Double-end cooperative emotion recognition method and device based on HRV sequence and electronic equipment

Abstract

The invention belongs to the technical field of artificial intelligence emotion recognition, and relates to a double-end cooperative emotion recognition method and device based on an HRV sequence and electronic equipment. The method comprises the steps that a heart rate variability sequence of a user is obtained through a mobile terminal, the mobile terminal sends a classifier obtaining request to a data processing end, and a corresponding user exclusive emotion recognition classifier continuously updated according to a user individual sample data set is received; and inputting the heart rate variability sequence into the user exclusive emotion recognition classifier, and outputting a corresponding emotion recognition result. By constructing a user exclusive emotion recognition classifier and adapting to the individual specificity and dynamic change characteristics of an HRV sequence, the emotion recognition accuracy in a real scene can reach 75% or above, and the problems of misjudgment and missed judgment are effectively avoided; the classifier is trained based on real scene data of the user and can adapt to crowds of all ages and diversified scenes such as daily scenes, sports scenes and sleep scenes, and the generalization ability is remarkably superior to that of a general model.

Description

Technical Field

[0001] This invention belongs to the field of artificial intelligence emotion recognition technology, and relates to a two-end collaborative emotion recognition method, device and electronic device based on HRV sequence. Background Technology

[0002] With the widespread adoption of smart terminal technology, emotion recognition based on physiological signals has gained significant attention due to its non-invasive and real-time advantages. Heart rate variability (HRV) sequences, as important physiological indicators reflecting the activity of the autonomic nervous system, have been proven to be closely related to emotional states, making them one of the core data sources for emotion recognition. Existing emotion recognition solutions mostly employ general classification models, failing to fully consider the impact of individual physiological differences on recognition accuracy, resulting in problems such as low recognition accuracy and poor adaptability in practical applications. Furthermore, traditional solutions often rely on dedicated detection equipment, making it difficult to meet the convenience needs of ordinary users in everyday scenarios. Therefore, how to conveniently acquire physiological data using mobile terminals, construct individualized emotion recognition models, and improve the accuracy and practicality of emotion recognition has become a pressing technical problem to be solved in this field. Summary of the Invention

[0003] The purpose of this invention is to address the aforementioned problems in existing technologies by proposing a two-end collaborative emotion recognition method based on HRV sequences.

[0004] The objective of this invention can be achieved through the following technical solution: a two-end collaborative emotion recognition method based on HRV sequences, comprising: Obtain the user's heart rate variability sequence through mobile terminals; The mobile terminal sends a classifier acquisition request to the data processing terminal and receives the corresponding user-specific emotion recognition classifier that is continuously updated based on the user's individual sample dataset. The heart rate variability sequence is input into the user-specific emotion recognition classifier, and the corresponding emotion recognition result is output.

[0005] As an optional embodiment of the present invention, a user-specific emotion recognition classifier that is continuously updated based on a user individual sample dataset includes: The user's heart rate variability sequence and corresponding emotion tags are obtained through mobile terminals; The mobile terminal constructs a user individual sample dataset based on the heart rate variability sequence, the emotion tag, and the user identifier, and sends it to the data processing terminal; The data processing terminal assigns classifier training tasks to the user individual sample dataset; Using the heart rate variability sequence as input and the emotion label as output, the assigned classifier is trained to obtain a user-specific emotion recognition classifier.

[0006] As an optional embodiment of the present invention, the classifier is implemented based on a machine learning model.

[0007] As an optional embodiment of the present invention, using the heart rate variability sequence as input and the emotion tag as output, a classifier is trained to obtain a user-specific emotion recognition classifier, including: The machine learning model is a support vector machine; The first and second parameters of the support vector machine are optimized by a grid search algorithm to generate parameter combinations. The optimal parameter combination is determined from the parameter combinations. The first parameter is used to balance the classification margin and the misclassification penalty, and the second parameter is used to adjust the distribution density of the heart rate variability sequence in the feature space. Construct a support vector machine for the user identifier based on the optimal parameter combination; Based on the training subset of the user individual sample dataset, with the heart rate variability sequence as input and the emotion label as output, a support vector machine for the user identifier is trained, and the accuracy is evaluated based on the evaluation subset of the user individual sample dataset. In response to the accuracy being greater than or equal to a preset accuracy, the current support vector machine is set as the dedicated emotion recognition support vector machine for the user identifier.

[0008] As an optional embodiment of the present invention, the first and second parameters of the support vector machine are optimized using a grid search algorithm to generate parameter combinations, and the optimal parameter combination is determined from the parameter combinations, including: Set the search ranges for the first parameter and the second parameter respectively; Construct a corresponding parameter grid based on the search range, and combine the first parameter and the second parameter in the parameter grid to generate a parameter combination; The support vector machine is trained for each parameter combination separately, and the accuracy is evaluated by cross-validation. The optimal parameter combination is determined based on the highest accuracy.

[0009] As an optional embodiment of the present invention, using the heart rate variability sequence as input and the emotion label as output, training the assigned classifier to obtain a user-specific emotion recognition model further includes: The machine learning model is a neural network model; Based on the training subset of the user individual sample dataset, the heart rate variability sequence is input into the neural network model to extract local temporal features; Using the emotion tag as output, the neural network model trained on the user identifier nonlinearly maps the extracted local temporal features to the emotion category probability distribution space; The neural network model is verified to meet the preset performance index based on the evaluation subset in the user individual sample dataset. In response to meeting the preset performance index, the current neural network model is set as the exclusive emotion recognition neural network model of the user identifier.

[0010] As an optional embodiment of the present invention, the process of training the neural network model of the user identifier includes: The difference between the sentiment labels output by the neural network model and the sentiment labels in the evaluation subset is evaluated using the cross-entropy loss function. The optimizer updates the parameters of the neural network model for the user identifier based on the differences.

[0011] As an optional embodiment of the present invention, after training the assigned classifier, the method further includes: The trained classifier is evaluated using a performance evaluation subset, and the optimal user-identified emotion recognition classifier is determined according to a preset evaluation criterion. The data processing terminal performs lightweight processing on the exclusive emotion recognition classifier of the optimal user identifier based on the data processing capability of the mobile terminal.

[0012] This invention also proposes a dual-end collaborative emotion recognition device based on HRV sequences, characterized in that it includes: The data acquisition module is used to acquire the user's heart rate variability sequence via the mobile terminal; The receiving model module is used for the mobile terminal to send a classifier acquisition request to the data processing terminal and receive the corresponding user-specific emotion recognition classifier that is continuously updated based on the user's individual sample dataset. The data analysis module is used to input the heart rate variability sequence into the user-specific emotion recognition classifier and output the corresponding emotion recognition result.

[0013] The present invention also provides an electronic device, comprising: processor; Memory used to store processor-executable instructions; The processor is configured to implement the aforementioned HRV sequence-based two-end collaborative emotion recognition method when executing executable instructions.

[0014] Compared with existing technologies, this invention has significant technical advantages: First, by constructing a user-specific emotion recognition classifier that adapts to the individual specificity and dynamic change characteristics of HRV sequences, the accuracy of emotion recognition in real-world scenarios can reach over 75%, effectively avoiding misjudgment and missed judgment. Second, the classifier is trained based on real-world user scenario data, making it adaptable to people of all ages and diverse scenarios such as daily life, exercise, and sleep, with significantly better generalization ability than general models. Third, relying on the dynamic optimization mechanism of "incremental training - model iteration," the classifier can continuously learn the user's HRV sequences and emotion mapping patterns, achieving a personalized emotional companionship experience upgrade. Fourth, adopting a lightweight design strategy, the classifier is small in size, fast in response, and low in power consumption, perfectly adapting to the computing power, storage, and power consumption limitations of mobile terminals, ensuring real-time interactive effects. Fifth, it supports flexible switching between various machine learning models such as neural network models and support vector machines, and reserves a multimodal data fusion interface, possessing excellent technical scalability. Attached Figure Description

[0015] Figure 1 This is a flowchart of the dual-end collaborative emotion recognition method based on HRV sequences according to an embodiment of the present invention; Figure 2 This is a block diagram of a dual-end collaborative emotion recognition device based on HRV sequences according to an embodiment of the present invention. Detailed Implementation

[0016] The following are specific embodiments of the present invention, which are described in conjunction with the accompanying drawings. However, the present invention is not limited to these embodiments.

[0017] Example 1 Based on the technical problems highlighted in the background, this embodiment proposes a two-end collaborative emotion recognition method based on HRV sequences, such as... Figure 1 As shown, it includes: S1, obtain the user's heart rate variability sequence through the mobile terminal; S2, the mobile terminal sends a classifier acquisition request to the data processing terminal and receives the corresponding user-specific emotion recognition classifier that is continuously updated based on the user's individual sample dataset; S3, input the heart rate variability sequence into the user-specific emotion recognition classifier and output the corresponding emotion recognition result.

[0018] This embodiment provides users with a dedicated emotion recognition classifier through a collaborative architecture of mobile terminal and data processing terminal. Specifically, the mobile terminal includes smartwatches, and the data processing terminal includes cloud servers. The smartwatch's built-in heart rate sensor collects the user's raw heart rate data in real time. Based on a 120Hz acquisition frequency, the RR interval sequence is extracted, and HRV sequence data is calculated to ensure real-time data accuracy. The HRV sequence data is then preprocessed, primarily for noise reduction and standardization. Wavelet transform algorithms are used to remove outliers caused by motion interference and environmental noise. The HRV sequence data is converted to a uniform scale using the Z-score standardization method, resulting in a standardized HRV sequence. The standardized HRV sequence includes time-domain parameters and frequency-domain parameters. The time-domain parameters include the standard deviation of the RR interval and the root mean square of the difference between adjacent RR intervals. The frequency-domain parameters include low-frequency components, high-frequency components, and the ratio of low-frequency to high-frequency components. If the mobile terminal has not yet deployed a user-specific emotion recognition classifier, it sends a classifier retrieval request to the data processing terminal. The data processing terminal will then distribute a user-specific emotion recognition classifier trained on the user's individual sample dataset, using the optimal user identifier. The mobile terminal can then deploy it. If the mobile terminal has already deployed a user-specific emotion recognition classifier, it inputs a standardized HRV sequence, and the user-specific emotion recognition classifier outputs the emotion recognition result based on the standardized HRV sequence. This dual-end collaborative architecture—combining mobile terminal HRV sequence acquisition, data processing terminal distribution of the user-specific classifier, and the classifier's output of the recognition result—directly adapts to the individual specificity and dynamic changes of HRV sequences, achieving a significant improvement in emotion recognition accuracy. Preferably, a user-specific emotion recognition classifier that is continuously updated based on individual user sample datasets includes: The user's heart rate variability sequence and corresponding emotion tags are obtained through mobile terminals; The mobile terminal constructs a user individual sample dataset based on the heart rate variability sequence, the emotion tag, and the user identifier, and sends it to the data processing terminal; The data processing terminal assigns classifier training tasks to the user individual sample dataset; Using the heart rate variability sequence as input and the emotion label as output, the assigned classifier is trained to obtain a user-specific emotion recognition classifier.

[0019] To adapt to a personalized emotion recognition classifier for each user, this embodiment uses a smartwatch as an example. First, the smartwatch's built-in heart rate sensor collects the user's raw heart rate data in real time and preprocesses it to obtain a standardized HRV sequence. Then, the smartwatch's interface guides the user to select an appropriate label for their current emotional state. This embodiment allows selection of four basic emotions: pleasure, calmness, tension, and anxiety. Combined with the user's unique identifier ID, a user-specific sample dataset is constructed: "Standardized HRV sequence - emotion label - user unique identifier ID," and then encrypted and uploaded to the data processing terminal. The data processing terminal uses a cloud server as an example. The cloud server assigns an independent classifier training task to each user's individual sample dataset. During classifier training, the standardized HRV sequence is used as input, and the emotion label is used as output. The training iteratively obtains a personalized emotion recognition classifier corresponding to the user's unique identifier ID, achieving personalized customization of the classifier. This is not limited by specific groups or scenarios and can adapt to users of different ages and diverse scenarios such as daily life, exercise, and sleep, significantly improving generalization ability.

[0020] Preferably, the classifier is implemented based on a machine learning model.

[0021] The classifier used in this embodiment is based on a machine learning model. What these machine learning models have in common is that they can learn the mapping relationship between input features and output categories through training, thereby making judgments on new input data.

[0022] Preferably, using the heart rate variability sequence as input and the emotion label as output, the assigned classifier is trained to obtain a user-specific emotion recognition classifier, including: The machine learning model is a support vector machine; The first and second parameters of the support vector machine are optimized by a grid search algorithm to generate parameter combinations. The optimal parameter combination is determined from the parameter combinations. The first parameter is used to balance the classification margin and the misclassification penalty, and the second parameter is used to adjust the distribution density of the heart rate variability sequence in the feature space. Construct a support vector machine for the user identifier based on the optimal parameter combination; Based on the training subset of the user individual sample dataset, with the heart rate variability sequence as input and the emotion label as output, a support vector machine for the user identifier is trained, and the accuracy is evaluated based on the evaluation subset of the user individual sample dataset. In response to the accuracy being greater than or equal to a preset accuracy, the current support vector machine is set as the dedicated emotion recognition support vector machine for the user identifier.

[0023] The machine learning model in this embodiment can employ a Support Vector Machine (SVM) based on the Radial Basis Function (RBF) kernel function. It uses standardized HRV sequences as feature input and sentiment labels as output to train the SVM. Since the relationship between sentiment labels and HRV sequence features is usually not linear, the RBF kernel function automatically maps the HRV sequence features to a high-dimensional space. During training, the first and second parameters need to be optimized. The first parameter is the penalty parameter C, and the second parameter is the kernel function parameter γ. Both are derived from the classic parameter system of SVM classification algorithms. C originates from the constraints of soft-margin SVM, specifically: Where w is the normal vector of the classification hyperplane. As a slack variable, the penalty parameter C is used to balance the classification margin and misclassification penalty. γ, derived from the Gaussian distribution parameter of the radial basis kernel function, is used to adjust the distribution density of the standardized HRV sequence in the high-dimensional feature space. During the support vector machine training process, the penalty parameter C and the kernel function parameter γ are optimized using a grid search algorithm. These two parameters are combined to determine the optimal parameter combination from all possible combinations. Then, a support vector machine for user identifiers is constructed using this optimal parameter combination. The user identifier support vector machine is trained on a pre-defined training subset from the user individual sample dataset. During training, the non-linear classification of the standardized HRV sequence features and emotion labels is achieved by maximizing the classification margin. After training, the performance of the trained user identifier support vector machine is evaluated using a pre-defined evaluation subset from the user individual sample dataset, such as a validation set. The preset accuracy is 85%. If the accuracy is greater than or equal to 85%, the currently trained user identifier support vector machine is set as a dedicated emotion recognition support vector machine for that user identifier and can be distributed to mobile terminals for user use. If the accuracy is less than 85%, it indicates that the performance of the support vector machine (SVM) for user identifiers constructed with the optimal parameter combination is insufficient. Further optimization of the penalty parameter C and kernel function parameter γ is needed until a suitable emotion recognition SVM for user identifiers is obtained. The first parameter (classification margin and misclassification penalty balance parameter) and the second parameter (feature space distribution density adjustment parameter) are optimized using a grid search algorithm to accurately select the optimal parameter combination, improving the fitting accuracy of the SVM to the mapping relationship between user HRV sequences and emotion labels. Combined with a dual verification mechanism of "training subset training + evaluation subset verification," the final generated dedicated SVM meets the preset accuracy requirements, further improving the accuracy and stability of emotion recognition. Preferably, the first and second parameters of the SVM are optimized using a grid search algorithm to generate parameter combinations, and the optimal parameter combination is determined from these combinations, including: Set the search ranges for the first parameter and the second parameter respectively; Construct a corresponding parameter grid based on the search range, and combine the first parameter and the second parameter in the parameter grid to generate a parameter combination; The support vector machine is trained for each parameter combination separately, and the accuracy is evaluated by cross-validation. The optimal parameter combination is determined based on the highest accuracy.

[0024] The specific process of optimizing the penalty parameter C and kernel function parameter γ using the grid search algorithm in this embodiment is as follows: The search range for the penalty parameter C is set to [0.01, 0.1, 1, 10, 100], and the search range for the kernel function parameter γ is set to [0.001, 0.01, 0.1, 1, 10]. A 5*5 parameter grid is constructed, generating 25 different parameter combinations. A support vector machine for each parameter combination is trained using user identifiers. For each parameter combination, k-fold cross-validation (e.g., k=5) is used to evaluate its performance: the user individual sample dataset is randomly divided into k subsets. One subset is used as the validation set, and the remaining k-1 subsets are used as the training set. The support vector machine is trained, and its accuracy on the validation set is calculated. After repeating this k times, the average of the k fold accuracies is used as the final performance estimate of the parameter combination. The parameter combination corresponding to the highest average of the k fold accuracies is determined as the optimal parameter combination. Those skilled in the art will understand that the number of folds k in the cross-validation can be adjusted according to the data size. By setting the parameter search range, constructing a parameter grid, combining verification, and selecting the best accuracy, the key parameters of the support vector machine are systematically optimized, thereby ensuring the performance stability of the dedicated classifier and improving the consistency of emotion recognition results.

[0025] Preferably, the process of training an assigned classifier using the heart rate variability sequence as input and the emotion label as output to obtain a user-specific emotion recognition model further includes: The machine learning model is a neural network model; Based on the training subset of the user individual sample dataset, the heart rate variability sequence is input into the neural network model to extract local temporal features; Using the emotion tag as output, the neural network model trained on the user identifier nonlinearly maps the extracted local temporal features to the emotion category probability distribution space; The neural network model is verified to meet the preset performance index based on the evaluation subset in the user individual sample dataset. In response to meeting the preset performance index, the current neural network model is set as the exclusive emotion recognition neural network model of the user identifier.

[0026] The machine learning model in this embodiment can also employ a neural network model, such as a Convolutional Neural Network (CNN) or a Recurrent Neural Network (RNN). Taking a CNN as an example, the CNN in this embodiment has three hidden layers, each of which may sequentially include a one-dimensional convolutional layer, a pooling layer, a batch normalization layer, and a fully connected layer. The one-dimensional convolutional layer is responsible for extracting local temporal features from the standardized HRV sequence. The pooling and batch normalization layers are used to reduce the dimensionality of the extracted local temporal features, suppress overfitting, and stabilize the data distribution. Finally, the fully connected layer fuses the local temporal features to construct a non-linear mapping relationship between the standardized HRV sequence and the sentiment label. Taking a recurrent neural network as an example, the recurrent neural network in this embodiment has three hidden layers, each of which may sequentially include an LSTM layer, a dropout layer, and a fully connected layer. The LSTM layer is used to capture long-term dependent features of the standardized HRV sequence, the dropout layer is used to suppress overfitting, and the fully connected layer integrates long-term dependent features and enhances the discriminative power of the sentiment label.

[0027] Before training the neural network model, the standardized HRV sequences in the user individual sample dataset are used as input, and the sentiment labels after one-hot encoding are used as output. The dataset is divided into training subset and evaluation subset according to the proportion. The training subset includes the training set, and the evaluation subset may include the validation set and the test set. The input layer matches multi-dimensional feature vectors, and the output layer has 4 neurons. Xavier initialization of weights and a bias term of 0.1 are used for training. The total number of iterations is 50-200 rounds. For the first 50 rounds, training data is input in batches of 16, and accuracy is evaluated using a validation set every 10 rounds. From rounds 51 to 200, the validation set accuracy is evaluated after each round. Convergence is triggered if the improvement is less than 0.5% for five consecutive rounds or if the accuracy is greater than or equal to 85%. During this period, the learning rate decays to 80% every 20 rounds. After training, the performance of the trained neural network model is evaluated using a test set. The preset performance metric is an accuracy greater than or equal to 85%. If the accuracy is greater than or equal to 85%, the trained neural network model is set as a user-identified emotion recognition neural network model and can be distributed to mobile devices for user use. If the accuracy is less than 85%, the number of neurons in the hidden layer is adjusted, and retraining is performed. This embodiment utilizes the nonlinear mapping capability of neural networks to effectively extract local temporal features of HRV sequences, adapting to the temporal data characteristics of HRV sequences. Compared with traditional machine learning models, it is better able to capture subtle correlations in the data. By combining the process of "feature extraction - emotion category probability mapping - evaluation subset verification", it ensures that the neural network model meets the preset performance indicators and improves the model's ability to process complex HRV sequence data.

[0028] Preferably, the process of training the neural network model of the user identifier includes: The difference between the sentiment labels output by the neural network model and the sentiment labels in the evaluation subset is evaluated using the cross-entropy loss function. The optimizer updates the parameters of the neural network model for the user identifier based on the differences.

[0029] In this embodiment, the process of training the neural network model calculates the prediction error using the cross-entropy loss function, which calculates the difference between the emotion label output by the neural network and the true emotion label corresponding to the evaluation subset. The weights are updated using the Adam optimizer based on backpropagation of the difference. The cross-entropy loss function evaluates the difference between the predicted label and the true label, accurately quantifying the training error of the neural network model. Combined with the optimizer's parameter update mechanism, efficient iterative optimization of the model parameters is achieved, accelerating the convergence speed of the neural network model, improving model training efficiency, and ensuring the model's prediction accuracy, thus ensuring that the final generated dedicated neural network model possesses excellent emotion recognition performance.

[0030] Preferably, after training the assigned classifier, the method further includes: The trained classifier is evaluated using a performance evaluation subset, and the optimal user-identified emotion recognition classifier is determined according to a preset evaluation criterion. The data processing terminal performs lightweight processing on the exclusive emotion recognition classifier of the optimal user identifier based on the data processing capability of the mobile terminal.

[0031] When assigning independent classifier training tasks, support vector machine (SVM) training tasks and neural network model training tasks can be assigned separately. Once the user-identified emotion recognition SVM and neural network model are trained, a performance evaluation subset can be used to evaluate the performance of these two classifiers, and the optimal emotion recognition classifier can be determined according to preset evaluation criteria. It should be noted that the performance evaluation subset and evaluation criteria can be adjusted by those skilled in the art based on actual performance requirements. Taking a smartwatch as the mobile terminal and a cloud server as the data processing end as an example, after determining the optimal emotion recognition classifier, the cloud server, considering the computing power limitations of the smartwatch, performs pruning optimization and quantization on the optimal emotion recognition classifier and transmits it to the smartwatch in encrypted form. The smartwatch completes decompression and deployment. After real-time collection and standardization of raw heart rate data, it inputs the deployed optimal emotion recognition classifier and outputs the corresponding emotion recognition result. The emotion recognition result can interact with other modules of the smartwatch. For example, the smartwatch can send the emotion recognition result to the interaction module, driving it to execute corresponding feedback, achieving precise linkage between emotion and interaction. The optimal classifier is selected through a performance evaluation subset to ensure that the classifier delivered to the mobile terminal is the current best classifier, thereby improving the emotion recognition effect. Simultaneously, lightweight processing is performed based on the data processing capabilities of mobile terminals, reducing model size and computational power consumption, perfectly adapting to the computing power, storage, and power consumption limitations of edge devices such as smartwatches, avoiding issues such as response latency and shortened battery life, and ensuring a real-time interactive experience. On the other hand, the cloud server establishes a dedicated classifier library for each user identifier, storing the trained, personalized emotion recognition classifier and recording version information, including training time, the number of individual user sample datasets, and recognition accuracy.

[0032] The smartwatch continuously collects feedback data consisting of "standardized HRV sequence - emotion label - emotion recognition result." Every 50-100 valid samples, it is automatically encrypted and uploaded to the cloud server. Based on these new valid samples, the cloud server incrementally trains a dedicated emotion recognition classifier. Once training is complete, the optimized classifier is pushed back to the smartwatch, enabling continuous iteration and upgrades. It should be noted that valid samples can be filtered using a data evaluation script integrated into the smartwatch.

[0033] The above methods offer significant technical advantages: First, by constructing a user-specific emotion recognition classifier that adapts to the individual specificity and dynamic changes of HRV sequences, the accuracy of emotion recognition in real-world scenarios can reach over 75%, effectively avoiding misjudgments and missed judgments. Second, the classifier is trained based on real-world user scenario data, making it adaptable to people of all ages and diverse scenarios such as daily life, exercise, and sleep, with significantly better generalization capabilities than general models. Third, relying on a dynamic optimization mechanism of "incremental training - model iteration," the classifier can continuously learn the user's HRV sequences and emotion mapping patterns, achieving a personalized emotional companionship experience upgrade. Fourth, adopting a lightweight design strategy, the classifier is small in size, fast in response, and low in power consumption, perfectly adapting to the computing power, storage, and power consumption limitations of mobile terminals, ensuring real-time interactive effects. Fifth, it supports flexible switching between various machine learning models such as neural network models and support vector machines, and reserves a multimodal data fusion interface, possessing excellent technical scalability.

[0034] Example 2 Based on the principles described in Embodiment 1, a dual-end collaborative emotion recognition device 100 based on HRV sequences is proposed, such as... Figure 2 As shown, it includes: The data acquisition module 110 is used to acquire the user's heart rate variability sequence through a mobile terminal; The receiving model module 120 is used for the mobile terminal to send a classifier acquisition request to the data processing terminal and receive the corresponding user-specific emotion recognition classifier that is continuously updated based on the user's individual sample dataset. The data analysis module 130 is used to input the heart rate variability sequence into the user-specific emotion recognition classifier and output the corresponding emotion recognition result.

[0035] Example 3 Furthermore, an electronic device is proposed, comprising: processor; Memory used to store processor-executable instructions; The processor is configured to implement a dual-end collaborative emotion recognition method based on HRV sequences in Embodiment 1 when executing executable instructions.

[0036] It should be noted that all directional indications (such as up, down, left, right, front, back, etc.) in the embodiments of the present invention are only used to explain the relative positional relationship and movement of each component in a certain specific posture (as shown in the figure). If the specific posture changes, the directional indication will also change accordingly.

[0037] Furthermore, it should be noted that the use of terms such as "first," "second," and "a" in this invention is for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this invention, "a plurality of" means at least two, such as two, three, etc., unless otherwise explicitly specified. The terms "connection," "fixed," etc., should be interpreted broadly. For example, "fixed" can mean a fixed connection, a detachable connection, or an integral part; it can mean a mechanical connection or an electrical connection; it can mean a direct connection or an indirect connection through an intermediate medium; it can mean the internal communication of two elements or the interaction between two elements, unless otherwise explicitly specified. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0038] Furthermore, the technical solutions of the various embodiments of the present invention can be combined with each other, but only if they are feasible for those skilled in the art. If the combination of technical solutions is contradictory or cannot be implemented, it should be considered that such combination of technical solutions does not exist and is not within the scope of protection claimed by the present invention.

[0039] The specific embodiments described herein are merely illustrative of the spirit of the invention. Those skilled in the art to which this invention pertains may make various modifications or additions to the described specific embodiments or use similar methods to substitute them, without departing from the spirit of the invention or exceeding the scope defined by the appended claims.

Claims

1. A two-end collaborative emotion recognition method based on HRV sequences, characterized in that, include: Obtain the user's heart rate variability sequence through mobile terminals; The mobile terminal sends a classifier acquisition request to the data processing terminal and receives the corresponding user-specific emotion recognition classifier that is continuously updated based on the user's individual sample dataset. The heart rate variability sequence is input into the user-specific emotion recognition classifier, and the corresponding emotion recognition result is output.

2. The method according to claim 1, characterized in that, A user-specific emotion recognition classifier, continuously updated based on individual user sample datasets, includes: The user's heart rate variability sequence and corresponding emotion tags are obtained through mobile terminals; The mobile terminal constructs a user individual sample dataset based on the heart rate variability sequence, the emotion tag, and the user identifier, and sends it to the data processing terminal; The data processing terminal assigns classifier training tasks to the user individual sample dataset; Using the heart rate variability sequence as input and the emotion label as output, the assigned classifier is trained to obtain a user-specific emotion recognition classifier.

3. The method according to claim 2, characterized in that, The classifier is implemented based on a machine learning model.

4. The method according to claim 3, characterized in that, Using the heart rate variability sequence as input and the emotion label as output, the assigned classifier is trained to obtain a user-specific emotion recognition classifier, including: The machine learning model is a support vector machine; The first and second parameters of the support vector machine are optimized by a grid search algorithm to generate parameter combinations. The optimal parameter combination is determined from the parameter combinations. The first parameter is used to balance the classification margin and the misclassification penalty, and the second parameter is used to adjust the distribution density of the heart rate variability sequence in the feature space. Construct a support vector machine for the user identifier based on the optimal parameter combination; Based on the training subset of the user individual sample dataset, with the heart rate variability sequence as input and the emotion label as output, a support vector machine for the user identifier is trained, and the accuracy is evaluated based on the evaluation subset of the user individual sample dataset. In response to the accuracy being greater than or equal to a preset accuracy, the current support vector machine is set as the dedicated emotion recognition support vector machine for the user identifier.

5. The method according to claim 4, characterized in that, The first and second parameters of the support vector machine are optimized using a grid search algorithm to generate parameter combinations. The optimal parameter combination is then determined from these combinations, including: Set the search ranges for the first parameter and the second parameter respectively; Construct a corresponding parameter grid based on the search range, and combine the first parameter and the second parameter in the parameter grid to generate a parameter combination; The support vector machine is trained for each parameter combination separately, and the accuracy is evaluated by cross-validation. The optimal parameter combination is determined based on the highest accuracy.

6. The method according to claim 3, characterized in that, Using the heart rate variability sequence as input and the emotion label as output, the assigned classifier is trained to obtain a user-specific emotion recognition model, which also includes: The machine learning model is a neural network model; Based on the training subset of the user individual sample dataset, the heart rate variability sequence is input into the neural network model to extract local temporal features; Using the emotion tag as output, the neural network model trained on the user identifier nonlinearly maps the extracted local temporal features to the emotion category probability distribution space; The neural network model is verified to meet the preset performance index based on the evaluation subset in the user individual sample dataset. In response to meeting the preset performance index, the current neural network model is set as the exclusive emotion recognition neural network model of the user identifier.

7. The method according to claim 6, characterized in that, The process of training the neural network model of the user identifier includes: The difference between the sentiment labels output by the neural network model and the sentiment labels in the evaluation subset is evaluated using the cross-entropy loss function. The optimizer updates the parameters of the neural network model for the user identifier based on the differences.

8. The method according to claim 2, characterized in that, After training the assigned classifier, the following is also included: The trained classifier is evaluated using a performance evaluation subset, and the optimal user-identified emotion recognition classifier is determined according to a preset evaluation criterion. The data processing terminal performs lightweight processing on the exclusive emotion recognition classifier of the optimal user identifier based on the data processing capability of the mobile terminal.

9. A dual-end collaborative emotion recognition device based on HRV sequences, characterized in that, include: The data acquisition module is used to acquire the user's heart rate variability sequence via the mobile terminal; The receiving model module is used for the mobile terminal to send a classifier acquisition request to the data processing terminal and receive the corresponding user-specific emotion recognition classifier that is continuously updated based on the user's individual sample dataset. The data analysis module is used to input the heart rate variability sequence into the user-specific emotion recognition classifier and output the corresponding emotion recognition result.

10. An electronic device, characterized in that, include: processor; Memory used to store processor-executable instructions; The processor is configured to implement the dual-end collaborative emotion recognition method based on HRV sequences as described in any one of claims 1-8 when executing executable instructions.

Images