Suspension control method and device based on emotional interaction, equipment and storage medium

Abstract

The application relates to a suspension control method and device based on emotional interaction, equipment and a storage medium. The suspension control method based on emotional interaction comprises the following steps: determining a current scene according to scene characteristics, and identifying user emotion through an emotional recognition model corresponding to the current scene, wherein different emotional recognition models are used to identify significant emotions in corresponding scenes; if the significant emotion is identified, determining a suspension control strategy corresponding to the significant emotion in the current scene based on an emotional strategy mapping table, and adjusting the current suspension parameter according to the suspension control strategy; if the non-significant emotion is identified, when the current scene is a dynamic scene, determining an optimization target corresponding to the current scene, and adjusting the current suspension parameter according to the optimization target corresponding to the current scene. Through the method of the application, the problems of complex emotional recognition model and insufficient strategy coverage in the conventional emotional control method can be improved, and the user experience is improved.

Description

Technical Field

[0001] This application relates to the field of active suspension control technology for vehicles, and in particular to a suspension control method, device, equipment, and storage medium based on emotion interaction. Background Technology

[0002] Current mainstream suspension control technologies include adjusting the suspension based on user preferences. This technology relies on preset fixed preferences or infers user preferences from driving data such as pedal input. However, this method lacks real-time performance and cannot directly reflect the user's actual suspension control needs.

[0003] To address this, active suspension control can be implemented based on emotional interaction. Unlike preferences, emotions can more directly, realistically, and quickly reflect a user's current state and needs. Through emotional interaction, the user's immediate feelings during driving can be captured more sensitively, providing a more targeted basis for suspension control adjustments. This method not only improves response speed but also, to some extent, compensates for the lag inherent in traditional preference inference methods.

[0004] However, active suspension control based on emotion interaction often fails to achieve the desired results in practical applications because emotions are complex and variable, with only subtle differences between them. To cover all potential emotions in all scenarios and achieve accurate recognition, the emotion recognition model must have extremely high accuracy. However, the emotion recognition process of complex models is time-consuming and computationally overloaded, making it unnecessary in certain scenarios. Simultaneously, suspension control strategies also struggle to cover all emotional states. Even if the user's emotion is accurately identified, there may be no corresponding strategy available, especially in dynamic scenarios. Insufficient strategy coverage can significantly impact the user's driving experience. Therefore, conventional emotion control methods still suffer from problems such as complex emotion recognition models and insufficient strategy coverage, leading to poor user experience. Summary of the Invention

[0005] Based on this, a suspension control method, device, equipment, and storage medium based on emotion interaction are provided to improve the poor user experience caused by the complexity of emotion recognition models and insufficient strategy coverage in conventional emotion control methods.

[0006] In a first aspect, this application provides a suspension control method based on emotion interaction. The method includes: determining the current scene based on scene characteristics, and identifying user emotions through an emotion recognition model corresponding to the current scene, wherein different emotion recognition models are used to identify salient emotions in the corresponding scene; if a salient emotion is identified, determining the suspension control strategy corresponding to the salient emotion in the current scene based on an emotion strategy mapping table, and adjusting the current suspension parameters according to the suspension control strategy; if a non-salient emotion is identified, when the current scene is a dynamic scene, determining the optimization target corresponding to the current scene, and adjusting the current suspension parameters according to the optimization target corresponding to the current scene.

[0007] In conjunction with the first aspect, in the first possible implementation of the first aspect, before performing the step of identifying user emotions through the emotion recognition model corresponding to the current scene, the method further includes: analyzing the salient emotions of users in each scene based on historical user emotion data in each scene; establishing the association between each scene and the corresponding salient emotions, and obtaining training data for each scene based on historical user emotion data, wherein the training data includes emotion data and emotion labels; and training the neural network using the training data so that the neural network, after training, can identify whether the user emotion in the corresponding scene is a salient emotion associated with the corresponding scene.

[0008] In conjunction with the first possible implementation of the first aspect, in the second possible implementation of the first aspect, the aforementioned step of training a neural network using the training data so that the neural network, after training, can identify whether a user's emotion in a corresponding scenario is a salient emotion associated with the corresponding scenario includes: inputting the emotion data of the training data for each scenario into the corresponding neural network; determining the user's emotion based on the probability that the emotion data output by the neural network belongs to a salient emotion; if the user's emotion does not match the emotion label of the training data, optimizing the neural network to obtain an emotion recognition model that can output a correct result consistent with the emotion label based on the emotion data, wherein the emotion label includes salient or non-salient emotions in the corresponding scenario.

[0009] In conjunction with the first aspect, in the third possible implementation of the first aspect, the aforementioned step of identifying user emotions through an emotion recognition model corresponding to the current scene includes: collecting a user dataset including facial expression data, voice data, and physiological data; generating a fusion feature of facial expression data, voice data, and physiological data based on the user dataset; inputting the fusion feature as emotion data into the emotion recognition model to output the probability that the user emotion is a significant emotion in the corresponding scene, and determining the user emotion based on the probability.

[0010] In conjunction with the third possible implementation of the first aspect, in the fourth possible implementation of the first aspect, the aforementioned step of inputting the fused features as emotional data into the emotional recognition model to output the probability of the user's emotion being a salient emotion in the corresponding scene, and determining the user's emotion based on the probability, includes: inputting the fused features as emotional data into the emotional recognition model to output the probability of the user's emotion being each of the at least one salient emotion in the corresponding scene; obtaining the salient emotion with the highest probability from the at least one salient emotion as the target emotion; if the probability of the target emotion is greater than a probability threshold, then determining the user's emotion as the target emotion; if the probability of the target emotion is less than or equal to the probability threshold, then determining the user's emotion as a non-salient emotion.

[0011] In conjunction with the first aspect, in the fifth possible implementation of the first aspect, after determining the optimization target corresponding to the current scenario and before adjusting the current suspension parameters according to the optimization target corresponding to the current scenario, the method further includes: monitoring the user's emotional data and evaluating the negative score of the user's emotions based on the emotional data; if the negative score is greater than the score threshold, it is determined that the current suspension control parameters need optimization, and the step of adjusting the current suspension parameters according to the optimization target corresponding to the current scenario is executed; if the negative score is less than or equal to the score threshold, the current suspension parameters are not adjusted.

[0012] In conjunction with the fifth possible implementation of the first aspect, in the sixth possible implementation of the first aspect, the aforementioned step of adjusting the current suspension parameters according to the optimization target corresponding to the current scenario includes: determining the optimization target corresponding to the current scenario based on the scenario target mapping table; adjusting the current suspension parameters according to the preset fine-tuning numerical gradient corresponding to the optimization target; if a decrease in the negative score of user emotion is detected, then the current suspension parameters are adjusted according to the optimization target corresponding to the current scenario until the negative score is less than or equal to the score threshold and then the adjustment is stopped; if the negative score of user emotion is detected to remain unchanged or increase, then the adjustment of the suspension parameters is stopped.

[0013] Secondly, this application provides a suspension control device, which includes: an identification unit, configured to determine the current scene based on scene characteristics and identify user emotions through an emotion recognition model corresponding to the current scene, wherein different emotion recognition models are used to identify salient emotions in the corresponding scene; a control unit, configured to determine the suspension control strategy corresponding to the salient emotion in the current scene based on an emotion strategy mapping table if the salient emotion is identified, and adjust the current suspension parameters according to the suspension control strategy; an optimization unit, configured to determine the optimization target corresponding to the current scene if the salient emotion is identified, when the current scene is a dynamic scene; the control unit is further configured to adjust the current suspension parameters according to the optimization target corresponding to the current scene.

[0014] In conjunction with the second aspect, in the first possible implementation of the second aspect, the suspension control device further includes a training unit, which is used to: analyze the salient emotions of users in each scenario based on historical user emotion data in each scenario; establish the correlation between each scenario and the corresponding salient emotions, and obtain training data for each scenario based on historical user emotion data, wherein the training data includes emotion data and emotion labels; and train the neural network using the training data so that the neural network can identify whether the user's emotion in the corresponding scenario is a salient emotion associated with the corresponding scenario after training.

[0015] In conjunction with the first possible implementation of the second aspect, in the second possible implementation of the second aspect, the aforementioned training unit is specifically used for: inputting the sentiment data of the training data of each scenario into the corresponding neural network; determining the user's sentiment based on the probability that the sentiment data output by the neural network belongs to a salient sentiment; if the user's sentiment does not match the sentiment label of the training data, optimizing the neural network to obtain a sentiment recognition model that can output a correct result consistent with the sentiment label based on the sentiment data, wherein the sentiment label includes salient or non-salient sentiments in the corresponding scenario.

[0016] In conjunction with the second aspect, in a third possible implementation of the second aspect, the aforementioned recognition unit is specifically used for: collecting a user dataset including facial expression data, voice data, and physiological data; generating a fusion feature of facial expression data, voice data, and physiological data based on the user dataset; inputting the fusion feature as emotion data into an emotion recognition model to output the probability that the user emotion is a significant emotion in the corresponding scene, and determining the user emotion based on the probability.

[0017] In conjunction with the third possible implementation of the second aspect, in the fourth possible implementation of the second aspect, the aforementioned identification unit is specifically used to: input the fused features as emotion data into the emotion recognition model, and output the probability of each salient emotion among at least one salient emotion in the corresponding scene; obtain the salient emotion with the highest probability from the at least one salient emotion as the target emotion; if the probability of the target emotion is greater than the probability threshold, then determine the user emotion as the target emotion; if the probability of the target emotion is less than or equal to the probability threshold, then determine the user emotion as a non-salient emotion.

[0018] In conjunction with the second aspect, in the fifth possible implementation of the second aspect, the aforementioned control unit is specifically used to: monitor the user's emotional data and evaluate the negative score of the user's emotions based on the emotional data; if the negative score is greater than the score threshold, determine that the current suspension control parameters need to be optimized, and execute the step of adjusting the current suspension parameters according to the optimization target corresponding to the current scenario; if the negative score is less than or equal to the score threshold, do not adjust the current suspension parameters.

[0019] In conjunction with the fifth possible implementation of the second aspect, in the sixth possible implementation of the second aspect, the aforementioned control unit is specifically used for: determining the optimization target corresponding to the current scene based on the scene target mapping table; adjusting the current suspension parameters according to the preset fine-tuning numerical gradient corresponding to the optimization target; if a decrease in the negative score of user emotion is detected, then the current suspension parameters are adjusted according to the optimization target corresponding to the current scene until the negative score is less than or equal to the score threshold and then the adjustment is stopped; if the negative score of user emotion is detected to remain unchanged or increase, then the adjustment of the suspension parameters is stopped.

[0020] Thirdly, this application provides a suspension control device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of any one of the first aspects to the first aspect.

[0021] Fourthly, this application provides a computer-readable storage medium, characterized in that the computer-readable storage medium stores a plurality of instructions adapted for loading by a processor and executing steps as described in any one of the first aspects to the first aspect.

[0022] In summary, this application provides a suspension control method, apparatus, device, and storage medium based on emotion interaction. Specifically, it employs scenario-specific emotion recognition models to identify salient emotions in corresponding scenarios. Salient emotions are those most likely to occur and require the most attention in a given scenario. Dedicated emotion recognition models can focus the recognition on salient emotions, reducing interference from non-salient emotions. Therefore, it can accurately identify emotions while reducing model complexity. Furthermore, after identifying whether a user's emotion is salient, this application can not only execute preset suspension control strategies for salient emotions covered by the strategy, but also optimize for non-salient emotions not covered by the strategy in dynamic scenarios according to the optimization objectives corresponding to the current scenario, thereby improving the problem of insufficient strategy coverage in dynamic scenarios. In general, this application can improve the problems of poor user experience caused by the complexity of emotion recognition models and insufficient strategy coverage in conventional emotion control methods. Attached Figure Description

[0023] Figure 1 This is a flowchart illustrating a suspension control method based on emotion interaction in one embodiment;

[0024] Figure 2 This is a schematic diagram of the suspension control device in one embodiment;

[0025] Figure 3 This is a schematic diagram of the suspension control device in one embodiment. Detailed Implementation

[0026] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0027] It should be noted that the illustrations provided in this embodiment are only schematic representations of the basic concept of this application. Therefore, the drawings only show components relevant to this application and are not drawn according to the actual number, shape, and size of components in implementation. In actual implementation, the form, quantity, and proportion of each component can be arbitrarily changed, and the component layout may also be more complex. The structures, proportions, sizes, etc., shown in the accompanying drawings are only used to complement the content disclosed in the specification for those skilled in the art to understand and read, and are not intended to limit the implementation conditions of this application. Therefore, they have no substantial technical significance. Any modification to the structure, change in the proportional relationship, or adjustment of the size, without affecting the effect and purpose that this application can produce, should still fall within the scope of the technical content disclosed in this application. At the same time, the terms such as "upper," "lower," "left," "right," "middle," and "one" used in this specification are only for clarity of description and are not intended to limit the scope of implementation of this application. Changes or adjustments in their relative relationships, without substantially changing the technical content, should also be considered within the scope of implementation of this application.

[0028] With the continuous improvement of automotive intelligence and the increasing demand from consumers for personalized driving experiences, the intelligent and personalized adjustment of vehicle suspension systems has become a key technological direction for enhancing overall vehicle comfort and driving pleasure. Traditional fixed-characteristic suspensions or single-mode adjustable suspensions are insufficient to meet the diverse needs of users in different scenarios. Therefore, developing intelligent suspension systems that can automatically identify user preferences and adjust suspension characteristics in real time is of great significance for optimizing user experience and enhancing product competitiveness.

[0029] Current mainstream personalized intelligent suspension technologies primarily rely on preset fixed preferences or the monitoring and analysis of user driving behavior to achieve preference recognition and suspension adjustment. Specifically, the system collects real-time user driving operation data (e.g., accelerator / brake pedal pressure, steering wheel angle, and yaw rate) through vehicle sensors (such as accelerator pedal position sensor, brake pedal pressure sensor, steering wheel angle sensor, and vehicle yaw rate sensor). Based on statistical analysis of this data (e.g., identifying tendencies towards aggressive or smooth driving), the system constructs a user driving style model and maps it to a preset suspension characteristic mode. Subsequently, the active or semi-active suspension system adjusts the damping force or vehicle height in real-time according to this mode setting, striving to make the vehicle's dynamic response match the inferred user preferences.

[0030] However, suspension adjustment technology based on user preferences lacks real-time performance and struggles to directly reflect the user's actual suspension control needs. On one hand, the system needs to accumulate driving behavior data over a certain period and in specific scenarios to infer user preferences. It cannot make immediate and proactive adjustments when the vehicle is stationary, at the start of a journey, or when user needs change instantaneously, exhibiting significant lag and passivity. This limits the practical application effectiveness of emotion recognition systems in vehicle dynamic control scenarios requiring rapid response, thus resulting in a lack of real-time performance. On the other hand, driving behaviors (such as rapid acceleration, sudden braking, and fast cornering) often simultaneously reflect the user's instantaneous driving intentions (such as overtaking and hazard avoidance) and long-term preferences, which are difficult to distinguish effectively. Temporary and situational intense operations are easily misidentified as the user's long-term preference for suspension control, leading to subsequent suspension adjustments that do not match actual needs. Therefore, it also suffers from the problem of failing to directly reflect the user's actual suspension control needs.

[0031] To address this, an emotion-based suspension control method can be adopted. However, the application of emotion interaction in the field of suspension control is not common because the emotion recognition model of conventional emotion-based suspension control methods is often very complex. It needs to accurately identify potential emotions in all scenarios. However, overly complex emotion recognition models have poor real-time performance, cannot provide timely feedback, and suffer from computational overload. Moreover, in certain scenarios, overly high-precision emotion recognition is unnecessary. After emotion recognition, there may be no strategy to execute, resulting in invalid emotion recognition and wasting computational resources. Therefore, the problems of complex emotion recognition models and insufficient strategy coverage still exist, leading to poor user experience.

[0032] This application offers a suspension control method based on emotion interaction that differs from conventional methods. Firstly, instead of employing a uniform, complex emotion recognition model, it sets different emotion recognition models for different scenarios. By identifying user emotions through the emotion recognition model corresponding to the current scenario, it can better identify salient emotions closely related to the scenario, reducing attention to insignificant emotions. This improves the accuracy and real-time performance of emotion recognition while mitigating computational overload and waste. Secondly, when a salient emotion is identified, the application adjusts the current suspension parameters according to the corresponding suspension control strategy for that emotion in the current scenario. This allows for the execution of different suspension control strategies under different scenarios or emotions, achieving refined suspension control and enhancing the user experience. Thirdly, when an insignificant emotion is identified and the current scenario is dynamic, the application adjusts the current suspension parameters according to the optimization objective corresponding to the current scenario, addressing the problem of insufficient strategy coverage leading to missing control strategies under insignificant emotions. In summary, the suspension control method based on emotion interaction provided in this application can quickly and accurately identify user emotions and accordingly execute refined and broad-coverage suspension control in the current scenario, significantly improving the user experience.

[0033] In short, the emotion recognition model provided in this application is a specialized model, not a general-purpose model. It does not need to accurately identify potential emotions in all scenarios; instead, it focuses on salient emotions that are more likely to occur in specific scenarios and require attention, reducing the focus on insignificant emotions. This allows the emotion recognition model to not only accurately identify emotions in specific scenarios but also to be simplified, thereby improving the problems of poor real-time performance and computational overload. Furthermore, this application can not only execute corresponding suspension control strategies for salient emotions covered by the policy but also optimize for insignificant emotions not covered by the policy in dynamic scenarios according to the optimization objective corresponding to the current scenario. Therefore, even in dynamic scenarios without policy coverage, suspension control can be optimized, improving the problem of insufficient policy coverage.

[0034] It should be noted that the suspension control method based on emotion interaction provided in this application can be executed by a suspension control device or suspension control equipment. The suspension control device or suspension control equipment can be any control device in a car that can implement the suspension control method based on emotion interaction of this application, such as a controller, processor, etc. In addition, the suspension control device or suspension control equipment can interact with other devices or equipment in other cars and execute the suspension control method based on emotion interaction proposed in this application.

[0035] refer to Figure 1 This is an embodiment of a suspension control method based on emotional interaction provided in this application. Next, this application will combine... Figure 1This paper describes the implementation process of an emotion-interaction-based suspension control method, using the suspension control device as the actuator. Specifically:

[0036] 110: Determine the current scene based on scene characteristics, and identify user emotions through the emotion recognition model corresponding to the current scene. Different emotion recognition models are used to identify the salient emotions of the corresponding scene.

[0037] 120: If a significant emotion is identified, the suspension control strategy corresponding to the significant emotion in the current scenario is determined based on the emotion strategy mapping table, and the current suspension parameters are adjusted according to the suspension control strategy;

[0038] In this context, salient emotions refer to those that are more likely to occur in a specific scenario and require special attention, while insignificant emotions refer to those that are less likely to occur in a specific scenario or do not require special attention. Insignificant emotions are all emotions other than salient emotions. For example, in the scenario of returning home, salient emotions include feelings of pleasure, while insignificant emotions include feelings of fear. In the scenario of off-roading, salient emotions include feelings of excitement, while insignificant emotions include feelings of calm.

[0039] 130: If the emotion is identified as non-significant, then when the current scene is a dynamic scene, determine the optimization objective corresponding to the current scene, and adjust the current suspension parameters according to the optimization objective corresponding to the current scene.

[0040] In step 110, the scene features include environmental features and vehicle features. Environmental features include at least one of location information, terrain information, weather information, personnel information, and activity information. Vehicle features include vehicle motion features, which include both moving and stationary states. The location information in the scene features can be obtained through GPS signal positioning combined with an in-vehicle navigation map. For example, the vehicle's latitude and longitude information is first obtained based on GPS, and then the location information (garage, suburbs, etc.) corresponding to the vehicle's latitude and longitude information is found on the in-vehicle navigation map. The terrain information in the scene features can be obtained through GPS signal positioning combined with pre-stored Digital Elevation Model (DEM) data or Digital Terrain Model (DTM) data in the in-vehicle navigation map. The analysis of DEM or DTM data is as follows: For example, first, the vehicle's latitude and longitude information is obtained based on GPS. Then, the corresponding DEM or DTM data in the navigation map is queried. Terrain information is calculated based on the DEM or DTM data, including altitude, slope, aspect, curvature, and ruggedness. Weather information in scene features can be obtained directly through sensor detection or by combining GPS signal positioning with cloud-based data. For example, first, the vehicle's latitude and longitude information is obtained based on GPS. Then, the weather information corresponding to the vehicle's latitude and longitude information is queried from the cloud. Weather information includes temperature, humidity, precipitation, wind speed, and visibility. Personnel information can be obtained through camera recognition, including facial expression data, appearance data, and height data. Activity information can be obtained by analyzing the audio and video data from external cameras and microphones. For example, audio and video data is collected from external cameras and microphones, and then analyzed to obtain activity information (parties, events, and film festivals, etc.).

[0041] For example, determining the current scene based on scene characteristics includes: when the location information shows that the vehicle is in a garage, the terrain information shows that the vehicle is on a flat road, and the motion characteristics show that the vehicle is stationary, the current scene is determined to be a homecoming scene; when the location information shows that the vehicle is in the suburbs, the terrain information shows that the road is rugged and bumpy, and the motion characteristics show that the vehicle is in motion, the current scene is determined to be an off-road scene.

[0042] After determining the current scene, the user's emotional data in the current scene is input into the emotion recognition model, so that the emotion recognition model processes the emotional data and outputs the probability that the user's emotion is a significant emotion in the current scene. The user's emotion is then recognized based on this probability.

[0043] Since different emotion recognition models are used to identify salient emotions in corresponding scenarios, it can be seen that the emotion recognition model used in this application is specifically designed for different scenarios and is dedicated to identifying salient emotions in corresponding scenarios. Therefore, it can identify salient emotions in corresponding scenarios more quickly and accurately, reducing model complexity.

[0044] Regarding step 110, this application also proposes a specific and implementable method, wherein the aforementioned step of identifying user emotions through an emotion recognition model corresponding to the current scenario includes:

[0045] 111: Collect user datasets including facial expression data, voice data, and physiological data;

[0046] 112: Generate fusion features of facial expression data, voice data, and physiological data based on user datasets;

[0047] 113: Input the fused features as emotion data into the emotion recognition model, output the probability of the user's emotion being the salient emotion of the corresponding scene, and determine the user's emotion based on the probability.

[0048] Among them, emotional data is multidimensional data that integrates facial expression data (such as video streams), voice data (such as audio streams), and physiological data (such as temperature / breathing information), which can comprehensively reflect the user's emotions. For example, spatiotemporal alignment is performed—using the timestamps of video frames of facial expression data as a reference, the speech data and physiological data are segmented by sliding a 0.5-second window, and aligned to the nearest frame through nearest neighbor interpolation to obtain trimodal information; then, the cross-correlation matrix of the trimodal information is calculated, and the cross-correlation matrix is ​​normalized by the softmax normalization function to generate attention weights. The attention weights are used to strengthen cross-modal correlation features. Finally, the cross-correlation matrix and attention weights are combined to output fused features (three-dimensional joint features). The fused features can be directly used as emotional data input into the emotion recognition model, or the fused features can be post-processed before being used as emotional data input into the emotion recognition model. For example, post-processing includes dimensionality reduction of the fused features through principal component analysis (PCA) to remove redundant information, retain the main features, and reduce computational complexity.

[0049] Regarding step 113, this application also provides a specific implementation method. The aforementioned step of inputting the fused features as emotional data into the emotional recognition model to output the probability of the user's emotion being a salient emotion in the corresponding scene, and determining the user's emotion based on the probability, includes: inputting the cross-relation matrix as emotional data into the emotional recognition model to output the probability of each salient emotion among at least one salient emotion in the corresponding scene; obtaining the salient emotion with the highest probability from the at least one salient emotion as the target emotion; if the probability of the target emotion is greater than a probability threshold, then determining the user's emotion as the target emotion; if the probability of the target emotion is less than or equal to the probability threshold, then determining the user's emotion as a non-salient emotion. For example, in a specific scenario, there are multiple salient emotions, including salient emotion A, salient emotion B, and salient emotion C. The cross-relation matrix is ​​input into the emotion recognition model for that specific scenario. The output shows that the user's emotion belongs to salient emotion A, salient emotion B, and salient emotion C respectively, with probabilities of 0.34, 0.22, and 0.44. The salient emotion C with the highest probability is selected. The probability of salient emotion C (0.44) is compared with a probability threshold of 0.5. Since the probability of salient emotion C is less than the probability threshold, the user's emotion is determined to be non-salient. However, if the probabilities of the user's emotion belonging to salient emotion A, salient emotion B, and salient emotion C are 0.28, 0.21, and 0.51 respectively, and the probability of salient emotion C (0.51) is greater than the probability threshold of 0.5, then the user's emotion is determined to be salient emotion C.

[0050] In this system, at least one salient emotion is defined for each scenario. When multiple salient emotions exist in a specific scenario, the salient emotion with the highest probability is selected as the target emotion. When a single salient emotion exists in a specific scenario, that single salient emotion, being the one with the highest probability, is directly selected as the target emotion. Furthermore, after outputting the probability of each salient emotion among the at least one salient emotion in the corresponding scenario, the average of the emotion probabilities of each salient emotion output N consecutively can be calculated. The salient emotion with the highest probability from the average of these N consecutive outputs is then selected as the target emotion. N is an integer greater than 1. Averaging is used to eliminate instantaneous noise.

[0051] After obtaining the target emotion, this application determines whether the target emotion is a salient emotion of the user in the current scenario by comparing the probability of the target emotion with a probability threshold. This reduces the probability of emotion recognition errors in the emotion recognition model, improves the accuracy of emotion recognition, and enhances the adaptability and reliability of suspension control in different scenarios. When the probability of the target emotion is greater than the probability threshold, the target emotion is determined to be a salient emotion of the user; when the probability of the target emotion is less than or equal to the probability threshold, the target emotion is determined to be a non-salient emotion of the user. The probability threshold is derived from a large amount of experimental data and is a key indicator that effectively reflects the salientity of user emotions. For example, it estimates the minimum probability of consistency between multiple sets of target emotions and their corresponding actual salient emotions by statistically analyzing whether they are consistent with each other, and uses this minimum probability as the probability threshold.

[0052] Regarding step 120, if the user's emotion is identified as a salient emotion, the suspension control strategy corresponding to the salient emotion in the current scenario is determined based on the emotion policy mapping table, and the current suspension parameters are adjusted according to the suspension control strategy. Different scenarios correspond to different emotion policy mapping tables, which record the suspension control strategies corresponding to salient emotions in different scenarios.

[0053] For example, in a homecoming scenario where the vehicle is stationary and in a family parking lot, if the user's emotion is the pleasant emotion specified for the homecoming scenario, then based on the emotion strategy mapping table corresponding to the homecoming scenario, a suspension control strategy of "the suspension making three small up-and-down movements" is obtained to simulate the vehicle greeting the user; in a party scenario where the vehicle is stationary and in a party venue, if the user's emotion is the excited emotion specified for the party scenario, then based on the emotion strategy mapping table corresponding to the party scenario, a suspension control strategy of "the suspension dancing according to the rhythm of the music" is obtained to simulate the scenario of the vehicle and the user dancing together; in a children's scenario where the vehicle is stationary and there are children inside, if the user's emotion... For scenarios involving children and specific emotions of anxiety, a suspension control strategy of "slowly rocking the entire suspension up and down" is obtained based on the emotional strategy mapping table corresponding to the children's scenario to soothe the child's mood. At the same time, the vehicle plays soothing music such as lullabies and activates the seat massage function. In off-road scenarios where the vehicle is moving and the road surface is rough and bumpy in the suburbs, if the user's emotion is the excitement specified in the off-road scenario, a suspension control strategy of "enhancing the handling feedback of the suspension" is obtained based on the emotional strategy mapping table corresponding to the off-road scenario. For example, by increasing the spring stiffness, compression damping, rebound damping, and anti-roll bar stiffness, the handling feel is enhanced, and the driver receives stronger feedback.

[0054] It's important to note that in the suspension control strategy of "enhancing suspension handling feedback," the anti-roll bar does not directly provide feedback on vehicle motion within the suspension. The anti-roll bar reduces body roll by resisting suspension deformation. Increased stiffness leads to a smaller body roll angle during cornering, resulting in more stable vehicle dynamics for the driver and making the grip limit more easily perceived, thus indirectly enhancing handling feel and feedback.

[0055] For example, taking the suspension control strategy of "three small up-and-down undulations" as an example, this explains how to adjust the suspension parameters to achieve the suspension control strategy. For example, the damping coefficient is adjusted to a value much smaller than the critical damping coefficient to ensure that the suspension is in an underdamped state; the target damping ratio is set between 0.1 and 0.3, with 0.2 being a typical starting point, to adjust the number of oscillations; in addition, the damping can be adjusted according to the actual number of oscillations. If the actual number of oscillations is more than 3 (the decay is too slow), the damping is slightly increased; if the actual number of oscillations is less than 3 (the decay is too fast), the damping is further decreased. This can be adjusted multiple times to ensure that the actual number of oscillations is 3.

[0056] Regarding step 130, if the emotion is identified as non-significant, then when the current scenario is dynamic, the optimization objective corresponding to the current scenario is determined, and the current suspension parameters are adjusted according to the optimization objective. Optimization objectives include safety-oriented, comfort-oriented, and performance-oriented, etc. The optimization objectives may differ for different scenarios. For example, in off-road scenarios, the user's emotion is not significant but non-significant, such as tension. In this case, the current suspension parameters are optimized according to the safety-oriented optimization objective corresponding to the off-road scenario. For example, by optimizing damping (ensuring tire contact with the ground, especially ensuring sufficient rebound damping and avoiding excessive compression damping), reducing anti-roll bar stiffness, and reducing spring stiffness, the sense of security is improved, thereby mitigating the user's non-significant emotions in the current scenario.

[0057] In step 130, after determining the optimization objective corresponding to the current scenario, and before adjusting the current suspension parameters according to the optimization objective corresponding to the current scenario, this application also provides an implementation method:

[0058] 131: Monitor users' emotional data and evaluate the negative rating of users' emotions based on the emotional data;

[0059] 132: If the negative score is greater than the score threshold, it is determined that the current suspension control parameters need to be optimized, and the step of adjusting the current suspension parameters according to the optimization target corresponding to the current scenario is executed;

[0060] 133: If the negative rating is less than or equal to the rating threshold, the current suspension parameters are not adjusted.

[0061] In this implementation, suspension parameters are not blindly optimized when the user's emotion is insignificant, because not all insignificant emotions require suspension optimization. Overly sensitive and frequent suspension optimization can degrade the user experience. Therefore, this implementation only optimizes the suspension when the user is experiencing a negative insignificant emotion. Specifically, this application determines whether the user's insignificant emotion is negative by evaluating the negative score of the user's emotion, and then determines whether the current suspension control parameters need optimization. If the negative score is greater than a scoring threshold, it is determined that the user's insignificant emotion is negative, and the current suspension control parameters need optimization. Conversely, if the negative score is less than or equal to the scoring threshold, it is determined that the user's insignificant emotion is not negative, and the current suspension control parameters do not need optimization, so no optimization is performed. By reducing suspension under insignificant emotions, unnecessary suspension control is reduced, thereby improving the user experience.

[0062] Regarding step 132, this application proposes a specific implementation method, wherein the aforementioned step of adjusting the current suspension parameters according to the optimization target corresponding to the current scenario includes: determining the optimization target corresponding to the current scenario based on the scenario target mapping table; adjusting the current suspension parameters according to the preset fine-tuning numerical gradient corresponding to the optimization target; if a decrease in the negative score of user emotion is detected, then the current suspension parameters are adjusted according to the optimization target corresponding to the current scenario until the negative score is less than or equal to the score threshold and then the adjustment is stopped; if the negative score of user emotion is detected to remain unchanged or increase, then the adjustment of the suspension parameters is stopped.

[0063] This application not only uses negative scores to determine whether suspension optimization is needed for insignificant emotions, but also uses negative scores to determine the limits of optimization. Since there is no preset suspension control strategy for insignificant emotions, determining the limits of optimization becomes a crucial issue in suspension optimization for insignificant emotions. To address this, this application uses a preset fine-tuning value corresponding to the optimization target as the adjustment step size, gradually adjusting the current suspension parameters. Each adjustment monitors the user's negative emotion score at least once. If the negative score decreases, it indicates that the current optimization target has a positive impact on improving the user's insignificant emotions. Therefore, the steps of adjusting the current suspension parameters and monitoring the negative score are repeated until the user's negative score increases, remains unchanged, or decreases to less than or equal to the score threshold, at which point adjustment stops. An increase in the negative score indicates that the current optimization target has a negative impact on improving the user's insignificant emotions, and adjustment is stopped to prevent the negative impact from expanding. A unchanged negative score indicates that the current optimization target has almost no impact on improving the user's insignificant emotions, and adjustment is stopped to reduce ineffective adjustment. A decrease in the negative score to less than or equal to the score threshold indicates that the optimization has reached its limit, and no further adjustment is needed to prevent over-adjustment.

[0064] Before performing step 110, this application also provides an implementation method that, before performing the aforementioned step of recognizing the user's emotion through the emotion recognition model corresponding to the current scene, further includes:

[0065] 141: Analyze users' significant emotions in each scenario based on historical user sentiment data;

[0066] 142: Establish the association between each scenario and the corresponding significant emotions, and obtain training data for each scenario based on historical user emotion data. The training data includes emotion data and emotion tags.

[0067] 143: The neural network is trained using the training data so that after training, the neural network can identify whether the user's emotion in the corresponding scenario is a significant emotion associated with the corresponding scenario.

[0068] Before identifying user emotions using the emotion recognition model corresponding to the current scenario, this application requires in-depth analysis and mining of historical user emotion data for each scenario. This can begin by using a complex and general emotion recognition model to statistically analyze a large amount of historical user emotion data, identifying the most frequently exhibited significant emotional features in different scenarios. Based on these analysis results, a mapping table between scenarios and significant emotions is established, thus establishing a clear association between each specific scenario and its corresponding significant emotion. Simultaneously, training data for each scenario is constructed based on the accumulated historical user emotion data. This training data not only contains rich emotion data but also manually labeled emotion tags. The neural network is then thoroughly trained using this training data, enabling it to accurately determine whether the user's emotions in a specific scenario match the features of the significant emotions in that scenario, i.e., possessing the ability to identify significant emotions in a specific scenario.

[0069] Regarding step 143, this application also provides a specific implementation method. The aforementioned step of training a neural network using the training data, so that the neural network can identify whether the user's emotion in a corresponding scene is a significant emotion associated with the corresponding scene after training, includes: inputting the emotion data of the training data for each scene into the corresponding neural network; determining the user's emotion based on the probability that the emotion data output by the neural network belongs to a significant emotion; if the user's emotion does not match the emotion label of the training data, optimizing the neural network to obtain an emotion recognition model that can output a correct result consistent with the emotion label based on the emotion data, wherein the emotion label includes significant or non-significant emotions in the corresponding scene.

[0070] This application uses sentiment data from training data corresponding to various scenarios as input, feeding them into a pre-constructed neural network for processing. The neural network includes an input layer, a fully connected layer, an activation function layer, and an output layer. The network extracts and analyzes features from the input sentiment data and outputs the probability that the sentiment data belongs to each salient sentiment. Based on this probability result, the current user's sentiment can be determined. During training, the output of the neural network is continuously monitored. If an inconsistency is found between the user's sentiment output by the neural network and the pre-labeled sentiment in the training data, the parameters of the neural network need to be optimized. Finally, through repeated iterations of the above process, the structure and parameters of the neural network are gradually optimized, ultimately resulting in a sentiment recognition model that can accurately output the correct result matching the sentiment label based on the input sentiment data.

[0071] It's important to note that the sentiment labels are pre-labeled based on scene characteristics, and include two types: one is typical and significant sentiment in that scene, and the other is atypical and insignificant sentiment in that scene. This distinction helps the model better learn the sentiment features of different scenes.

[0072] Emotional data, including facial expression data, voice data, and physiological data, can be collected through the following methods:

[0073] Facial expression data can be collected using devices such as high-definition cameras and / or infrared cameras used to capture users' facial expressions; voice data can be collected using devices such as microphones installed in the car's headliner or door trim panels; temperature information in physiological data can be collected using infrared cameras, while breathing information can be collected using millimeter-wave radar.

[0074] The visual acquisition device can be deployed above the dashboard, facing the driver and front passenger respectively. Specifically, it can collect facial expression data through a combination of a binocular depth camera and a near-infrared camera, with the binocular camera and near-infrared camera capturing images simultaneously. After acquiring the images from the binocular camera and near-infrared camera, a checkerboard calibration board is used to calculate the rotation / translation relationship between the cameras, achieving precise alignment of the coordinate system and pixels. During data fusion, in good lighting conditions, the RGB image from the binocular depth camera is used as the primary source, supplemented by a depth map for locating the face (e.g., filtering the background). In insufficient lighting conditions, the near-infrared camera is used as the primary source, supplemented by a depth map for determining the distance to the face, and the resulting image is fused to obtain a facial image. Image recognition is then performed on the facial image to obtain facial bounding boxes and the coordinates of five key points, among other facial expression data.

[0075] Image recognition during the aforementioned facial expression data acquisition process can be achieved using a Multi-task Cascaded Convolutional Network (MTCNN) network. The MTCNN network comprises a three-stage cascaded network: P-Net (for quickly generating candidate face bounding boxes by scanning the input image through a 12×12 pixel window), R-Net (for finely filtering out false positives and optimizing the face bounding box position), and O-Net (for locating the eyes, nose tip, and corners of the mouth, etc.). In addition to the aforementioned three-stage cascaded network, MTCNN can also detect faces of different sizes by scaling the image; merge candidate bounding boxes with high overlap using Non-Maximum Suppression (NMS); and directly predict facial feature point coordinates through regression.

[0076] After the audio acquisition device transmits the captured speech signal, it first performs high-frequency enhancement to improve the signal-to-noise ratio of the speech components. Then, it suppresses spectral leakage through framing (20-40ms / frame) and windowing (such as Hamming window). Next, it uses Mel-Frequency Cepstral Coefficients (MFCC) for feature extraction: a short-time Fourier transform (STFT) is performed on each frame of signal to obtain a linear spectrum. This spectrum is then mapped to a nonlinear spectrum by a Mel-scale filter bank to simulate the characteristics of human hearing. After taking the logarithmic energy, the cepstral coefficients are generated by decoupling through a discrete cosine transform (DCT). The low-frequency components represent the vocal tract transmission characteristics (timbre), while the high-frequency components reflect the excitation characteristics of the sound source (pronunciation content). Finally, speech data is generated.

[0077] The physiological data acquisition device uses infrared cameras to locate the region of interest (ROI) of the user's nose, calculates the average temperature within that region and establishes a baseline temperature, and monitors the temperature difference (ΔT) between the baseline and the baseline value in real time to output the local temperature change. Simultaneously, the physiological data acquisition device uses millimeter-wave radar to capture micro-movements in the chest and abdomen, uses a peak detection algorithm to identify extreme points in the respiratory cycle, and smooths the data using a moving average filter to finally output the respiratory rate (RR), thus generating physiological data.

[0078] join Figure 2 This application also provides a suspension control device. Embodiments of this application can divide the device into functional modules based on the above method examples. For example, each function can be divided into its own functional modules, or two or more functions can be integrated into one module. The integrated module can be implemented in hardware or as a software functional module. It should be noted that the module division in this application embodiment is illustrative and only represents one logical functional division; in actual implementation, there may be other division methods.

[0079] Specifically, the suspension control device includes: an identification unit 210, used to determine the current scene based on scene characteristics and identify user emotions through an emotion recognition model corresponding to the current scene, wherein different emotion recognition models are used to identify salient emotions in the corresponding scene; a control unit 220, used to determine the suspension control strategy corresponding to the salient emotion in the current scene based on an emotion strategy mapping table if the emotion is identified as salient, and adjust the current suspension parameters according to the suspension control strategy; an optimization unit 230, used to determine the optimization target corresponding to the current scene if the emotion is identified as insignificant, when the current scene is a dynamic scene; the control unit 220 is also used to adjust the current suspension parameters according to the optimization target corresponding to the current scene.

[0080] In one feasible embodiment, the suspension control device further includes a training unit 240, which is used to: analyze the salient emotions of users in each scenario based on historical user emotion data in each scenario; establish the correlation between each scenario and the corresponding salient emotions, and obtain training data for each scenario based on historical user emotion data, wherein the training data includes emotion data and emotion labels; and train the neural network using the training data so that the neural network can identify whether the user's emotion in the corresponding scenario is a salient emotion associated with the corresponding scenario after training.

[0081] In one feasible approach, the aforementioned training unit 240 is specifically used to: input the sentiment data of the training data for each scenario into the corresponding neural network; determine the user's sentiment based on the probability that the sentiment data output by the neural network belongs to a salient sentiment; if the user's sentiment does not match the sentiment label of the training data, optimize the neural network to obtain a sentiment recognition model that can output a correct result consistent with the sentiment label based on the sentiment data, wherein the sentiment label includes salient or non-salient sentiments in the corresponding scenario.

[0082] In one feasible implementation, the aforementioned recognition unit 210 is specifically used to: collect a user dataset including facial expression data, voice data, and physiological data; generate a fusion feature of facial expression data, voice data, and physiological data based on the user dataset; input the fusion feature as emotion data into an emotion recognition model to output the probability that the user emotion is a significant emotion in the corresponding scene, and determine the user emotion based on the probability.

[0083] In one feasible embodiment, the aforementioned identification unit 210 is specifically configured to: input the fused features as emotion data into an emotion recognition model, and output the probability of each salient emotion among at least one salient emotion in the corresponding scene for the user emotion; obtain the salient emotion with the highest probability from the at least one salient emotion as the target emotion; if the probability of the target emotion is greater than a probability threshold, then determine the user emotion as the target emotion; if the probability of the target emotion is less than or equal to the probability threshold, then determine the user emotion as a non-salient emotion.

[0084] In one implementable manner, the aforementioned control unit 220 is specifically used to: monitor the user's emotional data and evaluate the negative score of the user's emotions based on the emotional data; if the negative score is greater than a score threshold, determine that the current suspension control parameters need to be optimized, and execute the step of adjusting the current suspension parameters according to the optimization target corresponding to the current scenario; if the negative score is less than or equal to the score threshold, do not adjust the current suspension parameters.

[0085] In one feasible embodiment, the aforementioned control unit 220 is specifically used to: determine the optimization target corresponding to the current scene based on the scene target mapping table; adjust the current suspension parameters according to the preset fine-tuning numerical gradient corresponding to the optimization target; if the negative score of user emotion is detected to decrease, then continue to adjust the current suspension parameters according to the optimization target corresponding to the current scene until the negative score is less than or equal to the score threshold and then stop adjusting; if the negative score of user emotion is detected to remain unchanged or increase, then stop adjusting the suspension parameters.

[0086] The processor can be any unit capable of implementing adjustable signals, such as various microcontrollers, DSPs (digital signal processors), FPGAs (Field-Programmable Gate Arrays), host computers, or central processing units (CPUs). Among them, various control functions can be implemented by programming the microcontroller. For example, in this embodiment, the model can be used to recognize user emotions and adjust suspension parameters according to user emotions.

[0087] join Figure 3 This application also provides a suspension control device, which may include a processor 310 and a memory 320. The processor 310 and the memory 320 are connected via a bus 330. The processor 310 is used to execute multiple instructions; the memory is used to store multiple instructions, which are adapted to be loaded and executed by the processor as in the above embodiment of the remote testing method based on a human-machine interface. The processor may be an Electronic Control Unit (ECU), a Central Processing Unit (CPU), a general-purpose processor, a coprocessor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. The processor may also be a combination that implements computing functions, such as a combination of one or more microprocessors, a combination of a 5SP and a microprocessor, etc. In this embodiment, the processor may be a microcontroller. By programming the microcontroller, various control functions can be implemented. The processor has the advantages of powerful computing capabilities and fast processing speed.

[0088] Specifically, the processor 310 is used to execute the functions of the recognition unit 210, which are used to determine the current scene based on scene features and to identify user emotions through the emotion recognition model corresponding to the current scene, wherein different emotion recognition models are used to identify salient emotions in the corresponding scene; it is also used to execute the functions of the control unit 220, which are used to determine the suspension control strategy corresponding to the salient emotion in the current scene based on the emotion strategy mapping table if the emotion is identified as salient, and to adjust the current suspension parameters according to the suspension control strategy; it is also used to execute the functions of the optimization unit 230, which are used to determine the optimization target corresponding to the current scene if the emotion is identified as insignificant, when the current scene is a dynamic scene; the processor 310 is also used to adjust the current suspension parameters according to the optimization target corresponding to the current scene.

[0089] In one feasible embodiment, the aforementioned processor 310 is further configured to perform the functions of the training unit 240, for: analyzing the salient emotions of users in each scenario based on historical user emotion data in each scenario; establishing the association between each scenario and the corresponding salient emotions, and obtaining training data for each scenario based on historical user emotion data, wherein the training data includes emotion data and emotion labels; and training the neural network using the training data so that the neural network, after training, can identify whether the user's emotion in the corresponding scenario is a salient emotion associated with the corresponding scenario.

[0090] In one feasible implementation, the aforementioned processor 310 is specifically configured to: input the sentiment data of the training data for each scenario into the corresponding neural network; determine the user's sentiment based on the probability that the sentiment data output by the neural network belongs to a salient sentiment; if the user's sentiment does not match the sentiment label of the training data, optimize the neural network to obtain a sentiment recognition model that can output a correct result consistent with the sentiment label based on the sentiment data, wherein the sentiment label includes salient or non-salient sentiments in the corresponding scenario.

[0091] In one implementable manner, the aforementioned processor 310 is specifically configured to: collect a user dataset including facial expression data, voice data, and physiological data; generate a fusion feature of the facial expression data, voice data, and physiological data based on the user dataset; input the fusion feature as emotion data into an emotion recognition model to output the probability that the user emotion is a significant emotion in the corresponding scene, and determine the user emotion based on the probability.

[0092] In one implementable manner, the aforementioned processor 310 is specifically configured to: input the fused features as emotion data into an emotion recognition model, and output the probability of each salient emotion among at least one salient emotion in the corresponding scene as the user emotion; obtain the salient emotion with the highest probability from the at least one salient emotion as the target emotion; if the probability of the target emotion is greater than a probability threshold, then determine the user emotion as the target emotion; if the probability of the target emotion is less than or equal to the probability threshold, then determine the user emotion as a non-salient emotion.

[0093] In one implementable manner, the aforementioned processor 310 is specifically configured to: monitor the user's emotional data and evaluate the negative score of the user's emotion based on the emotional data; if the negative score is greater than a score threshold, determine that the current suspension control parameters need optimization, and execute the step of adjusting the current suspension parameters according to the optimization target corresponding to the current scenario; if the negative score is less than or equal to the score threshold, then do not adjust the current suspension parameters.

[0094] In one implementable manner, the aforementioned processor 310 is specifically used to: determine the optimization target corresponding to the current scene based on the scene target mapping table; adjust the current suspension parameters according to the preset fine-tuning numerical gradient corresponding to the optimization target; if the negative score of user emotion is detected to decrease, then continue to adjust the current suspension parameters according to the optimization target corresponding to the current scene until the negative score is less than or equal to the score threshold and then stop adjusting; if the negative score of user emotion is detected to remain unchanged or increase, then stop adjusting the suspension parameters.

[0095] This application also provides a computer-readable storage medium storing a plurality of instructions adapted for loading by a processor and executing the methods in any of the foregoing embodiments. A processor is provided for executing the plurality of instructions; a memory is provided for storing the plurality of instructions, which are loaded by the processor and executed by the processor to perform the suspension control method based on emotion interaction in the above embodiments.

[0096] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0097] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.

Claims

1. A suspension control method based on emotional interaction, characterized in that, include: The current scene is determined based on scene characteristics, and the user's emotions are identified through the emotion recognition model corresponding to the current scene. Different emotion recognition models are used to identify the salient emotions of the corresponding scene. If a significant emotion is identified, the suspension control strategy corresponding to the significant emotion in the current scenario is determined based on the emotion strategy mapping table, and the current suspension parameters are adjusted according to the suspension control strategy. If the emotion is identified as non-significant, then if the current scene is dynamic, the optimization objective corresponding to the current scene is determined, and the current suspension parameters are adjusted according to the optimization objective corresponding to the current scene. After determining the optimization objective for the current scenario, the process before adjusting the current suspension parameters based on that objective includes: Monitor users' emotional data and evaluate the negative rating of users' emotions based on the emotional data; If the negative score is greater than the score threshold, it is determined that the current suspension control parameters need to be optimized, and the step of adjusting the current suspension parameters according to the optimization target corresponding to the current scenario is executed. If the negative rating is less than or equal to the rating threshold, the current suspension parameters will not be adjusted.

2. The method according to claim 1, characterized in that, Before performing the step of identifying user emotions using the emotion recognition model corresponding to the current scene, the method further includes: Analyze users' significant emotions in each scenario based on historical user sentiment data. Establish the association between each scenario and the corresponding significant emotions, and obtain training data for each scenario based on historical user emotion data, wherein the training data includes emotion data and emotion tags; The neural network is trained using the training data, enabling it to identify whether a user's emotion in a given scenario is a significant emotion associated with that scenario.

3. The method according to claim 2, characterized in that, The step of training the neural network using the training data, so that the neural network can identify whether the user's emotion in a corresponding scene is a significant emotion associated with the corresponding scene after training, includes: The sentiment data from the training data of each scenario is input into the corresponding neural network; The user's emotion is determined based on the probability that the emotion data output by the neural network belongs to a significant emotion; If the user's emotion does not match the emotion label in the training data, the neural network is optimized to obtain an emotion recognition model that can output the correct result consistent with the emotion label based on the emotion data. The emotion label includes significant or non-significant emotions in the corresponding scene.

4. The method according to claim 1, characterized in that, The step of identifying user emotions using an emotion recognition model corresponding to the current scene includes: Collect user datasets that include facial expression data, voice data, and physiological data; Generate fusion features of facial expression data, voice data, and physiological data based on the user dataset; The fused features are input into the emotion recognition model as emotion data, and the probability of the user's emotion being a significant emotion in the corresponding scene is output. The user's emotion is then determined based on the probability.

5. The method according to claim 4, characterized in that, The step of inputting the fused features as emotion data into the emotion recognition model, outputting the probability that the user's emotion is a significant emotion in the corresponding scene, and determining the user's emotion based on the probability includes: The fused features are input into the emotion recognition model as emotion data, and the output is the probability of each salient emotion in at least one salient emotion of the corresponding scene. From the at least one significant emotion, select the significant emotion with the highest probability as the target emotion; If the probability of the target emotion is greater than the probability threshold, then the user emotion is determined to be the target emotion; if the probability of the target emotion is less than or equal to the probability threshold, then the user emotion is determined to be a non-significant emotion.

6. The method according to claim 1, characterized in that, The step of adjusting the current suspension parameters according to the optimization objective corresponding to the current scenario includes: Determine the optimization objective corresponding to the current scenario based on the scenario objective mapping table; Adjust the current suspension parameters according to the preset fine-tuning numerical gradient corresponding to the optimization objective; If a decrease in the negative score of user emotion is detected, the current suspension parameters will continue to be adjusted according to the optimization goal corresponding to the current scenario until the negative score is less than or equal to the score threshold and then the adjustment will stop. If the negative rating of the user's emotions remains unchanged or increases, the suspension parameters will not be adjusted.

7. A suspension control device, characterized in that, The device includes: The recognition unit is used to determine the current scene based on scene features and to identify the user's emotions through the emotion recognition model corresponding to the current scene. Different emotion recognition models are used to identify the salient emotions of the corresponding scene. The control unit is used to determine the suspension control strategy corresponding to the significant emotion in the current scene based on the emotion strategy mapping table if the significant emotion is identified, and to adjust the current suspension parameters according to the suspension control strategy. The optimization unit is used to determine the optimization objective for the current scene if the identified emotion is non-significant and the current scene is dynamic. The control unit is also used to adjust the current suspension parameters according to the optimization target corresponding to the current scenario; After determining the optimization objective for the current scenario, the process before adjusting the current suspension parameters based on that objective includes: Monitor users' emotional data and evaluate the negative rating of users' emotions based on the emotional data; If the negative score is greater than the score threshold, it is determined that the current suspension control parameters need to be optimized, and the step of adjusting the current suspension parameters according to the optimization target corresponding to the current scenario is executed. If the negative rating is less than or equal to the rating threshold, the current suspension parameters will not be adjusted.

8. A suspension control device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the suspension control method according to any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a plurality of instructions adapted for loading by a processor and executing the steps of the suspension control method as described in any one of claims 1 to 6.

Images