Method and device for real-time decoding of electroencephalogram signals and intention recognition of attention glasses

Abstract

The present application discloses a method and device for real-time decoding of electroencephalogram signals and intention recognition of attention glasses. The method includes: obtaining a standardized signal sequence corresponding to an initial signal sequence collected by the attention glasses; obtaining frequency domain components corresponding to the standardized signal sequence and energy analysis results corresponding to the frequency domain components, constructing a dual-domain feature sequence for correlation mapping, obtaining feature breakpoints, establishing a time-frequency relationship graph according to the feature breakpoints, determining a mapping interval according to the time-frequency relationship graph, and enhancing the feature sequence according to the mapping interval to obtain an enhanced feature sequence; performing mapping transformation on the enhanced feature sequence to obtain initial components, determining main components in the initial components, and generating a feature representation according to the main components; obtaining a feature interval and interval features corresponding to the feature interval to generate an intention feature sequence; obtaining change nodes and timing features corresponding to the intention feature sequence to generate an intention change sequence; and obtaining an intention recognition result corresponding to the intention change sequence.

Description

Technical Field

[0001] This application relates to the technical field of brain-computer interfaces, and particularly to a method and device for real-time decoding of electroencephalogram (EEG) signals and intention recognition of attention glasses. Background Art

[0002] As a portable EEG acquisition and analysis device, the core function of attention glasses is to real-time decode the user's EEG signals and recognize their cognitive intentions. The traditional EEG signal decoding methods mainly have the following problems: First, fixed sampling parameters and filtering strategies are adopted in the signal acquisition and preprocessing links, making it difficult to cope with the dynamic changes in signal quality; second, the feature extraction process fails to effectively combine the complementary information of time-domain and frequency-domain features, resulting in incomplete feature expression; third, the feature mapping and enhancement links lack an adaptive mechanism and cannot be dynamically adjusted according to the feature importance; finally, the intention recognition process ignores the temporal correlation of the feature sequence, affecting the judgment accuracy of state transitions.

[0003] In practical applications, the factors affecting the decoding effect are mainly reflected in three aspects: One is the interference at the signal level, including baseline drift, channel noise, etc., which affects the stability of feature extraction; the second is the changes at the feature level, where users exhibit diverse feature patterns and complex temporal dependence relationships under different cognitive tasks; the third is the differences at the intention level, where the user's attention state and intention expression are affected by factors such as emotions and fatigue, showing significant individual differences and time-varying characteristics. Existing technologies often adopt independent processing modules and lack the collaborative optimization among various links, making it difficult to systematically solve the above challenges.

[0004] In addition, the processing power of portable devices is limited. Existing methods generally adopt complex offline analysis algorithms, which have a large amount of calculation and high latency, and are difficult to meet the requirements of real-time interaction. At the same time, the adaptability of the algorithms is insufficient and they cannot dynamically adjust the processing strategy according to the device resources.

[0005] Therefore, there is an urgent need for a method to solve at least one of the above problems. Summary of the Invention

[0006] The embodiments of this application provide a method and device for real-time decoding of EEG signals and intention recognition of attention glasses, aiming to solve the problems that existing methods generally adopt complex offline analysis algorithms, which have a large amount of calculation and high latency, and are difficult to meet the requirements of real-time interaction. At the same time, the adaptability of the algorithms is insufficient and they cannot dynamically adjust the processing strategy according to the device resources.

[0007] In a first aspect, the embodiments of this application provide a method for real-time decoding of EEG signals and intention recognition of attention glasses, including:

[0008] Obtain the multi-channel EEG time-domain signals collected by the attention glasses, perform multi-point sampling on the multi-channel EEG time-domain signals to obtain the original sampling data, and generate an initial signal sequence according to the original sampling data;

[0009] Obtain the signal segments that need to be corrected in the initial signal sequence, perform compensation adjustment on the signal segments to generate a standardized signal sequence; obtain the time-domain features corresponding to the standardized signal sequence, perform frequency-domain conversion on the time-domain features to obtain the frequency-domain components and the energy analysis results corresponding to the frequency-domain components, so as to construct a dual-domain feature sequence according to the energy analysis results;

[0010] Perform correlation mapping on the dual-domain feature sequence to obtain feature breakpoints, establish a time-frequency relationship graph according to the feature breakpoints, determine the mapping interval according to the time-frequency relationship graph, and enhance the feature sequence according to the mapping interval to obtain an enhanced feature sequence;

[0011] Perform mapping conversion on the enhanced feature sequence to obtain initial components, determine the main components in the initial components, and generate a feature representation according to the main components; perform segmented splitting on the feature representation to obtain feature intervals and the interval features corresponding to the feature intervals, and generate an intention feature sequence according to the interval features;

[0012] Obtain the change nodes corresponding to the intention feature sequence and the corresponding timing features, and generate an intention change sequence according to the multiple timing features;

[0013] Obtain the preliminary marks corresponding to the intention change sequence, and obtain the intention recognition results corresponding to the preliminary marks.

[0014] In a second aspect, the present application also provides an EEG signal real-time decoding and intention recognition device, including:

[0015] A signal acquisition module, configured to obtain the multi-channel EEG time-domain signals collected by the attention glasses, perform multi-point sampling on the multi-channel EEG time-domain signals to obtain the original sampling data, and generate an initial signal sequence according to the original sampling data;

[0016] A compensation adjustment module, configured to obtain the signal segments that need to be corrected in the initial signal sequence, perform compensation adjustment on the signal segments to generate a standardized signal sequence; obtain the time-domain features corresponding to the standardized signal sequence, perform frequency-domain conversion on the time-domain features to obtain the frequency-domain components and the energy analysis results corresponding to the frequency-domain components, so as to construct a dual-domain feature sequence according to the energy analysis results;

[0017] An association mapping module, configured to perform association mapping on the dual-domain feature sequence, obtain feature breakpoints, establish a time-frequency relationship graph according to the feature breakpoints, determine a mapping interval according to the time-frequency relationship graph, and enhance the feature sequence according to the mapping interval to obtain an enhanced feature sequence;

[0018] A mapping conversion module, configured to perform mapping conversion on the enhanced feature sequence, obtain an initial component, determine a main component in the initial component, and generate a feature representation according to the main component; perform segmented slicing on the feature representation to obtain a feature interval and interval features corresponding to the feature interval, and generate an intention feature sequence according to the interval features;

[0019] A feature acquisition module, configured to acquire change nodes corresponding to the intention feature sequence and corresponding timing features, and generate an intention change sequence according to multiple said timing features;

[0020] A result acquisition module, configured to acquire a preliminary label corresponding to the intention change sequence and obtain an intention recognition result corresponding to the preliminary label.

[0021] In a third aspect, the present application further provides a computer device, including a processor and a memory, where the memory is used to store a computer program, and when the computer program is executed by the processor, it implements the method for real-time decoding and intention recognition of electroencephalogram signals of the attention glasses as described in the first aspect.

[0022] In a fourth aspect, the present application further provides a computer-readable storage medium, which stores a computer program, and when the computer program is executed by a processor, it implements the method for real-time decoding and intention recognition of electroencephalogram signals of the attention glasses as described in the first aspect.

[0023] Compared with the prior art, the present application at least has the following beneficial effects:

[0024] 1. The decoding robustness is improved through a multi-level signal processing strategy. Adaptive sampling and dynamic baseline correction are adopted in the signal acquisition stage, effectively suppressing signal interference; time-domain and frequency-domain features are combined in the feature extraction stage, enhancing the integrity of feature expression; adaptive weight adjustment is adopted in the feature enhancement stage, improving the discriminative ability of features. This multi-level processing method significantly enhances the anti-interference ability of the system corresponding to the method (collectively referred to as the system in the specification) in a complex environment, making the decoding result more reliable.

[0025] 2. The recognition accuracy is improved through the dynamic mapping and state optimization of the feature sequence. The system performs multi-scale analysis on the feature sequence to accurately capture the key nodes of state changes; through state merging and optimization processing, the instability of state judgment is effectively eliminated; a hierarchical intention recognition strategy is adopted to achieve precise distinction of different types of intentions. This method based on sequence processing greatly improves the recognition accuracy of the system for user intention changes.

[0026] 3. High-efficient data processing is achieved through a pipeline processing architecture. The system adopts a segmented processing and parallel computing strategy, and an adaptive computing strategy is adopted in each processing link to dynamically adjust the processing depth according to the feature importance; through feature screening and dimension optimization, data redundancy is reduced; a streaming processing architecture is used to achieve high-efficient data transmission and processing. This optimized design improves the system processing efficiency, ensures the processing effect, and meets the actual application requirements at the same time.

[0027] It should be understood that the above general description and the following detailed description are only exemplary and explanatory, and cannot limit this application. Brief Description of the Drawings

[0028] Figure 1 It is a schematic flowchart of the method for real-time decoding and intention recognition of electroencephalogram signals of the attention glasses shown in the embodiments of this application;

[0029] Figure 2 It is a schematic structural diagram of the device for real-time decoding and intention recognition of electroencephalogram signals shown in the embodiments of this application;

[0030] Figure 3 It is a schematic structural diagram of the computer device shown in the embodiments of this application. Detailed Description of the Embodiments

[0031] In the following description, specific details such as specific system structures and technologies are presented for the purpose of illustration rather than limitation, so as to thoroughly understand the embodiments of this application. However, those skilled in the art should clearly understand that this application can also be implemented in other embodiments without these specific details. In other cases, the detailed descriptions of well-known systems, devices, circuits, and methods are omitted to avoid unnecessary details from interfering with the description of this application.

[0032] It should be understood that when used in the specification and the appended claims of this application, the term "comprising" indicates the presence of the 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 their combinations.

[0033] It should also be understood that the term "and / or" as used in the specification and appended claims of this application refers to any combination and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0034] As used in the specification and appended claims of this application, the term "if" can be interpreted as "when" or "once" or "in response to determining" or "in response to detecting" depending on the context. Similarly, the phrase "if determined" or "if [the described condition or event] is detected" can be interpreted as meaning "once determined" or "in response to determining" or "once [the described condition or event] is detected" or "in response to detecting [the described condition or event]" depending on the context.

[0035] In addition, in the description of the specification and appended claims of this application, the terms "first", "second", "third", etc. are only used for distinguishing descriptions and cannot be understood as indicating or implying relative importance.

[0036] Reference to "one embodiment" or "some embodiments" etc. described in the specification of this application means that a specific feature, structure or characteristic described in connection with that embodiment is included in one or more embodiments of this application. Thus, statements such as "in one embodiment", "in some embodiments", "in other some embodiments", "in still other embodiments" etc. that appear in different places in this specification do not necessarily all refer to the same embodiment, but mean "one or more but not all embodiments", unless otherwise specifically emphasized in other ways. The terms "comprising", "including", "having" and their variants all mean "including but not limited to", unless otherwise specifically emphasized in other ways.

[0037] The technical solutions of the embodiments of this application will be introduced below.

[0038] As a portable electroencephalogram (EEG) acquisition and analysis device, the core function of the attention glasses is to decode the user's EEG signals in real time and recognize their cognitive intentions. The traditional EEG signal decoding methods mainly have the following problems: First, fixed sampling parameters and filtering strategies are adopted in the signal acquisition and preprocessing links, making it difficult to cope with the dynamic changes in signal quality; Second, the feature extraction process fails to effectively combine the complementary information of time-domain and frequency-domain features, resulting in incomplete feature expression; Third, the feature mapping and enhancement links lack an adaptive mechanism and cannot be dynamically adjusted according to feature importance; Finally, the intention recognition process ignores the temporal correlation of the feature sequence, affecting the accuracy of state transition judgment.

[0039] In practical applications, the factors affecting the decoding effect are mainly reflected in three aspects: First, interference at the signal level, including baseline drift, channel noise, etc., affects the stability of feature extraction; second, changes at the feature level, where users exhibit diverse feature patterns and complex temporal dependencies under different cognitive tasks; third, differences at the intention level, where the user's attention state and intention expression are affected by factors such as emotion and fatigue, showing significant individual differences and time-varying characteristics. Existing technologies often adopt independent processing modules, lacking collaborative optimization among various links, and it is difficult to achieve a systematic solution to the above challenges.

[0040] In addition, the processing power of portable devices is limited. Existing methods generally use complex offline analysis algorithms, which have a large amount of computation and high latency, and it is difficult to meet the requirements of real-time interaction. At the same time, the adaptability of the algorithms is insufficient, and they cannot dynamically adjust the processing strategy according to the device resources. Therefore, there is a need to develop an efficient decoding framework that can rationally utilize computing resources while ensuring recognition accuracy through feature optimization and sequence processing.

[0041] Please refer to Figure 1 , Figure 1 which is a schematic flowchart of a method for real-time decoding and intention recognition of electroencephalogram signals of an attention glasses provided by an embodiment of this application. The method for real-time decoding and intention recognition of electroencephalogram signals of the attention glasses in the embodiment of this application can be applied to a computer device, and the computer device includes but is not limited to devices such as smart phones, laptop computers, tablet computers, desktop computers, physical servers, and cloud servers. As Figure 1 shown, the method for real-time decoding and intention recognition of electroencephalogram signals of the attention glasses in this embodiment includes steps S101 to S106, which are described in detail as follows:

[0042] Step S101, obtain multi-channel electroencephalogram time-domain signals collected by the attention glasses, perform multi-point sampling on the multi-channel electroencephalogram time-domain signals to obtain original sampling data, and generate an initial signal sequence according to the original sampling data.

[0043] Specifically, the implementation process of obtaining the multi-channel EEG time-domain signals of the attention glasses is first based on a multi-channel signal acquisition system, which includes a signal acquisition unit, a data conversion unit, and a data transmission unit. The signal acquisition unit uses 4-8 dry electrodes to form an acquisition array. The electrodes are made of Ag / AgCl material, with a diameter of 10 mm. The electrode surface is specially treated to improve the conductivity. The electrode surface is coated with a nanoscale conductive material, with a material thickness of 0.5-1.0 μm and a surface resistivity of less than 0.1 Ω / cm². The electrode layout follows the international 10-20 system, mainly distributed in cognitive activity-related areas such as the frontal lobe (at positions Fp1 and Fp2) and the temporal lobe (at positions T3 and T4). The electrode spacing is maintained within the range of 3-5 cm. At the same time, a reference electrode is placed at the position of the mastoid behind the left ear. Each electrode is subjected to signal conditioning through an independent preamplifier. The amplifier uses a low-noise operational amplifier chip, with an adjustable gain range of 1000-10000 times, usually set to 5000 times. The input impedance of the preamplifier is greater than 100 MΩ, the common-mode rejection ratio is greater than 100 dB, and the input reference noise is less than 1 μV (p-p). The data conversion unit uses a 24-bit high-precision ADC, with a quantization accuracy of 0.1 μV and a maximum sampling rate supporting 2000 Hz. The standard working sampling rate is set to 500 Hz. The signal-to-noise ratio of the ADC is greater than 100 dB, the integral nonlinear error is less than 0.001%, and the differential nonlinear error is less than 0.0005%. Before analog-to-digital conversion, the signal is processed by an analog band-pass filter, using a fourth-order Butterworth filter, with a passband range of 0.1-100 Hz, a stopband attenuation greater than 60 dB, and a passband ripple less than 0.5 dB. The data transmission unit uses a high-speed serial interface, with a transmission rate not less than 1 Mbps, using a differential signal transmission method, supporting real-time data stream transmission and status monitoring. The transmission protocol includes a data header (synchronization word, packet sequence number, timestamp), a data payload (multi-channel sampled data), and a check field (CRC32 check code).

[0044] In some embodiments, generating the initial signal sequence according to the original sampled data includes: obtaining the sampling point distribution corresponding to the multi-point sampling; marking the valid sampling segments in the original sampled data according to the sampling point distribution; and generating the initial signal sequence according to the valid sampling segments.

[0045] After completing the basic configuration of the signal acquisition system, multi-point sampling is performed on the acquired time-domain signals. This process samples multi-channel signals based on the time-division multiplexing principle, and each sampling period includes the data acquisition and conversion processes of all channels. Taking a sampling rate of 500 Hz as an example, the system completes a full-channel sampling every 2 ms, and the sampling time interval for each channel does not exceed 50 μs. The system uses a high-precision crystal oscillator to provide a reference clock with a clock frequency of 20 MHz and a frequency stability better than ±20 ppm. The sampling process adopts a synchronous sampling mechanism, and all channels complete data acquisition under the control of the same clock signal. The sampling timing deviation between channels is less than 1 μs, ensuring the time consistency of multi-channel data. Each sampling point record contains a timestamp (64-bit precision, resolution of 1 μs), a channel number (8 bits), and a voltage value (24 bits). The sampled data undergoes preliminary digital filtering processing using a 100th-order FIR filter. The filter coefficients are designed by the Hamming window function method, and the passband cut-off frequency matches that of the analog filter, with a transition band width of 2 Hz. The filtered data is organized into data frames in chronological order, and each frame contains 50 ms of data volume, that is, 25 sampling points. The data frame structure includes a frame header (32 bytes, including a synchronization word, frame number, and timestamp), a data area (variable according to the number of channels, with each sampling point occupying 4 bytes), and a frame tail (4 bytes, including a CRC32 checksum). The system uses a double-buffer mechanism to manage data frames. The main buffer maintains a data length of 10 seconds and adopts a circular queue structure. The buffer size is dynamically allocated according to the number of channels. The secondary buffer is used for data preprocessing and has a size of 1 / 5 of the main buffer. The marking of valid sampling points is based on signal amplitude and spectral characteristics. The system calculates the voltage difference between adjacent sampling points (threshold set to 50 μV) and the local frequency distribution (mainly focusing on the energy distribution in the 0.1 - 30 Hz frequency band), and determines the validity of sampling points in combination with preset judgment rules.

[0046] After obtaining the valid sampling point markers, the system generates the initial signal sequence. This process first normalizes the amplitude of the valid sampling data. The piecewise linear mapping method is used to unify the signal amplitudes of different channels into the range of [-1, 1]. The normalization process is divided into three steps: First, calculate the signal amplitude range of each channel to determine the slope and bias of the mapping function; then truncate the outliers outside the normal range, and the truncation threshold is set to the mean ± 3 times the standard deviation; finally, perform the linear mapping transformation. The normalized data constitutes the standardized time-domain sequence, and each data point contains the normalized voltage value, the original timestamp, and the channel identifier. For the data gaps between adjacent valid sampling segments, different interpolation methods are used according to the gap length: cubic spline interpolation is used for gaps less than 100 ms, and the interpolation node interval is 1 / 4 of the original sampling interval; linear interpolation is used for gaps in the range of 100 - 300 ms, and the number of interpolation points is equal to the original sampling rate; gaps exceeding 300 ms are marked as signal breakpoints, and a marker field is added at the breakpoints. The interpolation process is carried out in a per-channel manner to ensure the continuity of the signal, and an interpolation marker bit is added in the interpolation interval. The final initial signal sequence is stored in a structured manner, and the data structure includes: the main data area (multi-channel time-domain data, indexed by channel number), the time information area (including the original timestamp and the interpolation point timestamp), and the marker information area (valid segment marker, breakpoint marker, interpolation marker). The sequence is stored in segments of 1 second, with an overlap rate of 50% between adjacent segments, and segment numbers and time range information are added to each segment of data. The signal sequence is stored in a compressed format, using a lossless compression algorithm, with a compression ratio of not less than 2:1, and at the same time ensuring that the decompression speed meets the requirements of real-time processing.

[0047] Step S102, obtain the signal segments that need to be corrected in the initial signal sequence, perform compensation adjustment on the signal segments to generate a standardized signal sequence; obtain the time-domain characteristics corresponding to the standardized signal sequence, perform frequency-domain conversion on the time-domain characteristics to obtain the frequency-domain components and the energy analysis results corresponding to the frequency-domain components, so as to construct a dual-domain feature sequence according to the energy analysis results.

[0048] Specifically, the process of detecting baseline drift in the initial signal sequence obtained in step S101 first uses a multi-scale analysis method. The system segments the initial signal sequence into segments of 2-second length, with an overlap rate of 50% between adjacent segments. For each signal segment, the local mean curve is calculated using a sliding window of 100 ms with a sliding step of 20 ms. The local mean curve reflects the slow-changing trend of the signal and contains baseline drift information. The system performs wavelet decomposition on the local mean curve using the db4 wavelet basis function with 4 decomposition levels. In the decomposition result of each level, the low-frequency coefficients are extracted as the baseline drift features. A threshold is set for the extracted feature sequence, and the threshold value is adaptively determined according to the statistical characteristics of the signal segment, generally taking 2 times the local standard deviation. When the feature value exceeds the threshold, the corresponding time point is marked as the starting point of baseline drift. The system also calculates the slope and duration of the baseline drift to establish a drift feature vector. The drift detection result is represented by a binary sequence, where 1 indicates the presence of drift and 0 indicates no drift. The detection process uses a parallel processing architecture and supports simultaneous analysis of multiple channels. The system transmits the detection result together with the original signal and the drift feature vector to the next processing stage.

[0049] In some embodiments, obtaining the signal segment to be corrected in the initial signal sequence, compensating and adjusting the signal segment, and generating a standardized signal sequence includes: detecting baseline drift in the initial signal sequence to obtain the signal segment to be corrected; performing the compensation and adjustment on the signal segment to be corrected; and generating a standardized signal sequence according to the adjustment result corresponding to the compensation and adjustment.

[0050] Based on the detection result, the system marks the signal segments to be corrected. The marking process first obtains the binary sequence and the drift feature vector of drift detection from the previous stage to determine the boundaries of each drift interval. Extend 100 ms forward from the starting point of drift and backward until the drift feature returns to normal. For each marked interval, the main feature parameters of the drift are calculated: drift amplitude (deviation from the local mean), drift rate (change amount per unit time), and drift pattern (linear / non-linear). The system classifies the drift intervals according to these feature parameters into two categories: fast drift (duration less than 500 ms) and slow drift (duration greater than 500 ms). For fast drift, the system adds transition marks at the boundaries of the drift interval for subsequent smoothing processing. For slow drift, the system further divides the drift interval into multiple sub-intervals, each sub-interval having a length of 200 ms and an overlap of 50 ms between adjacent sub-intervals. When performing interval marking, the system also records the statistical characteristics of the interval, including mean, standard deviation, maximum and minimum values, etc. This information is transmitted to the next stage as a reference for compensation and adjustment. The marking information is stored in a compressed format and includes time stamps, interval types, feature parameters, etc.

[0051] The system compensates and adjusts the marked signal segments to be corrected. According to the interval type and characteristic parameters transmitted in the previous stage, the system selects different compensation strategies: for fast drift, an adaptive filtering method is adopted, using a 50th-order FIR filter, and the filter coefficients are updated in real time through the LMS algorithm, and the step size parameter is dynamically adjusted according to the drift amplitude. For slow drift, a piecewise polynomial fitting method is adopted, and a 3rd-order polynomial is used for baseline estimation for each sub-interval, and the polynomial coefficients are solved by the least squares method. A weighted average method is used at the interval boundary to achieve smooth transition. The compensation process is iterative. After each iteration, the residual drift amount is calculated, and the iteration stops when the residual amount is less than a preset threshold (usually 5% of the original drift amplitude). The compensation adjustment also considers the frequency characteristics of the signal to ensure that the effective components of the signal are not affected during the compensation process. The system evaluates the quality of the compensated signal and calculates the stability index and the change in signal-to-noise ratio of the signal. The compensation results are stored in a temporary buffer, including the original signal, compensation values, corrected signal, and various evaluation indicators, providing input for the next-stage standardization process.

[0052] After completing the compensation adjustment, the system generates a standardized signal sequence. The standardization process first reads the corrected signal data from the temporary buffer, including the compensated time-domain signal and related evaluation indicators. The corrected signal read is amplitude-normalized using a piecewise linear mapping method, and the mapping interval is [-1,1]. The influence of the compensation adjustment is considered during the normalization process, and weighted mapping is used for regions with larger compensation amplitudes. The signal sequence is segmented, with a segment length of 1 second and an overlapping rate of 50% between adjacent segments. Each signal segment contains a complete data structure: signal data (corrected time-domain data), time information (original timestamp and processing timestamp), feature markers (drift features, compensation parameters), and quality indicators (signal-to-noise ratio, stability score). The standardized sequence adopts a hierarchical storage structure, including a data layer (signal data), a feature layer (processing parameters and results), and a control layer (processing flow information). The system checks the integrity of the standardized sequence to ensure data continuity and feature consistency. After passing the check, the standardized sequence is transmitted to the next processing link. The sequence transmission adopts a block transmission method, with each block of data being 2 seconds in size, and the transmission process includes a data verification mechanism.

[0053] In some embodiments, obtaining the time-domain features corresponding to the standardized signal sequence, performing a frequency-domain conversion on the time-domain features, obtaining the frequency-domain components and the energy analysis results corresponding to the frequency-domain components, and constructing a dual-domain feature sequence according to the energy analysis results includes: segmenting the standardized signal sequence to obtain a plurality of signal segments; obtaining the time-domain features corresponding to the signal segments; performing a frequency-domain conversion on the time-domain features to obtain the frequency-domain components; obtaining the energy analysis results corresponding to the frequency-domain components; and constructing the dual-domain feature sequence according to the time-domain features, frequency-domain components, and energy analysis results.

[0054] Exemplarily, segmenting the standardized signal sequence to obtain multiple signal segments includes: segmenting the standardized signal according to a preset initial segment length to obtain multiple initial segments; the overlapping rate corresponding to adjacent initial segments is within a preset overlapping rate range; iteratively optimizing the initial segment length according to a preset segment range, and segmenting the standardized signal sequence according to the optimized initial segment length to obtain multiple signal segments.

[0055] Adaptive segmentation strategy is adopted for segmenting the standardized signal sequence. The system initially segments the signal sequence with a basic length of 2 seconds, and maintains a 50% overlapping rate between adjacent segments to ensure the continuity of signal features. Calculate the signal statistical characteristics for each basic segment, including mean, variance, peak-to-peak value, and zero-crossing rate. These characteristics respectively reflect the baseline level, fluctuation intensity, amplitude range, and frequency change of the signal. Set an adaptive threshold based on the statistical characteristics for boundary determination. When the feature difference degree between adjacent sub-segments exceeds the threshold, it is marked as a paragraph boundary. The segmentation process adopts an iterative optimization method. The initial segment length is 2 seconds, and in each iteration, the segment length can be adjusted to 1 / 2 or 2 times the original according to the feature difference, and the adjustment range is limited between 1 - 4 seconds. The system introduces an adjustment limit mechanism to prevent over-segmentation. The segmentation result includes the start time, segment length, and segment feature description of each segment, and is stored in a structured manner. After segmentation, time-domain features are extracted for each signal segment: amplitude features include mean, variance, skewness, kurtosis, maximum value, and minimum value; morphological features include zero-crossing rate, number of wave peaks, number of wave valleys, and waveform factor; statistical features include autocorrelation coefficient, cross-correlation coefficient, and entropy value. Feature extraction adopts a sliding window method with a window length of 200ms and a step size of 50ms to ensure the time continuity of features.

[0056] Based on the time-domain feature sequence extracted in the previous stage, the system performs frequency-domain transformation. First, the Fourier transform is performed on the time-domain feature sequence (including statistical features such as mean, variance, skewness, kurtosis, etc.) to obtain the frequency distribution characteristics of the features. At the same time, the fast Fourier transform is performed on the original signal segment to obtain the basic spectral features. Before the transformation, the Hanning window is used to reduce spectral leakage, and the amplitude spectrum and phase spectrum of the signal are obtained. Then, the wavelet transform is performed on the signal segment. The db4 wavelet basis function is selected for 5-layer decomposition, and the decomposition depth covers the key frequency bands. In each layer of decomposition, the system performs correlation analysis between the change trend of the time-domain feature sequence and the distribution of wavelet coefficients, and calculates and saves the approximation coefficients and detail coefficients. At the same time, the Hilbert transform is applied to obtain the instantaneous frequency and instantaneous amplitude characteristics of the signal, and a corresponding relationship is established with the time-domain morphological features that have been extracted. The system organizes the results of various transformations into a frequency-domain feature matrix. The rows of the matrix correspond to time points, and the columns correspond to different frequency-domain features. All calculation processes use high-precision numerical representations, and the features obtained by each transformation method are timestamped to ensure synchronization with the time-domain feature sequence.

[0057] After obtaining the frequency-domain transformation results, the system performs energy analysis. Energy statistics are carried out in the standard EEG frequency bands, including the δ band, θ band, α band, β band, and γ band. The absolute energy and relative energy of each frequency band are calculated, and normalization is performed to make the energies of different frequency bands comparable. The energy ratios between frequency bands are analyzed, including θ / α, β / α, (θ + α) / β, etc. These ratios reflect the relative intensities of different EEG rhythms. For the wavelet transform results, the energy distribution of each scale coefficient is calculated separately, and an energy-scale diagram is constructed to show the energy distribution at different decomposition scales. The system also calculates the time-varying characteristics of energy, including the short-time energy change rate and the energy fluctuation range. All energy characteristics are organized into a feature vector as a complete description of the signal energy characteristics. The energy analysis results also carry a time index, which is convenient for tracking the time process of energy changes.

[0058] Based on the analysis results of time-domain features, frequency-domain features, and energy features, the system constructs a dual-domain feature sequence. First, standardization processing is performed on various types of features to make different types of features comparable. Feature fusion adopts a weighted combination method, and the weight coefficients are determined by analyzing the significance of the features. For each signal segment, a feature descriptor is generated, which includes a time-domain feature subset, a frequency-domain feature subset, and an energy feature subset. Adjacent feature descriptors are established corresponding relationships through timestamps to ensure the continuity of the feature sequence. The system removes redundant features through feature analysis to reduce the dimension of the feature space. The final feature data is organized into a multi-layer structure: the basic feature matrix stores the time-domain and frequency-domain features of all channels, the frequency-band energy matrix records the energy distribution of each frequency band, and the channel correlation matrix describes the feature relationships between channels. The feature sequence is stored in time segments, and each segment contains complete feature descriptions and time range information.

[0059] In this way, the system completes the process of feature extraction and representation of the standardized signal sequence. The formed dual-domain feature sequence preserves the complete feature information of the signal in the time domain and frequency domain, including descriptions in multiple dimensions such as amplitude, morphology, frequency, and energy. The feature data is stored in a hierarchical structure, ensuring the correspondence with the original signal through time indexing, and providing a data basis for subsequent feature correlation mapping processing. Each component of this feature sequence can be accessed independently, facilitating the extraction and use of specific types of feature information in subsequent processing.

[0060] Step S103: Perform correlation mapping on the dual-domain feature sequence to obtain feature breakpoints, establish a time-frequency relationship graph based on the feature breakpoints, determine the mapping interval according to the time-frequency relationship graph, and enhance the feature sequence according to the mapping interval to obtain an enhanced feature sequence.

[0061] Specifically, to perform correlation mapping on the dual-domain feature sequence, first establish a feature correspondence relationship. The system divides the dual-domain feature sequence into basic processing units with a length of 2 seconds, and adjacent units overlap by 1 second to ensure the continuity of feature mapping. For each processing unit, extract the time-domain feature matrix and the frequency-domain feature matrix. The rows of the matrix represent time sampling points, and the columns represent different feature types. The time-domain features within the processing unit include waveform features (peak value, valley value, zero-crossing point, waveform slope, wave amplitude mean), statistical features (mean, standard deviation, kurtosis, skewness, variance, quartile difference), and morphological features (waveform factor, pulse factor, margin factor, waveform steepness). The frequency-domain features include spectral features (main frequency, frequency center, bandwidth, frequency concentration, spectral kurtosis), energy features (band energy, energy ratio, power spectral density), and phase features (phase spectrum, group delay, phase coherence). The system establishes a feature mapping table using a multi-level index structure. The first-level index distinguishes time-domain and frequency-domain features. The second-level index groups by feature category. The third-level index points to specific feature attributes. The bottom-level index links the feature value and the timestamp. The correlation strength between features is calculated by the correlation coefficient, and three levels of correlation thresholds are set: strong correlation (above 0.8), medium correlation (0.5 - 0.8), and weak correlation (below 0.5), and the correlation type, strength value, and time persistence are recorded.

[0062] Based on the results of feature correlation analysis, the system marks the breakpoints in the feature sequence. In the marking process, the established feature mapping table is first used, and the correlation strength is used as an important basis for breakpoint judgment. The feature breakpoint detection adopts an adaptive sliding window method. The basic window length is set to 500 ms, and the window length can be dynamically adjusted according to the signal change rate, ranging from 300 - 700 ms, and the step size is 1 / 5 of the window length. Within each sliding window, the analysis is carried out by combining the change rate of feature correlation degree (based on the calculated correlation coefficient), the fluctuation amplitude of feature values, and the dispersion degree of feature distribution. When these indicators simultaneously exceed their respective thresholds and the duration exceeds 3 sampling points, the center point of this time period is marked as a candidate breakpoint. The system uses a hierarchical clustering method to process the candidate breakpoints. The clustering feature vector includes the time position of the breakpoint, the feature change amplitude, and the similarity of the front and back features. Each breakpoint determines its final position by the centroid method and calculates the significance score.

[0063] According to the marked feature breakpoint distribution and feature mapping results, the system constructs a time - frequency relationship graph. The relationship graph adopts a four - layer matrix structure, including a time layer, a feature layer, a correlation layer, and a control layer. The time layer records the timestamps of sampling points and breakpoint positions. The feature layer stores the numerical values and change trends of various features. The correlation layer integrates the feature mapping relationship and correlation strength. The control layer manages the processing parameters and status marks. For each feature breakpoint, the system analyzes the feature change trends of the 50 sampling points before and after it, and combines the associated features to perform change curve fitting. In the relationship graph, the weights of feature connections are set based on the correlation strength. The dominant features are marked with solid lines of different thicknesses, and the secondary features are marked with dashed lines. The system analyzes the feature propagation characteristics between breakpoints and constructs a propagation parameter model.

[0064] Using the established time - frequency relationship graph, the system determines the feature mapping intervals. The mapping interval division first identifies the stable regions in the relationship graph. The judgment basis includes the coefficient of variation of feature correlation strength, the stability index of feature values, and the consistency measure of feature distribution. The system sets the stability threshold by combining the associated features. When these indicators simultaneously meet the threshold requirements within a continuous time period, this time period is marked as a candidate interval. The system optimizes the boundaries of the candidate intervals, and the boundary point positions are determined by calculating the feature gradients. Each mapping interval generates a feature descriptor, which includes the time range, the feature set, the correlation map, and the stability score.

[0065] Based on the determined mapping interval, the system performs three-level feature enhancement processing based on associated features. Feature-level enhancement processes individual features, including median filtering to remove burst noise, trend term extraction to enhance long-term changes, and outlier detection and correction. Association-level enhancement optimizes feature relationships according to the association strength, including non-linear mapping of the association strength, selective enhancement of the association pattern, and smoothing of time-varying associations. Sequence-level enhancement optimizes the feature sequence globally, achieving time consistency optimization and spatial consistency enhancement of the features. After three-level feature enhancement, an enhanced feature sequence is finally obtained.

[0066] In some embodiments, enhancing the feature sequence according to the mapping interval to obtain an enhanced feature sequence includes: enhancing the sequence according to the mapping interval; performing paragraph division on the enhanced feature sequence to obtain signal segments; calculating the feature intensity values corresponding to the signal segments; marking key feature segments according to the feature intensity values; and generating the enhanced feature sequence according to the key feature segments.

[0067] When performing paragraph division on the feature sequence enhanced in the previous step, a multi-level division strategy is adopted. The system first initially segments the feature sequence with a basic length of 1 second, and maintains an overlap rate of 30% between adjacent segments to ensure feature continuity. Calculate feature distribution metrics for each basic segment, including time-domain feature distribution (mean, variance, skewness, kurtosis, waveform factor, impulse factor), frequency-domain feature distribution (energy distribution, frequency center, bandwidth, spectral peak distribution, spectral entropy), and association feature distribution (feature correlation, association strength, synchronization, coupling degree). The calculation of the distribution metrics uses a sliding window method with a window length of 200 ms and a step size of 50 ms to ensure capturing the dynamic changes of the features. The feature distribution within each window is obtained by the kernel density estimation method, and the kernel function selects the Gaussian kernel, and the bandwidth is determined by cross-validation. The system uses these distribution metrics to segment the feature sequence through an improved hierarchical clustering method, and the clustering feature vector includes all distribution metrics and their first-order differences. The clustering distance adopts a weighted combination of the Euclidean distance and the Mahalanobis distance of the feature distribution, with a weight ratio of 7:3. The system performs post-processing on the clustering results, merging overly short paragraphs (less than 100 ms) and splitting overly long paragraphs (greater than 3 seconds). The finally obtained signal segments include paragraph boundary timestamps, paragraph feature descriptions, paragraph type markers, and paragraph quality scores.

[0068] Based on the acquired signal segments, the system calculates the characteristic intensity index for each segment. According to the segmentation results and characteristic distributions, the system designs three types of characteristic intensity calculation schemes. The time-domain intensity index utilizes the time-domain characteristic distribution to calculate waveform amplitude intensity (peak-to-peak value, root mean square value, waveform skewness, waveform steepness), morphological intensity (curvature, sharpness, flatness, irregularity), and statistical intensity (variance contribution rate, entropy value, complexity, prediction error). The frequency-domain intensity index is based on the frequency-domain characteristic distribution, including band energy intensity (the proportion of energy in each frequency band and its time-varying characteristics), spectral peak intensity (main frequency amplitude, number of spectral peaks, spectral peak spacing, spectral peak morphology), and modulation intensity (sideband energy ratio, modulation depth, modulation frequency, modulation index). The correlation intensity index combines the correlation characteristic distribution to calculate the characteristic correlation intensity, synchronization intensity, and coupling intensity. The intensity calculation process adopts an adaptive window method with a basic window length of 500 ms, and the window length can be adjusted within the range of 300 - 700 ms according to the signal change rate. For each intensity index, the system calculates a set of statistical characteristics and conducts correlation analysis with the distribution index to ensure the accuracy of the intensity calculation.

[0069] Using the calculated characteristic intensity values, the system performs characteristic marking. In the marking process, all intensity indexes of each segment are first combined into a high-dimensional feature vector, which contains the calculation results of three types of indexes: time-domain intensity, frequency-domain intensity, and correlation intensity. The principal component analysis method with regularization constraints is used for dimensionality reduction to ensure that more than 90% of the information is retained after dimensionality reduction. In the feature space after dimensionality reduction, the improved DBSCAN density clustering algorithm is used to classify the segments. The system calculates the significance score of the pattern based on three dimensions: intensity level (using the absolute value of intensity and its stability), feature stability (based on the smoothness of the calculated feature changes), and pattern uniqueness (Mahalanobis distance from other patterns). The significance score is calculated by weighted summation, and the weight coefficients are optimized and determined through grid search. The system marks the segments with significance scores exceeding the threshold as key feature segments and further identifies their dominant features and secondary features.

[0070] Based on the feature marking results, the system generates the final enhanced feature sequence. The generation process adopts a hierarchical enhancement strategy, and different differential processing schemes are used for different types of feature segments. For key feature segments: the dominant features are selectively amplified through an adaptive non-linear function, and the gain coefficient is dynamically adjusted between 1.2 and 1.5 according to the calculated feature intensity; the secondary features are subjected to relative suppression processing, and the suppression coefficient ranges from 0.6 to 0.8, and the suppression intensity is inversely proportional to the significance of the dominant features; the feature associations are optimized and mapped through an S-shaped function to strengthen the significant association patterns and weaken the unstable associations. For non-key feature segments, basic feature retention processing is performed, and only necessary noise suppression and smoothing processing are carried out to ensure the overall continuity of the sequence. The system establishes a feature time-series association model, including short-range associations and long-range associations. During the enhancement process, a feature consistency check is also performed to ensure that the enhanced features maintain their physical meanings. The finally obtained enhanced feature sequence has a clearer feature expression, more prominent key patterns, and more stable time-series associations.

[0071] In step S104, the enhanced feature sequence is subjected to mapping transformation to obtain initial components, the main components are determined from the initial components, and feature representations are generated based on the main components; the feature representations are segmented to obtain feature intervals and the interval features corresponding to the feature intervals, and an intention feature sequence is generated based on the interval features.

[0072] Specifically, when performing mapping transformation on the enhanced feature sequence, the system adopts a multi-level transformation strategy. First, time alignment is performed on the enhanced feature sequence using a sliding window method with a window length of 500 ms and an overlap rate of 50%. Within each window, preprocessing of normalizing the feature data is performed, including amplitude normalization and variance normalization. Then, a non-linear mapping method is used for feature space transformation, and the kernel function is used to map the features into a high-dimensional space. The kernel function selects the RBF kernel, and the kernel parameters are determined through cross-validation. The time-series relationship of the features is considered during the mapping process, and a time weight factor is introduced, with a larger weight for recent features and a gradually decaying weight for distant features. After the mapping transformation, the system obtains the initial feature components, and each component contains feature expressions in three aspects: time domain information, frequency domain information, and association information. The system sorts the initial components according to the contribution degree and establishes a mapping parameter index table for each component to record its time attributes and corresponding weight coefficients.

[0073] In some embodiments, the performing mapping transformation on the enhanced feature sequence to obtain initial components includes: performing time alignment on the enhanced feature sequence based on a sliding window method; performing preprocessing of normalizing the aligned enhanced feature sequence; the normalization processing includes amplitude normalization and variance normalization; performing feature space transformation on the normalized enhanced feature sequence according to a non-linear mapping method to map the enhanced feature sequence into a high-dimensional space to obtain the initial components.

[0074] Based on the generated initial feature components, the system performs feature screening. In the screening process, the significance index is calculated first by using the mapping parameter index table and combining the time weights of each component. The significance evaluation includes variance contribution rate (based on component ranking), information gain (calculated using feature expressions), and stability score (weighted calculation based on time weights). The system combines these three indicators with weighted combination, and the weight coefficients are dynamically adjusted according to the mapping effect. Then an adaptive threshold is set, and the threshold value is determined according to the statistical characteristics of the score distribution, generally taking the 75th percentile. The components exceeding the threshold are retained, and the components below the threshold are filtered out. At the same time, the system analyzes the correlation between components. Using the established feature expressions, for component pairs with a correlation coefficient exceeding 0.8, the component with the higher score is retained, and the other component is deleted to avoid feature redundancy. The screening result contains the complete feature information of the retained components and the correlation analysis result.

[0075] Using the feature screening result, the system determines the main components. First, cluster analysis is performed on the set of screened components. The spectral clustering algorithm is adopted, and the similarity matrix is constructed based on the calculated component scores. The optimal number of clusters is determined through eigenvalue analysis, generally choosing the inflection point position of the eigenvalue descending curve. For each cluster, the system calculates the centrality indicators of the components within the cluster using the significance index, including degree centrality, closeness centrality, and betweenness centrality. The component with the highest centrality indicator is selected as the representative component of the cluster. Then, combining the correlation analysis result, the importance ranking of all representative components is carried out based on their projection area and coverage range in the original feature space. Finally, the components ranked in the top 60% in terms of importance are selected as the main components.

[0076] Based on the determined main components, the system generates a feature representation. First, the basic framework of the feature representation is constructed. The framework is designed with a tensor structure, including a time dimension (sequence of sampling points), a feature dimension (sequence of main components), and a correlation dimension (relationship between components). The original sampling frequency is maintained in the time dimension, and each time point corresponds to a complete set of feature vectors. In the feature dimension, the main components are organized according to the determined importance ranking, and orthogonality is maintained between the components. In the correlation dimension, a component correlation graph is constructed using the correlation analysis result. Then, the main components are combined and optimized, and an improved autoencoder structure is designed. The network parameters of the encoder and decoder are configured based on the component characteristics, and an L1 regularization term is introduced during the training process to promote the sparsity of the feature representation. The encoded features are reconstructed, and the reconstruction process adopts an iterative optimization strategy and uses the screening criteria for quality evaluation. The reconstructed features are smoothed in time series using the Kalman filter method, and the filter parameters are adaptively adjusted according to the dynamic characteristics of the signal. The finally generated feature representation can effectively capture the time-varying characteristics of the original features, maintain the correlation structure between features, and reduce data redundancy.

[0077] When segmenting the feature representation, the system adopts an adaptive segmentation strategy. First, the feature representation is initially divided according to a basic time length of 1 second, and a 40% overlap rate is maintained between adjacent segments to ensure the continuity of features and the complete capture of the change trend. Within each basic segment, statistical indicators of the features are calculated, including the mean vector (reflecting the central tendency of the features), the covariance matrix (describing the correlation between features), the autocorrelation coefficient (characterizing the temporal dependence of features), the entropy value (quantifying the uncertainty of features), and the complexity (measuring the degree of change of features). Based on these statistical indicators, an improved dynamic programming algorithm is used to determine the optimal segmentation points. The cost function of the algorithm is designed as: Cost = w1×Consistency + w2×Difference + w3×Length, where Consistency represents the consistency of features within a segment, Difference represents the difference of features between segments, Length is the segment length penalty term, and the weight coefficients [w1, w2, w3] are determined by grid search optimization. The algorithm is implemented in a sliding window manner, with a basic window length of 200 ms, which can be adaptively adjusted within the range of 100 - 300 ms according to the feature change rate. For each segment, the system records its time range (start time, end time, duration), feature statistics (mean, variance, kurtosis, skewness), and boundary features (boundary point feature values, boundary gradients). The system organizes the segmentation results into a segmented feature table, which contains the complete statistical indicators and boundary information of each segment.

[0078] Based on the segmented feature table, the system performs feature marking and extraction for each interval. In the marking process, an interval feature descriptor is first constructed using the calculated statistical indicators. The descriptor contains three types of information: statistical features, morphological features, and dynamic features. Statistical features inherit the calculated statistics such as mean, variance, skewness, and kurtosis; morphological features are based on boundary features, and morphological indicators such as the waveform factor (reflecting the complexity of the waveform), the impulse factor (characterizing the impact of the waveform), and the margin factor (describing the smoothness of the waveform) are calculated; dynamic features utilize temporal information, including dynamic parameters such as the change rate (the instantaneous change speed of features), the fluctuation range (the change amplitude of features), and the trend direction (the development trend of features). Feature extraction adopts a sliding window method with a window length of 100 ms and a step size of 20 ms to ensure the capture of detailed changes in features. The feature significance score is calculated for each interval. The score calculation integrates statistical indicators and adopts a weighted combination method: Score = α×Amplitude + β×Stability + γ×Uniqueness, where the weight coefficients [α, β, γ] are determined by cross-validation. The system analyzes the feature patterns of key intervals, identifies dominant features and secondary features through hierarchical clustering methods, and constructs a hierarchical relationship graph of features.

[0079] Continuing to utilize the segmentation results, the system conducts multi-scale interval feature extraction. During the extraction process, according to the time range information, the changing patterns of features are captured at different time scales. At the short time scale (200 ms), the system extracts local features, including instantaneous feature values (based on statistics), local changing trends (using boundary gradients), and feature jump points (identified through boundary features). The short-time feature extraction adopts the sliding window method, with a window length of 100 ms and a step size of 20 ms. The first-order difference and second-order difference of features are calculated for each window, reflecting the changing speed and acceleration of features. At the medium time scale (500 ms), the evolution patterns of features are analyzed by combining the autocorrelation coefficient, including the periodicity, stationarity, and correlation of the feature sequence. The system conducts time-frequency analysis on each feature sequence, using the wavelet transform method, selecting the Morlet wavelet as the basis function, and setting the decomposition scale to 5 layers to obtain the energy distribution of features in different frequency bands. At the long time scale (above 1 s), global features are extracted based on complexity metrics, including the long-term trend, abnormal patterns, and repetitive patterns of features. The system fuses the feature information at different scales, adopting a weighted combination strategy, and the weight coefficients are dynamically allocated according to the discriminative ability of the features at each scale.

[0080] Using the feature information extracted in the previous three paragraphs, the system generates an intention feature sequence. The sequence generation process first establishes a feature transformation model based on the segmentation structure. The model adopts a three-layer structure design: a feature mapping layer, a feature combination layer, and a feature integration layer. The feature mapping layer uses an improved kernel function method to map the original feature space to the intention feature space. The RBF kernel is selected as the kernel function, and the kernel parameters are optimized through feature significance scores. The mapping process considers the temporal dependence of features and introduces a calculated time decay factor to give greater weight to recent features. The feature combination layer uses the identified dominant and secondary features for non-linear combination. The combination methods include weighted summation, max pooling, and average pooling, and are adaptively selected according to the feature type. The weight coefficients are optimized by the least squares method, and the objective function includes a reconstruction error term and a regularization term. The feature integration layer arranges the combined features in chronological order and uses sequence alignment technology to process feature sequences of different lengths. The alignment reference point is determined based on the results of multi-scale feature analysis. The system smooths the generated sequence using an improved Kalman filter algorithm, and the filter parameters are dynamically adjusted according to the temporal features. At the same time, outlier detection and correction are performed. The determination of outliers is based on the feature significance threshold, and the correction uses the local interpolation method. The finally obtained intention feature sequence achieves temporal continuity and consistency of feature expression while maintaining the discriminability of features.

[0081] Step S105: Obtain the change nodes corresponding to the intention feature sequence and the corresponding temporal features, and generate an intention change sequence based on multiple temporal features.

[0082] In some embodiments, obtaining the change nodes corresponding to the intent feature sequence and the corresponding timing features, and generating an intent change sequence according to the plurality of timing features includes: segmenting and marking the intent feature sequence to obtain a plurality of the change nodes; obtaining the timing features corresponding to each of the change nodes; performing state merging according to each of the timing features to obtain a merging result corresponding to the state merging; and obtaining the intent change sequence corresponding to the merging result.

[0083] When segmenting and marking the intent feature sequence, the system adopts a dynamic segmentation strategy of multi-feature fusion. First, the intent feature sequence is initially divided according to a basic length of 500 ms, and a 30% overlapping interval is set between adjacent segments to ensure the complete capture of change nodes and the continuous expression of change trends. For each basic segment, the system calculates change metrics based on the feature sequence passed in step, including: 1) amplitude change rate, obtained by calculating the local derivative of the feature sequence, reflecting the change speed of feature intensity; 2) direction change rate, describing the trajectory change in the feature space using the angle change of the feature vector; 3) shape change rate, calculated based on the shape features (such as curvature, torsion) of the feature curve; 4) correlation change rate, characterized by the change speed of the cross-correlation function between features. The change rate calculation adopts an adaptive window method, and the window length is adjusted within the range of 100 - 300 ms according to the severity of feature changes, and the window overlap rate is set to 50%. The system combines the feature mapping results in step. When the feature changes are relatively severe, the window length is automatically shortened to improve the time resolution; when the feature changes are gentle, the window length is increased to improve the statistical reliability. For the calculated change rate sequence, the system uses a multi-threshold detection method to identify change nodes, and the threshold value is dynamically set based on the statistical characteristics of the sequence, generally taking 2 - 3 times the local standard deviation. Each detected change node records information such as its time position, change amplitude, change direction, and correlation strength, forming a change node feature vector.

[0084] Based on the changed nodes obtained from the markings, the system extracts temporal features. The extraction process is divided into two levels: the node level and the interval level. The node-level features describe the characteristics of individual changed nodes, making full use of the calculated change rate information, and focusing on the analysis of: 1) node strength, calculated by integrating the amplitude change rate and the direction change rate; 2) node polarity, judging the positive and negative directions of the change based on the morphological change characteristics; 3) node sharpness, measuring the degree of rapidity or slowness of the change using local curvature information; 4) node duration, determined according to the window analysis results. The system builds a feature description model for each node, recording the feature changes within the 50 ms range before and after the node, and the model parameters are optimized based on the change rate sequence. The interval-level features focus on the change patterns between adjacent nodes, and the analysis includes: 1) change trend, judged by the combination of various calculated change rates; 2) change rate, integrating the change information of multiple time scales; 3) change law, based on the periodic analysis results; 4) stability assessment, judged using the associated change rate. The system captures the feature evolution process within the interval through the sliding window method, with a window length of 200 ms and a step size of 50 ms, to achieve continuous monitoring of the feature changes.

[0085] Using the extracted temporal features, the system performs state merging processing. The merging process first constructs a state description table, and the content recorded in the table integrates the change indicators and temporal features: 1) basic features, including the statistics of the node-level features; 2) duration, based on the node duration analysis; 3) influence range, determined by the interval-level features. For each state, the system calculates two key indicators: 1) stability indicator, evaluated by combining the change rate sequence and the interval features; 2) influence indicator, analyzed based on the node strength and the interval change trend. The state merging adopts an adaptive method, and the merging conditions are determined according to the similarity of the state features and the temporal constraints. During the merging process, the system focuses on: 1) the time continuity of the state, ensured by the interval-level features; 2) the feature consistency of the state, verified by the multi-dimensional change rate; 3) the transition characteristics of the state, analyzed based on the node features. The system sets a dynamic merging threshold, and the threshold value comprehensively considers the change significance and the feature stability, varying within the range of 0.5 - 1.5 times the state spacing.

[0086] Based on the result of state merging, the system generates an intention change sequence. The generation process first establishes a state transition framework, which integrates the analysis results of the first three stages: 1) State description, including change characteristics, timing characteristics, and merging characteristics; 2) State parameters, combining node characteristics and state indicators; 3) State evolution rules, based on merging conditions and state transition constraints. The system generates state descriptors for each time point, recording in detail: 1) The characteristic components of the current state, from multi-dimensional change rates; 2) Statistical indicators, calculated based on timing characteristics; 3) Timing characteristics, inheriting interval characteristics; 4) Association characteristics, using state relationship analysis. The sequence generation process pays particular attention to the persistence requirements of states, ensuring smooth transitions between adjacent states by setting a minimum duration (based on node duration statistics) and a state inertia coefficient (using stability indicators). The system performs three-layer optimization on the generated sequence: 1) Local optimization, using change rate information to eliminate sudden noises; 2) Regional optimization, enhancing the main change trend based on interval characteristics; 3) Global optimization, using state relationships to repair possible state breakpoints. The finally generated intention change sequence can accurately depict the dynamic change process of the user's cognitive state, achieving precise expression and tracking of the intention process.

[0087] Step S106, obtain the preliminary label corresponding to the intention change sequence, and obtain the intention recognition result corresponding to the preliminary label.

[0088] When performing feature matching on the intention change sequence of the step, the system adopts a hierarchical matching strategy. First, the intention change sequence is segmented according to a basic length of 1 second, with a 25% overlapping interval between adjacent segments to ensure the continuity of the matching process. For each segment, matching features are calculated based on the state descriptor of the step, including: 1) State features, inheriting the basic feature attributes in the state description of the step, including state value, state duration, state stability, state distribution characteristics, and state boundary characteristics; 2) Change features, using the state transition information of the step, including change amplitude, change rate, change direction, change acceleration, and change periodicity; 3) Association features, based on the state association analysis of the step, including state transition probability, state similarity, state dependence, state coupling degree, and state transitivity. Feature matching adopts a multi-level comparison method. First, a rough match is performed based on the main state features transmitted by the step to quickly screen candidate items; then, the state transition information is used for a fine match to compare in detail the secondary features and change features of the state. During the matching process, the system sets weight coefficients for different features. The weight of the main features is in the range of 0.4 - 0.6, the weight of the secondary features is in the range of 0.2 - 0.4, and the weight of the auxiliary features is in the range of 0.1 - 0.2. The preliminary labeling result is obtained from the matching process, and each label contains a matching degree score, a list of matching features, and a reliability assessment. The system performs a preliminary screening on the matching results, removing the labels with a matching degree lower than the threshold, and the threshold value is dynamically set according to the statistical characteristics of the feature distribution.

[0089] In some embodiments, obtaining the intention recognition result corresponding to the preliminary tag includes: enhancing the features of the preliminary tag to obtain enhanced features; classifying and screening the enhanced features; and generating the intention recognition result according to the screening result corresponding to the classification and screening.

[0090] Based on the generated preliminary tagging result, the system performs feature enhancement processing. In the enhancement process, reliability analysis is first performed on each tag, and the confidence index of the tag is calculated based on the matching result, including: 1) Feature matching degree, inheriting the matching degree score; 2) Temporal consistency, evaluated based on state transition features; 3) Context relevance, calculated using correlation features. The feature matching degree is obtained by calculating the cosine similarity of the feature vectors, the temporal consistency is determined based on the autocorrelation analysis of the sequence, and the context relevance is obtained through the cross-correlation calculation of the sliding window. For tags with low confidence (confidence < 0.6), the system activates the feature supplement mechanism, and derives supplementary features by analyzing the feature distribution of adjacent tags. Feature supplementation considers the temporal continuity constraint, combines the state transition rule, and uses interpolation and smoothing methods to ensure the smooth transition between the supplementary features and the original features. For tags with high confidence (confidence > 0.8), the system performs feature strengthening, differentiates and enhances according to the importance of the matching features, highlights its dominant features, and suppresses the secondary features. The enhancement process adopts an adaptive weight strategy, and the weight coefficient is dynamically adjusted according to the confidence of the tag. At the same time, the system conducts a consistency check on the enhanced features to ensure consistency with the matching standard.

[0091] Using the enhanced features, the system performs classification and screening. The screening process is designed with a three-layer screening mechanism: 1) Feature layer screening, based on the discriminative ability of enhanced single features, calculates the discrimination index of features, including eigenvalue distribution, feature stability, and feature uniqueness; 2) Pattern layer screening, focuses on the recognition effect of the generated feature combinations, and analyzes the integrity, typicality, and representativeness of the feature patterns; 3) Sequence layer screening, evaluates the temporal coherence of the feature sequences, considering the smoothness, periodicity, and trend of the sequences. The system calculates the information gain and feature importance scores for each feature, and performs feature screening in combination with confidence evaluation. Feature layer screening is based on the discriminative ability of single features, calculates the discrimination index of features, including eigenvalue distribution (mean, variance, skewness, kurtosis), feature stability (short-term stability, long-term stability), and feature uniqueness (degree of distinction from other features). The information gain and feature importance scores are calculated for each feature, and the features with scores lower than the threshold are filtered. Pattern layer screening focuses on the recognition effect of feature combinations, and analyzes the integrity (completeness of pattern components), typicality (degree of matching with the standard pattern), and representativeness (ability to express the current intention) of the feature patterns. The system performs clustering analysis on the feature patterns to identify the main pattern categories and variant pattern categories. Sequence layer screening evaluates the temporal coherence of the feature sequences, considering the smoothness (degree of change between adjacent points), periodicity (regularity of repeating patterns), and trend (long-term change direction) of the sequences.

[0092] For the screening results, the system outputs the final intention recognition results. In the result generation process, first, the high-confidence markers are feature-aggregated, and the markers with similar time (time interval < 200ms) and similar features (similarity > 0.8) are combined into intention units. Each intention unit inherits the enhanced features, including: 1) Core feature set, based on the key features determined by screening; 2) Auxiliary feature set, including the retained secondary features; 3) Time attribute set, integrating the timing information. The core features include three categories: state features, change features, and association features, and the feature weights are determined based on the screening scores. The auxiliary features provide the context information of the intention, including environmental features, historical features, and prediction features. The time attribute describes the duration characteristics of the intention, and combines the timing consistency analysis to record the start time, duration, and change moment of the intention. The system performs temporal organization on the intention units, and establishes the association relationships between intention units according to the sequence screening results, including causal association (front-back causality), parallel association (occurring simultaneously), and hierarchical association (inclusion relationship). By analyzing the temporal patterns of the intention units, the system can accurately capture the evolution process and transformation rules of the intention. The finally output intention recognition result sequence not only includes the intention judgment at each time point, but also includes the complete feature support information and reliability evaluation.

[0093] To implement the method for real-time decoding of EEG signals and intention recognition of the attention glasses corresponding to the above method embodiments, so as to achieve the corresponding functions and technical effects. Refer to Figure 2 , Figure 2 FIG. Figure 2 shows a structural block diagram of a device 200 for real-time decoding of EEG signals and intention recognition provided by an embodiment of the present application. For ease of description, only the parts related to this embodiment are shown. The device 200 for real-time decoding of EEG signals and intention recognition provided by the embodiment of the present application includes:

[0094] A signal acquisition module 201, configured to acquire multi-channel EEG time-domain signals collected by the attention glasses, perform multi-point sampling on the multi-channel EEG time-domain signals to obtain original sampling data, and generate an initial signal sequence according to the original sampling data;

[0095] A compensation and adjustment module 202, configured to acquire a signal segment to be corrected in the initial signal sequence, perform compensation and adjustment on the signal segment to generate a standardized signal sequence; acquire time-domain features corresponding to the standardized signal sequence, perform frequency-domain conversion on the time-domain features to obtain frequency-domain components and energy analysis results corresponding to the frequency-domain components, so as to construct a dual-domain feature sequence according to the energy analysis results;

[0096] An association mapping module 203, configured to perform association mapping on the dual-domain feature sequence to obtain feature breakpoints, establish a time-frequency relationship graph according to the feature breakpoints, determine a mapping interval according to the time-frequency relationship graph, so as to enhance the feature sequence according to the mapping interval to obtain an enhanced feature sequence;

[0097] A mapping conversion module 204, configured to perform mapping conversion on the enhanced feature sequence to obtain initial components, determine main components in the initial components, and generate a feature representation according to the main components; perform segmented slicing on the feature representation to obtain a feature interval and interval features corresponding to the feature interval, and generate an intention feature sequence according to the interval features;

[0098] A feature acquisition module 205, configured to acquire change nodes corresponding to the intention feature sequence and corresponding timing features, and generate an intention change sequence according to multiple timing features;

[0099] A result acquisition module 206, configured to acquire a preliminary label corresponding to the intention change sequence, and acquire an intention recognition result corresponding to the preliminary label.

[0100] The above device 200 for real-time decoding of EEG signals and intention recognition can implement the method for real-time decoding of EEG signals and intention recognition of the attention glasses in the above method embodiments. The optional items in the above method embodiments are also applicable to this embodiment, which will not be elaborated here. The remaining content of the embodiment of the present application can refer to the content of the above method embodiment, and will not be repeated in this embodiment.

[0101] Figure 3 This is a schematic structural diagram of a computer device provided by an embodiment of the present application. As Figure 3 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. When the processor 30 executes the computer program 32, the steps in any of the above method embodiments are implemented.

[0102] The computer device 3 may be a computing device such as a smart phone, a tablet computer, a desktop computer, and a cloud server. The computer device may include but is not limited to the processor 30 and the memory 31. Those skilled in the art can understand that Figure 3 this is only an example of the computer device 3 and does not constitute a limitation on the computer device 3. It may include more or fewer components than shown in the figure, or combine certain components, or different components. For example, it may also include input / output devices, network access devices, etc.

[0103] The so-called processor 30 may be a central processing unit (CPU). The processor 30 may also be other general-purpose processors, digital signal processors (DSPs), application specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0104] In some embodiments, the memory 31 may be an internal storage unit of the computer device 3, such as the 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 the internal storage unit and the external storage device of the computer device 3. The memory 31 is used to store an operating system, application programs, a 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.

[0105] In addition, an embodiment of the present application also provides a computer-readable storage medium storing a computer program, and when the computer program is executed by a processor, the steps in any of the above method embodiments are implemented.

[0106] An embodiment of the present application provides a computer program product, and when the computer program product runs on a computer device, the computer device is caused to execute the steps in each of the above method embodiments.

[0107] In several embodiments provided by the present application, it can be understood that each block in the flowchart or block diagram may represent a module, a program segment, or a part of code, and the module, the program segment, or the part of code includes one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than marked in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in the reverse order, depending on the functions involved.

[0108] If the above-mentioned 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 this application, in essence, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device to execute all or part of the steps of the methods described in various embodiments of this application. The aforementioned storage medium includes: various media that can store program codes, such as USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, or optical discs.

[0109] The specific embodiments described above further elaborate on the purpose, technical solution, and beneficial effects of this application. It should be understood that the above description is only the specific embodiments of this application and is not used to limit the protection scope of this application. It is particularly pointed out that for those skilled in the art, any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of this application shall be included in the protection scope of this application.

Claims

1. A method for real-time decoding of electroencephalogram signals and intention recognition of an attention glasses, characterized in that, Including: Obtain the multi-channel electroencephalogram time-domain signals collected by the attention-getting glasses, perform multi-point sampling on the multi-channel electroencephalogram time-domain signals, obtain the original sampling data, and generate an initial signal sequence according to the original sampling data; Obtain the signal segments that need to be corrected in the initial signal sequence, perform compensation adjustment on the signal segments to generate a standardized signal sequence; obtain the time-domain features corresponding to the standardized signal sequence, perform frequency-domain conversion on the time-domain features to obtain the frequency-domain components and the energy analysis results corresponding to the frequency-domain components, so as to construct a dual-domain feature sequence according to the energy analysis results; Perform correlation mapping on the dual-domain feature sequence to obtain feature breakpoints, establish a time-frequency relationship graph according to the feature breakpoints, determine the mapping interval according to the time-frequency relationship graph, so as to enhance the dual-domain feature sequence according to the mapping interval to obtain an enhanced feature sequence; Perform mapping conversion on the enhanced feature sequence to obtain initial components, determine the main components in the initial components, and generate a feature representation according to the main components, including: performing feature screening based on the generated initial components; using the feature screening results to determine the main components, performing cluster analysis on the filtered component set, determining the optimal number of clusters through eigenvalue analysis, and for each cluster, calculating the centrality index of the components within the cluster using a significance index; selecting the component with the highest centrality index as the representative component; combining the correlation analysis results, performing importance ranking on the representative components, and selecting the components with the top 60% importance ranking as the main components; segmenting the feature representation to obtain feature intervals and the interval features corresponding to the feature intervals, and generating an intention feature sequence according to the interval features; Obtain the change nodes corresponding to the intention feature sequence and the corresponding timing features, and generate an intention change sequence according to the multiple timing features; Obtain the preliminary marks corresponding to the intention change sequence, and obtain the intention recognition results corresponding to the preliminary marks.

2. The method according to claim 1, wherein The generating the initial signal sequence according to the original sampling data includes: Obtain the sampling point distribution corresponding to the multi-point sampling; Mark the valid sampling segments in the original sampling data according to the sampling point distribution; Generate the initial signal sequence according to the valid sampling segments.

3. The method according to claim 1, characterized in that, The obtaining the signal segments that need to be corrected in the initial signal sequence, performing compensation adjustment on the signal segments to generate a standardized signal sequence includes: Perform baseline drift detection on the initial signal sequence to obtain the signal segments that need to be corrected; Perform the compensation adjustment on the signal segments to be corrected; Generate a standardized signal sequence according to the adjustment results corresponding to the compensation adjustment.

4. The method according to claim 1, wherein The enhancing the dual-domain feature sequence according to the mapping interval to obtain an enhanced feature sequence includes: Enhance the dual-domain feature sequence according to the mapping interval; Perform paragraph division on the enhanced dual-domain feature sequence to obtain signal segments; Calculate the feature intensity values corresponding to the signal segments; Mark the key feature segments according to the feature intensity values; Generate the enhanced feature sequence according to the key feature segments.

5. The method according to claim 1, wherein Obtaining the change nodes corresponding to the intention feature sequence and the corresponding timing features, and generating an intention change sequence according to the multiple timing features, includes: Performing segment marking on the intention feature sequence to obtain multiple change nodes; Obtaining the timing features corresponding to each change node; Performing state merging according to each timing feature to obtain the merging result corresponding to the state merging; Obtaining the intention change sequence corresponding to the merging result.

6. The method according to claim 1, wherein Obtaining the intention recognition result corresponding to the preliminary marking, includes: Performing feature enhancement on the preliminary marking to obtain enhanced features; Performing classification and screening on the enhanced features; Generating the intention recognition result according to the screening result corresponding to the classification and screening.

7. The method according to claim 1, wherein Obtaining the time domain features corresponding to the standardized signal sequence, performing frequency domain conversion on the time domain features, obtaining the frequency domain components and the energy analysis results corresponding to the frequency domain components, so as to construct a dual-domain feature sequence according to the energy analysis results, includes: Segmenting the standardized signal sequence to obtain multiple signal segments; Obtaining the time domain features corresponding to the signal segments; Performing frequency domain conversion on the time domain features to obtain the frequency domain components; Obtaining the energy analysis results corresponding to the frequency domain components; Constructing the dual-domain feature sequence according to the time domain features, frequency domain components and energy analysis results.

8. The method according to claim 7, wherein Segmenting the standardized signal sequence to obtain multiple signal segments, includes: Segmenting the standardized signal sequence according to a preset initial segment length to obtain multiple initial segments; the overlapping rate between adjacent initial segments is within a preset overlapping rate range; Iteratively optimizing the initial segment length according to a preset segment range, and segmenting the standardized signal sequence according to the optimized initial segment length to obtain multiple signal segments.

9. The method according to claim 1, wherein Performing mapping conversion on the enhanced feature sequence to obtain initial components, includes: Performing time alignment on the enhanced feature sequence based on a sliding window method; Performing normalization preprocessing on the aligned enhanced feature sequence; the normalization preprocessing includes amplitude normalization and variance normalization; Performing feature space conversion on the normalized enhanced feature sequence according to a non-linear mapping method, mapping the enhanced feature sequence to a high-dimensional space to obtain the initial components.

10. A computer device, characterized in that, Including 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 9 when executing the computer program.

Images