A method and device for electroencephalogram emotion recognition guided by synergistic optimization of network

Abstract

The application provides a kind of synergic optimization network guidance electroencephalogram emotion recognition method and device, it is related to medical instrument technical field.The method comprises: obtaining brain wave signal;Extract the band feature of brain wave signal, calculate interval time piece difference entropy, obtain the time-frequency representation of brain wave;The neural network architecture comprising long short-term memory network and full connection layer is constructed;The knowledge-driven architecture optimization of neural network is carried out using large language model, and the network parameters are automatically adjusted in combination with particle swarm algorithm;The spatio-temporal characteristics extracted are multi-objective optimized, and high-precision emotion classification is realized.The application can improve the accuracy of emotion recognition through intelligent architecture search guided by large language model.

Description

Technical Field

[0001] This invention relates to the field of medical device technology, and in particular to a method and device for EEG emotion recognition guided by a collaboratively optimized network. Background Technology

[0002] Current applications of this technology primarily focus on emotion recognition based on electroencephalography (EEG) and the design and optimization of neural networks. This research explores the application of neural architecture search in emotion recognition, demonstrating how network optimization improves model performance and design efficiency. However, existing neural architecture search methods face limitations when applied to EEG-based emotion recognition, including: the search space design often lacks domain-specific knowledge about EEG signal characteristics and emotion recognition principles; the computational overhead of exploring multiple architecture possibilities limits practical applicability; and current architectures often struggle to effectively extract the complex spatiotemporal relationships between different frequency bands in EEG signals. Furthermore, while large language models demonstrate significant capabilities in understanding and reasoning about domain-specific knowledge, their direct application to EEG-based emotion recognition presents challenges due to the limited size of available datasets and the complexity of EEG signals. Summary of the Invention

[0003] To address the technical problems of existing neural architectures lacking domain-specific knowledge regarding EEG signal characteristics and emotion recognition principles in their search space design, thus increasing computational overhead and making it difficult for current architectures to effectively constrain the complex spatiotemporal relationships between different frequency bands in EEG signals, this invention provides a collaboratively optimized network-guided EEG emotion recognition method and apparatus. The technical solution is as follows:

[0004] On the one hand, a novel EEG emotion recognition method guided by a collaborative optimization network is provided. This method is implemented by a novel EEG emotion recognition device guided by a collaborative optimization network, and includes:

[0005] S1. Acquire brainwave signals; preprocess and extract features from the brainwave signals to obtain time-frequency representation data of brainwaves;

[0006] S2. Construct the basic neural network architecture; the basic neural network architecture includes: a long short-term memory network module, an attention enhancement module, a fully connected feature fusion module, and a classification output module;

[0007] S3. The parameters of the basic neural network architecture are optimized using a large language model with prior knowledge and a particle swarm optimization algorithm to obtain an optimized collaborative optimization network.

[0008] S4. Input the time-frequency representation data of the EEG into the optimized collaborative optimization network, extract the temporal-dependent features through the long short-term memory network module; input the temporal-dependent features into the attention enhancement module for feature enhancement to obtain key features; input the key features into the fully connected feature fusion module for feature fusion to obtain fused features; input the fused features into the classification output module, perform emotion classification based on the pre-constructed multi-objective loss function, and obtain the final emotion recognition result.

[0009] Optionally, S1 performs preprocessing and feature extraction on the EEG signal, including:

[0010] S11. Filter the EEG signals to the delta (1-4Hz), theta (4-8Hz), alpha (8-13Hz), beta (13-30Hz), and gamma (30-75Hz) frequency bands respectively;

[0011] S12. Convert each EEG channel into a corresponding frequency band feature channel;

[0012] S13. Divide the EEG signal into 16 equal intervals in the time dimension and calculate the differential entropy features of each interval.

[0013] S14. Generate time-frequency representation data of EEG based on the differential entropy characteristics of each interval.

[0014] Optionally, the search space of the basic neural network architecture is configured as: S = {S lstm , S dense}, where the configuration parameter S of the Long Short-Term Memory network module lstm This includes: number of network layers, number of neurons, activation function, and random inactivation rate;

[0015] Among them, the configuration parameter S of the fully connected feature fusion module dense This includes: the number of neurons, random inactivation rate, activation function, and fixed classification output layer.

[0016] Optionally, the initial structure construction guided by the knowledge of the large language model with prior knowledge includes: domain knowledge K. domain The structured organization is used for signal processing knowledge, emotion recognition principles, and architectural design rules;

[0017] The process of optimizing the search strategy guided by the knowledge of the large language model with prior knowledge includes: evaluating the architecture based on historical performance records, generating architecture optimization suggestions, and guiding the direction of parameter adjustment.

[0018] Optionally, the pre-constructed multi-objective optimization loss function is represented by the following formula (1):

[0019] L=αL acc +βL ite +γL emo (1)

[0020] Where L represents a pre-constructed multi-objective optimization loss function; L acc For classification accuracy loss; L ite To calculate the efficiency loss; L emo α represents the loss for emotion feature extraction; β represents the weighting coefficient for the loss for classification accuracy; and γ represents the weighting coefficient for the loss for emotion feature extraction.

[0021] Optionally, the parameter update method of the particle swarm optimization algorithm is expressed by the following formula (2):

[0022] v t+1 =w t v t +c 1t r1(p best -x t )+c 2t r2(g best -x t (2)

[0023] x t+1 =x t +v t+1

[0024] Among them, v t w represents the particle's current velocity. t c represents the inertial weight. 1t c is the first learning factor; 2t r1 and r2 are random numbers in the range [0,1], representing the second learning factor; p best For the individual's optimal position; g best The optimal position for the group; v t+1 Represented as the optimized velocity of the particle; x t Indicates the current particle position; x t+1 This indicates the optimized particle position.

[0025] Optionally, step S3 employs a large language model with prior knowledge and a particle swarm optimization algorithm to optimize the parameters of the basic neural network architecture, obtaining an optimized collaborative optimization network, including:

[0026] S31. Using a large language model with prior knowledge to provide knowledge guidance for the basic neural network architecture includes: integration of structured domain knowledge, design of architecture search space, and adjustment of optimization strategies.

[0027] S32. The parameters of the basic neural network architecture are optimized using the particle swarm optimization algorithm. The optimized architecture parameters are obtained by updating the particle velocity and position, dynamically adjusting the learning factor, and evaluating the optimization performance. Based on the optimized architecture parameters, the optimized collaborative optimization network is obtained.

[0028] On the other hand, a novel EEG emotion recognition device guided by a collaborative optimization network is provided. This device is applied to a novel EEG emotion recognition method guided by a collaborative optimization network. The device includes:

[0029] The acquisition and preprocessing unit is used to acquire brainwave signals; preprocess and extract features from the brainwave signals to obtain time-frequency representation data of brainwaves;

[0030] The building blocks are used to construct the basic neural network architecture; the basic neural network architecture includes: a long short-term memory network module, an attention enhancement module, a fully connected feature fusion module, and a classification output module;

[0031] The optimization unit is used to optimize the parameters of the basic neural network architecture using a large language model with prior knowledge and a particle swarm optimization algorithm to obtain an optimized collaborative optimization network.

[0032] The recognition unit is used to input the time-frequency representation data of the EEG into the optimized collaborative optimization network, extract time-dependent features through the long short-term memory network module; input the time-dependent features into the attention enhancement module for feature enhancement to obtain key features; input the key features into the fully connected feature fusion module for feature fusion to obtain fused features; and input the fused features into the classification output module for emotion classification based on the pre-constructed multi-objective loss function to obtain the final emotion recognition result.

[0033] Optionally, the acquisition and preprocessing unit is used for:

[0034] The EEG signals were filtered to the delta (1-4Hz), theta (4-8Hz), alpha (8-13Hz), beta (13-30Hz), and gamma (30-75Hz) frequency bands, respectively.

[0035] Each EEG channel is converted into a corresponding frequency band feature channel;

[0036] The EEG signal was divided into 16 equal intervals along the time dimension, and the differential entropy feature of each interval was calculated.

[0037] Based on the differential entropy characteristics of each interval, time-frequency representation data of EEG are generated.

[0038] Optionally, the search space of the basic neural network architecture is configured as: S = {S lstm , Sdense}, where the configuration parameter S of the Long Short-Term Memory network module lstm This includes: number of network layers, number of neurons, activation function, and random inactivation rate;

[0039] Among them, the configuration parameter S of the fully connected feature fusion module dense This includes: the number of neurons, random inactivation rate, activation function, and fixed classification output layer.

[0040] Optionally, the initial structure construction guided by the knowledge of the large language model with prior knowledge includes: domain knowledge K. domain The structured organization is used for signal processing knowledge, emotion recognition principles, and architectural design rules;

[0041] The process of optimizing the search strategy guided by the knowledge of the large language model with prior knowledge includes: evaluating the architecture based on historical performance records, generating architecture optimization suggestions, and guiding the direction of parameter adjustment.

[0042] Optionally, the pre-constructed multi-objective optimization-based loss function is represented by the following formula (1):

[0043] L=αL acc +βL ite +γL emo (1)

[0044] Where L represents a pre-constructed multi-objective optimization loss function; L acc For classification accuracy loss; L ite To calculate the efficiency loss; L emo α represents the loss for emotion feature extraction; β represents the weighting coefficient for the loss for classification accuracy; and γ represents the weighting coefficient for the loss for emotion feature extraction.

[0045] Optionally, the parameter update method of the particle swarm optimization algorithm is expressed by the following formula (2):

[0046] v t+1 =w t v t +c 1t r1(p best -x t )+c 2t r2(g best -x t (2)

[0047] x t+1 =x t +v t+1

[0048] Among them, v tw represents the particle's current velocity. t c represents the inertial weight. 1t c is the first learning factor; 2t r1 and r2 are random numbers in the range [0,1], representing the second learning factor; p best For the individual's optimal position; g best The optimal position for the group; v t+1 Represented as the optimized velocity of the particle; x t Indicates the current particle position; x t+1 This indicates the optimized particle position.

[0049] Optionally, the optimization unit is used for:

[0050] The use of large language models with prior knowledge to guide the basic neural network architecture includes: integration of structured domain knowledge, design of architecture search space, and adjustment of optimization strategies.

[0051] The parameters of the basic neural network architecture are optimized using the particle swarm optimization algorithm. The optimized architecture parameters are obtained by updating particle velocity and position, dynamically adjusting the learning factor, and evaluating the optimization performance. Based on the optimized architecture parameters, the optimized collaborative optimization network is obtained.

[0052] On the other hand, a novel collaborative optimization network-guided EEG emotion recognition device is provided, the novel collaborative optimization network-guided EEG emotion recognition device comprising: a processor; a memory, the memory storing computer-readable instructions, which, when executed by the processor, implement any of the methods described above for novel collaborative optimization network-guided EEG emotion recognition.

[0053] On the other hand, a computer-readable storage medium is provided, wherein at least one instruction is stored in the storage medium, the at least one instruction being loaded and executed by a processor to implement any of the above-described novel collaboratively optimized network-guided EEG emotion recognition methods.

[0054] The beneficial effects of the technical solutions provided in the embodiments of the present invention include at least the following:

[0055] This invention first acquires brainwave signals; preprocesses and extracts features from the brainwave signals to obtain time-frequency representation data of the brainwaves; secondly, it constructs a basic neural network architecture, which includes a long short-term memory network module, an attention enhancement module, a fully connected feature fusion module, and a classification output module; it optimizes the parameters of the basic neural network architecture using a large language model with prior knowledge and a particle swarm optimization algorithm to obtain an optimized collaborative optimization network; finally, it inputs the time-frequency representation data of the brainwaves into the optimized collaborative optimization network, extracts temporal dependent features through the long short-term memory network module; it inputs the temporal dependent features into the attention enhancement module for feature enhancement to obtain key features; it inputs the key features into the fully connected feature fusion module for feature fusion to obtain fused features; and it inputs the fused features into the classification output module for emotion classification based on a pre-constructed multi-objective loss function to obtain the final emotion recognition result.

[0056] The embodiments of the present invention maintain reasonable computational overhead, effectively utilize domain knowledge, and can identify key neural features in the forehead and temporal regions. The embodiments of the present invention can automatically discover and utilize complex relationships in EEG signals, thereby improving model performance, and perform well in binary, multi-class emotion classification, and four-class emotion recognition tasks, thus improving its broad applicability. The embodiments of the present invention can improve the accuracy of emotion recognition. Attached Figure Description

[0057] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0058] Figure 1 This is a flowchart of a novel EEG emotion recognition method guided by a collaborative optimization network, provided by an embodiment of the present invention.

[0059] Figure 2 This is a schematic diagram of the structure of a high-efficiency collaborative optimization network provided in an embodiment of the present invention;

[0060] Figure 3 This is a schematic diagram of a structure illustrating an electroencephalogram (EEG) topological pattern, provided by an embodiment of the present invention.

[0061] Figure 4 This is a block diagram of a novel EEG emotion recognition device guided by a collaborative optimization network, provided in an embodiment of the present invention.

[0062] Figure 5This is a schematic diagram of the structure of a novel EEG emotion recognition device guided by a collaborative optimization network, provided in an embodiment of the present invention. Detailed Implementation

[0063] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0064] In embodiments of the present invention, words such as "exemplarily," "for example," etc., are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" in the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the word "exemplary" is intended to present the concept in a concrete manner. Furthermore, in embodiments of the present invention, the meaning expressed by "and / or" can be both, or either one.

[0065] In the embodiments of this invention, the terms "image" and "picture" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning. Similarly, the terms "of," "corresponding (relevant)," and "corresponding" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning.

[0066] In this embodiment of the invention, sometimes a subscript such as W1 may be written in a non-subscript form such as W1. When the difference is not emphasized, the meaning they express is the same.

[0067] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0068] This invention provides a novel EEG emotion recognition method guided by a collaboratively optimized network. This method can be implemented using a novel EEG emotion recognition device guided by a collaboratively optimized network, which can be a terminal or a server. Figure 1 The flowchart shown is for a novel collaborative optimization network-guided EEG emotion recognition method. The processing flow of this method may include the following steps:

[0069] S1. Acquire brainwave signals; preprocess and extract features from the brainwave signals to obtain time-frequency representation data of the brainwaves.

[0070] Before obtaining the time-frequency characterization data of the brain waves, the brain wave signals are processed to remove power line noise and reduce systematic artifacts to obtain preprocessed brain wave signals.

[0071] Optionally, the specific implementation process of S1 may include S11-S14:

[0072] S11. Filter the EEG signals to the delta (1-4Hz), theta (4-8Hz), alpha (8-13Hz), beta (13-30Hz), and gamma (30-75Hz) frequency bands respectively;

[0073] S12. Convert each EEG channel into a corresponding frequency band feature channel;

[0074] Each EEG channel contributes five spectral components, providing 310 62×5 feature channels for SEED and 160 32×5 feature channels for DEAP.

[0075] S13. Divide the EEG signal into 16 equal intervals in the time dimension and calculate the differential entropy features of each interval.

[0076] S14. Generate time-frequency representation data of EEG based on the differential entropy characteristics of each interval.

[0077] The time-frequency representation data of EEG is expressed as: X input ∈ R T × S × F Where T represents the time period, referring to the number of segments into which the signal is divided in the time dimension, and each segment contains information about the EEG signal within that time window; S represents the spatial features extracted from the EEG channels, referring to the signals recorded by different EEG channels, each channel can be regarded as a spatial location; F represents the spectral features extracted from the EEG channels, referring to the characteristics of the signal in different frequency ranges, including: delta waves: 1-4 Hz, theta waves: 4-8 Hz, alpha waves: 8-13 Hz, beta waves: 13-30 Hz, and gamma waves: 30-75 Hz, etc. R represents the set of real numbers, indicating that each feature value is a real number.

[0078] In one feasible implementation, fine-grained temporal dynamics of EEG signals in each frequency band are captured while maintaining computational efficiency. Specifically, a 16×310 feature representation is used for the SEED emotion dataset, and a 16×160 feature representation is used for the DEAP dataset, which can effectively characterize the spectral-spatial patterns and temporal evolution of emotional states.

[0079] S2. Construct the basic neural network architecture; the basic neural network architecture includes: a long short-term memory network module, an attention enhancement module, a fully connected feature fusion module, and a classification output module;

[0080] The Long Short-Term Memory (LSTM) network module is used to capture temporally dependent features.

[0081] The attention enhancement module is used to highlight key features.

[0082] The optimized collaborative optimization network, through a large language model with prior knowledge and neural architecture search, can advance EEG-based emotion recognition. The overall architecture of the optimized collaborative optimization network can be represented as: M opt =BEACON(D EEG , K domain , S, Θ LLM ); where BEACON represents the optimized architecture of the neural network; M opt D represents the optimized neural architecture. EEG K represents the preprocessed EEG dataset. domain S defines the architecture search space, encompassing knowledge from the fields of electroencephalography and emotion recognition, and Θ. LLM This represents the parameters of the large language model that guide the optimization process.

[0083] Optionally, the search space of the basic neural network architecture is configured as: S = {S lstm , S dense}, where the configuration parameter S of the Long Short-Term Memory network module lstm This includes: number of network layers, number of neurons, activation function, and random inactivation rate;

[0084] Among them, the configuration parameter S of the fully connected feature fusion module dense This includes: the number of neurons, random inactivation rate, activation function, and fixed classification output layer.

[0085] The configuration parameter of the Long Short-Term Memory (LSTM) network module is represented as S. lstm = {n layers ∈ [1, 4], n units ∈ [32, 256], func ∈ Func, p drop ∈ [0.1, 0.5]}; where Func includes a comprehensive set of activation functions;

[0086] The configuration parameters of the fully connected feature fusion module are represented as S. dense = {n units ∈[32, 256], p drop ∈[0.1, 0.5], func∈Func}, with a fixed output layer of 1×4.

[0087] In one feasible implementation, the above configuration can ensure a reduction in the dimensionality of features obtained from the Long Short-Term Memory network to the final sentiment classification layer.

[0088] S3. The parameters of the basic neural network architecture are optimized using a large language model with prior knowledge and a particle swarm optimization algorithm to obtain an optimized collaborative optimization network.

[0089] Optionally, the specific implementation process of S3 may include S31-S32:

[0090] S31. Using a large language model with prior knowledge to provide knowledge guidance for the basic neural network architecture includes: integration of structured domain knowledge, design of architecture search space, and adjustment of optimization strategies.

[0091] Among them, the large language model with prior knowledge is achieved through domain knowledge K. domain This application utilizes iFlytek's Spark 3.0 large language model to intelligently optimize architecture selection.

[0092] S32. The parameters of the basic neural network architecture are optimized using the particle swarm optimization algorithm. The optimized architecture parameters are obtained by updating the particle velocity and position, dynamically adjusting the learning factor, and evaluating the optimization performance. Based on the optimized architecture parameters, the optimized collaborative optimization network is obtained.

[0093] The optimized architecture parameters are represented as: θ t = Particle Swarm Optimization (S, H) t Large Language Model (K) domain )); where θ t Let H represent the architecture parameters at iteration t; S represents the search space of the architecture; H represents the architecture parameters at iteration t. t Indicates search history; K domain Represents domain knowledge.

[0094] Optionally, the initial structure construction guided by the knowledge of a large language model with prior knowledge includes: domain knowledge K domain The structured organization is used for signal processing knowledge, emotion recognition principles, and architectural design rules;

[0095] The process of optimizing a search strategy guided by knowledge from a large language model with prior knowledge includes: evaluating the architecture based on historical performance records, generating architecture optimization suggestions, and guiding the direction of parameter adjustments.

[0096] In one feasible implementation, a large language model with prior knowledge is used to guide the optimization of parameters of the basic neural architecture. This includes: using the large language model for knowledge guidance, the system constructs prompts by integrating the following: current architecture state and performance; historical performance patterns and trends; domain-specific constraints and requirements; and previous large model recommendations and their results. The large model processes the search history by: identifying promising architectural patterns; detecting potential performance bottlenecks; suggesting parameter adjustments; and guiding the exploration direction based on domain knowledge. The large language model with prior knowledge guides the dynamic adjustment of the particle swarm optimization algorithm, considering factors such as: current convergence rate; the balance between exploration and utilization; and historical performance improvement trends. The optimization search process includes updating particle velocity and position, dynamically adjusting learning factors, and evaluating optimization performance. Through this collaborative optimization method, domain expertise is effectively combined with efficient search strategies to obtain the optimal neural architecture suitable for EEG-based emotion recognition.

[0097] S4. Input the time-frequency representation data of EEG into the optimized collaborative optimization network, and extract the temporal-dependent features through the long short-term memory network; input the temporal-dependent features into the attention enhancement module for feature enhancement to obtain key features; input the key features into the fully connected feature fusion module for feature fusion to obtain fused features; input the fused features into the classification output module, and perform emotion classification based on the pre-constructed multi-objective loss function to obtain the final emotion recognition result.

[0098] The process of obtaining the final emotion recognition result can be represented as: Y = f dense (f lstm (f att (X input ))), where X input f represents the input data of the model. att f represents attention-enhancing features. lstm Indicates capturing time dependency, f dense Perform sentiment classification; Y represents the model's final output.

[0099] Optionally, the pre-constructed multi-objective optimization-based loss function is represented by the following formula (1):

[0100] L=αL acc +βL ite +γL emo (1)

[0101] Where L represents a pre-constructed multi-objective optimization loss function; L acc For classification accuracy loss; L ite To calculate the efficiency loss; L emoα represents the loss for emotion feature extraction; β represents the weighting coefficient for the loss for classification accuracy; and γ represents the weighting coefficient for the loss for emotion feature extraction.

[0102] Optionally, the parameter update method of the particle swarm optimization algorithm is expressed by the following formula (2):

[0103] v t+1 =w t v t +c 1t r1(p best -x t )+c 2t r2(g best -x t (2)

[0104] x t+1 =x t +v t+1

[0105] Among them, v t w represents the particle's current velocity. t c represents the inertial weight. 1t c is the first learning factor; 2t r1 and r2 are random numbers in the range [0,1], representing the second learning factor; p best For the individual's optimal position; g best The optimal position for the group; v t+1 Represented as the optimized velocity of the particle; x t Indicates the current particle position; x t+1 This indicates the optimized particle position.

[0106] According to embodiments of the present invention, a lightweight real-time emotion recognition system can be designed. By optimizing the efficient architecture and coordinating the network structure and parameters, and by employing model compression techniques, including pruning and quantization, the computational complexity and memory usage of the model can be reduced, thereby improving the real-time performance of the system. Through the above methods, an efficient and low-power emotion recognition system can be obtained, meeting the needs of practical scenarios.

[0107] Among them, such as Figure 2The diagram illustrates the structure of a high-efficiency collaborative optimization network provided in this embodiment of the invention. This framework is a system integrating a long short-term memory network, a fully connected network, a large language model, and a particle swarm optimization algorithm, used to process EEG data for emotion recognition. The raw EEG signals are first converted into data and then input into the basic neural network architecture to extract time-frequency representations. The particle swarm optimization algorithm is used to optimize the parameters of the basic neural network architecture, updating particle velocity and position, dynamically adjusting learning factors, and evaluating optimization performance. The optimization process of the particle swarm optimization algorithm utilizes the large language model for knowledge guidance through structured domain knowledge integration, architecture search space design, and optimization strategy adjustment. Finally, the optimized model, after performance verification, outputs classification results, achieving the classification of different emotional states.

[0108] Among them, such as Figure 3 The diagram illustrates a structural representation of an EEG topology pattern according to an embodiment of the present invention. In one feasible implementation, the topology reveals distinct activation patterns of different emotion categories in various electrode regions. This embodiment of the invention can identify key neural features in the frontal and temporal regions, which are highly consistent with emotion-related brain activity frequently reported in neuroscience literature. The visualization demonstrates that the efficient collaborative optimization network architecture search successfully discovered and utilized neurologically meaningful features, thereby effectively distinguishing complex emotional states.

[0109] In one feasible implementation, to demonstrate the effectiveness of the architecture search method guided by a large language model with prior knowledge, this application was compared with existing methods on the SEED dataset and the DEAP dataset, and the comparison results are shown in Table 1 below.

[0110] Table 1

[0111]

[0112] As shown in Table 1, the proposed method achieves an accuracy of 96.42% on the SEED dataset, 85.10% on the DEAP arousal binary classification task, 86.40% on the DEAP pleasure binary classification task, and 77.65% on the DEAP four-class arousal-pleasure joint classification task. These results significantly outperform existing NAS methods, demonstrating the powerful capabilities of the proposed method in complex emotion recognition tasks.

[0113] As shown in Table 1, the EEG emotion recognition method guided by the collaborative optimization network consistently outperforms existing methods across all evaluation metrics. On the SEED dataset, the method achieves 96.42% accuracy, surpassing state-of-the-art methods MD-AGCN and RGNN. More significant improvements are observed on the DEAP dataset, where the method achieves 85.10% accuracy in arousal and 86.40% in valence classification, significantly outperforming the state-of-the-art method eNAS. In the DEAP arousal-valence joint classification task, the method achieves 77.65% accuracy, demonstrating a significant improvement over eNAS. The collaborative optimization network-guided EEG emotion recognition method also shows significant improvement in handling complex emotion patterns across different architectural approaches. Traditional spatiotemporal methods, such as TSception (63.75% on DEAP-arousal) and feature learning methods, such as EGGFuseNet (59.06% on SEED), typically rely on manually designed architectures and cannot effectively capture the complex relationships within EEG signals.

[0114] Among them, the EEG emotion recognition method guided by the collaborative optimization network maintains reasonable computational overhead thanks to its knowledge-driven evolutionary optimization mechanism. Furthermore, the framework of the collaborative optimization network-guided EEG emotion recognition method effectively utilizes domain knowledge, enabling it to identify key neural features in the prefrontal and temporal regions, which are highly consistent with emotion-related brain activity frequently reported in neuroscience literature. The collaborative optimization network-guided EEG emotion recognition method can also automatically discover and utilize complex relationships in EEG signals, thereby improving model performance. It performs excellently in binary, multi-class, and four-class emotion recognition tasks, demonstrating its broad applicability.

[0115] This invention first acquires brainwave signals; preprocesses and extracts features from the brainwave signals to obtain time-frequency representation data of the brainwaves; secondly, it constructs a basic neural network architecture, which includes a long short-term memory network module, an attention enhancement module, a fully connected feature fusion module, and a classification output module; it optimizes the parameters of the basic neural network architecture using a large language model with prior knowledge and a particle swarm optimization algorithm to obtain an optimized collaborative optimization network; finally, it inputs the time-frequency representation data of the brainwaves into the optimized collaborative optimization network, extracts temporal dependent features through the long short-term memory network module; it inputs the temporal dependent features into the attention enhancement module for feature enhancement to obtain key features; it inputs the key features into the fully connected feature fusion module for feature fusion to obtain fused features; and it inputs the fused features into the classification output module for emotion classification based on a pre-constructed multi-objective loss function to obtain the final emotion recognition result.

[0116] The embodiments of the present invention can maintain reasonable computational overhead. The embodiments of the present invention effectively utilize domain knowledge and can identify key neural features in the forehead and temporal regions. The embodiments of the present invention can automatically discover and utilize complex relationships in EEG signals, thereby improving model performance. They perform well in binary classification, multi-class emotion classification, and four-class emotion recognition tasks, which can improve their wide applicability. The embodiments of the present invention can improve the accuracy of emotion recognition.

[0117] Figure 4 This is a block diagram illustrating a novel EEG emotion recognition device guided by a collaboratively optimized network, according to an exemplary embodiment. The device is used in a novel EEG emotion recognition method guided by a collaboratively optimized network. (Refer to...) Figure 4 The device includes an acquisition and preprocessing unit 410, a construction unit 420, an optimization unit 430, and an identification unit 440. Wherein:

[0118] The acquisition and preprocessing unit is used to acquire brainwave signals; preprocess and extract features from the brainwave signals to obtain time-frequency representation data of brainwaves;

[0119] The building blocks are used to construct the basic neural network architecture; the basic neural network architecture includes: a long short-term memory network module, an attention enhancement module, a fully connected feature fusion module, and a classification output module;

[0120] The optimization unit is used to optimize the parameters of the basic neural network architecture using a large language model with prior knowledge and a particle swarm optimization algorithm to obtain an optimized collaborative optimization network.

[0121] The recognition unit is used to input the time-frequency representation data of the EEG into the optimized collaborative optimization network, extract time-dependent features through the long short-term memory network module; input the time-dependent features into the attention enhancement module for feature enhancement to obtain key features; input the key features into the fully connected feature fusion module for feature fusion to obtain fused features; and input the fused features into the classification output module for emotion classification based on the pre-constructed multi-objective loss function to obtain the final emotion recognition result.

[0122] Optionally, the acquisition and preprocessing unit 410 is used for:

[0123] The EEG signals were filtered to the delta (1-4Hz), theta (4-8Hz), alpha (8-13Hz), beta (13-30Hz), and gamma (30-75Hz) frequency bands, respectively.

[0124] Each EEG channel is converted into a corresponding frequency band feature channel;

[0125] The EEG signal was divided into 16 equal intervals along the time dimension, and the differential entropy feature of each interval was calculated.

[0126] Based on the differential entropy characteristics of each interval, time-frequency representation data of EEG are generated.

[0127] Optionally, the search space of the basic neural network architecture is configured as: S = {S lstm , S dense}, where the configuration parameter S of the Long Short-Term Memory network module lstm This includes: number of network layers, number of neurons, activation function, and random inactivation rate;

[0128] Among them, the configuration parameter S of the fully connected feature fusion module dense This includes: the number of neurons, random inactivation rate, activation function, and fixed classification output layer.

[0129] Optionally, the initial structure construction guided by the knowledge of the large language model with prior knowledge includes: domain knowledge K. domain The structured organization is used for signal processing knowledge, emotion recognition principles, and architectural design rules;

[0130] The process of optimizing the search strategy guided by the knowledge of the large language model with prior knowledge includes: evaluating the architecture based on historical performance records, generating architecture optimization suggestions, and guiding the direction of parameter adjustment.

[0131] Optionally, the pre-constructed multi-objective optimization-based loss function is represented by the following formula (1):

[0132] L=αL acc +βL ite +γL emo (1)

[0133] Where L represents a pre-constructed multi-objective optimization loss function; L acc For classification accuracy loss; L ite To calculate the efficiency loss; L emo α represents the loss for emotion feature extraction; β represents the weighting coefficient for the loss for classification accuracy; and γ represents the weighting coefficient for the loss for emotion feature extraction.

[0134] Optionally, the parameter update method of the particle swarm optimization algorithm is expressed by the following formula (2):

[0135] v t+1 =w t v t +c 1t r1(p best -x t )+c2t r2(g best -x t (2)

[0136] x t+1 =x t +v t+1

[0137] Among them, v t w represents the particle's current velocity. t c represents the inertial weight. 1t c is the first learning factor; 2t r1 and r2 are random numbers in the range [0,1], representing the second learning factor; p best For the individual's optimal position; g best The optimal position for the group; v t+1 Represented as the optimized velocity of the particle; x t Indicates the current particle position; x t+1 This indicates the optimized particle position.

[0138] Optionally, the optimization unit 430 is used to:

[0139] The use of large language models with prior knowledge to guide the basic neural network architecture includes: integration of structured domain knowledge, design of architecture search space, and adjustment of optimization strategies.

[0140] The parameters of the basic neural network architecture are optimized using the particle swarm optimization algorithm. The optimized architecture parameters are obtained by updating particle velocity and position, dynamically adjusting the learning factor, and evaluating the optimization performance. Based on the optimized architecture parameters, the optimized collaborative optimization network is obtained.

[0141] This invention first acquires brainwave signals; preprocesses and extracts features from the brainwave signals to obtain time-frequency representation data of the brainwaves; secondly, it constructs a basic neural network architecture, which includes a long short-term memory network module, an attention enhancement module, a fully connected feature fusion module, and a classification output module; it optimizes the parameters of the basic neural network architecture using a large language model with prior knowledge and a particle swarm optimization algorithm to obtain an optimized collaborative optimization network; finally, it inputs the time-frequency representation data of the brainwaves into the optimized collaborative optimization network, extracts temporal dependent features through the long short-term memory network module; it inputs the temporal dependent features into the attention enhancement module for feature enhancement to obtain key features; it inputs the key features into the fully connected feature fusion module for feature fusion to obtain fused features; and it inputs the fused features into the classification output module for emotion classification based on a pre-constructed multi-objective loss function to obtain the final emotion recognition result.

[0142] The embodiments of the present invention maintain reasonable computational overhead, effectively utilize domain knowledge, and can identify key neural features in the forehead and temporal regions. The embodiments of the present invention can automatically discover and utilize complex relationships in EEG signals, thereby improving model performance, and perform well in binary, multi-class emotion classification, and four-class emotion recognition tasks, thus improving its broad applicability. The embodiments of the present invention can improve the accuracy of emotion recognition.

[0143] Figure 5 This is a schematic diagram of the structure of a novel EEG emotion recognition device guided by a collaborative optimization network, as provided in an embodiment of the present invention. Figure 5 As shown, the novel collaborative optimization network-guided EEG emotion recognition device may include the above-mentioned... Figure 4 The novel EEG emotion recognition device guided by a collaborative optimization network is shown. Optionally, the novel EEG emotion recognition device 510 guided by a collaborative optimization network may include a first processor 2001.

[0144] Optionally, the novel collaboratively optimized network-guided EEG emotion recognition device 510 may also include a memory 2002 and a transceiver 2003.

[0145] The first processor 2001, memory 2002, and transceiver 2003 can be connected via a communication bus.

[0146] The following is combined Figure 5 A detailed introduction to each component of the novel collaborative optimization network-guided EEG emotion recognition device 510 is provided below:

[0147] The first processor 2001 is the control center of the novel collaborative optimization network-guided EEG emotion recognition device 510. It can be a single processor or a collective term for multiple processing elements. For example, the first processor 2001 can be one or more central processing units (CPUs), application-specific integrated circuits (ASICs), or one or more integrated circuits configured to implement embodiments of the present invention, such as one or more digital signal processors (DSPs), or one or more field-programmable gate arrays (FPGAs).

[0148] Optionally, the first processor 2001 can perform various functions of the novel collaborative optimization network-guided EEG emotion recognition device 510 by running or executing software programs stored in the memory 2002 and calling data stored in the memory 2002.

[0149] In a specific implementation, as one example, the first processor 2001 may include one or more CPUs, for example... Figure 5 CPU0 and CPU1 are shown in the diagram.

[0150] In a specific implementation, as one example, the novel collaborative optimization network-guided EEG emotion recognition device 510 may also include multiple processors, such as... Figure 5 The first processor 2001 and the second processor 2004 are shown in the diagram. Each of these processors can be a single-core processor or a multi-core processor. Here, a processor can refer to one or more devices, circuits, and / or processing cores used to process data (such as computer program instructions).

[0151] The memory 2002 is used to store the software program that executes the present invention, and is controlled by the first processor 2001 to execute it. The specific implementation method can be referred to the above method embodiment, and will not be repeated here.

[0152] Optionally, the memory 2002 may be a read-only memory (ROM) or other type of static storage device capable of storing static information and instructions, random access memory (RAM) or other type of dynamic storage device capable of storing information and instructions, or electrically erasable programmable read-only memory (EEPROM), compact disc read-only memory (CD-ROM) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital universal optical discs, Blu-ray discs, etc.), magnetic disk storage media or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but not limited thereto. The memory 2002 may be integrated with the first processor 2001 or may exist independently, and may be connected to the interface circuit of the brainwave emotion recognition device 510 guided by the novel collaborative optimization network. Figure 5 (Not shown in the image) is coupled to the first processor 2001, and this embodiment of the invention does not specifically limit this.

[0153] The transceiver 2003 is used to communicate with network devices or with terminal devices.

[0154] Alternatively, transceiver 2003 may include a receiver and a transmitter. Figure 5 (Not shown separately). The receiver is used to implement the receiving function, and the transmitter is used to implement the transmitting function.

[0155] Optionally, the transceiver 2003 can be integrated with the first processor 2001 or exist independently, and can be connected to the interface circuit of the EEG emotion recognition device 510 guided by the novel collaborative optimization network. Figure 5 (Not shown in the image) is coupled to the first processor 2001, and this embodiment of the invention does not specifically limit this.

[0156] It should be noted that, Figure 5 The structure of the novel collaborative optimization network-guided EEG emotion recognition device 510 shown does not constitute a limitation on the router. Actual knowledge structure recognition devices may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0157] Furthermore, the technical effects of the novel collaborative optimization network-guided EEG emotion recognition device 510 can be referenced from the technical effects of the novel collaborative optimization network-guided EEG emotion recognition method described in the above method embodiments, and will not be repeated here.

[0158] It should be understood that the first processor 2001 in the embodiments of the present invention may be a central processing unit (CPU), or it may 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 any conventional processor, etc.

[0159] It should also be understood that the memory in the embodiments of the present invention can be volatile memory or non-volatile memory, or may include both volatile and non-volatile memory. The non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. The volatile memory can be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of random access memory (RAM) are available, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate synchronous DRAM (DDR SDRAM), enhanced synchronous DRAM (ESDRAM), synchronous linked DRAM (SLDRAM), and direct rambus RAM (DR RAM).

[0160] The above embodiments can be implemented, in whole or in part, by software, hardware (such as circuits), firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of the present invention are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. A semiconductor medium can be a solid-state drive.

[0161] It should be understood that the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. A and B can be singular or plural. Additionally, the character " / " in this article generally indicates an "or" relationship between the preceding and following related objects, but it can also represent an "and / or" relationship. Please refer to the context for a more accurate understanding.

[0162] In this invention, "at least one" means one or more, and "more than one" means two or more. "At least one of the following" or similar expressions refer to any combination of these items, including any combination of a single item or a plurality of items. For example, at least one of a, b, or c can represent: a, b, c, ab, ac, bc, or abc, where a, b, and c can be a single item or multiple items.

[0163] It should be understood that, in various embodiments of the present invention, the order of the above-mentioned process numbers does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0164] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0165] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the devices, apparatuses, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0166] In the several embodiments provided by this invention, it should be understood that the disclosed devices, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another device, or some features may be ignored or not executed. Furthermore, the displayed or discussed mutual couplings, direct couplings, or communication connections may be through some interfaces; indirect couplings or communication connections between devices or units may be electrical, mechanical, or other forms.

[0167] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0168] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0169] If the aforementioned functions are implemented as software functional units 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 invention, or the part that contributes to the prior art, or a part of the 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 to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0170] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for EEG emotion recognition guided by a collaborative optimization network, characterized in that, The method includes: S1. Acquire brainwave signals; preprocess and extract features from the brainwave signals to obtain time-frequency representation data of brainwaves; S1 performs preprocessing and feature extraction on the EEG signal, including: S11. Filter the EEG signals to the delta (1-4Hz), theta (4-8Hz), alpha (8-13Hz), beta (13-30Hz), and gamma (30-75Hz) frequency bands respectively; S12. Convert each EEG channel into a corresponding frequency band feature channel; S13. Divide the EEG signal into 16 equal intervals in the time dimension and calculate the differential entropy features of each interval. S14. Generate time-frequency representation data of EEG based on the differential entropy characteristics of each interval; S2. Construct the basic neural network architecture; the basic neural network architecture includes: a long short-term memory network module, an attention enhancement module, a fully connected feature fusion module, and a classification output module; The search space configuration of the basic neural network architecture is: S = {S lstm , S dense }, where the configuration parameter S of the Long Short-Term Memory network module lstm This includes: number of network layers, number of neurons, activation function, and random inactivation rate; Among them, the configuration parameter S of the fully connected feature fusion module dense This includes: the number of neurons, random inactivation rate, activation function, and fixed classification output layer; S3. The parameters of the basic neural network architecture are optimized using a large language model with prior knowledge and a particle swarm optimization algorithm to obtain an optimized collaborative optimization network. The initial structure construction guided by the knowledge of the large language model with prior knowledge includes: domain knowledge K. domain The structured organization is used for signal processing knowledge, emotion recognition principles, and architectural design rules; The process of optimizing the search strategy guided by the knowledge-guided large language model with prior knowledge includes: evaluating the architecture based on historical performance records, generating architecture optimization suggestions and guiding the direction of parameter adjustment; Specifically, S3 employs a large language model with prior knowledge and a particle swarm optimization algorithm to optimize the parameters of the basic neural network architecture, obtaining an optimized collaborative optimization network, including: S31. Using a large language model with prior knowledge to provide knowledge guidance for the basic neural network architecture includes: integration of structured domain knowledge, design of architecture search space, and adjustment of optimization strategies. S32. The parameters of the basic neural network architecture are optimized using the particle swarm optimization algorithm. The optimized architecture parameters are obtained by updating the particle velocity and position, dynamically adjusting the learning factor, and evaluating the optimization performance. Based on the optimized architecture parameters, the optimized collaborative optimization network is obtained. The parameter update method of the particle swarm optimization algorithm is expressed by the following formula (1): v t+1 =w t v t +c 1t r1(p best -x t )+c 2t r2(g best -x t )(1) x t+1 =x t +v t+1； Among them, v t w represents the particle's current velocity. t c represents the inertial weight. 1t c is the first learning factor; 2t r1 and r2 are random numbers in the range [0,1], representing the second learning factor; p best For the individual's optimal position; g best The optimal position for the group; v t+1 The optimized velocity for the particle; x t x represents the current particle position. t+1 Indicates the optimized particle position; S4. Input the time-frequency representation data of the EEG into the optimized collaborative optimization network, and extract the temporal-dependent features through the long short-term memory network module; input the temporal-dependent features into the attention enhancement module for feature enhancement to obtain key features; input the key features into the fully connected feature fusion module for feature fusion to obtain fused features; input the fused features into the classification output module, and perform emotion classification based on the pre-constructed multi-objective loss function to obtain the final emotion recognition result; The pre-constructed multi-objective optimization loss function is expressed by the following formula (2): L=αL acc +βL ite +γL emo (2) Where L represents a pre-constructed multi-objective optimization loss function; L acc For classification accuracy loss; L ite To calculate the efficiency loss; L emo α represents the loss for emotion feature extraction; β represents the weighting coefficient for the loss for classification accuracy; and γ represents the weighting coefficient for the loss for emotion feature extraction.

2. A collaboratively optimized network-guided EEG emotion recognition device, wherein the collaboratively optimized network-guided EEG emotion recognition device is used to implement the collaboratively optimized network-guided EEG emotion recognition method as described in claim 1, characterized in that, The device includes: The acquisition and preprocessing unit is used to acquire brainwave signals; preprocess and extract features from the brainwave signals to obtain time-frequency representation data of brainwaves; The building blocks are used to construct the basic neural network architecture; the basic neural network architecture includes: a long short-term memory network module, an attention enhancement module, a fully connected feature fusion module, and a classification output module; The optimization unit is used to optimize the parameters of the basic neural network architecture using a large language model with prior knowledge and a particle swarm optimization algorithm to obtain an optimized collaborative optimization network. The recognition unit is used to input the time-frequency representation data of the EEG into the optimized collaborative optimization network, extract time-dependent features through the long short-term memory network module; input the time-dependent features into the attention enhancement module for feature enhancement to obtain key features; input the key features into the fully connected feature fusion module for feature fusion to obtain fused features; and input the fused features into the classification output module for emotion classification based on the pre-constructed multi-objective loss function to obtain the final emotion recognition result.

3. A brainwave emotion recognition device guided by a collaborative optimization network, characterized in that, The collaboratively optimized network-guided EEG emotion recognition device includes: processor; A memory storing computer-readable instructions that, when executed by the processor, implement the method as described in claim 1.

4. A computer-readable storage medium, characterized in that, The computer-readable storage medium contains program code that can be invoked by a processor to execute the method as described in claim 1.

Images