Vehicle-mounted electroencephalogram driver emotion safety monitoring method and system based on meta-learning fast adaptation

Abstract

The invention provides a vehicle-mounted electroencephalogram driver emotion safety monitoring method and system based on meta-learning fast adaptation. The method comprises the steps that multi-channel electroencephalogram data of a target vehicle-mounted driver are collected through a vehicle-mounted electroencephalogram collection system; data preprocessing is conducted on the multi-channel electroencephalogram data, processing is conducted according to a segmentation and sub-segmentation strategy, unified input is formed, and the length and the overlapping degree of segments and sub-segments are configurable parameters; performing adaptive mapping and hierarchical multi-branch fusion on the unified input to obtain fusion representation; carrying out few-sample rapid individualized adaptation and online adaptive calibration processing on the fusion representation to obtain calibration features; and the calibration features are input into a lightweight Transform for time sequence modeling, after pooling, classification identification is completed, the early warning level is judged, an intervention instruction is output according to a preset linkage rule, subsystems and systems in the vehicle are driven to run online, and events are recorded. The method provided by the invention meets the requirements of low time delay, high robustness and long-term stable use of a vehicle-mounted scene.

Description

Technical Field

[0001] This invention relates to the field of intelligent connected vehicle technology, and in particular to an in-vehicle EEG driver emotion safety monitoring method and system based on meta-learning rapid adaptation. Background Technology

[0002] With the development of intelligent connected vehicles and advanced driver assistance systems, human factors related to emotions in driving safety are becoming increasingly prominent. Existing in-vehicle EEG emotion monitoring faces three main engineering bottlenecks in practical applications: First, significant differences exist across subjects, and traditional solutions rely on long-term individual calibration, lacking rapid individualization capabilities under limited sample conditions; second, online operation is susceptible to distribution drift and electrode state changes, lacking stable real-time calibration and adaptive mechanisms, making long-term stability difficult; third, automotive-grade environments have stringent requirements for low latency, small models, and stability, necessitating real-time inference on the device side and forming a closed-loop control system of "recognition-warning-linkage-feedback" with multi-level in-vehicle warning strategies. Therefore, there is an urgent need for an integrated EEG emotion safety monitoring method for in-vehicle scenarios that simultaneously possesses rapid individualization with limited samples, stable online calibration, and automotive-grade real-time deployment capabilities. Summary of the Invention

[0003] In view of this, the present invention provides a method and system for monitoring the driver's emotions and safety based on meta-learning and rapid adaptation using in-vehicle EEG, in order to solve the above problems.

[0004] This invention provides a vehicle-mounted EEG driver emotion safety monitoring method based on meta-learning and rapid adaptation, comprising: acquiring multi-channel EEG data of a target vehicle driver through an in-vehicle EEG acquisition system; preprocessing the multi-channel EEG data according to a segmentation and sub-segmentation strategy to form a unified input, wherein the length and overlap of segments and sub-segments are configurable parameters; performing adaptive graphing and hierarchical multi-branch fusion on the unified input to obtain a fused representation; performing rapid individualized adaptation and online adaptive calibration on the fused representation to obtain calibration features; inputting the calibration features into a lightweight Transformer for temporal modeling, performing pooling to complete classification and recognition and determine the warning level, outputting intervention commands according to preset linkage rules to drive the in-vehicle subsystem, and the system running online and recording events.

[0005] In another implementation of the present invention, the unified input is:

[0006] in, For batch, For timing window number, For the number of channels, The rPSD feature dimensions are defined for seven predefined frequency bands.

[0007] In another implementation of the present invention, the fusion is represented as:

[0008]

[0009] in, To learn the contribution of each branch in the current sample.

[0010] In another implementation of the present invention, the rapid individualized adaptation with few samples includes: constructing individual emotion prototypes based on a small number of support samples and fusion them with global prototypes in a weighted manner to generate an individualized prototype set; generating feature modulation vectors based on the individualized prototype set, rapidly modulating query features, and adaptively weighting them according to their distance from each prototype to form prototype guidance enhancement; and using sliding statistical updates and adaptive weight modulation in online adaptive calibration to suppress distribution drift.

[0011] In another implementation of the present invention, the individualized prototype is:

[0012] in, According to the emotion category; This is the modulation vector.

[0013] In another implementation of the present invention, the enhanced query feature is:

[0014] in, This is a coefficient used to control the guiding intensity; Indicates the query features Perform amplitude modulation. For weights.

[0015] In another implementation of the present invention, the warning level is multi-level, and the linkage trigger adopts a preset threshold and hysteresis strategy; the intervention command supports event recording, secondary confirmation and cancellation.

[0016] Another aspect of the present invention provides a vehicle-mounted EEG driver emotion safety monitoring system based on meta-learning rapid adaptation, comprising: a data acquisition module: acquiring multi-channel EEG data of a target vehicle driver through a vehicle-mounted EEG acquisition system; a data processing module: preprocessing the multi-channel EEG data, processing it according to a segmentation and sub-segmentation strategy to form a unified input, wherein the length and overlap of the segments and sub-segments are configurable parameters; performing adaptive graphing and hierarchical multi-branch fusion on the unified input to obtain a fusion representation; performing few-sample rapid individualized adaptation and online adaptive calibration on the fusion representation to obtain calibration features; and a result output module: inputting the calibration features into a lightweight Transformer for temporal modeling, performing pooling to complete classification and recognition and determine the warning level, outputting intervention commands according to preset linkage rules to drive the in-vehicle subsystem, and the system running online and recording events.

[0017] In another aspect, the present invention provides an electronic device comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the computer program, implements the steps of a vehicle-mounted EEG driver emotion safety monitoring method based on meta-learning rapid adaptation as described in any of the preceding claims. In another aspect, the present invention provides a computer storage medium storing a computer program, which, when executed by a processor, implements the steps of a vehicle-mounted EEG driver emotion safety monitoring method based on meta-learning rapid adaptation as described in any of the preceding claims.

[0018] This invention presents a vehicle-mounted EEG driver emotion safety monitoring method based on meta-learning rapid adaptation. It introduces meta-learning rapid adaptation and prototype-guided enhancement to achieve rapid individualization of new drivers with a limited number of support samples, eliminating the need for lengthy calibration and improving deployment efficiency and generalization ability. It employs an integrated model (MAGT) combining adaptive mapping, hierarchical multi-branch fusion, and lightweight Transformer collaboration, ensuring recognition accuracy while achieving low latency and small model size at the edge, meeting automotive-grade real-time requirements and deployment constraints. A unified input is constructed based on segmentation / sub-segmentation and Welch seven-band rPSD. Adaptive mapping and online calibration jointly suppress the effects of electrode contact changes and scene noise, maintaining stable recognition in complex in-vehicle environments. The communication interface for vehicle cockpit / domain control is easy to integrate, supporting localized processing and minimal retention, log reporting, and OTA policy deployment, facilitating long-term operation and maintenance and policy optimization, and complying with privacy and compliance requirements. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. By reading the detailed description of the embodiments below, the advantages and benefits of the solutions will become clear to those skilled in the art. The accompanying drawings are only for illustrating preferred embodiments and are not intended to limit the present invention. In the accompanying drawings: Figure 1 This is a schematic flowchart of a vehicle-mounted EEG driver emotion safety monitoring method based on meta-learning rapid adaptation, according to an embodiment of the present invention.

[0020] Figure 2 This is a control block diagram of an in-vehicle EEG driver emotion safety monitoring system based on meta-learning rapid adaptation, according to an embodiment of the present invention. Detailed Implementation

[0021] To enable those skilled in the art to better understand the technical solutions in the embodiments of the present invention, the technical solutions in the embodiments of the present invention will be clearly and thoroughly described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art should fall within the protection scope of the present invention.

[0022] Figure 1 This is a flowchart illustrating a method for monitoring driver emotions and safety based on meta-learning and rapid adaptation using in-vehicle EEG, as provided in an embodiment of the present invention. Figure 1 As shown, this embodiment mainly includes: S101. Collect multi-channel EEG data of the target vehicle driver through the vehicle-mounted EEG acquisition system.

[0023] For example, a multi-channel device conforming to an approximate distribution of 10–20 can be selected (the number of channels C can be 16–128, preferably about 62), and the sampling rate can be 128–512 Hz (preferably about 200 Hz); dry / wet electrodes are allowed, and basic contact optimization and channel health detection are performed before acquisition (impedance control ≤10 kΩ, preferably ≤5 kΩ); tags are acquired and recorded in scenarios such as stationary / driving / nighttime / congestion.

[0024] S102. Perform data preprocessing on the multi-channel EEG data, and process it according to the segmentation and sub-segmentation strategy to form a unified input, wherein the length and overlap of the segments and sub-segments are configurable parameters.

[0025] For example, a sliding window is used to divide the data according to preset segment length, overlap, sub-segment length, and sub-segment overlap. The Welch method is used to estimate the spectrum of each sub-segment, and the relative power spectral density (rPSD) is calculated and normalized in seven predefined frequency bands [1,3], [4,8], [8,12], [12,16], [16,20], [20,28], and [30,45] Hz to form a 7-dimensional feature per channel. The data is organized as (batch B, time window S, channel C, feature F=7) and provided as a unified input for subsequent integrated modeling.

[0026] S103. Adaptive graph construction and hierarchical multi-branch fusion are performed on the unified input to obtain a fused representation.

[0027] For example, adaptive graphing (AGE1 / AGE2) and hierarchical multi-branch fusion (HFF) are performed sequentially in the model (MAGT) to obtain a fused representation. The adaptive graphing is used to characterize the topological relationships between channels, and the hierarchical multi-branch fusion is used to perform attention-weighted convergence of multi-path features and hierarchical enhancement to obtain a robust multi-scale fused representation.

[0028] Channel topology representation is performed based on learnable or conditionally adaptive adjacency constraints to capture stable and discriminative channel relationships. Subsequently, attention weighting and hierarchical enhancement are applied to multi-path features in HFF to output a robust multi-scale fusion representation, which serves as input for subsequent individualization and calibration.

[0029] S104. Perform few-sample fast individualized adaptation and online adaptive calibration on the fused representation to obtain calibration features.

[0030] For example, when a new driver or a change in scenario occurs, a small number of labeled support samples (e.g., 3–5 short windows per class) are collected to construct an individual emotion prototype and generate modulation vectors for rapid individualized enhancement. In the online phase, a momentum strategy is used to update the global prototype in small steps to maintain long-term adaptability. At the same time, normalization based on sliding statistics and adaptive weight modulation suppress the distribution drift caused by changes in cross-session, cross-device, and electrode states, and output stable calibration features.

[0031] S105. Input the calibration features into the lightweight Transformer (Lite-Trans) for time-series modeling, perform classification and recognition after pooling, determine the warning level, output intervention commands according to preset linkage rules, drive in-vehicle voice, lighting, seat, air conditioning and other subsystems, the system runs online and records events.

[0032] For example, multiple warning levels (such as L1 / L2 / L3) are output, and threshold and hysteresis strategies are used to reduce frequent switching; the in-vehicle voice, lighting, seat, air conditioning and other subsystems are linked according to rules to implement intervention, and the intervention events and recognition results are recorded with timestamps; when the recognition is uncertain or an anomaly is detected, it automatically switches to a conservative strategy (degenerate operation) and triggers re-adaptation or strategy adjustment in the background to restore performance.

[0033] It communicates with the vehicle cockpit / domain controller, outputs intervention commands and records events according to preset linkage rules, performs operation logs and policy management (including OTA) and privacy compliance maintenance, forming a continuous closed loop of "collection - processing - identification - linkage - feedback".

[0034] Secure communication is achieved via CAN / Ethernet / IPC, with authentication / signature and prevention of accidental touches and timeout retries; localized processing and minimal retention are implemented to meet privacy compliance; OTA canary deployment and rollback of policies and models are supported, and long-term operation is subject to performance auditing, anomaly tracking, and continuous optimization.

[0035] This invention presents a vehicle-mounted EEG driver emotion safety monitoring method based on meta-learning rapid adaptation. It introduces meta-learning rapid adaptation and prototype-guided enhancement to achieve rapid individualization of new drivers with a limited number of support samples, eliminating the need for lengthy calibration and improving deployment efficiency and generalization ability. It employs an integrated model (MAGT) combining adaptive mapping, hierarchical multi-branch fusion, and lightweight Transformer collaboration, ensuring recognition accuracy while achieving low latency and small model size at the edge, meeting automotive-grade real-time requirements and deployment constraints. A unified input is constructed based on segmentation / sub-segmentation and Welch seven-band rPSD. Adaptive mapping and online calibration jointly suppress the effects of electrode contact changes and scene noise, maintaining stable recognition in complex in-vehicle environments. The communication interface for vehicle cockpit / domain control is easy to integrate, supporting localized processing and minimal retention, log reporting, and OTA policy deployment, facilitating long-term operation and maintenance and policy optimization, and complying with privacy and compliance requirements.

[0036] In another implementation of the present invention, the unified input is:

[0037] in, For batch, For timing window number, For the number of channels, The rPSD feature dimensions are defined for seven predefined frequency bands; the time series and batch data are merged into... This facilitates unified graph modeling on the channel dimension.

[0038] For example, after collecting multi-channel EEG data from the target vehicle driver, the data is organized into a four-dimensional tensor according to preset segments and sub-segments. The relative power spectral density (rPSD) is extracted from seven predefined frequency bands using the Welch method to form a unified input for subsequent integrated processing. To facilitate channel topology calculation, the temporal and batch data are temporarily merged to obtain a three-dimensional representation, which is used for graph modeling and fusion in the channel dimension.

[0039] In another implementation of the present invention, the fusion is represented as:

[0040]

[0041] in, To learn the contribution of each branch in the current sample.

[0042] For example, the original projection and the two adaptive representations are stitched together after a lightweight transformation. Learning the contribution of each branch under the current sample, softmax normalizes the result on the branch dimension. Then, the branch features are weighted and summed to obtain the fused representation. .

[0043] Specifically, adaptive graph encoding (AGE1 / AGE2) is first performed to obtain a robust channel topology representation. This step uses a learnable (or conditionally adaptive) adjacency matrix to perform neighborhood aggregation in the channel dimension and introduces "relative difference" to enhance the characterization of local changes. Then, the basic channel mapping and the relative difference path are adaptively fused according to the channel importance, thereby reducing the impact of electrode contact changes and in-vehicle noise on the representation.

[0044]

[0045] in, For learnable or conditionally adaptive adjacencies, Ensure non-negative edges are connected. To make the graph symmetrical, we need to increase the self-loop strength (increase the proportion of its own information). Improve overall topology robustness.

[0046]

[0047] This indicates the weighted convergence of channels within their neighborhood. This represents the change in the neighborhood response relative to the original value. Using relative differential energy can mitigate baseline drift caused by absolute amplitude shift and changes in electrode contact.

[0048] The first item in the left parentheses is "Basic Channel Mapping", and the second item is "Relative Differential Enhancement". Control the ratio of the two paths; right side Generate channel-by-channel weights and perform adaptive modulation according to channel importance.

[0049] Subsequently, hierarchical multi-branch fusion (HFF) is performed to improve multi-source robustness and separability. In this step, the original projection branch and the two adaptive mapping outputs are spliced ​​together after undergoing a lightweight transformation within the scale. The weights of each branch under the current sample are learned using an attention approach. Furthermore, the hierarchical enhancement structure is used to compress redundancy and highlight discriminative features. The output serves as a unified representation for subsequent temporal modeling and individualized adaptation.

[0050] In another implementation of the present invention, the rapid individualized adaptation with few samples includes: constructing individual emotion prototypes based on a small number of support samples and fusion them with global prototypes in a weighted manner to generate an individualized prototype set; generating feature modulation vectors based on the individualized prototype set, rapidly modulating query features, and adaptively weighting them according to their distance from each prototype to form prototype guidance enhancement; and using sliding statistical updates and adaptive weight modulation in online adaptive calibration to suppress distribution drift.

[0051] For example, Meta-Learning Fast Adaptation (MLA) is performed when there are a few supporting samples to achieve individualization of new drivers under limited sample conditions. This step first calculates the intra-class individual prototype from the supporting samples and then weights and fuses it with the global prototype to obtain the prototype set of the current individual; then, a lightweight modulation vector is generated using the aggregated features of the supporting samples to quickly modulate the query (real-time) features, and prototype-guided enhancement is formed by adaptively weighting the features according to their distance from each prototype; finally, a momentum strategy is used to make small-step updates to the global prototype to maintain long-term stability.

[0052] In another implementation of the present invention, the individualized prototype is:

[0053] in, According to the emotion category; This is the modulation vector.

[0054] For example, for each emotion category use By merging the global prototype with the individual prototypes of the supporting samples of this type, an individualized prototype is obtained. The modulation vector is generated using the mean features of the support samples. It is used to perform lightweight, continuously differentiable scaling on query features.

[0055] In another implementation of the present invention, the enhanced query feature is:

[0056] in, This is a coefficient used to control the guiding intensity; Indicates the query features Perform amplitude modulation. For weights.

[0057] For example, first define the query features Amplitude modulation is obtained ( (Control modulation intensity):

[0058] As a learnable distance metric, the smaller the distance, the greater the weight. The larger the value, the more weighted prototypes are used. Perform residual guidance (coefficient) (Controlling the guidance intensity) to obtain individualized enhancement .

[0059] Next, online adaptive calibration (OAC) is performed on the individualized enhanced features to suppress distribution drift caused by cross-session / device states. The system standardizes the features using sliding statistics and further modulates the output with adaptive weights, ensuring that the features entering the time-series model remain stable under different driving environments and electrode states.

[0060]

[0061] Using momentum By performing an exponential moving average, new data is gradually integrated into the global prototype to ensure steady-state stability and convergence during long-term use.

[0062]

[0063] To control the response speed of statistical updates, first perform standardization to suppress distribution drift, then use... The feature-wise weights are generated and re-modulated to output a stable time-series modeling input.

[0064] Finally, lightweight temporal modeling (Lite-Trans) and classification output are performed. The calibrated features are restored to their temporal form, and lightweight strategies such as low-head attention and depthwise separable one-dimensional convolution are used to complete global temporal dependency modeling on the edge. After a few layers of stacking, temporal aggregation is performed and fed into a lightweight classification head to output the emotional state and warning level. When the confidence is insufficient or abnormal, a conservative degradation strategy is activated, and the event is recorded and re-adaptation is triggered in the background, forming a sustainable "collection-processing-recognition-linkage" closed loop.

[0065]

[0066] in, From the temporal characteristics linear mapping, For head dimension; This represents the correlation distribution between temporal locations. This is a weighted sum after attention. Combined with depthwise separable one-dimensional convolutions (grouped convolutions on the channel / head dimension) and compressed feedforward layers, low-latency global temporal dependency modeling can be achieved at the edge.

[0067]

[0068] Timing representation after stacking several layers The data is averaged over time and fed into a two-layer perceptron (with ReLU / GELU activation, etc.). The softmax function outputs the probability of each emotion category. Then, it is mapped to a multi-level warning system based on thresholds / hysteresis, which is linked to the in-vehicle subsystems and records the events. When there is uncertainty or anomalies, it switches to a conservative strategy, triggers a re-adaptation in the background, and completes the closed loop of "identification - warning - intervention - feedback".

[0069] In another implementation of the present invention, the warning level is multi-level, and the linkage trigger adopts a preset threshold and hysteresis strategy; the intervention command supports event recording, secondary confirmation and cancellation.

[0070] Through the above steps, the model sequentially completes channel topology adaptation, branch-level robust fusion, few-sample individualization enhancement, online calibration, and lightweight temporal modeling on a unified input engineering interface. This not only meets the requirements of vehicle real-time performance and small model deployment, but also improves the stable recognition capability in complex in-vehicle environments and across subject conditions, and achieves safe linkage with multiple in-vehicle subsystems.

[0071] Compared with the prior art, the present invention has the following beneficial effects: (1) Rapid adaptation across subjects: Meta-learning is introduced to rapidly adapt and prototype-guided enhancement, enabling rapid individualization of new drivers with a small number of supporting samples, without the need for long-term calibration, thus improving online efficiency and generalization ability.

[0072] (2) Online low latency and lightweight: The integrated model (MAGT) with adaptive mapping, hierarchical multi-branch fusion and lightweight Transformer collaboration is adopted to achieve low latency and small model size on the edge while ensuring recognition accuracy, thus meeting the automotive-grade real-time requirements and deployment constraints.

[0073] (3) Robustness and robustness: Based on the segmentation / sub-segmentation and Welch seven-band rPSD to construct a unified input, adaptive mapping and online calibration jointly suppress the influence of electrode contact changes and scene noise, and maintain stable recognition in complex in-vehicle environments.

[0074] (4) Easy integration and maintenance: The communication interface for vehicle cockpit / domain control is easy to integrate, supports localized processing and minimal retention, log reporting and OTA policy distribution, which facilitates long-term operation and maintenance and policy optimization, and meets privacy and compliance requirements.

[0075] Another aspect of the present invention, such as Figure 2 As shown, a vehicle-mounted EEG driver emotion safety monitoring system based on meta-learning and rapid adaptation is provided, comprising: Data acquisition module: Collects multi-channel EEG data of the target vehicle driver through the vehicle-mounted EEG acquisition system; Data processing module: performs data preprocessing on the multi-channel EEG data, processes it according to segmentation and sub-segmentation strategies to form a unified input, wherein the length and overlap of segments and sub-segments are configurable parameters; performs adaptive mapping and hierarchical multi-branch fusion on the unified input to obtain a fused representation; performs few-sample fast individualized adaptation and online adaptive calibration on the fused representation to obtain calibration features; The result output module inputs the calibration features into a lightweight Transformer for time-series modeling, performs classification and recognition after pooling, determines the warning level, outputs intervention commands based on preset linkage rules, drives in-vehicle voice, lighting, seat, air conditioning and other subsystems, and the system runs online and records events.

[0076] This invention presents an in-vehicle EEG driver emotion safety monitoring system based on meta-learning rapid adaptation. It introduces meta-learning rapid adaptation and prototype-guided enhancement to achieve rapid individualization of new drivers with a limited number of support samples, eliminating the need for lengthy calibration and improving deployment efficiency and generalization ability. It employs an integrated model (MAGT) combining adaptive mapping, hierarchical multi-branch fusion, and lightweight Transformer collaboration, ensuring recognition accuracy while achieving low latency and small model size at the edge, meeting automotive-grade real-time requirements and deployment constraints. A unified input is constructed based on segmented / sub-segmented data and Welch seven-band rPSD. Adaptive mapping and online calibration jointly suppress the effects of electrode contact variations and scene noise, maintaining stable recognition in complex in-vehicle environments. The communication interface for in-vehicle cockpit / domain control is easy to integrate, supporting localized processing and minimal retention, log reporting, and OTA policy deployment, facilitating long-term operation and maintenance and policy optimization, and complying with privacy and compliance requirements.

[0077] In another aspect of the present invention, the electronic device includes: a processor, a memory, and a communication bus and a communication interface.

[0078] in: The processor, memory, and communication interface communicate with each other via a communication bus.

[0079] A communication interface is used to communicate with other electronic devices or servers.

[0080] The processor is used to execute programs, specifically, to perform any of the steps of the vehicle-mounted EEG driver emotion safety monitoring method based on meta-learning rapid adaptation in the above embodiments.

[0081] Specifically, the program may include program code, which includes computer operation instructions.

[0082] The processor may be a central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of this application. The one or more processors included in the smart device may be processors of the same type, such as one or more CPUs; or they may be processors of different types, such as one or more CPUs and one or more ASICs.

[0083] Memory is used to store programs. Memory may include high-speed RAM, and may also include non-volatile memory, such as at least one disk drive.

[0084] Specifically, the program can be used to cause the processor to execute the steps of any of the vehicle-mounted EEG driver emotion safety monitoring methods based on meta-learning rapid adaptation described in the embodiments. The specific implementation of each step in the program can be found in the corresponding descriptions of the steps and units executed in any of the above-described vehicle-mounted EEG driver emotion safety monitoring methods based on meta-learning rapid adaptation, and will not be repeated here. Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the devices and modules described above can be referred to the corresponding process descriptions in the foregoing method embodiments.

[0085] An exemplary embodiment of this application also provides a non-transitory computer-readable storage medium storing computer instructions, wherein the computer instructions are used to cause a computer to perform the methods of various embodiments of this application.

[0086] The methods described above according to embodiments of the present invention can be implemented in hardware, firmware, or as software or computer code that can be stored in a recording medium (such as a CD-ROM, RAM, floppy disk, hard disk, or magneto-optical disk), or as computer code originally stored on a remote recording medium or a non-transitory machine-readable medium and subsequently stored on a local recording medium, downloaded via a network. Thus, the methods described herein can be processed by software stored on a recording medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware (such as an ASIC or FPGA). It is understood that the computer, processor, microprocessor controller, or programmable hardware includes storage components (e.g., RAM, ROM, flash memory, etc.) capable of storing or receiving software or computer code, which, when accessed and executed by the computer, processor, or hardware, implements the methods described herein. Furthermore, when a general-purpose computer accesses code used to implement the methods shown herein, the execution of the code transforms the general-purpose computer into a dedicated computer for executing the methods shown herein.

[0087] Specific embodiments of the present invention have now been described. Other embodiments are within the scope of the appended claims. In some cases, the actions described in the claims can be performed in a different order and still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily require a specific or sequential order to achieve the desired result.

[0088] It should be noted that all directional indications (such as up, down, left, right, back, etc.) in the embodiments of the present invention are only used to explain the relative positional relationship between the components in a certain order (as shown in the figure). If the specific order changes, the directional indication will also change accordingly.

[0089] In the description of this invention, the terms "first" and "second" are used only for convenience in describing different components or names, and should not be construed as indicating or implying a sequential relationship, relative importance, or implicitly specifying the number of technical features indicated. Thus, a feature defined with "first" and "second" may explicitly or implicitly include at least one of that feature.

[0090] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention.

[0091] It should be noted that although specific embodiments of the present invention have been described in detail with reference to the accompanying drawings, this should not be construed as limiting the scope of protection of the present invention. Various modifications and variations that can be made by those skilled in the art without inventive effort within the scope described in the claims still fall within the scope of protection of the present invention.

[0092] The examples of the embodiments of the present invention are intended to concisely illustrate the technical features of the embodiments of the present invention, so that those skilled in the art can intuitively understand the technical features of the embodiments of the present invention, and are not intended to be an improper limitation of the embodiments of the present invention.

[0093] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for monitoring driver emotions and safety in vehicles based on meta-learning and rapid adaptation, characterized in that, include: Multi-channel EEG data of the target vehicle driver was collected using an in-vehicle EEG acquisition system. The multi-channel EEG data is preprocessed according to a segmentation and sub-segmentation strategy to form a unified input, wherein the length and overlap of the segments and sub-segments are configurable parameters. Adaptive graph construction and hierarchical multi-branch fusion are performed on the unified input to obtain a fused representation; The fused representation is subjected to rapid individualized adaptation and online adaptive calibration with few samples to obtain calibration features; The calibration features are input into a lightweight Transformer for time-series modeling. After pooling, classification and recognition are completed, and the warning level is determined. Intervention commands are output according to preset linkage rules to drive the in-vehicle subsystems. The system runs online and records events.

2. The method according to claim 1, characterized in that, The unified input is: in, For batch, For timing window number, For the number of channels, The rPSD feature dimensions are defined for seven predefined frequency bands.

3. The method according to claim 1, characterized in that, The fusion is represented as: in, To learn the contribution of each branch in the current sample.

4. The method according to claim 1, characterized in that, The rapid individualized adaptation with few samples includes: Individual emotion prototypes are constructed based on a small number of supporting samples and then weighted and fused with the global prototype to generate an individualized prototype set. Based on the individualized prototype set, a feature modulation vector is generated to quickly modulate the query features, and prototype-guided enhancement is formed by adaptively weighting the features according to their distance from each prototype. Online adaptive calibration employs sliding statistical updates and adaptive weight modulation to suppress distribution drift.

5. The method according to claim 4, characterized in that, The individualized prototype is: in, According to the emotion category; This is the modulation vector.

6. The method according to claim 5, characterized in that, The enhanced query features are: in, This is a coefficient used to control the guiding intensity; Indicates the query features Perform amplitude modulation. For weights.

7. The method according to claim 1, characterized in that, The warning levels are multi-level, and the linkage triggering adopts a preset threshold and hysteresis strategy. The intervention instructions support event logging, secondary confirmation, and revocation.

8. A vehicle-mounted EEG driver emotion safety monitoring system based on meta-learning and rapid adaptation, characterized in that, include: Data acquisition module: Collects multi-channel EEG data of the target vehicle driver through the vehicle-mounted EEG acquisition system; Data processing module: performs data preprocessing on the multi-channel EEG data, processes it according to segmentation and sub-segmentation strategies to form a unified input, wherein the length and overlap of segments and sub-segments are configurable parameters; performs adaptive mapping and hierarchical multi-branch fusion on the unified input to obtain a fused representation; performs few-sample fast individualized adaptation and online adaptive calibration on the fused representation to obtain calibration features; The result output module inputs the calibration features into a lightweight Transformer for time-series modeling, performs pooling to complete classification and identification, determines the warning level, outputs intervention commands according to preset linkage rules, drives the in-vehicle subsystems, and the system runs online and records events.

9. An electronic device, characterized in that, include: The system includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the vehicle-mounted EEG driver emotion safety monitoring method based on meta-learning rapid adaptation as described in any one of claims 1 to 7.

10. A computer storage medium, characterized in that, The computer storage medium stores a computer program, which, when executed by a processor, implements the steps in the vehicle-mounted EEG driver emotion safety monitoring method based on meta-learning rapid adaptation as described in any one of claims 1 to 7.

Images