Pet emotion regulation generation system under perspective of human-pet interaction psychology

Abstract

The invention discloses a pet emotion regulation generation system under the perspective of human-pet interaction psychology, and relates to the technical field of artificial intelligence, and the system comprises a multi-mode emotion perception module, an emotion state recognition module, a personalized regulation strategy generation module, a human-pet bidirectional interaction execution module and a dynamic optimization learning module. By fusing multi-modal data and a deep neural network model, high-precision recognition of the pet emotional state is realized, and a personalized interaction adjustment strategy is generated. The invention aims to solve the problem of insufficient accuracy and individuation of pet emotion regulation in the prior art, and can significantly improve the accuracy and generalization ability of emotion recognition, improve the pertinence and continuity of the regulation strategy, and enhance the overall coordination and adaptability of the system.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to a pet emotion regulation and generation system from the perspective of human-pet interaction psychology. Background Technology

[0002] With the increasing popularity of the human-pet cohabitation lifestyle and the growing attention to pet mental health, pet emotion regulation technology has gradually become a research hotspot in the field of smart pet products. As an important part of the family, the emotional state of pets directly affects the quality of human-pet interaction and the harmonious atmosphere of the family. Currently, pet emotion recognition and intervention technologies mainly rely on behavioral observation, physiological signal monitoring, and environmental stimulus responses, attempting to build an interactive system oriented towards the psychological needs of pets. However, existing technological systems have significant limitations in terms of theoretical foundation, perception dimensions, and regulatory mechanisms, making it difficult to achieve accurate understanding and effective intervention of pets' emotional states.

[0003] Among these, the psychology of human-pet interaction provides important theoretical support for understanding pet emotions. This field emphasizes inferring a pet's internal emotional state by analyzing its behavioral patterns, physiological responses, and environmental adaptability in specific social situations, and then designing interaction strategies that align with its cognitive patterns. Based on this theory, the core objective of an emotion regulation system is to establish a closed-loop mechanism that can dynamically perceive changes in a pet's emotions, identify emotional triggers, and generate appropriate intervention plans, thereby improving the pet's psychological well-being and the stability of the human-pet relationship.

[0004] Existing technologies face multiple bottlenecks in achieving the above goals: First, emotion recognition models generally lack guidance from psychological theories, relying mostly on single behavioral features or isolated physiological indicators for emotion classification, ignoring the multi-factor coupling mechanism of emotion generation, resulting in highly subjective recognition results and weak generalization ability; Second, the interaction strategy generation process is disconnected from individual pet differences, failing to combine breed characteristics, growth stage, and past interaction history to construct personalized adjustment logic, making intervention measures lack specificity and sustained effectiveness; Third, the system lacks the ability to model the dynamics of two-way human-pet interaction, designing adjustment schemes only from the pet's perspective, ignoring the significant impact of the owner's behavior, language, and emotions on the pet's psychological state, resulting in limited adjustment effects; Finally, existing systems mostly use static rule bases or simple feedback mechanisms, unable to evolve and optimize strategies based on real-time interaction data, making it difficult to adapt to complex and ever-changing family life scenarios. Summary of the Invention

[0005] The purpose of this invention is to provide a pet emotion regulation generation system from the perspective of human-pet interaction psychology, so as to solve the problems of insufficient accuracy and personalization in the existing pet emotion regulation technology.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: A pet emotion regulation generation system from the perspective of human-pet interaction psychology, the system includes the following components: The system comprises the following modules: a multimodal emotion perception module, which collects pet behavior data, physiological signal data, and environmental interaction data, and constructs an emotion perception vector based on multi-source feature fusion according to human-pet interaction psychology theory; an emotion state recognition module, connected to the multimodal emotion perception module, which receives the emotion perception vector, classifies and identifies the pet's emotional state using a deep neural network model, and outputs an emotion state label and confidence score; a personalized regulation strategy generation module, connected to the emotion state recognition module, which generates an appropriate interaction regulation strategy based on the emotion state label and confidence score, combined with the pet's individual profile data; a human-pet two-way interaction execution module, connected to the personalized regulation strategy generation module, which executes the generated interaction regulation strategy and monitors the pet's feedback response to the regulation strategy in real time; and a dynamic optimization learning module, connected to both the human-pet two-way interaction execution module and the personalized regulation strategy generation module, which continuously optimizes the emotion recognition model and regulation strategy generation logic based on feedback response data.

[0007] Preferably, the multimodal emotion perception module includes a behavior perception unit, a physiological perception unit, and an environmental perception unit; the behavior perception unit uses a multi-view visual sensor array to collect pet limb movement sequences, facial expression changes, and movement trajectory data, and extracts behavioral dynamic features through a spatiotemporal feature extraction network; the physiological perception unit integrates a heart rate variability sensor, a skin conductance sensor, and a body temperature monitoring sensor, and continuously collects physiological signal data at a sampling frequency of 128Hz; the environmental perception unit deploys a sound acquisition device, an ambient light intensity sensor, and a temperature and humidity sensor to record environmental acoustic characteristics, changes in light intensity, and microenvironment parameters.

[0008] Furthermore, the emotion state recognition module employs a hierarchical attention mechanism neural network architecture, including a feature-level attention subnetwork and a temporal-level attention subnetwork. The feature-level attention subnetwork assigns feature weights to the emotion perception vector output by the multimodal emotion perception module and calculates the contribution of each modality feature to emotion classification. The temporal-level attention subnetwork captures key time segments in the emotional state evolution process, identifying emotion trigger points and duration patterns. The neural network architecture uses a weighted combination of cross-entropy loss function and focus loss function for model training, with the loss function expression being: in, , , , For the number of categories, For real labels, To predict probabilities.

[0009] Furthermore, the personalized adjustment strategy generation module includes a strategy knowledge base and an individual adaptation engine. The strategy knowledge base stores adjustment strategy templates based on human-pet interaction psychology theory, including three major categories: soothing strategies, incentive strategies, and transfer strategies. The individual adaptation engine calculates the strategy matching score based on the pet's breed characteristics, age stage, historical interaction records, and current environmental context, and selects adjustment strategies with a matching score higher than 0.85 for output. The individual adaptation engine adopts a multi-objective optimization algorithm, simultaneously considering three optimization objectives: strategy effectiveness, execution feasibility, and pet acceptance.

[0010] Preferably, the human-pet two-way interaction execution module includes a sound interaction unit, a tactile interaction unit, and a visual interaction unit; the sound interaction unit generates acoustic stimulation signals in a specific frequency range (125Hz-16kHz), including soothing music, natural soundscapes, and owner voice simulation; the tactile interaction unit generates tactile feedback of different intensities (0.1-2.0m / s²) and patterns (continuous, intermittent, gradual) through a controllable vibration device; the visual interaction unit uses an LED light source array with adjustable color temperature (2700K-6500K) and brightness (10-500lux) to generate dynamic light and shadow patterns.

[0011] Furthermore, the dynamic optimization learning module adopts a deep reinforcement learning framework, including a policy evaluation network and a value function network. The policy evaluation network calculates the immediate reward value of the policy adjustment based on the pet feedback data after the interaction is executed. The value function network predicts the long-term cumulative reward and guides the parameter update of the policy generation module. The deep reinforcement learning framework adopts a proximal policy optimization algorithm and updates the model parameters every 24 hours with an update step size of 0.002.

[0012] On the other hand, a method for generating pet emotion regulation from the perspective of human-pet interaction psychology is presented. The specific steps of this method are as follows: Step S110, simultaneously collecting pet behavior data, physiological signal data, and environmental interaction data through a multimodal sensor array to construct a multi-source heterogeneous data stream; Step S120, preprocessing and extracting features from the multi-source heterogeneous data stream to generate an emotion perception vector containing spatiotemporal features, frequency domain features, and statistical features; Step S130, inputting the emotion perception vector into a trained deep neural network model to output the pet emotion state classification result and the corresponding confidence score; Step S140, based on the emotion state classification result... In step S150, based on confidence scores and pet individual profile data, a matching regulation strategy template is retrieved from the strategy knowledge base; in step S160, the parameters of the retrieved regulation strategy template are fine-tuned using a multi-objective optimization algorithm to generate personalized interactive regulation instructions; in step S170, the personalized interactive regulation instructions are executed, and regulatory stimuli are output through multiple channels including acoustic, tactile, and visual; in step S180, the pet's feedback response to the regulatory stimuli is monitored in real time, and behavioral change data and physiological signal change data are collected; in step S190, the effectiveness index of the regulation strategy is calculated based on the feedback response data, and the strategy knowledge base and deep neural network model parameters are updated.

[0013] Compared with the prior art, the beneficial technical effects of the present invention are as follows: This invention achieves multi-factor coupled analysis of pet emotional states by integrating a multimodal emotion perception mechanism guided by human-pet interaction psychology theory, which significantly improves the accuracy and generalization ability of emotion recognition.

[0014] This invention provides a personalized regulation strategy generation method based on individual difference modeling, which fully considers variety characteristics, growth stage and historical interaction data, fundamentally improving the targeting and sustainability of the regulation strategy.

[0015] This invention introduces a two-way interactive dynamic modeling mechanism between humans and pets, enabling quantitative analysis and integration of the impact on the owner's behavior, and effectively improving the overall coordination and adaptability of the emotion regulation system.

[0016] This invention employs a dynamic optimization architecture based on deep reinforcement learning, enabling the system to continuously evolve and adjust its strategies based on real-time interactive data, significantly enhancing the system's robustness and intelligence in complex home environments. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the overall technical solution architecture of a pet emotion regulation and generation system from the perspective of human-pet interaction psychology proposed in this invention. Figure 2 This is a schematic diagram of the core principle framework for generating multimodal emotion perception and personalized regulation strategies based on human-pet interaction psychology in this invention. Detailed Implementation

[0018] The features and exemplary embodiments of various aspects of the present invention will now be described in detail. To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are merely intended to explain the present invention and not to limit the present invention. For those skilled in the art, the present invention can be practiced without some of these specific details. The following description of the embodiments is merely to provide a better understanding of the present invention by illustrating examples of the invention.

[0019] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising..." does not exclude the presence of additional identical elements in the process, method, article, or apparatus that includes said element.

[0020] In the embodiments of the present invention, the same reference numerals denote the same components, and for the sake of brevity, detailed descriptions of the same components are omitted in different embodiments. It should be understood that the thickness, length, width, and other dimensions of various components in the embodiments of the present invention shown in the accompanying drawings, as well as the overall thickness, length, width, and other dimensions of the integrated device, are merely illustrative and should not constitute any limitation on the present invention; the term "multiple" in the present invention refers to two or more (including two).

[0021] Example 1 In the daily care of family pets, the multimodal emotion perception module monitors the pet's emotional state comprehensively through a distributed sensor network. The behavior perception unit uses three 1080P high-definition cameras to form a multi-view visual sensor array, deployed at the top, sides, and ground level of the pet's activity area, simultaneously capturing the pet's limb movement sequences at a sampling rate of 30 frames per second. The system tracks the 3D coordinate changes of 17 major joints of the pet in real time using a skeletal keypoint detection algorithm, calculating the amplitude and frequency characteristics of limb movements. Facial expression change analysis employs a hybrid architecture of local binary mode and convolutional neural networks, focusing on capturing micro-expression features such as eye opening degree, ear posture angle, and the frequency of muscle contraction in the mouth and nose. Motion trajectory data, through a combination of background subtraction and optical flow, records the pet's movement path, dwell time, and activity intensity in space, forming a spatiotemporal trajectory matrix containing position coordinates, velocity vectors, and acceleration changes.

[0022] The physiological sensing unit integrates medical-grade biosignal acquisition equipment. Its heart rate variability sensor employs photoplethysmography (PPG) technology, continuously monitoring heart rate interval fluctuations via an ear clip probe at a sampling frequency of 128Hz. The skin conductance sensor uses two silver-silver chloride electrodes placed on the pet's abdomen to measure skin conductance changes ranging from 0.02 to 100 microsiemens, with a resolution of 0.001 microsiemens. The body temperature monitoring sensor uses a non-contact infrared temperature measurement module, monitoring a range of 35-42 degrees Celsius with an accuracy of 0.1 degrees Celsius, acquiring surface temperature data every 5 seconds. All physiological signal data undergoes power frequency noise suppression using a fourth-order Butterworth filter, and motion artifact interference is eliminated using wavelet transform. Time-domain features include mean, standard deviation, and root mean square value; frequency-domain features include the low-frequency power to high-frequency power ratio and total power spectral density.

[0023] The environmental sensing unit includes a high-sensitivity microphone array, a digital illuminance sensor, and a combined temperature and humidity probe. The sound acquisition device employs a 4-microphone ring array with a frequency response range of 20Hz-20kHz, using beamforming technology to focus on the pet's vocalization area and record acoustic parameters such as sound pressure level, fundamental frequency, and formants. The ambient light intensity sensor measures from 0-2000 lux with an accuracy of ±5%, monitoring the periodic changes in natural light and the on / off status of artificial light sources. The temperature and humidity sensor measures from 0-50 degrees Celsius and 0-100% relative humidity, recording environmental parameter changes every 10 seconds. All environmental data and behavioral / physiological data are synchronized at the microsecond level via hardware timestamps, constructing a multi-source data stream with a unified time reference.

[0024] The emotion state recognition module receives an emotion perception vector from the multimodal emotion perception module. This vector has 256 dimensions, including 83 behavioral features, 97 physiological features, and 76 environmental features. The feature-level attention subnetwork adopts a three-layer fully connected architecture: 256 nodes in the input layer, 128 nodes in the hidden layer, and 6 nodes in the output layer corresponding to six basic emotion states. This subnetwork calculates the importance weights of each feature dimension using a trainable parameter matrix. The weights for behavioral features range from 0.15 to 0.35, for physiological features from 0.25 to 0.45, and for environmental features from 0.20 to 0.40. The temporal-level attention subnetwork uses a long short-term memory network structure with 64 hidden units. It identifies key time segments in the evolution of emotion states through a gating mechanism, focusing on capturing feature change patterns within a time window from 5 seconds before the emotion trigger to 10 seconds after the trigger.

[0025] The neural network was trained using an adaptive moment estimation algorithm with an initial learning rate of 0.001, a batch size of 32, and 500 training epochs. The loss function combination used cross-entropy loss weight α=0.7, focus loss weight β=0.3, and focus parameter γ=2. The model achieved a classification accuracy of 92.3% on the validation set, with a 95.1% recognition rate for pleasant states, 89.7% for anxious states, and 91.2% for fear states. The confidence score output used the maximum probability of softmax, with a threshold of 0.75. Samples below this threshold were marked as uncertain states and triggered a re-evaluation mechanism.

[0026] The personalized regulation strategy generation module stores regulation strategy templates based on animal behavior theory, including 12 soothing strategies, 9 incentive strategies, and 7 transfer strategies. Soothing strategies mainly include low-frequency sound wave stimulation, gentle tactile feedback, and warm-colored lighting environments; incentive strategies include interactive game instructions, food reward mechanisms, and the introduction of novel objects; transfer strategies involve attention guidance, environmental changes, and task orientation. The individual adaptation engine uses an elite-preserving genetic algorithm with a population size of 50, a crossover probability of 0.85, a mutation probability of 0.15, and 100 evolutionary iterations per generation.

[0027] The individual adaptation engine considers four dimensions and their weights when calculating strategy matching: breed characteristics (0.30), age group (0.25), historical interaction records (0.35), and environmental context (0.10). The breed characteristics dimension establishes a breed feature matrix, including 15 evaluation indicators such as body size, activity level, and social needs. The age group is divided into four stages: infancy (0-1 years), adolescence (1-3 years), adulthood (3-8 years), and old age (8 years and above), with different stimulus intensity parameters set for each stage. Historical interaction records analyze the success rate, response delay, and duration of strategy execution under the same emotional state over the past 30 days. The environmental context assesses the impact of current time, location, and the presence of people on strategy execution.

[0028] The human-pet two-way interaction module's sound interaction unit uses digital signal processing technology to generate acoustic stimuli within a specific frequency range. Soothing music is generated using a 432Hz base frequency, with harmonic progressions following an I-IV-VI pattern and a tempo controlled at 60-70 BPM. Natural soundscapes include rain, stream sounds, and birdsong, with sound pressure levels controlled at 45-55 dB. Owner voice simulation reconstructs the owner's specific intonation using speech synthesis technology, with a fundamental frequency range of 120-250Hz and a speech rate slowed to 0.8 times the normal speed. All acoustic stimuli are output through dual-channel speakers, with a phase difference controlled at 5-15 milliseconds to create a spatial sound field effect.

[0029] The haptic interaction unit uses a linear resonant actuator to generate precisely controlled haptic feedback. Vibration intensity is controlled via PWM modulation, with an acceleration range of 0.1-2.0 m / s² and a resolution of 0.01 m / s². Vibration mode programming includes a continuous mode with a frequency of 1-5 Hz, an intermittent mode with an on / off time of 2-5 seconds and an off time of 1-3 seconds, and a gradual intensity linear change period of 10-30 seconds. The application area of ​​the haptic feedback is adjusted according to the pet breed; for dogs, it is mainly applied to the chest and back, while for cats, it is mainly applied to the head and neck.

[0030] The visual interaction unit employs a full-spectrum adjustable LED light source array, with a color temperature adjustment range of 2700K-6500K and a brightness adjustment range of 10-500 lux. Dynamic lighting modes include a breathing mode with a brightness change cycle of 5-15 seconds, a gradient mode with a color temperature change rate of 1-3K / second, and a flashing mode with a frequency of 0.5-2Hz. The illuminated area dynamically adjusts based on the pet's current position, using an infrared positioning system to track the pet's coordinates in real time and control the corresponding area's lights to output a predetermined light mode.

[0031] The dynamic optimization learning module employs a proximal policy optimization algorithm framework. Both the policy evaluation network and the value function network are three-layer fully connected structures with 128 and 64 hidden layer nodes, respectively. The immediate reward function design includes four components: a reward weight of 0.4 for mood improvement, a reward weight of 0.3 for behavioral response, a reward weight of 0.2 for physiological indicators, and a reward weight of 0.1 for environmental adaptation. The mood improvement reward is based on the output change of the mood state recognition module: a change from negative to positive mood earns +1, maintaining a positive mood earns +0.5, and mood deterioration incurs a penalty of -1. The behavioral response reward assesses the pet's proactive responses to stimuli, including approach behavior, interactive behavior, and exploratory behavior.

[0032] The value function network predicts the cumulative reward over the next 10 time steps, with a discount factor γ = 0.99. Model parameters are updated every 24 hours, with an update step size ε = 0.002 and a clipping range of [0.95, 1.05]. Each update is based on 300-500 sets of interaction data collected in the past 24 hours, using mini-batch gradient descent with a batch size of 64 and a learning rate of 0.0001. Policy improvement is evaluated using relative performance metrics; the new policy must demonstrate an expected reward improvement of more than 5% compared to the old policy before being accepted for an update.

[0033] The system operation begins in step S110, where the multimodal sensor array simultaneously starts data acquisition, with a sampling rate of 30Hz for behavioral data, 128Hz for physiological data, and 0.1Hz for environmental data. Step S120 preprocesses the multi-source heterogeneous data stream, including sensor data calibration, time alignment, and imputation of missing values. In the feature extraction stage, principal component analysis is used for dimensionality reduction, retaining the principal components with a variance contribution rate of over 95%, ultimately generating a 256-dimensional emotion perception vector.

[0034] Step S130 inputs the emotion perception vector into the trained deep neural network model, with a forward propagation computation time of less than 50 milliseconds. The model outputs a six-dimensional probability vector corresponding to six basic emotion states, and the maximum value is taken as the classification result, while recording the confidence score. When the confidence score is lower than 0.75, the system activates the redundant sensor data fusion mechanism, extends the observation time window to 30 seconds, and recalculates.

[0035] Step S140 retrieves matching regulation strategy templates from the strategy knowledge base based on the emotional state classification results and confidence scores, combined with pet individual profile data. The retrieval algorithm uses an inverted index structure to establish a mapping relationship between emotional states and strategy types, with a response time of less than 100 milliseconds. Step S150 optimizes the parameters of the retrieved regulation strategy templates using a multi-objective optimization algorithm. The optimization objectives include a strategy effectiveness weight of 0.5, an execution feasibility weight of 0.3, and a pet acceptability weight of 0.2, with the solution time controlled within 2 seconds.

[0036] Step S160 executes personalized interactive adjustment commands, outputting adjustment stimuli through multiple channels including acoustic, tactile, and visual. Multi-channel synchronous control accuracy reaches 10 milliseconds, and the stimulation duration is adjusted according to the emotional state: 5-10 minutes for anxiety, 3-5 minutes for fear, and 1-3 minutes for pleasure. Step S170 monitors the pet's feedback response to the adjustment stimuli in real time, focusing on collecting behavioral change data including activity trajectory, posture changes, and vocalization frequency, as well as physiological signal change data including heart rate variability, skin conductance response, and body temperature fluctuations.

[0037] Step S180 calculates the effectiveness indicators of the adjustment strategy based on the feedback response data, including the immediate effect index, the sustained effect index, and the adaptation effect index. The strategy knowledge base is updated using an incremental learning mechanism, increasing the weight of newly added successful strategy cases by 0.1 and decreasing the weight of failed cases by 0.05. The deep neural network model parameters are updated using an elastic weight consolidation method to protect important parameters from being overwritten, retaining 90% of the performance from previous tasks.

[0038] The above description is merely a preferred embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Therefore, all equivalent changes made in accordance with the structure, shape, and principle of the present invention should be covered within the scope of protection of the present invention.

Claims

1. A pet emotion regulation generation system under the perspective of human-pet interaction psychology, characterized in that, The system comprises the following components: A multi-modal emotion perception module for collecting pet behavior data, physiological signal data and environmental interaction data, and constructing a multi-source feature fusion emotion perception vector based on human-pet interaction psychology theory; An emotion state recognition module connected to the multi-modal emotion perception module for receiving the emotion perception vector, classifying and recognizing the pet emotion state through a deep neural network model, and outputting an emotion state label and a confidence score; An individualized adjustment strategy generation module connected to the emotion state recognition module for generating an adaptive interaction adjustment strategy based on the emotion state label and the confidence score, and combining pet individual profile data; A human-pet bidirectional interaction execution module connected to the individualized adjustment strategy generation module for executing the generated interaction adjustment strategy and monitoring the pet's feedback response to the adjustment strategy in real time; A dynamic optimization learning module connected to the human-pet bidirectional interaction execution module and the individualized adjustment strategy generation module for continuously optimizing the emotion recognition model and the adjustment strategy generation logic based on the feedback response data.

2. The pet emotion regulation generation system under the perspective of human-pet interaction psychology according to claim 1, characterized in that, The individualized adjustment strategy generation module includes a strategy knowledge base and an individual adaptation engine; the strategy knowledge base stores adjustment strategy templates constructed based on human-pet interaction psychology theory, including three categories of soothing strategies, incentive strategies and diversion strategies; The individual adaptation engine calculates a strategy matching degree score based on pet breed characteristics, age stage, historical interaction records and current environmental context, and selects an adjustment strategy with a matching degree higher than 0.85 for output; The individual adaptation engine uses a multi-objective optimization algorithm to consider three optimization objectives of strategy effectiveness, execution feasibility and pet acceptance.

3. The pet emotion regulation generation system under the perspective of human-pet interaction psychology of claim 1, wherein, The human-pet bidirectional interaction execution module includes a sound interaction unit, a tactile interaction unit and a visual interaction unit; the sound interaction unit generates acoustic stimulation signals in a specific frequency range, including soothing music, natural soundscapes and master voice simulation; the tactile interaction unit generates continuous, intermittent and gradually changing tactile feedback of different intensities and patterns through a controllable vibration device; the visual interaction unit uses an adjustable color temperature and brightness LED light source array to generate dynamic light and shadow patterns.

4. The pet emotion regulation generation system under the perspective of human-pet interaction psychology according to claim 1, wherein, The dynamic optimization learning module uses a deep reinforcement learning framework, including a policy evaluation network and a value function network; the policy evaluation network calculates the immediate reward value of the adjustment strategy based on the pet feedback data after interaction; The value function network predicts long-term cumulative rewards to guide parameter updates of the strategy generation module; the deep reinforcement learning framework uses a proximal policy optimization algorithm.

5. The pet emotion regulation generation system under the perspective of human-pet interaction psychology according to claim 2, characterized in that, The individual adaptation engine considers the weight allocation of four dimensions when calculating the strategy matching degree: the four dimensions include breed characteristics, age stage, historical interaction records and environmental context; the breed characteristics dimension establishes a breed characteristic matrix; The age stage is divided into four stages: infancy, youth, adulthood and old age, and different stimulation intensity parameters are set for each stage.

6. The pet emotion regulation generation system under the perspective of human-pet interaction psychology according to claim 3, characterized in that, The sound interaction unit uses digital signal processing technology to generate acoustic stimuli in a specific frequency range; soothing music generation uses a 432Hz reference frequency, and the harmony follows the I-IV-V-I pattern; natural soundscapes include rain sounds, stream sounds and bird songs; The master voice simulation reconstructs the master's specific tone through voice synthesis technology.

7. The pet emotion regulation generation system under the perspective of human-pet interaction psychology of claim 3, wherein, The haptic interaction unit adopts linear resonant actuators to generate precisely controlled haptic feedback; vibration intensity is controlled by PWM modulation; vibration mode programming includes continuous mode, intermittent mode, and gradual mode.

8. The pet emotion regulation generation system under the perspective of human-pet interaction psychology according to claim 1, wherein, The emotion state recognition module receives an emotion perception vector from the multi-modal emotion perception module, which has a dimension of 256, including 83-dimensional behavioral characteristics, 97-dimensional physiological characteristics, and 76-dimensional environmental characteristics. The feature-level attention subnetwork adopts a three-layer fully connected architecture, with 256 nodes in the input layer, 128 nodes in the hidden layer, and 6 nodes in the output layer corresponding to the six basic emotional states; the subnetwork calculates the importance weight of each feature dimension through a trainable parameter matrix.

9. The pet emotion regulation generation system under the perspective of human-pet interaction psychology according to claim 4, wherein, The dynamic optimization learning module adopts a proximal policy optimization algorithm framework, and the policy evaluation network and the value function network are both three-layer fully connected structures; the instant reward function design includes four components: emotion improvement reward, behavior response reward, physiological indicator reward, and environment adaptation reward. The emotion improvement reward is based on the output change of the emotion state recognition module, with a reward of +1 for a negative emotion turning into a positive emotion, a reward of +0.5 for maintaining a positive emotion, and a penalty of -1 for an emotional deterioration.

Images