature space. In actual application scenarios, for example, when users use brain-computer headphones in subway cars, the feature data will show typical mixed distribution characteristics: the low-frequency vibration noise generated by the subway operation forms a stable Gaussian distribution in the range of 0.5-3 Hz, and the confidence level is usually higher than 90%; while the environmental noise caused by passengers walking and talking is Poisson distributed in a wider frequency band, and the confidence level fluctuates between 75% and 85%. When the user is in the office, the power frequency interference from the computer monitor forms a discrete peak distribution near 50 Hz, and its feature parameter value shows a strong periodicity in the time series. The method adopts the kernel density estimation method and uses adaptive bandwidth parameters to model the distribution of noise features in different dimensions. By analyzing the cluster center, discrete degree and time correlation of the characteristic data, the method finally obtains a distribution map, which records the spatial distribution position, probability density value and time evolution characteristics of each type of noise.

[0048] The noise distribution recorded in the above distribution map is mapped into the statistical domain and a feature expression is formed. The method first establishes a statistical feature space based on the spatial distribution position in the distribution map. The area with a probability density value greater than 0.1 is selected as the focus of attention, and a multidimensional feature space including statistics such as mean, variance, skewness, and kurtosis is constructed. In actual scenarios, for example, when users use equipment in an office environment, the power frequency interference in the distribution map shows significant periodicity, and its statistical characteristics are stable distribution with a variance less than 0.05; while the impact noise generated by keyboard tapping is shown as discrete peak points in the distribution map, and its statistical characteristics have high peaks (greater than 5) and irregular time intervals. When the user is in a shopping mall environment, the distribution map shows that the background music forms a stable power spectrum distribution, and its statistical characteristics show a normal distribution with a variance less than 0.1. The method calculates the projection of various types of noise in the distribution map on the statistical dimension to form a feature vector, and uses the principal component analysis method to reduce the dimension, retaining the main components with an explained variance rate of more than 95%, and finally forming a feature expression containing noise types and main statistical characteristics.

[0049] Based on this feature expression to generate noise structure, the method first establishes a hierarchical framework according to the time scale differences of the features in each dimension in the matrix. The feature expression is layered according to the time scale, and fast-changing features (variance greater than 1) are classified into the instantaneous layer (milliseconds), medium-changing features (variance between 0.1 and 1) are classified into the short-term layer (seconds), and slowly changing features (variance less than 0.1) are classified into the long-term layer (minutes). For example, when the user is in a home office environment, the air conditioner operation noise and computer fan noise with a variance less than 0.1 in the feature expression matrix form the bottom layer structure; the printer working noise and microwave heating noise with a variance between 0.3-0.8 form the middle layer structure; the doorbell and talking sound with a variance greater than 1 form the top layer structure. In the library environment, the feature expression matrix shows that the variance of the constant air conditioning ventilation sound is less than 0.05, forming the bottom layer; the environmental noise of page turning and whispering conversation has a variance between 0.2-0.4, forming the middle layer; the variance of sudden noise such as chair movement is greater than 1.2, forming the top layer. The method analyzes the correlation coefficients between different layers of features in the feature expression matrix and establishes a complete noise structure including the time series model, state transition matrix and inter-layer correlation weights. Finally, the method outputs a complete noise structure, which includes the feature distribution, transition probability and time series law of each layer of noise.

[0050] Step S102, obtain the noise level corresponding to the noise structure, build a suppression chain according to the noise level, perform noise reduction decomposition in the suppression chain, determine the processing node according to the processing result corresponding to the noise reduction decomposition, generate a process chain according to the processing node, and determine the scheduling plan in the process chain.

[0051] Specifically, data scheduling is performed on the noise reduction processing results output by the step to determine the processing nodes. Since the step has completed the noise suppression of the instantaneous layer (L1), the short-term layer (L2) and the long-term layer (L3), the method needs to configure corresponding processing nodes according to the signal characteristics of different levels.

[0052] In some embodiments, the noise level includes at least an instantaneous layer, a short-time layer and a long-time layer; a suppression chain is constructed according to the noise level, including: configuring a high-speed sampling node in the instantaneous layer; configuring a feature extraction node in the short-time layer; configuring a pattern recognition node in the long-time layer; and constructing a suppression chain according to the high-speed sampling node, the feature extraction node and the pattern recognition node.

[0053] In practical applications, for example, when users use devices in public transportation scenarios, the method configures high-speed sampling node N1 (sampling rate 500Hz) according to the fast fluctuation characteristics of the L1 layer; sets feature extraction node N2 (processing window 100ms) based on the intermediate frequency characteristics of the L2 layer; and arranges pattern recognition node N3 (refresh cycle 200ms) for the stable characteristics of the L3 layer. When users use devices on the subway, the N1 node processes the suppressed instantaneous interference signal in real time; on the bus, the N2 node processes the intermediate frequency characteristics after noise reduction; in the elevator, the N3 node analyzes the long-term trend after processing. In the hospital scenario, the method adjusts the node configuration according to the characteristics of different areas: in the noisy waiting hall, the real-time processing capability of the N1 node is strengthened; in the relatively quiet consulting room, the feature extraction function of the N2 node is highlighted; in the long-term care ward, the state tracking capability of the N3 node is strengthened. In the airport environment, the method configures processing nodes according to the characteristics of different areas: the security inspection area focuses on configuring the N1 node to process rapidly changing signals; the waiting area strengthens the N2 node to extract stable features; the boarding channel area enhances the state recognition of the N3 node. Through this node planning corresponding to the denoising level, the method forms a three-level processing network structure.

[0054] In view of the processing characteristics of each node in this three-level processing network, the method designs a process chain around the processing node. When applied in an office environment, the method configures a direct memory access channel for the N1 node based on the high-speed sampling characteristics of the N1 node; sets a data cache mechanism based on the feature extraction requirements of the N2 node; and establishes a parallel processing framework to adapt to the pattern recognition characteristics of the N3 node. For example, when it is detected that the user is doing focused work, such as programming development, the method strengthens the feature extraction priority of the N2 node based on the stable signal flow collected by the N1 node, while maintaining the regular recognition rhythm of the N3 node. In the educational scenario, the method adjusts the task allocation according to the processing capabilities of different nodes: arranges the data acquisition timing based on the sampling characteristics of the N1 node, allocates processing tasks using the feature extraction capabilities of the N2 node, and tracks the learning status through the recognition mechanism of the N3 node. This process design based on node characteristics forms a dynamically adaptive processing chain structure.

[0055] Based on the task allocation characteristics of this processing chain structure, the method uses the process chain to arrange data allocation. In terms of data flow control, the method sets the basic data block size according to the sampling characteristics of the N1 node in the processing chain, determines the feature extraction batch according to the processing capacity of the N2 node, and arranges data flow according to the recognition cycle of the N3 node. In the home use scenario, the method flexibly adjusts in different time periods according to the characteristics of the processing chain: the feature extraction capability of the N2 node is mainly used in the early morning period, the processing load of the three nodes is balanced in the noon period, and the overall processing intensity is reduced in the night period. In the gym scenario, the method adjusts the strategy according to the real-time status of the processing chain: when the exercise intensity changes, the processing priority of each node is adjusted accordingly. This data allocation mechanism based on the characteristics of the processing chain generates a complete task scheduling record.

[0056] According to the execution of this task scheduling record, the method forms the final scheduling plan. The scheduling plan establishes a hierarchical scheduling structure by analyzing the processing efficiency and resource utilization of each node in the record: the top layer is responsible for global resource allocation, the middle layer performs task scheduling, and the bottom layer processes data cache. In the subway commuting scenario, the method dynamically adjusts the processing strategy according to the signal quality change trend reflected in the scheduling record: when the subway starts or brakes, the front-end processing priority is increased; when it runs smoothly, the standard configuration is adopted. In the shopping mall environment, the method optimizes the processing flow for different time periods based on the load analysis of the scheduling record: strengthen concurrent processing during crowded periods and reduce processing intensity during sparse periods. In the conference room scenario, the method adjusts the resource allocation strategy according to the pattern characteristics of the scheduling record: strengthen feature extraction in key scenarios and maintain balanced processing in conventional scenarios. In large sports stadiums, the method adjusts the processing strategy according to the characteristics of different areas: in the crowded entrance channel, give priority to ensuring the stability of signal acquisition; in the audience area, focus on extracting the user's attention characteristics; in the rest area, balance the resource allocation of various processing tasks. In the museum scenario, the method optimizes the processing flow according to the characteristics of different exhibition areas: in the interactive experience area, the real-time response capability is enhanced; in the static display area, the accuracy of feature extraction is highlighted; in the guided tour area, the continuity of state recognition is optimized. In the library environment, the method adjusts the processing strategy according to different functional areas: in the reading area, the stability of the processing task is maintained, in the discussion area, the real-time performance of signal processing is enhanced, and in the learning area, the accuracy of feature extraction is optimized. Through this data-driven scheduling scheme, the method achieves the real-time requirements of brain-computer signal processing.

[0057] Data scheduling is implemented for the noise reduction results output by the step to determine the processing nodes. The step uses a suppression chain to hierarchically process the noise in the signal: the instantaneous layer suppresses power frequency interference and motion artifacts, and outputs a steady-state fast signal with a 30% improvement in waveform smoothness; the short-time layer filters environmental noise and equipment interference, and obtains a clear feature signal with a signal-to-noise ratio of 15dB; the long-time layer eliminates temperature drift and baseline drift, and generates a reference signal with a stability of 95%. Methods Processing schemes are configured for the three types of signal characteristics after noise reduction: a high-speed sampling node N1 is set for the smooth signal of the L1 layer, and a double buffer mechanism is used to achieve stable acquisition to ensure signal integrity; a feature extraction node N2 is configured for the high signal-to-noise ratio signal of the L2 layer, and a sliding window method is used to extract features to ensure feature significance; a pattern recognition node N3 is arranged for the stable signal of the L3 layer, and classification is performed through a state machine to achieve reliable recognition. In the office scenario, this node configuration fully utilizes the characteristics of the noise-reduced signal: when working independently, the N1 node can accurately collect the user's attention fluctuations, when discussing in a meeting, the N2 node can reliably extract the participation characteristics, and when working in a team, the N3 node can continuously evaluate the cognitive state. Based on this hierarchical processing of the noise-reduced signal, the method constructs a multi-level processing network.

[0058] Around this processing network derived from the characteristics of noise reduction signals, the method designs a process chain according to the characteristics of each node. The N1 node is oriented to high-smoothness signals, and a 1024-point circular buffer and DMA transmission channel are designed to achieve a data throughput of 1MB / s to ensure the continuous acquisition of smooth signals; the N2 node processes high signal-to-noise ratio features, builds an 8-level pipeline structure, configures a 100ms sliding window, and improves feature extraction efficiency through parallel computing; the N3 node analyzes stable reference signals, establishes a state evaluation framework, and sets a 200ms refresh cycle to achieve reliable pattern recognition. An asynchronous communication mechanism is used between nodes to ensure the real-time nature of signal processing. In the learning scenario, this process design shows excellent performance: during classroom teaching, the buffer mechanism of the N1 node captures the instantaneous changes in attention, the pipeline of the N2 node analyzes the changing trend of the degree of understanding, and the evaluation framework of the N3 node tracks the overall learning status. During self-study, the method can accurately capture the fluctuations in concentration, identify the early characteristics of fatigue, and warn of the downward trend in learning efficiency. This hierarchical process design forms a stable data processing architecture, which includes three levels of data flow: buffering, pipelining, and evaluation, and millisecond-level processing timing.

[0059] According to the processing requirements of this three-level data flow, the method formulates a resource allocation strategy. The 1024-point buffer and 1MB / s throughput of the N1 node require higher processing resources, so 30% of the basic resources are allocated; the 8-level pipeline and 100ms window calculation of the N2 node require more computing power support, and 45% of dynamic resources are configured; the 200ms refresh cycle of the N3 node has less processing pressure, and 25% of the resources are reserved to meet the needs. In terms of priority setting, the real-time buffer of the N1 node requires the highest response speed and is configured with the highest priority; the pipeline processing of the N2 node allows short delays and is set to a medium priority; the state evaluation of the N3 node can be slightly delayed and uses the basic priority. In the commuting scenario, this allocation mechanism based on processing requirements shows good results: when traveling on the subway, the high priority ensures the stable acquisition of EEG signals in a mobile environment, sufficient computing resources support the real-time analysis of the attention level, and reasonable evaluation resources realize the dynamic tracking of cognitive states. Through this precise resource configuration, the method forms an execution framework that includes priority management and resource allocation.

[0060] Based on the running status of this execution framework, the method establishes an adaptive scheduling scheme. The monitoring module sets thresholds in a targeted manner: considering that 30% of the resource quota of the N1 node is mainly used for buffer management, the buffer occupancy exceeding 80% is set as a warning point; 45% of the resources of the N2 node mainly support pipeline operations, and the processing queue exceeding 64 frames is set as a switching point; 25% of the resources of the N3 node must ensure the refresh requirement of 200ms, and the response delay exceeding 100ms is set as an optimization point. In leisure scenarios, this scheduling scheme can adapt to different usage states: when the user works in a cafe, the method automatically adjusts the pipeline priority of the N2 node to ensure the accuracy of concentration calculation in a noisy environment; when reading quietly in the library, the refresh cycle of the N3 node is optimized to achieve continuous evaluation of the reading comprehension level; in the home scenario, the method can flexibly adjust the processing resources according to the user's activity status to balance the monitoring needs during work and rest. This adaptive scheduling based on resource allocation achieves the optimal utilization of the method's processing capacity.

[0061] Step S103, performing scenario analysis on the scheduling scheme, obtaining environmental features, constructing a response chain according to the environmental features, and obtaining adaptation parameters corresponding to the response chain.

[0062] Specifically, the scheduling scheme output by the step is analyzed in the scenario and the environmental features are marked. The key thresholds set by the monitoring module in the step (buffer occupancy 80%, processing queue 64 frames and response delay 100ms) provide an environmental assessment benchmark for the method. When the buffer occupancy is close to the threshold, it indicates that there is persistent interference; the backlog of the processing queue indicates that complex interference occurs; and the increase in response delay indicates that the interference intensity is large. The method extracts and classifies features for different scenarios based on these indicators: low-frequency electromagnetic interference in the office environment is marked as E1-type features, marking its stability and periodicity; crowd activity noise in public places is marked as E2-type features, recording its randomness and fluctuation range; vibration interference from transportation tools is marked as E3-type features, and its frequency and amplitude characteristics are extracted. In the office scenario, the method identifies the power frequency interference of the display as the E1 dominant feature, the impact interference generated by keyboard tapping as the E2 secondary feature, and the vibration of the air conditioning equipment as the E3 background feature. In the commuting scenario, the vibration of the subway operation becomes the E3 dominant feature, the activity noise of the surrounding crowd is the E2 main feature, and the power frequency interference of the car lighting is the E1 background feature. In the library environment, the power frequency interference of fluorescent lamps is the E1 feature, the sound of turning pages constitutes the E2 feature, and the vibration of air conditioners forms the E3 feature. These environmental features are organized into a multidimensional feature matrix.

[0063] In some embodiments, environmental features include dominant features, secondary features and background features; a response chain is constructed based on the environmental features, including: constructing a notch filter group based on the dominant features, and the center frequency of the notch filter group is automatically adjusted following the interference source; constructing an adaptive threshold detector based on the secondary features, and the sensitivity of the adaptive threshold detector changes dynamically with the environmental noise level; constructing a frequency tracking compensator based on the background features, and the compensation parameters of the frequency tracking compensator are updated in real time with the vibration intensity; and constructing a response chain based on the notch filter group, the adaptive threshold detector and the frequency tracking compensator.

[0064] Based on this multidimensional feature matrix, the method establishes a response chain. For the E1 type power frequency feature, a notch filter group is designed, and the center frequency is automatically adjusted following the interference source; for the E2 type random feature, an adaptive threshold detector is constructed, and the sensitivity changes dynamically with the environmental noise level; corresponding to the E3 type vibration feature, a frequency tracking compensator is implemented, and the compensation parameters are updated in real time with the vibration intensity. In the coffee shop environment, the response chain shows good adaptability: it automatically filters out the power frequency noise of the coffee machine, dynamically handles the random interference of customers walking, and stably compensates for the low-frequency vibration of the seat shaking. In the online learning scenario, the method can accurately respond to the interference characteristics of electronic devices, effectively filter the random noise of the environment, and continuously monitor the user's state changes. In the business center scenario, the response chain can handle the comprehensive interference of various office equipment: the mechanical noise of the printer is classified as the E2 feature and filtered in time, the low-frequency vibration of the central air conditioner is identified as the E3 feature and dynamically compensated, and the power frequency interference of the power equipment is marked as the E1 feature and accurately filtered. Through this feature-driven response mechanism, the method generates a response data set containing filtering parameters, detection thresholds, and compensation coefficients.

[0065] Based on the processing effect of this response data set, the method implements parameter conversion. The filter parameters of E1 features are mapped to the normalized space of 0-1 to achieve accurate positioning of power frequency interference; the detection threshold of E2 features is converted to the probability distribution domain to optimize the recognition accuracy of random noise; the compensation coefficient of E3 features is transformed to the frequency response range to improve the accuracy of vibration suppression. In the home scene, this parameter conversion mechanism can adapt to environmental changes in different areas: accurately convert the power frequency parameters of home appliances in the living room, accurately map the random characteristics of human activities in the bedroom, and effectively transform the vibration coefficient of the air conditioning method in the study. In the shared office space, the method realizes intelligent parameter conversion: the human voice interference in the open area is mapped to a dynamic probability threshold, the interference of electronic equipment is converted to accurate filtering parameters, and the building vibration is converted to a real-time compensation coefficient. In the remote conference scene, the method continuously adjusts the parameter mapping: the vibration characteristics of the computer fan are converted to adaptive compensation parameters, the keyboard sound is mapped to a flexible detection threshold, and the display interference is converted to optimized filtering parameters. Through this multi-dimensional parameter mapping, the method establishes a standardized feature parameter space.

[0066] According to this standardized feature parameter space, the method outputs adaptive parameters. For the power frequency feature parameters, adaptive filter group parameters are generated, including center frequency, bandwidth and attenuation depth; for the random feature parameters, dynamic detector parameters are output, including decision threshold, time window and sensitivity coefficient; for the vibration feature parameters, real-time compensator parameters are generated, including frequency range, amplitude coefficient and phase compensation. In an open office environment, this adaptive parameter can effectively support the noise reduction processing of brain-computer headsets: automatic adjustment to maintain the clarity of speech features during conference discussions, optimization of parameters to improve the stability of attention signals during independent work, and balance of parameters to ensure comfort during rest time. In the collaborative learning space, the method adjusts parameters according to different learning modes: improve the tolerance of random noise during group discussions, enhance the protection of attention signals during autonomous learning, and optimize the interference suppression of audio equipment during online courses. In the silent office area, the method achieves the best working experience through fine parameter adjustment: accurately filter out the interference of electronic devices, effectively suppress low-frequency vibrations in the environment, and maintain the high fidelity of brain-computer signals. This scenario-based parameter adaptation mechanism realizes the intelligent matching of the method to different environments.

[0067] Step S104, perform signal analysis on the characteristic data according to the adaptation parameters, obtain the characteristic area, build a protection chain according to the characteristic area, reorganize the signal according to the protection chain to generate protection data, determine the control position in the protection data, generate a control chain according to the control position, and obtain control parameters corresponding to the control chain.

[0068] Specifically, the output adaptation parameters are used for signal analysis to extract feature regions. The adaptive filter group parameters (center frequency, bandwidth, attenuation depth) generated in the steps are used to define the frequency band range of the signal, the detector parameters (decision threshold, time window, sensitivity coefficient) guide the time domain division of the features, and the compensator parameters (frequency range, amplitude coefficient, phase compensation) determine the compensation interval of the signal. The wavelet analysis method is adopted, and the db4 wavelet basis is selected for 5-layer decomposition to achieve multi-resolution analysis of the signal. On this basis, a three-level feature extraction framework is constructed: the alpha band related to concentration is extracted in the 8-13Hz range using a bandpass filter group, and its instantaneous amplitude and phase are calculated by Hilbert transform, which is marked as the F1 feature region; the beta band related to cognitive load is identified in the 14-30Hz range using an adaptive threshold, and its time-frequency characteristics are obtained by combining short-time Fourier transform, which is divided into the F2 feature region; the theta band related to relaxation state is defined in the 4-7Hz range using morphological operators, and its energy distribution is extracted based on wavelet packet analysis, which is determined as the F3 feature region. In the online meeting scenario, this parameter-guided partitioning method accurately captures various features: the F1 region records the transient changes of attention fluctuations, the F2 region preserves the dynamic characteristics of cognitive load, and the F3 region reflects the gradual trend of relaxation. The method integrates these feature region data and their extraction parameters into a feature map.

[0069] In some embodiments, the characteristic area includes a first area, a second area and a third area, the first area records the transient changes of attention fluctuations, the second area preserves the dynamic characteristics of cognitive load, and the third area reflects the gradual trend of the relaxation state; constructing a protection chain according to the characteristic area, including: obtaining an energy tracking protector based on Kalman filtering corresponding to the first area; obtaining a multi-scale morphological protector corresponding to the second area; obtaining a spectrum protector corresponding to the third area; constructing a protection chain according to the energy tracking protector, the multi-scale morphological protector and the spectrum protector.

[0070] Based on the regional division of this feature mapping table, the method constructs a protection chain structure. First, for the α band in the F1 area of ​​the mapping table, an energy tracking protector based on Kalman filtering is designed to achieve ±0.5dB energy stability control. The state equation of the filter contains two components: energy level and change rate. The observation equation uses the root mean square value of the sliding window, and the adaptive adjustment of the process noise covariance achieves a fast response to energy mutations. Secondly, corresponding to the β band in the F2 area, a multi-scale morphological protector is established to ensure that the waveform integrity reaches 95%. The burst interference is removed by cascading opening and closing operations, and the details of the characteristic waveform are maintained by using adaptive structural elements. Finally, for the θ band in the F3 area, a spectrum protector based on the Welch method is implemented to maintain the band characteristic deviation less than 2%. The improved periodogram method is used to estimate the power spectral density, and the spectrum leakage is reduced by combining adaptive segmented averaging. In the focused office scenario, this mapping-based protection mechanism shows excellent performance: the stable tracking of the α wave energy achieves an accuracy of 0.2dB, the maintenance of the β wave morphology ensures 98% integrity, and the maintenance of the θ wave spectrum reaches 1.5% accuracy. Through this targeted protection strategy, the method generates a protection feature library containing waveform characteristics, energy distribution, and spectral structure.

[0071] Using the data indicators of this protected feature library, the method implements signal reconstruction. The method first establishes a three-dimensional reconstruction space, including the time domain dimension, frequency domain dimension and feature dimension. In the time domain dimension, the alpha wave energy index of the F1 area is least squares aligned with the original signal, and the phase deviation is compensated by the interpolation algorithm to achieve accurate reconstruction of attention features, and the reconstruction error is controlled within 5%; in the frequency domain dimension, the beta wave morphology parameters of the F2 area are wavelet fused with the baseline signal, and the soft threshold method is used to remove the fusion artifacts to complete the accurate recovery of cognitive load, and the feature fidelity reaches 92%; in the feature dimension, the theta wave spectrum data of the F3 area is adaptively combined with the reference signal in the spectral domain, and the spectrum estimation is optimized using the improved MUSIC algorithm to achieve reliable restoration of the relaxation state, and the spectrum correlation exceeds 0.95. In a team collaboration environment, this feature-driven reconstruction method achieves high-precision signal recovery: the change of concentration during meeting discussion is reproduced with a time resolution of 0.1s, the thinking activity during program design is reconstructed with a feature retention rate of 95%, and the degree of relaxation during rest communication is restored with a frequency accuracy of 0.2Hz. Through this multi-level signal reconstruction process, the method forms a reconstructed data scheme including time domain reconstruction results, feature recovery parameters and state evaluation indicators.

[0072] According to the processing results of this reorganized data scheme, the method generates protection data. The method constructs a hierarchical data protection architecture: the first layer is the real-time protection layer. For the reconstructed α wave signal in the scheme, a concentration evaluation mechanism based on a recurrent neural network is established. The attention change trend is tracked through long short-term memory units to achieve an evaluation delay of 200ms and a detection accuracy of 95%; the second layer is the feature protection layer. Corresponding to the reconstructed β wave signal, a cognitive load monitoring system based on a support vector machine is constructed. The radial basis kernel function is used for nonlinear mapping to generate a trend map with 85% confidence; the third layer is the state protection layer. For the restored θ wave signal, a relaxation tracking strategy based on fuzzy logic is designed. The state change is evaluated using adaptive inference rules to form a progress indicator with a reliability of 90%. In the online office environment, this multi-dimensional protection data provides a stable performance guarantee for the brain-computer headset: the attention index during the meeting report achieves a fluctuation control of ±2%, the cognitive load evaluation during document processing maintains a feature correlation of 0.8, and the relaxation monitoring during rest adjustment achieves a resolution accuracy of 0.1. This protection mechanism based on feature reconstruction ensures the reliability and stability of brain-computer signals in complex environments.

[0073] In some embodiments, the control position includes a first position, a second position, and a third position, the first position is used to adjust the acquisition gain in real time to adapt to the dynamic changes of the home environment, the second position is used to optimize the feature window to capture the attention fluctuations in the online meeting, and the third position is used to continuously evaluate the working status to ensure the effect of remote collaboration; a control chain is generated according to the control position, including: constructing an adaptive gain controller according to the real-time control parameters corresponding to the first position; the adaptive gain controller includes a feedforward compensation unit and a feedback correction unit; constructing a dynamic window controller according to the feature control parameters corresponding to the second position; constructing a fuzzy state controller according to the state control parameters corresponding to the third position, and the fuzzy state controller adopts a two-layer reasoning mechanism; constructing a control chain according to the gain controller, the dynamic window controller and the fuzzy state controller.

[0074] Parameter allocation is performed on the protection data output by the step to establish the control position. The three-layer protection data formed by the step provides a control benchmark: the 95% detection accuracy of the real-time layer shows the dynamic characteristics of the α-wave attention signal, the 85% confidence of the feature layer reflects the stability requirements of the β-wave cognitive load, and the 90% reliability of the state layer indicates the evaluation criteria of the θ-wave relaxation. Methods A hierarchical control architecture is established based on these characteristic indicators: in view of the real-time characteristics of α-wave concentration, the P1 control position is set at the signal acquisition end, and a proportional-integral controller is used. The integral time constant is set to 50ms and the feedforward gain is 0.8 to achieve fast response; in view of the stability requirements of the β-wave cognitive load, the P2 control position is configured at the feature extraction end, and a model predictive controller is used. The prediction step is 100ms and the control cycle is 60ms to ensure feature stability; corresponding to the evaluation requirements of the θ-wave relaxation, the P3 control position is arranged at the state evaluation end to realize the adaptive state controller, with an adjustment cycle of 200ms and a learning rate of 0.05 to ensure a stable state. In the remote office scenario, this control layout embodies good hierarchical characteristics: the P1 position adjusts the acquisition gain in real time to adapt to the dynamic changes of the home environment, the P2 position optimizes the feature window to capture the attention fluctuations in the online meeting, and the P3 position continuously evaluates the working status to ensure the effect of remote collaboration. These control positions and their parameters, control algorithms, and performance indicators are integrated into a position control matrix.

[0075] Based on this position control matrix containing control algorithms and performance indicators, the method organizes a multi-level control chain. First, according to the real-time control parameters of the P1 position in the matrix, an adaptive gain controller is constructed, and a feedforward compensation unit (gain range 0-40dB) and a feedback correction unit (correction step size 0.5dB) are designed. The signal change is predicted in real time through the improved Kalman estimator (state dimension 4, observation dimension 2), and the response time is controlled within 10ms; secondly, according to the characteristic control parameters of the P2 position, a dynamic window controller is designed, combining the two strategies of sliding average (window base length 50ms) and exponential smoothing (smoothing coefficient 0.3), the window length is adaptively adjusted in the range of 50-200ms, and the feature extraction delay is kept within 30ms; finally, the state control parameters of the P3 position are used to realize the fuzzy state controller, using a two-level reasoning mechanism (5 levels of input fuzzification and 3 levels of output defuzzification), the input variables are state deviation and change rate, and the output variables are control gain and adjustment step size. This multi-level control mechanism ultimately generates a control rule library containing control strategies, adjustment parameters and optimization indicators.

[0076] Relying on the adjustment mechanism of this control rule base, the method performs threshold adjustment. According to the gain control strategy in the rule base, a three-level threshold adjustment scheme is designed: the signal amplitude threshold is adjusted piecewise linearly within the range of ±2dB, and the segmentation points are set at -1dB and +1dB; the sampling rate threshold is adaptively switched in the range of 500-1000Hz, and the switching points are set at 700Hz and 900Hz; the quantization accuracy threshold is dynamically configured in the range of 16-24bit, and the configuration points are 18bit and 22bit. Corresponding to the feature control strategy, adaptive threshold optimization is achieved: the energy threshold is automatically adjusted according to the signal-to-noise ratio, and the adjustment step is 0.5dB; the waveform threshold changes dynamically with the feature significance, and the change rate is limited to 5% / s; the spectrum threshold is updated in real time according to the bandwidth utilization, and the update cycle is 100ms. For the state control strategy, a partition threshold management is established: the attention threshold interval is divided into three levels: low, medium and high, and the width of each level is 20%; the cognitive load threshold is set to change gradually, and the change step is 10%; the relaxation threshold is defined by a fuzzy interval, and the overlap is set to 15%. In the library scenario, this threshold adjustment mechanism is particularly effective: in a quiet reading environment, the method can capture weaker EEG changes and achieve an accurate assessment of reading concentration. Through this multi-dimensional threshold adjustment, the method establishes a complete parameter optimization solution.

[0077] According to the indicator requirements of this parameter optimization scheme, the method outputs control parameters. In response to the signal control requirements in the scheme, the acquisition control parameter set is generated: including variable gain coefficient (range 0-40dB, step 0.5dB, response time 5ms), dynamic sampling rate (basic 500Hz, maximum 1kHz, switching delay 2ms), adaptive quantization bit (minimum 16bit, maximum 24bit, switching time 10ms). Corresponding to the feature control requirements, the feature control parameter group is output: covering energy smoothing coefficient (time constant 10-50ms, adjustment step 5ms), morphological operator parameters (structural elements 3-7 points, update cycle 20ms), spectrum analysis parameters (resolution 0.1Hz, update rate 10Hz). Facing the state control goal, the state control parameter set is generated: setting the attention scoring standard (5 levels, 20 points per level, update cycle 100ms), cognitive load level (3 levels of classification, each level with a protection interval, response time 50ms), relaxation index (percentage system, set buffer zone, smoothing time 200ms). In an open office environment, these control parameters show excellent adaptability: they can accurately capture changes in participation during meeting discussions, reliably assess cognitive engagement during document processing, and accurately track recovery status during rest adjustments. This parameter adjustment based on the optimization scheme enables the method to operate reliably in complex environments.

[0078] Step S105, performing noise reduction mapping on the control parameter, obtaining the noise reduction interval and converting the noise reduction interval into a mapping chain, and obtaining a mapping result corresponding to the mapping chain.

[0079] Specifically, the control parameters output by the step are mapped for noise reduction, and the noise reduction interval is located. The three types of control parameters generated by the step provide the basis for noise reduction mapping: the acquisition control parameter set (gain range, sampling rate, quantization bit number) guides the division of the signal layer noise reduction interval, the feature control parameter group (time constant, structural element, update cycle) determines the feature layer noise reduction range, and the state control parameter set (scoring standard, classification level) defines the state layer noise reduction boundary. Methods Firstly, the least squares spectrum estimation method is used to analyze the signal characteristics, the window length is set to 512 points, the overlap rate is 50%, and the noise distribution characteristics are obtained. Then, based on these characteristics, a three-level noise reduction mapping structure is constructed: the Z1 noise reduction interval is set at the signal layer, the improved Wiener filtering algorithm is used, the adaptive noise power estimation is introduced, the estimation window is 100ms, and the tracking step is 10ms; the Z2 noise reduction interval is divided at the feature layer, the wavelet threshold is used for noise reduction, the db4 wavelet basis function is selected, the 5-layer decomposition is performed, and the adaptive threshold is designed; the Z3 noise reduction interval is defined at the state layer, and a multi-channel adaptive notch filter is realized, with a center frequency range of 45-55Hz and adaptive bandwidth. In the remote office scenario, this partition noise reduction strategy effectively copes with different types of interference: the Z1 interval suppresses the power frequency interference of household appliances, the Z2 interval removes the random noise of human activities, and the Z3 interval eliminates the background noise of the environment. The algorithm configuration and performance indicators of these noise reduction intervals are integrated into interval mapping data.

[0080] Based on the configuration information of the interval mapping data, the method constructs a denoising mapping chain. First, for the signal layer denoising of the Z1 interval, an adaptive filter group is designed: the power frequency notch unit adopts a second-order IIR structure, the gain suppression is -40dB, and the phase compensation adopts an all-pass network; the high-pass filter unit uses an elliptic filter, the cutoff frequency is 0.1Hz, and the passband ripple is 0.1dB; the low-pass filter unit implements a Chebyshev structure, the cutoff frequency is 100Hz, and the stopband attenuation is 50dB. Secondly, for the feature layer denoising of the Z2 interval, a wavelet soft threshold processor is implemented: the threshold of each decomposition layer adopts the SURE criterion, combined with the minimum maximum method and unbiased risk estimation, and the optimal threshold is adaptively calculated; the signal reconstruction adopts the improved Mallat algorithm to optimize the boundary effect in the reconstruction process. Finally, for the state layer denoising of the Z3 interval, an autoregressive predictor is constructed: the model order is 4, the recursive least squares algorithm is used to update the parameters, and the forgetting factor is set to 0.98. Through this multi-level denoising structure, the method generates a mapping chain parameter library containing filter parameters, threshold coefficients and prediction models.

[0081] Through the configuration of this mapping chain parameter library, the method carries out noise reduction effect evaluation. For the processing results of the signal layer Z1 interval, a three-dimensional evaluation index is designed: the signal-to-noise ratio improvement adopts a segmented calculation method, and the expected improvement is 15dB; the phase distortion uses cross-correlation analysis, and the allowable deviation is ±5°; the group delay evaluation is based on energy-weighted average, which is controlled within 2ms. For the noise reduction output of the feature layer Z2 interval, a feature preservation evaluation system is constructed: the feature retention rate is based on wavelet energy distribution, which is required to be greater than 90%; the waveform similarity adopts correlation coefficient analysis, which is required to be greater than 0.85; the energy loss rate is estimated by power spectral density and is limited to less than 10%. For the processing effect of the state layer Z3 interval, a triple evaluation mechanism is implemented: the state recognition accuracy is based on confusion matrix analysis, with a target of 95%; the real-time performance is evaluated by delay characteristics, which is required to be less than 100ms; the anti-interference ability is evaluated by Monte Carlo method, and the reliability requirement is 90%. In the library learning scenario, this evaluation system accurately measures the performance of the method: the signal-to-noise ratio in a quiet environment is improved by 18dB, the feature retention rate is maintained at more than 95%, and the state recognition accuracy is close to 97%. Through this methodological effect evaluation, a complete evaluation report including data statistics and performance indicators is formed.

[0082] Following the specific requirements of this performance evaluation report, the method outputs the final mapping results. According to the evaluation indicators of the signal layer in the report, the filter parameter configuration is optimized: the notch filter uses phase-locked loop technology to achieve dynamic tracking, the high-pass filter adds transition band compensation, and the low-pass filter adopts Kaiser window design. Corresponding to the performance requirements of the feature layer, the wavelet processing strategy is improved: the threshold calculation introduces an adaptive factor, the reconstruction algorithm adds boundary processing, and the energy compensation adopts a segmentation strategy. For the evaluation results of the state layer, the prediction model structure is optimized: the model order is adaptively adjusted, the prediction step size changes dynamically, and the parameter update adopts a variable forgetting factor. In an open office environment, these optimized mapping results significantly improve the performance of the method: the voice during telephone conferences is clearer, the state recognition during team collaboration is more accurate, and the attention assessment during long-term work is more reliable. This evaluation-driven mapping optimization ensures that the method maintains a stable noise reduction effect in complex environments.

[0083] Step S106, determining the output region according to the mapping result, constructing an output chain according to the output region, completing noise reduction on the signal data in the output chain, and outputting a pure signal.

[0084] Specifically, the mapping results output by the step are integrated to determine the output area. The step generates performance mapping results for the signal layer (signal-to-noise ratio 15dB, phase deviation ±5°), feature layer (retention rate 90%, similarity 0.85) and state layer (accuracy 95%, delay 100ms). The method performs three-level region division: First, the O1 output area is divided for the signal layer results, and the fusion strategy is designed: a 256-point sliding window is used in the time domain level to perform data segmentation and local correlation analysis; a 512-point FFT transform is performed in the frequency domain level to calculate the power spectrum density distribution; and the instantaneous phase features are extracted by Hilbert transform at the phase level. Secondly, the O2 output area is set corresponding to the feature layer results to realize feature recombination: first, wavelet decomposition is performed, db4 wavelet basis function is selected, and 5-layer decomposition is performed; then the energy distribution of each scale coefficient is calculated, and the adaptive threshold is set; finally, the coefficient is screened by the soft threshold method. Finally, the O3 output area is divided for the state layer results, and a state synthesis method is constructed: including a state recognition unit based on a mixed Gaussian model, a state tracking unit using a Kalman filter, and a state prediction unit using a recursive neural network. Through this three-level region division, the method determines the signal integration region O1, the feature fusion region O2 and the state synthesis region O3, each of which contains specific processing strategies and performance indicators.

[0085] Based on these three determined output regions, the method constructs an output chain. A signal integrator is constructed in the O1 region, which includes three processing units: the time domain alignment unit adopts an improved cross-correlation algorithm, designs a multi-scale pyramid structure, calculates correlation coefficients at different scales, and realizes fast signal alignment; the amplitude normalization unit uses a 128-point sub-window based on an adaptive threshold technology to calculate local statistical features and dynamically update the normalization coefficient; the phase correction unit uses a 64-order FIR all-pass filter group to achieve a linear phase response in the range of 0-100Hz. A feature fusion device is implemented for the O2 region: the feature extraction layer uses wavelet packet decomposition, constructs a four-layer decomposition tree, and calculates the terminal node energy; the feature selection layer adopts an improved principal component analysis to adaptively determine the feature dimension; the feature mapping layer designs a three-layer neural network, and the hidden layer uses the ReLU activation function. A state integrator is constructed in the O3 region: a five-state Markov model is designed to describe state transitions, the forward-backward algorithm is used to estimate state probabilities, and the optimal state sequence is determined by the Viterbi algorithm. Through this hierarchical construction, the method forms a complete output chain structure, including three processing levels: signal integration, feature fusion, and state integration.

[0086] Using the output chain of this three-layer processing, the method performs noise reduction processing. The specific configuration of the signal integrator: time domain alignment uses a 512-point main window and a 64-point subwindow to calculate multi-scale correlation coefficients; amplitude normalization uses a 128-point sliding window with a 50% overlap rate and adaptively updates statistical parameters; phase correction is achieved through a 64-order FIR filter bank, and 18 subband filters are designed to cover the 0-100Hz frequency range. The processing parameters of the feature fusion are: wavelet packet decomposition selects the db4 basis function, the energy threshold is set to 85%, and a four-layer decomposition tree is constructed; principal component analysis retains features with a 95% contribution rate, and the feature dimension reduction matrix is ​​updated in real time; the neural network uses a [64-32-16] layer structure, the batch size is set to 32, and the learning rate is 0.001. The operating parameters of the state integrator are: the state transfer matrix is ​​updated every 500ms, the state probability estimation uses a 200ms time window, and the state sequence smoothing uses a 5-point median filter. After these three layers of serial processing, the method generates a stable signal output stream, which contains clear signals after noise reduction, prominent features and continuous state information.

[0087] According to this signal output stream, the method generates the final pure brain-computer interaction signal. The method first uses an improved Savitzky-Golay filter to smooth the signal. The window length of the filter is dynamically adjusted. The processing process is divided into three steps: the first step is to segment the original data, and each segment uses a 21-point window for local polynomial fitting; the second step is to calculate the optimal polynomial coefficients in each window, and the order is adaptively selected from 2 to 4; the third step is to achieve signal smoothing through convolution operation. Then the method performs intelligent edge processing and adopts a three-level boundary optimization strategy: first, the signal is extended at both ends of the boundary, and the extension length is half of the window; then the weighted average method is used to fuse the original data and the extended data; finally, the boundary smooth transition is achieved through nonlinear interpolation. The method then performs the final amplitude adjustment. The implementation process includes: first calculating the signal envelope and extracting the amplitude change characteristics of the signal; then designing the dynamic threshold based on the envelope characteristics and determining the gain adjustment range; finally, smooth amplitude adjustment is achieved through piecewise linear mapping. For EEG signals of different frequency bands, the method adopts differentiated processing strategies: the α band (8-13Hz) focuses on maintaining waveform details by adjusting the filter coefficients; the β band (14-30Hz) focuses on phase characteristics and is processed using zero-phase filtering technology; the θ band (4-7Hz) focuses on baseline stability and uses a high-order polynomial fitting method. In complex usage environments, the method also designs an adaptive processing mechanism: dynamically adjust processing parameters according to the local characteristics of the signal, track signal change trends in real time, and ensure the continuity and stability of the output signal. Through this multi-level, multi-strategy signal processing, the method ultimately outputs a clear, stable, and pure brain-computer interaction signal.

[0088] In order to implement the dynamic noise reduction method of the brain-computer AI headset corresponding to the above method embodiment, so as to achieve the corresponding functions and technical effects. Figure 2 , Figure 2 The structural block diagram of a dynamic noise reduction device 200 provided in an embodiment of the present application is shown. For the convenience of description, only the parts related to the present embodiment are shown. The dynamic noise reduction device 200 provided in an embodiment of the present application includes:

[0089] The structure acquisition module 201 is used to acquire the signal data collected by the brain-computer AI headset and the corresponding noise segment, acquire the characteristic data corresponding to the noise segment, and acquire the noise structure corresponding to the characteristic data;

[0090] A level acquisition module 202 is used to acquire a noise level corresponding to the noise structure, construct a suppression chain according to the noise level, perform noise reduction decomposition in the suppression chain, determine a processing node according to a processing result corresponding to the noise reduction decomposition, generate a process chain according to the processing node, and determine a scheduling scheme in the process chain;

[0091] A scenario analysis module 203 is used to perform scenario analysis on the scheduling scheme, obtain environmental features, build a response chain according to the environmental features, and obtain adaptation parameters corresponding to the response chain;

[0092] A parameter acquisition module 204 is used to perform signal analysis on the characteristic data according to the adaptation parameter, obtain a characteristic area, build a protection chain according to the characteristic area, perform signal reorganization according to the protection chain to generate protection data, determine a control position in the protection data, generate a control chain according to the control position, and obtain control parameters corresponding to the control chain;

[0093] A noise reduction mapping module 205 is used to perform noise reduction mapping on the control parameter, obtain a noise reduction interval and convert the noise reduction interval into a mapping chain, and obtain a mapping result corresponding to the mapping chain;

[0094] The clean output module 206 is used to determine an output region according to the mapping result, construct an output chain according to the output region, perform noise reduction on the signal data in the output chain, and output a clean signal.

[0095] The above-mentioned dynamic noise reduction device 200 can implement the dynamic noise reduction method of the brain-computer AI headset of the above-mentioned method embodiment. The optional items in the above-mentioned method embodiment are also applicable to this embodiment and will not be described in detail here. The rest of the contents of the embodiment of this application can refer to the contents of the above-mentioned method embodiment, and will not be repeated in this embodiment.

[0096] Figure 3 This is a schematic diagram of the structure of a computer device provided in one embodiment of the present application. Figure 3As shown, the computer device 3 of this embodiment includes: at least one processor 30 ( Figure 3 Only one is shown in the figure), a memory 31 and a computer program 32 stored in the memory 31 and executable on the at least one processor 30, wherein the processor 30 implements the steps of any of the above method embodiments when executing the computer program 32.

[0097] The computer device 3 may be a computing device such as a smart phone, a tablet computer, a desktop computer, a cloud server, etc. The computer device may include but is not limited to a processor 30 and a memory 31. Those skilled in the art will understand that Figure 3 It is only an example of computer device 3 and does not constitute a limitation on computer device 3. It may include more or fewer components than shown in the figure, or a combination of certain components, or different components, for example, it may also include input and output devices, network access devices, etc.

[0098] The processor 30 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor, etc.

[0099] In some embodiments, the memory 31 may be an internal storage unit of the computer device 3, such as a hard disk or memory of the computer device 3. In other embodiments, the memory 31 may also be an external storage device of the computer device 3, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. equipped on the computer device 3. Further, the memory 31 may also include both an internal storage unit and an external storage device of the computer device 3. The memory 31 is used to store an operating system, an application program, a boot loader (BootLoader), data, and other programs, such as the program code of the computer program. The memory 31 may also be used to temporarily store data that has been output or is to be output.

[0100] In addition, an embodiment of the present application further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps in any of the above method embodiments are implemented.

[0101] An embodiment of the present application provides a computer program product. When the computer program product is run on a computer device, the computer device implements the steps in the above-mentioned method embodiments when executing the computer device.

[0102] In several embodiments provided in the present application, it is understood that each box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and the module, a program segment or a part of a code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from that marked in the accompanying drawings. For example, two consecutive boxes can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved.

[0103] If the functions are implemented in the form of software function modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application can be essentially or partly embodied in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium and includes several instructions for a computer device to execute all or part of the steps of the methods described in the various embodiments of the present application. The aforementioned storage media include: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, and other media that can store program codes.

[0104] The specific embodiments described above further describe the purpose, technical solutions and beneficial effects of the present application in detail. It should be understood that the above description is only a specific embodiment of the present application and is not intended to limit the scope of protection of the present application. It is particularly pointed out that for those skilled in the art, any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included in the scope of protection of the present application.

Claims

1. A dynamic noise reduction method for brain-computer AI headphones, characterized in that: include: Obtain signal data collected by the brain-computer AI headset and a corresponding noise segment, obtain feature data corresponding to the noise segment, and obtain a noise structure corresponding to the feature data; Acquire a noise level corresponding to the noise structure, construct a suppression chain according to the noise level, perform noise reduction decomposition in the suppression chain, determine a processing node according to a processing result corresponding to the noise reduction decomposition, generate a process chain according to the processing node, and determine a scheduling scheme in the process chain; Performing scenario analysis on the scheduling scheme, obtaining environmental features, constructing a response chain according to the environmental features, and obtaining adaptation parameters corresponding to the response chain; Performing signal analysis on the characteristic data according to the adaptation parameter to obtain a characteristic area, constructing a protection chain according to the characteristic area, performing signal reorganization according to the protection chain to generate protection data, determining a control position in the protection data, generating a control chain according to the control position, and obtaining control parameters corresponding to the control chain; Performing noise reduction mapping on the control parameter, obtaining a noise reduction interval and converting the noise reduction interval into a mapping chain, and obtaining a mapping result corresponding to the mapping chain; An output region is determined according to the mapping result, an output chain is constructed according to the output region, noise reduction of the signal data is completed in the output chain, and a pure signal is output.

2. The method according to claim 1, characterized in that The acquiring characteristic data corresponding to the noise section includes: Extracting features from the noise segment to obtain a feature set corresponding to the noise segment; generating a noise spectrum according to the feature set; The characteristic data is extracted from the noise spectrum.

3. The method according to claim 1, characterized in that The obtaining of the noise structure corresponding to the characteristic data includes: Performing distribution analysis on the characteristic data to obtain noise distribution corresponding to the characteristic data; Mapping the noise distribution into a statistical domain to obtain a feature expression; The noise structure is generated according to the characteristic expression.

4. The method according to claim 1, characterized in that The noise level at least includes an instantaneous layer, a short-term layer and a long-term layer; the step of constructing a suppression chain according to the noise level includes: configuring high-speed sampling nodes in the instantaneous layer; configuring a feature extraction node in the short-term layer; configuring a pattern recognition node in the long-term layer; The suppression chain is constructed according to the high-speed sampling node, the feature extraction node and the pattern recognition node.

5. The method according to claim 1, characterized in that The environmental characteristics include dominant characteristics, secondary characteristics and background characteristics; The step of constructing a response chain according to the environmental characteristics includes: Constructing a notch filter group according to the dominant characteristics, wherein the center frequency of the notch filter group is automatically adjusted following the interference source; constructing an adaptive threshold detector based on the secondary features, wherein the sensitivity of the adaptive threshold detector changes dynamically with the ambient noise level; Constructing a frequency tracking compensator according to the background characteristics, wherein the compensation parameters of the frequency tracking compensator are updated in real time with the vibration intensity; The response chain is constructed according to the notch filter bank, the adaptive threshold detector and the frequency tracking compensator.

6. The method according to claim 1, characterized in that The characteristic area includes a first area, a second area and a third area, the first area records the transient changes of attention fluctuations, the second area stores the dynamic characteristics of cognitive load, and the third area reflects the gradual trend of relaxation state; The step of constructing a protection chain according to the characteristic area includes: Acquire a Kalman filter-based energy tracking protector corresponding to the first area; Obtaining a multi-scale morphological protector corresponding to the second area; Obtaining a spectrum protector corresponding to the third area; The protection chain is constructed according to the energy tracking protector, the multi-scale morphology protector and the spectrum protector.

7. The method according to claim 1, characterized in that The control position includes a first position, a second position and a third position, wherein the first position is used to adjust the acquisition gain in real time to adapt to the dynamic changes of the home environment, the second position is used to optimize the feature window to capture the attention fluctuations in the online meeting, and the third position is used to continuously evaluate the working status to ensure the effect of remote collaboration; generating a control chain according to the control position includes: Constructing an adaptive gain controller according to the real-time control parameters corresponding to the first position; the adaptive gain controller includes a feedforward compensation unit and a feedback correction unit; constructing a dynamic window controller according to the characteristic control parameters corresponding to the second position; Constructing a fuzzy state controller according to the state control parameters corresponding to the third position, wherein the fuzzy state controller adopts a double-layer reasoning mechanism; The control chain is constructed according to the gain controller, the dynamic window controller and the fuzzy state controller.

8. A dynamic noise reduction device, characterized in that: include: A structure acquisition module, used to acquire signal data collected by the brain-computer AI headset and the corresponding noise segment, acquire feature data corresponding to the noise segment, and acquire the noise structure corresponding to the feature data; A level acquisition module, used for acquiring a noise level corresponding to the noise structure, constructing a suppression chain according to the noise level, performing noise reduction decomposition in the suppression chain, determining a processing node according to a processing result corresponding to the noise reduction decomposition, generating a process chain according to the processing node, and determining a scheduling scheme in the process chain; A scenario analysis module, used to perform scenario analysis on the scheduling scheme, obtain environmental features, build a response chain according to the environmental features, and obtain adaptation parameters corresponding to the response chain; a parameter acquisition module, configured to perform signal analysis on the characteristic data according to the adaptation parameter, obtain a characteristic area, construct a protection chain according to the characteristic area, perform signal reorganization according to the protection chain to generate protection data, determine a control position in the protection data, generate a control chain according to the control position, and obtain a control parameter corresponding to the control chain; A noise reduction mapping module, used to perform noise reduction mapping on the control parameter, obtain a noise reduction interval and convert the noise reduction interval into a mapping chain, and obtain a mapping result corresponding to the mapping chain; A clean output module is used to determine an output area according to the mapping result, build an output chain according to the output area, complete noise reduction on the signal data in the output chain, and output a clean signal.

9. A computer device, characterized in that: The method comprises a memory and a processor; the memory is used to store a computer program; the processor is used to execute the computer program and implement the method according to any one of claims 1 to 7 when executing the computer program.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, which, when executed by a processor, causes the processor to implement the method according to any one of claims 1 to 7.

Images