Pet personalized physical rehabilitation method and system based on multi-modal data fusion

Abstract

The invention discloses a pet personalized physical rehabilitation method and system based on multi-modal data fusion. The method comprises the following steps: establishing a digital file of a target pet; according to the method, multi-modal data such as behavior characteristics and physiological indexes are fused, and the space-time correlation is mined by using a multi-head self-attention mechanism, so that the limitation of a single data source can be overcome, the pain level and the joint limitation degree of the pet can be accurately identified, and an objective basis is provided for scheme formulation; based on medical history and rehabilitation sensitivity base lines in a pet digital file, in combination with real-time environment and expression recognition, rehabilitation strength, frequency and aromatic therapy formula are dynamically adjusted, accurate rehabilitation of'one pet and one strategy 'is achieved, and secondary damage or stress to pets caused by a standardized process is avoided; and in combination with expert-algorithm closed-loop verification and concept drift detection, the system can automatically adapt to new cases and environment changes, and the accuracy of rehabilitation effect evaluation and the scientificity of scheme recommendation are continuously improved.

Description

Technical Field

[0001] This invention relates to the field of pet rehabilitation technology, specifically to a personalized physical rehabilitation method and system for pets based on multimodal data fusion. Background Technology

[0002] With the growing trend towards more humanized pets, the demand for pet health management services is increasing. Currently, common pet services on the market suffer from the following shortcomings: 1. Service fragmentation and non-standardization: Traditional pet grooming and care services are mostly single and isolated projects, lacking systematic long-term health management plans based on the individual health status of pets. Service quality is highly dependent on the personal experience of service personnel, making it difficult to standardize and replicate.

[0003] 2. Lack of data-driven and personalized services: Existing service models cannot track and analyze pets' long-term physiological and behavioral data, making it impossible to achieve truly "personalized" customized service solutions. The effects are difficult to quantify, evaluate, and continuously optimize.

[0004] 3. Talent training is out of sync with industry: The training content of pet-related majors in vocational schools lags behind market demand. Students lack opportunities to make decisions and practice in real and complex multimodal data environments, resulting in a disconnect between talent training and the high-end service market.

[0005] 4. Low resource allocation efficiency: The lack of an efficient intelligent matching platform between offline service providers' technician resources and customer needs leads to idle human resources or difficulties in service appointments, resulting in low overall operational efficiency.

[0006] Therefore, there is an urgent need for a new pet health management solution that can integrate online and offline resources, achieve standardized services, personalized solutions, and efficient operations. Summary of the Invention

[0007] To address the aforementioned technical problems, this technical solution provides a personalized physical rehabilitation method and system for pets based on multimodal data fusion, which solves the problems mentioned in the background section.

[0008] To achieve the above objectives, the technical solution adopted by the present invention is as follows: In a first aspect of the present invention, a method for personalized physical rehabilitation of pets based on multimodal data fusion using electronic devices is provided, comprising: Establish a digital profile for the target pet, which includes the pet's basic physiological attributes and historical health records; Multimodal monitoring data of the target pet during the current service period is collected through offline service terminals. The multimodal monitoring data includes at least behavioral characteristic data and physiological indicator data. Multimodal monitoring data is uploaded to the cloud data center, and the data analysis engine performs feature extraction and fusion processing on behavioral feature data and physiological indicator data to generate pet status feature vectors. Based on the pet's state feature vector, matching and calculation are performed in a preset physical rehabilitation program algorithm library to generate a personalized physical rehabilitation execution instruction set for the target pet. The execution instruction set includes aromatherapy formula parameters, physiotherapy equipment operating parameters, manual operation standards, and rehabilitation duration. The personalized physical rehabilitation instruction set is pushed to the technician's terminal, and the technician performs physical rehabilitation services on the pet according to the standardized operation instructions displayed on the terminal; During and after the service, real-time reaction data and post-rehabilitation data of the pets are collected again. The rehabilitation effect index is calculated based on the preset evaluation algorithm, and all process data is sent back to the cloud data center for model iteration and optimization.

[0009] Preferably, establishing the digital profile of the target pet specifically includes: The breed, age, weight, microchip ID, and past medical history of the pets are collected as basic static data, denoted as a vector. ; Construct a pet health knowledge graph by using natural language processing technology to extract entities and map relationships from historical health records and medical texts, transforming them into structured medical record vectors. ; We used a time-series database to store the physiological index change curves during historical rehabilitation processes, and calculated the decay factor of historical rehabilitation effects based on a sliding window algorithm. Generate pet-specific baseline parameters for rehabilitation sensitivity. ; Among them, attenuation factor The calculation follows the logic of exponentially weighted moving average: ; In the formula, This is the attenuation factor of the rehabilitation effect at the current moment. The decay factor is the value from the previous time step. The assessment score reflects the rehabilitation progress of the previous cycle. The smoothing coefficient is 0 < <1, used to adjust the weight of the influence of historical data on the current value; Baseline parameters of rehabilitation sensitivity Defined as the expected value of the ratio of historical rehabilitation stimulus to physiological response: ; In the formula, This refers to changes in physiological indicators. This represents the change in the intensity of the rehabilitation intervention. This represents the mathematical expectation operation.

[0010] Preferably, the step of collecting multimodal monitoring data of the target pet within the current service period through offline service terminals specifically includes: Kinematic data, including three-dimensional acceleration, of pets is collected using inertial measurement units deployed on rehabilitation equipment. angular velocity Euler angles To generate gait feature sequences ; The surface temperature distribution cloud map and heart rate variability data sequence of the pet were simultaneously acquired by a non-contact infrared thermal imager and a photoplethysmography pulse wave sensor. Edge computing nodes are introduced at the data acquisition end to perform real-time denoising on the raw data stream. Specifically, a wavelet threshold denoising algorithm is used to remove motion artifacts, and wavelet coefficients are used to... The processing logic is as follows: ; in, These are the original wavelet coefficients. The number of decomposition layers, The translation factor is... For the first The threshold of the layer, and , For the first Layer noise standard deviation The signal length; Kalman filtering is used to perform microsecond-level synchronization and alignment of the timestamps of heterogeneous sensors. The state update equation is as follows: ; in, for The posterior state estimate at time t. These are the prior state estimates. For Kalman gain, For the observed values, This is the observation matrix.

[0011] Preferably, the specific algorithm logic for feature extraction and fusion processing of behavioral feature data and physiological indicator data through the data analysis engine is as follows: behavioral characteristic data Perform multi-resolution wavelet packet decomposition to extract the energy entropy of each frequency band. kurtosis coefficient and approximate entropy As a characteristic of nonlinear dynamics; The formula for calculating energy entropy is as follows: ; ; in, Total number of frequency bands For the first Signal energy of each frequency band This refers to the proportion of energy in this frequency band to the total energy. Physiological indicator data Perform time-domain statistical analysis and extract the mean. Standard deviation skewness and kurtosis Power spectral density features were extracted by combining frequency domain analysis. ; Construct a multimodal fusion network based on a multi-head self-attention mechanism to integrate behavioral feature vectors. With physiological feature vector Mapped to a high-dimensional latent space, using the learned weight matrix , , The contribution of each modality is dynamically allocated to generate a high-dimensional pet state feature vector that includes spatiotemporal correlation. ; The formula for calculating attention weights is: ; In the formula, For querying the matrix, The key matrix, For value matrices, Let be the dimension of the key vector. It is a normalized exponential function; The multi-head mechanism output is: ; .

[0012] In the formula, For the number of attention heads, For the first Output of size , , For the first The weight matrix corresponding to each head This is for outputting the projection matrix.

[0013] Preferably, the matching and calculation based on the pet's state feature vector within a pre-defined physical rehabilitation algorithm library specifically includes: The currently generated pet state feature vector The data is fed into a trained deep neural network classifier to identify the pet's current pain level and the degree of limitation in joint mobility. ; Based on the identification results, the rehabilitation movement sets of similar cases are retrieved from the physical rehabilitation program algorithm library, and the current gait sequence is calculated using the dynamic time warping algorithm. With standard rehabilitation gait sequence bending distance ; ; in, For the optimal alignment path, For path length, ( , () represents pairs of coordinate points on the path. The function for calculating Euclidean distance. and These are the feature vectors of the two sequences at corresponding points; Introduce a constrained optimization function to maximize the rehabilitation effect. Minimize pet stress response To achieve a dual objective, a genetic algorithm is used to automatically adjust the force threshold of rehabilitation exercises. Duration and rest time between groups Generate a personalized set of execution instructions that includes curves showing specific joint angle changes; The objective function is defined as: ; in, To synthesize the objective function value, , These are the weighting coefficients for rehabilitation effect and stress response, respectively. Maximum heart rate variability during rehabilitation; The constraints are: ; in, , These represent the lower and upper limits of safety, respectively. The threshold for muscle fatigue time in pets.

[0014] Preferably, the specific calculation model for calculating rehabilitation effect indicators based on a preset evaluation algorithm includes: Construct a rehabilitation effect evaluation matrix, including the improvement rate of objective physiological indicators, subjective behavioral scores, and predicted values ​​of pet stress hormone levels; The weight coefficients of each level of indicators were determined by using the analytic hierarchy process (AHP) combined with an expert experience database. ,satisfy ; The fuzzy comprehensive evaluation algorithm is used to process mixed qualitative and quantitative data, and the membership level of rehabilitation effect is output. ,in For the weight vector, It is a fuzzy relation matrix. This represents a fuzzy composition operation; And calculate the rate of recovery progress in the current service period relative to the previous period: ; In the formula, For the first The rate of recovery progress over time, This is the overall evaluation score for the current period. This is the overall evaluation score for the previous period.

[0015] Preferably, the specific mechanism for transmitting all process data back to the cloud data center for model iterative optimization is as follows: A federated learning framework is established, where each offline service terminal fine-tunes the basic rehabilitation model locally using private data, uploading only the model gradient parameters. Instead of sending the raw data to the cloud, the cloud aggregates gradients from various terminals to update the global model. ; The aggregation formula is: ; in, For the updated global model parameters, For the number of terminals, For the first Number of samples per terminal The total number of samples, For the first Local model parameters of each terminal at time t; A concept drift detection mechanism is introduced to monitor the distribution changes of input data in real time, and the KS test statistic is used. Determine the differences in distribution: ; in, and Let be the empirical cumulative distribution functions of data samples from two different time periods, and let sup denote the supremum; when When the threshold is exceeded, the incremental learning process is automatically triggered; Establish an "expert-algorithm" closed-loop verification system. When the deviation between the rehabilitation plan recommended by the algorithm and the plan actually corrected by the technician exceeds a preset threshold, the abnormal case is marked as a difficult sample and added to the negative sample mining pool for retraining.

[0016] Preferably, the generation of a personalized physical rehabilitation instruction set for the target pet further includes adaptive adjustment logic based on environmental perception data: The temperature, humidity, and noise level in the rehabilitation room are collected by environmental sensors at offline service terminals. The model uses environmental parameters as covariates to calculate rehabilitation intensity. When the ambient temperature is higher than the preset comfort range, the frequency of rehabilitation movements is automatically reduced and a hydration reminder is added. By combining the video facial expression recognition results of pets during rehabilitation, a convolutional neural network is used to identify micro-expressions, such as changes in ear position, eye avoidance, and frequency of nose licking. If fear or resistance signals are identified, alternative action instructions with reduced difficulty are immediately generated and pushed to the technician's terminal.

[0017] In a second aspect of the invention, a personalized physical rehabilitation system for pets based on multimodal data fusion is also provided, comprising: A creation module is used to create a digital profile of the target pet, which includes the pet's basic physiological attributes and historical health records. The data acquisition module is used to collect multimodal monitoring data of the target pet during the current service period through an offline service terminal. The multimodal monitoring data includes at least behavioral characteristic data and physiological indicator data. The extraction module is used to upload multimodal monitoring data to the cloud data center, and use the data analysis engine to extract and fuse behavioral feature data and physiological indicator data to generate pet status feature vectors. The generation module is used to match and calculate based on the pet's state feature vector in a preset physical rehabilitation scheme algorithm library to generate a personalized physical rehabilitation execution instruction set for the target pet. The execution instruction set includes aromatherapy formula parameters, physiotherapy equipment operating parameters, manual operation standards, and rehabilitation duration. The push module is used to push personalized physical rehabilitation execution instruction sets to the technician's terminal, and the technician performs physical rehabilitation services on the pet according to the standardized operation instructions displayed on the terminal; The optimization module is used to collect real-time reaction data and post-rehabilitation data of the pet again during and after the service execution, calculate rehabilitation effect indicators based on a preset evaluation algorithm, and send all process data back to the cloud data center for model iteration and optimization.

[0018] In a third aspect of the invention, an electronic device is also provided. The electronic device includes at least one processor; and a memory communicatively connected to the at least one processor; the memory stores instructions executable by the at least one processor to enable the at least one processor to perform the method of the first aspect of the invention.

[0019] Compared with existing technologies, this invention provides a personalized physical rehabilitation method and system for pets based on multimodal data fusion, which has the following beneficial effects: 1. This invention overcomes the limitations of single data sources by fusing multimodal data such as behavioral characteristics (e.g., gait) and physiological indicators (e.g., heart rate, electromyography) and utilizing multi-head self-attention mechanisms to mine spatiotemporal correlations. It accurately identifies the pet's pain level and joint limitation, providing an objective basis for treatment plan development. Based on the pet's medical history and rehabilitation sensitivity baseline in its digital archive, combined with real-time environmental and facial expression recognition, it dynamically adjusts rehabilitation intensity, frequency, and aromatherapy formulas to achieve precise "one pet, one policy" rehabilitation, avoiding secondary harm or stress to pets caused by standardized procedures. Through edge computing and cloud collaboration, it monitors changes in physiological indicators in real time. Once abnormal heart rate or muscle fatigue signals are detected, the system automatically triggers a pop-up alarm and pauses the command, forcing entry into a safe observation mode, significantly improving the safety of the rehabilitation process. Furthermore, by using a federated learning framework to transmit process data, it achieves multi-terminal model iteration while protecting privacy. Combined with "expert-algorithm" closed-loop verification and concept drift detection, the system can automatically adapt to new cases and environmental changes, continuously improving the accuracy of rehabilitation effect evaluation and the scientific nature of treatment plan recommendations.

[0020] 2. This invention achieves precise rehabilitation for each pet with a tailored approach. For example, when the feature value is matched with the "aromatherapy solution algorithm library", if the feature value shows "high anxiety index", the algorithm will automatically generate a standardized and executable set of digital instructions, including "using lavender and chamomile compound essential oil diluted in a 1:3 ratio, combined with slow massage techniques, and simultaneously activating a low-intensity soothing pulse magnetic field for 15 minutes", which meets the needs of staff.

[0021] 3. This invention can also be applied in schools, where students can learn to use the data collection and evaluation tools of this system, simulate the development of healing plans for virtual pets in the system, and obtain system-based algorithm scores. Students can also access real, desensitized electronic health records and plan libraries for case study learning, thereby achieving deep integration of industry and education. Attached Figure Description

[0022] Figure 1 This is a schematic diagram of the method flow of S101-S106 in this invention; Figure 2This is a schematic diagram of the method flow for S201-S203 in this invention; Figure 3 This is a schematic diagram of the method flow for S301-S304 in this invention; Figure 4 This is a schematic diagram of the method flow for S401-S403 in this invention. Detailed Implementation

[0023] The following description is intended to disclose the invention and enable those skilled in the art to implement it. The preferred embodiments described below are merely examples, and other obvious variations will occur to those skilled in the art.

[0024] Example 1 Please refer to Figure 1 As shown, in a first aspect of the present invention, a method for personalized physical rehabilitation of pets based on multimodal data fusion using electronic devices is provided, comprising: S101. Establish a digital profile for the target pet, which includes the pet's basic physiological attributes and historical health records. S102. Collect multimodal monitoring data of the target pet during the current service period through offline service terminals. The multimodal monitoring data shall include at least behavioral characteristic data and physiological indicator data. S103. Upload the multimodal monitoring data to the cloud data center, and use the data analysis engine to extract and fuse the behavioral feature data and physiological indicator data to generate a pet status feature vector. S104. Based on the pet's state feature vector, match and calculate in the preset physical rehabilitation scheme algorithm library to generate a personalized physical rehabilitation execution instruction set for the target pet. The execution instruction set includes aromatherapy formula parameters, physiotherapy equipment operating parameters, manual operation standards and rehabilitation duration. S105. Push the personalized physical rehabilitation instruction set to the technician's terminal, and the technician performs physical rehabilitation services on the pet according to the standardized operation instructions displayed on the terminal; S106. During and after the service, collect the pet's real-time reaction data and post-rehabilitation data again, calculate the rehabilitation effect index based on the preset evaluation algorithm, and send all process data back to the cloud data center for model iteration and optimization.

[0025] Those skilled in the art will understand that this invention, by fusing multimodal data such as behavioral characteristics (e.g., gait) and physiological indicators (e.g., heart rate, electromyography), and utilizing multi-head self-attention mechanisms to mine spatiotemporal correlations, overcomes the limitations of single data sources, accurately identifies the level of pain and the degree of joint limitation in pets, and provides an objective basis for program development. Based on the pet's medical history and rehabilitation sensitivity baseline in its digital archive, combined with real-time environmental and facial expression recognition, it dynamically adjusts the intensity, frequency, and aromatherapy formula for rehabilitation, achieving precise rehabilitation tailored to each pet and avoiding secondary harm or stress caused to pets by standardized procedures. Through edge computing and cloud collaboration, it monitors changes in physiological indicators in real time. Once abnormal heart rate or muscle fatigue signals are detected, the system automatically triggers a pop-up alarm and pauses the command, forcing entry into a safe observation mode, significantly improving the safety of the rehabilitation process. Furthermore, by using a federated learning framework to transmit process data, it achieves multi-terminal model iteration while protecting privacy. Combined with expert-algorithm closed-loop verification and concept drift detection, the system can automatically adapt to new cases and environmental changes, continuously improving the accuracy of rehabilitation effect evaluation and the scientific nature of program recommendations.

[0026] Please refer to Figure 2 As shown, creating a digital profile for the target pet includes: S201. Collect the pet's breed, age, weight, microchip ID, and past medical history as basic static data, denoted as a vector. ; S202. Construct a pet health knowledge graph, transforming the medical records from historical health records into structured medical record vectors by extracting entities and mapping relationships using natural language processing techniques. ; S203. Utilize a time-series database to store physiological indicator change curves during historical rehabilitation processes, and calculate the attenuation factor of historical rehabilitation effects based on a sliding window algorithm. Generate pet-specific baseline parameters for rehabilitation sensitivity. ; Among them, attenuation factor The calculation follows the logic of exponentially weighted moving average: ; In the formula, This is the attenuation factor of the rehabilitation effect at the current moment. The decay factor is the value from the previous time step. The assessment score reflects the rehabilitation progress of the previous cycle. The smoothing coefficient is 0 < <1, used to adjust the weight of the influence of historical data on the current value; Baseline parameters of rehabilitation sensitivity Defined as the expected value of the ratio of historical rehabilitation stimulus to physiological response: ; In the formula, This refers to changes in physiological indicators. This represents the change in the intensity of the rehabilitation intervention. This represents the mathematical expectation operation.

[0027] Please refer to Figure 3 As shown, multimodal monitoring data of the target pet is collected through offline service terminals during the current service period, specifically including: S301. Collect kinematic data of the pet, including three-dimensional acceleration, using an inertial measurement unit deployed on rehabilitation equipment. angular velocity Euler angles To generate gait feature sequences ; S302. Simultaneously acquire the pet's body surface temperature distribution cloud map and heart rate variability data sequence using a non-contact infrared thermal imager and a photoplethysmography pulse wave sensor. S303. Introduce edge computing nodes at the data acquisition end to perform real-time denoising processing on the raw data stream. Specifically, wavelet threshold denoising algorithm is used to remove motion artifacts, and wavelet coefficients are used to... The processing logic is as follows: ; in, These are the original wavelet coefficients. The number of decomposition layers, The translation factor is... For the first The threshold of the layer, and , For the first Layer noise standard deviation The signal length; S304, and Kalman filtering is used to perform microsecond-level synchronization and alignment of the timestamps of heterogeneous sensors. The state update equation is: ; in, for The posterior state estimate at time t. These are the prior state estimates. For Kalman gain, For the observed values, This is the observation matrix.

[0028] The specific algorithm logic for feature extraction and fusion processing of behavioral characteristic data and physiological indicator data through the data analysis engine is as follows: behavioral characteristic data Perform multi-resolution wavelet packet decomposition to extract the energy entropy of each frequency band. kurtosis coefficient and approximate entropy As a characteristic of nonlinear dynamics; The formula for calculating energy entropy is as follows: ; ; in, Total number of frequency bands For the first Signal energy of each frequency band This refers to the proportion of energy in this frequency band to the total energy. Physiological indicator data Perform time-domain statistical analysis and extract the mean. Standard deviation skewness and kurtosis Power spectral density features were extracted by combining frequency domain analysis. ; Construct a multimodal fusion network based on a multi-head self-attention mechanism to integrate behavioral feature vectors. With physiological feature vector Mapped to a high-dimensional latent space, using the learned weight matrix , , The contribution of each modality is dynamically allocated to generate a high-dimensional pet state feature vector that includes spatiotemporal correlation. ; The formula for calculating attention weights is: ; In the formula, For querying the matrix, The key matrix, For value matrices, Let be the dimension of the key vector. It is a normalized exponential function; The multi-head mechanism output is: ; .

[0029] In the formula, For the number of attention heads, For the first Output of size , , For the first The weight matrix corresponding to each head This is for outputting the projection matrix.

[0030] Based on the pet's state feature vector, matching and calculation are performed in a pre-defined physical rehabilitation algorithm library, specifically including: The currently generated pet state feature vector The data is fed into a trained deep neural network classifier to identify the pet's current pain level and the degree of limitation in joint mobility. ; Based on the identification results, the rehabilitation movement sets of similar cases are retrieved from the physical rehabilitation program algorithm library, and the current gait sequence is calculated using the dynamic time warping algorithm. With standard rehabilitation gait sequence bending distance ; ; in, For the optimal alignment path, For path length, ( , () represents pairs of coordinate points on the path. The function for calculating Euclidean distance. and These are the feature vectors of the two sequences at corresponding points; Introduce a constrained optimization function to maximize the rehabilitation effect. Minimize pet stress response To achieve a dual objective, a genetic algorithm is used to automatically adjust the force threshold of rehabilitation exercises. Duration and rest time between groups Generate a personalized set of execution instructions that includes curves showing specific joint angle changes; The objective function is defined as: ; in, To synthesize the objective function value, , These are the weighting coefficients for rehabilitation effect and stress response, respectively. Maximum heart rate variability during rehabilitation; The constraints are: ; in, , These represent the lower and upper limits of safety, respectively. The threshold for muscle fatigue time in pets.

[0031] The specific calculation model for calculating rehabilitation effect indicators based on a preset evaluation algorithm includes: Construct a rehabilitation effect evaluation matrix, including the improvement rate of objective physiological indicators, subjective behavioral scores, and predicted values ​​of pet stress hormone levels; The weight coefficients of each level of indicators were determined by using the analytic hierarchy process (AHP) combined with an expert experience database. ,satisfy ; The fuzzy comprehensive evaluation algorithm is used to process mixed qualitative and quantitative data, and the membership level of rehabilitation effect is output. ,in For the weight vector, It is a fuzzy relation matrix. This represents a fuzzy composition operation; And calculate the rate of recovery progress in the current service period relative to the previous period: ; In the formula, For the first The rate of recovery progress over time, This is the overall evaluation score for the current period. This is the overall evaluation score for the previous period.

[0032] The specific mechanism for transmitting all process data back to the cloud data center for model iteration and optimization is as follows: A federated learning framework is established, where each offline service terminal fine-tunes the basic rehabilitation model locally using private data, uploading only the model gradient parameters. Instead of sending the raw data to the cloud, the cloud aggregates gradients from various terminals to update the global model. ; The aggregation formula is: ; in, For the updated global model parameters, For the number of terminals, For the first Number of samples per terminal The total number of samples, For the first Local model parameters of each terminal at time t; A concept drift detection mechanism is introduced to monitor the distribution changes of input data in real time, and the KS test statistic is used. Determine the differences in distribution: ; in, and Let be the empirical cumulative distribution functions of data samples from two different time periods, and let sup denote the supremum; when When the threshold is exceeded, the incremental learning process is automatically triggered; Establish an "expert-algorithm" closed-loop verification system. When the deviation between the rehabilitation plan recommended by the algorithm and the plan actually corrected by the technician exceeds a preset threshold, the abnormal case is marked as a difficult sample and added to the negative sample mining pool for retraining.

[0033] Please refer to Figure 4 As shown, when generating a personalized physical rehabilitation instruction set for the target pet, it also includes adaptive adjustment logic based on environmental perception data: S401. Collect the temperature, humidity and noise level of the rehabilitation room through environmental sensors at offline service terminals; S402. Input environmental parameters as covariates into the rehabilitation intensity calculation model. When the ambient temperature is higher than the preset comfort range, automatically reduce the frequency of rehabilitation movements and add a hydration reminder. S403. Combining the video facial expression recognition results of the pet during the rehabilitation process, the convolutional neural network is used to identify micro-expressions, such as changes in ear position, eye avoidance, and frequency of nose licking. If fear or resistance signals are identified, alternative action instructions with reduced difficulty are immediately generated and pushed to the technician's terminal.

[0034] In a second aspect of the invention, a personalized physical rehabilitation system for pets based on multimodal data fusion is also provided, comprising: The creation module is used to create a digital profile for the target pet, which includes the pet's basic physiological attributes and historical health records. The data collection module is used to collect multimodal monitoring data of the target pet during the current service period through offline service terminals. The multimodal monitoring data includes at least behavioral characteristic data and physiological indicator data. The extraction module uploads multimodal monitoring data to the cloud data center and uses the data analysis engine to extract and fuse behavioral feature data and physiological indicator data to generate pet status feature vectors. The generation module is used to match and calculate based on the pet's state feature vector in a preset physical rehabilitation program algorithm library to generate a personalized physical rehabilitation execution instruction set for the target pet. The execution instruction set includes aromatherapy formula parameters, physiotherapy equipment operating parameters, manual operation standards, and rehabilitation duration. The push module is used to push personalized physical rehabilitation instruction sets to the technician's terminal, and the technician performs physical rehabilitation services on the pet according to the standardized operation instructions displayed on the terminal; The optimization module is used to collect real-time reaction data and post-rehabilitation data of the pet during and after the service execution. It calculates rehabilitation effect indicators based on a preset evaluation algorithm and sends all process data back to the cloud data center for model iteration and optimization.

[0035] In a third aspect of the invention, an electronic device is also provided. The electronic device includes at least one processor; and a memory communicatively connected to the at least one processor; the memory stores instructions executable by the at least one processor, which, when executed by the at least one processor, enable the at least one processor to perform the method of the first aspect of the invention.

[0036] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claimed invention. The scope of protection claimed by the appended claims and their equivalents is defined.

Claims

1. A personalized physical rehabilitation method for pets based on multimodal data fusion, characterized in that, include: Establish a digital profile for the target pet, which includes the pet's basic physiological attributes and historical health records; Multimodal monitoring data of the target pet during the current service period is collected through offline service terminals. The multimodal monitoring data includes at least behavioral characteristic data and physiological indicator data. Multimodal monitoring data is uploaded to the cloud data center, and the data analysis engine performs feature extraction and fusion processing on behavioral feature data and physiological indicator data to generate pet status feature vectors. Based on the pet's state feature vector, matching and calculation are performed in a preset physical rehabilitation program algorithm library to generate a personalized physical rehabilitation execution instruction set for the target pet. The execution instruction set includes aromatherapy formula parameters, physiotherapy equipment operating parameters, manual operation standards, and rehabilitation duration. The personalized physical rehabilitation instruction set is pushed to the technician's terminal, and the technician performs physical rehabilitation services on the pet according to the standardized operation instructions displayed on the terminal; During and after the service, real-time reaction data and post-rehabilitation data of the pets are collected again. The rehabilitation effect index is calculated based on the preset evaluation algorithm, and all process data is sent back to the cloud data center for model iteration and optimization.

2. The personalized physical rehabilitation method for pets based on multimodal data fusion according to claim 1, characterized in that, The establishment of the target pet's digital profile specifically includes: The breed, age, weight, microchip ID, and past medical history of the pets are collected as basic static data, denoted as a vector. ; Construct a pet health knowledge graph by using natural language processing technology to extract entities and map relationships from historical health records and medical texts, transforming them into structured medical record vectors. ; We used a time-series database to store the physiological index change curves during historical rehabilitation processes, and calculated the decay factor of historical rehabilitation effects based on a sliding window algorithm. Generate pet-specific baseline parameters for rehabilitation sensitivity. ; Among them, attenuation factor The calculation follows the logic of exponentially weighted moving average: ； In the formula, This is the attenuation factor of the rehabilitation effect at the current moment. The decay factor is the value from the previous time step. The assessment score reflects the rehabilitation progress of the previous cycle. The smoothing coefficient is 0 < <1, used to adjust the weight of the influence of historical data on the current value; Baseline parameters of rehabilitation sensitivity Defined as the expected value of the ratio of historical rehabilitation stimulus to physiological response: ； In the formula, This refers to changes in physiological indicators. This represents the change in the intensity of the rehabilitation intervention. This represents the mathematical expectation operation.

3. The personalized physical rehabilitation method for pets based on multimodal data fusion according to claim 2, characterized in that, The collection of multimodal monitoring data of the target pet within the current service period through offline service terminals specifically includes: Kinematic data, including three-dimensional acceleration, of pets is collected using inertial measurement units deployed on rehabilitation equipment. angular velocity Euler angles To generate gait feature sequences ; The surface temperature distribution cloud map and heart rate variability data sequence of the pet were simultaneously acquired by a non-contact infrared thermal imager and a photoplethysmography pulse wave sensor. Edge computing nodes are introduced at the data acquisition end to perform real-time denoising on the raw data stream. Specifically, a wavelet threshold denoising algorithm is used to remove motion artifacts, and wavelet coefficients are used to... The processing logic is as follows: ； in, These are the original wavelet coefficients. The number of decomposition layers, The translation factor is... For the first The threshold of the layer, and , For the first Layer noise standard deviation The signal length; Kalman filtering is used to perform microsecond-level synchronization and alignment of the timestamps of heterogeneous sensors. The state update equation is as follows: ； in, for The posterior state estimate at time t. These are the prior state estimates. For Kalman gain, For the observed values, This is the observation matrix.

4. The personalized physical rehabilitation method for pets based on multimodal data fusion according to claim 3, characterized in that, The specific algorithm logic for feature extraction and fusion processing of behavioral feature data and physiological indicator data through the data analysis engine is as follows: behavioral characteristic data Perform multi-resolution wavelet packet decomposition to extract the energy entropy of each frequency band. kurtosis coefficient and approximate entropy As a characteristic of nonlinear dynamics; The formula for calculating energy entropy is as follows: ； ； in, Total number of frequency bands For the first Signal energy of each frequency band This refers to the proportion of energy in this frequency band to the total energy. Physiological indicator data Perform time-domain statistical analysis and extract the mean. Standard deviation skewness and kurtosis Power spectral density features were extracted by combining frequency domain analysis. ; Construct a multimodal fusion network based on a multi-head self-attention mechanism to integrate behavioral feature vectors. With physiological feature vector Mapped to a high-dimensional latent space, using the learned weight matrix , , The contribution of each modality is dynamically allocated to generate a high-dimensional pet state feature vector that includes spatiotemporal correlation. ; The formula for calculating attention weights is: ； In the formula, For querying the matrix, The key matrix, For value matrices, Let be the dimension of the key vector. It is a normalized exponential function; The multi-head mechanism output is: ； ； In the formula, For the number of attention heads, For the first Output of size , , For the first The weight matrix corresponding to each head This is for outputting the projection matrix.

5. The personalized physical rehabilitation method for pets based on multimodal data fusion according to claim 4, characterized in that, The matching and calculation based on the pet's state feature vector within a pre-defined physical rehabilitation algorithm library specifically includes: The currently generated pet state feature vector The data is fed into a trained deep neural network classifier to identify the pet's current pain level and the degree of limitation in joint mobility. ; Based on the identification results, the rehabilitation movement sets of similar cases are retrieved from the physical rehabilitation program algorithm library, and the current gait sequence is calculated using the dynamic time warping algorithm. With standard rehabilitation gait sequence bending distance ; ； in, For the optimal alignment path, For path length, ( , () represents a pair of coordinate points on the path. The function for calculating Euclidean distance. and These are the feature vectors of the two sequences at corresponding points; Introduce a constrained optimization function to maximize the rehabilitation effect. Minimize pet stress response To achieve a dual objective, a genetic algorithm is used to automatically adjust the force threshold of rehabilitation exercises. Duration and rest time between groups Generate a personalized set of execution instructions that includes curves showing specific joint angle changes; The objective function is defined as: ； in, To synthesize the objective function value, , These are the weighting coefficients for rehabilitation effect and stress response, respectively. Maximum heart rate variability during rehabilitation; The constraints are: ； in, , These represent the lower and upper limits of safety, respectively. The threshold for muscle fatigue time in pets.

6. The personalized physical rehabilitation method for pets based on multimodal data fusion according to claim 5, characterized in that, The specific calculation model for calculating rehabilitation effect indicators based on a preset evaluation algorithm includes: Construct a rehabilitation effect evaluation matrix, including the improvement rate of objective physiological indicators, subjective behavioral scores, and predicted values ​​of pet stress hormone levels; The weight coefficients of each level of indicators were determined by using the analytic hierarchy process (AHP) combined with an expert experience database. ,satisfy ; The fuzzy comprehensive evaluation algorithm is used to process mixed qualitative and quantitative data, and the membership level of rehabilitation effect is output. ,in For the weight vector, It is a fuzzy relation matrix. This represents a fuzzy composition operation; And calculate the rate of recovery progress in the current service period relative to the previous period: ； In the formula, For the first The rate of recovery progress over time, This is the overall evaluation score for the current period. This is the overall evaluation score for the previous period.

7. The personalized physical rehabilitation method for pets based on multimodal data fusion according to claim 6, characterized in that, The specific mechanism for transmitting all process data back to the cloud data center for model iteration and optimization is as follows: A federated learning framework is established, where each offline service terminal fine-tunes the basic rehabilitation model locally using private data, uploading only the model gradient parameters. Instead of sending the raw data to the cloud, the cloud aggregates gradients from various terminals to update the global model. ; The aggregation formula is: ； in, For the updated global model parameters, For the number of terminals, For the first Number of samples per terminal The total number of samples, For the first Local model parameters of each terminal at time t; A concept drift detection mechanism is introduced to monitor the distribution changes of input data in real time, and the KS test statistic is used. Determine the differences in distribution: ； in, and Let be the empirical cumulative distribution functions of data samples from two different time periods, and sup denote the supremum; when When the threshold is exceeded, the incremental learning process is automatically triggered; Establish an "expert-algorithm" closed-loop verification system. When the deviation between the rehabilitation plan recommended by the algorithm and the plan actually corrected by the technician exceeds a preset threshold, the abnormal case is marked as a difficult sample and added to the negative sample mining pool for retraining.

8. The personalized physical rehabilitation method for pets based on multimodal data fusion according to claim 7, characterized in that, The generation of a personalized physical rehabilitation instruction set for the target pet also includes adaptive adjustment logic based on environmental perception data. The temperature, humidity, and noise level in the rehabilitation room are collected by environmental sensors at offline service terminals. The model uses environmental parameters as covariates to calculate rehabilitation intensity. When the ambient temperature is higher than the preset comfort range, the frequency of rehabilitation movements is automatically reduced and a hydration reminder is added. By combining the video facial expression recognition results of pets during rehabilitation, a convolutional neural network is used to identify micro-expressions, such as changes in ear position, eye avoidance, and frequency of nose licking. If fear or resistance signals are identified, alternative action instructions with reduced difficulty are immediately generated and pushed to the technician's terminal.

9. A personalized physical rehabilitation system for pets based on multimodal data fusion, used to implement the personalized physical rehabilitation method for pets based on multimodal data fusion as described in any one of claims 1-8, characterized in that, include: A creation module is used to create a digital profile of the target pet, which includes the pet's basic physiological attributes and historical health records. The data acquisition module is used to collect multimodal monitoring data of the target pet during the current service period through an offline service terminal. The multimodal monitoring data includes at least behavioral characteristic data and physiological indicator data. The extraction module is used to upload multimodal monitoring data to the cloud data center, and use the data analysis engine to extract and fuse behavioral feature data and physiological indicator data to generate pet status feature vectors. The generation module is used to match and calculate based on the pet's state feature vector in a preset physical rehabilitation scheme algorithm library to generate a personalized physical rehabilitation execution instruction set for the target pet. The execution instruction set includes aromatherapy formula parameters, physiotherapy equipment operating parameters, manual operation standards, and rehabilitation duration. The push module is used to push personalized physical rehabilitation execution instruction sets to the technician's terminal, and the technician performs physical rehabilitation services on the pet according to the standardized operation instructions displayed on the terminal; The optimization module is used to collect real-time reaction data and post-rehabilitation data of the pet again during and after the service execution, calculate rehabilitation effect indicators based on a preset evaluation algorithm, and send all process data back to the cloud data center for model iteration and optimization.

10. An electronic device, comprising at least one processor; and a memory communicatively connected to said at least one processor; characterized in that, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-8.

Images