AI pet cat behavior monitoring method and system

Abstract

The invention discloses an AI pet cat behavior monitoring method and system, and relates to the technical field of pet behavior monitoring, and the method comprises the steps: carrying out the DWT of a fusion feature, obtaining a multi-scale wavelet coefficient, generating a fractional order enhanced wavelet coefficient through a fractional order derivative operator, constructing a causal enhanced rhythm feature through the combination of a Granger causal value and a gray correlation degree, and carrying out the recognition of the behavior of an AI pet cat. Topological entropy and conditional entropy are calculated, and final features are generated in combination with biological rhythm modulation and entanglement weight; according to the method, multi-scale wavelet coefficient analysis is introduced, periodicity and rhythm characteristics of behavior data are extracted in combination with fractional order enhanced wavelet coefficients, behavior mode changes of the pet cat under different time scales are effectively captured, and the accuracy of the pet cat behavior model is improved through rhythm adjustment weight and dynamic confidence calculation in combination with biological rhythm modulation and environmental factors. The accuracy of behavior classification and health assessment is improved, and the behavior recognition precision and the reliability of health early warning are improved.

Description

Technical Field

[0001] This invention relates to the field of pet behavior monitoring technology, and in particular to an AI-based method and system for monitoring the behavior of pet cats. Background Technology

[0002] The application of artificial intelligence technology in pet behavior monitoring has developed rapidly. In particular, driven by smart pet devices and data analysis technology, pet health and behavior analysis has gradually become a research hotspot. Smart collars, touch point monitoring devices, and environmental sensing devices based on sensor technology are widely used to collect pet behavior data. Through the fusion of multimodal data and combined with machine learning and deep learning algorithms, it is possible to achieve preliminary classification of pet behavior and monitoring of health status. Time-domain and frequency-domain feature extraction technology based on acceleration and angular velocity, attitude sequence analysis technology of video frames, and time-spectrum analysis technology of acoustic signals have shown good results in pet behavior recognition. Deep learning models have shown strong feature extraction capabilities in processing multimodal data, providing technical support for behavior classification and health assessment.

[0003] However, existing technologies still have shortcomings. The fusion of multimodal data remains at the level of simple feature splicing, lacking in-depth mining of causal relationships and topological structures between data, resulting in insufficient accuracy in behavior classification. It is easy to make misjudgments when distinguishing subtle behaviors. Existing technologies rarely introduce methods such as fractional-order enhancement or interactive topological entropy in the feature extraction process, making it difficult to capture the complex nonlinear features of behavioral data and limiting the system's ability to identify abnormal behaviors in the early stage. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides an AI pet cat behavior monitoring method and system, which solves the problem that the fusion of multimodal data remains at the level of simple feature splicing, lacking in-depth mining of causal relationships and topological structures between data, resulting in insufficient accuracy of behavior classification and easy misjudgment when distinguishing subtle behaviors. Existing technologies rarely introduce methods such as fractional-order enhancement or interactive topological entropy in the feature extraction process, making it difficult to capture the complex nonlinear features of behavioral data and limiting the system's ability to identify abnormal behaviors in the early stage.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: In a first aspect, the present invention provides an AI-based method for monitoring the behavior of pet cats, comprising, Behavioral data is collected and preprocessed to obtain fusion features. Multi-scale wavelet coefficients are generated through DWT and enhanced to obtain fractional-order enhanced wavelet coefficients. Rhythmic energy is calculated, dominant frequency components and periodic intensity are extracted, Granger causality value and grey relational degree are calculated, and causal enhanced rhythmic features are obtained by combining dominant frequency components and periodic intensity. Rhythm adjustment weights are calculated to obtain classification input features. The classification input features are input into a random forest classifier to generate behavior category probabilities. An optimization objective function is constructed and iteratively optimized to obtain optimized behavior category probabilities. Predicted behavior labels are generated, and dynamic confidence is calculated. Based on fused features and causal enhanced rhythm features, the k-nearest neighbor algorithm is used to establish topological neighbor relationships to obtain joint probability density. Single feature topological entropy, interaction topological entropy, and conditional entropy are calculated, and biological rhythm modulation is applied to generate modulated interaction topological features. Entanglement weights are calculated in combination with daily habits to obtain the final features. Based on the final features, a health score is calculated using a gradient boosting tree. The health score is then modulated by combining the weights of environmental and habit interaction features. A dynamically adjusted threshold is calculated, and a health alert is triggered through screening. The predicted behavior label and dynamic confidence level are then uploaded.

[0007] As a preferred embodiment of the AI ​​pet cat behavior monitoring method of the present invention, the step of combining the dominant frequency component and the periodic intensity to obtain causal enhanced rhythm features, calculating rhythm adjustment weights, and obtaining classification input features includes: The fusion features are normalized, multi-scale wavelet coefficients are generated by DWT and enhanced to obtain fractional-order enhanced wavelet coefficients. The rhythm energy is calculated and the main frequency component and period intensity are extracted. Calculate Granger causality value and grey relational degree, combine dominant frequency component and periodic intensity to obtain causal enhanced rhythm features, and combine rhythm adjustment weights to obtain classification input features.

[0008] As a preferred embodiment of the AI ​​pet cat behavior monitoring method of the present invention, the step of obtaining the optimized behavior category probability, generating predicted behavior labels, and calculating dynamic confidence includes: Based on the input features, the probability of behavior categories is obtained through a random forest classifier. An optimization objective function is constructed, and the optimal random forest classifier parameters are output through iterative optimization to obtain the optimized behavior category probability. The category with the maximum value is set as the predicted behavior label, and the dynamic confidence level is calculated.

[0009] As a preferred embodiment of the AI ​​pet cat behavior monitoring method of the present invention, the step of generating modulated interaction topological features, combining daily habits to calculate entanglement weights, and obtaining the final features includes: Based on the fusion feature combined with the causal enhanced rhythm feature, the k-nearest neighbor algorithm is used to construct the topological neighbor relationship, calculate the joint probability density between features, calculate the topological entropy of a single feature, the interaction topological entropy and the conditional entropy, apply biological rhythm modulation to obtain the modulated interaction topological feature, calculate the entanglement weight, and generate the final feature.

[0010] As a preferred embodiment of the AI ​​pet cat behavior monitoring method of the present invention, the step of triggering health alerts through screening and uploading predicted behavior tags and dynamic confidence scores includes: Based on the final features, a gradient boosting tree is used to calculate the health score, and the health score is modulated. Calculate the dynamic adjustment threshold. If the modulated health score is greater than the dynamic adjustment threshold, a health alert is triggered, and the predicted behavior label and dynamic confidence score are uploaded.

[0011] As a preferred embodiment of the AI ​​pet cat behavior monitoring method of the present invention, the step of collecting and preprocessing behavioral data includes: Behavioral data is collected and preprocessed, including high-pass filtering for denoising acceleration and angular velocity, median filtering for denoising claw contact area, band-pass filtering for denoising acoustic signals, inter-frame difference for extracting dynamic regions from video frames, one-hot encoding for species, and BERT model for converting daily habits into text vectors.

[0012] As a preferred embodiment of the AI ​​pet cat behavior monitoring method of the present invention, the obtained fusion features include: Based on behavioral data, motion feature vectors, claw contact change rate, posture features, Mel frequency cepstral coefficients, volume change rate, and environmental change rate are calculated, input into the VIT model, and spatiotemporal feature vectors and motion feature vectors are obtained to generate fused features.

[0013] Secondly, this invention provides an AI pet cat behavior monitoring system, comprising, The data collection and preprocessing module is used to collect behavioral data and perform preprocessing. The feature extraction and fusion module is used to extract features from behavioral data and then fuse them. The behavior analysis and classification module is used to generate predicted behavior labels by calculating rhythm energy, dominant frequency components and Granger causality values, combined with a random forest classifier. The health monitoring and early warning module is used to calculate a health score based on the final characteristics and combine environmental and habitual factors to generate modulated health warnings and dynamic confidence levels. The optimization and feedback module is used to iteratively optimize classifier parameters, optimize behavior category probabilities, dynamically adjust health scores, and ultimately trigger health alerts and upload data.

[0014] Thirdly, the present invention provides a computer device including a memory and a processor, wherein the memory stores a computer program, wherein when the computer program is executed by the processor, it implements any step of the AI ​​pet cat behavior monitoring method as described in the first aspect of the present invention.

[0015] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the AI ​​pet cat behavior monitoring method as described in the first aspect of the present invention.

[0016] The beneficial effects of this invention are as follows: This invention obtains multi-scale wavelet coefficients by using DWT on the fused features, generates fractional-order enhanced wavelet coefficients by fractional derivative operators, constructs causal enhanced rhythm features by combining Granger causality values ​​and grey relational degrees, calculates topological entropy and conditional entropy, and generates the final features by combining biological rhythm modulation and entanglement weights; effectively capturing changes in the behavioral patterns of pet cats at different time scales, improving the accuracy of behavioral classification and health assessment, and enhancing the precision of behavioral recognition and the reliability of health warnings. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart of the AI ​​pet cat behavior monitoring method in Example 1.

[0019] Figure 2 This is a schematic diagram of the AI ​​pet cat behavior monitoring system in Example 1. Detailed Implementation

[0020] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0021] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0022] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0023] Example 1, referring to Figure 1 and Figure 2 This is the first embodiment of the present invention, which provides an AI-based method for monitoring the behavior of pet cats, including the following steps: S1. Collect behavioral data and preprocess it to obtain fusion features. Generate multi-scale wavelet coefficients through DWT and enhance them to obtain fractional-order enhanced wavelet coefficients. Calculate rhythm energy, extract the main frequency component and period intensity, calculate Granger causality value and gray relational degree, combine the main frequency component and period intensity to obtain causal enhanced rhythm features, calculate rhythm adjustment weights, and obtain classification input features. Specifically, collecting and preprocessing behavioral data includes: Collect behavioral data, including deploying smart collars for pet cats, including triaxial accelerometers and gyroscopes, to collect raw motion data, including acceleration and angular velocity; Deploy FTIR touch points in the cat's bed or on the floor to collect the paw contact area, using the following formula: ; in, For time The contact area of ​​the claw. For time pixels Light intensity, This refers to the number of pixels. Obtain pet cat information, including age and daily habits, through API interface; Deploy temperature and humidity sensors, microphones, and infrared cameras to collect temperature, humidity, acoustic signals, and video frames; The collected data is preprocessed, including high-pass filtering for denoising acceleration and angular velocity, median filtering for denoising claw contact area, band-pass filtering for denoising acoustic signals, inter-frame difference extraction for dynamic regions of video frames, and conversion of daily habits into text vectors using the BERT model.

[0024] By integrating a three-axis accelerometer and gyroscope into a smart collar, the subtle movement trajectories of pet cats can be captured, such as the slight tremors when running, jumping, or stationary. This not only provides high-resolution raw motion data but also lays a solid foundation for subsequent behavior pattern recognition. The application of FTIR touch points on the cat's bed or floor innovatively quantifies paw contact dynamics, revealing gait changes in pet cats under different emotional or health states, such as the uneven distribution of paw pressure when anxious. This provides a new dimension for non-invasive behavior monitoring. Age and daily habit information obtained through the API interface, combined with environmental and multimedia data from temperature and humidity sensors, microphones, and infrared cameras, forms a comprehensive ecological data chain. Effective noise reduction through high-pass filtering, median filtering, and band-pass filtering ensures signal purity, while the application of inter-frame difference and BERT models further extracts dynamic regions and semantic vectors, enhancing the semantic richness of the data.

[0025] Furthermore, the fusion features are obtained, including: Based on acceleration and angular velocity, the mean and variance of angular velocity and acceleration are calculated, and the time domain features are concatenated to obtain the time domain features. The frequency domain features are calculated through fast Fourier transform, and the motion feature vector is obtained by horizontal stacking. Based on the claw contact area, the claw contact change rate is calculated using the following formula: ; in, For time The rate of change of claw contact, For time The contact area of ​​the claw. For time intervals; Based on video frames, keypoints such as the head, tail, and limbs are extracted using the ResNet-50 model. A pose sequence is constructed, and the pose change rate is calculated to obtain pose features. The formula is as follows: ; ; in, For time pose sequence, For time Key points x-coordinate For time Key points The ordinate, For the number of key points, For time The rate of change of attitude, For time Key points x-coordinate For time Key points The ordinate; The coordinate system is constructed by defining the top left corner of the video frame as the far point of the coordinate system, defining the horizontal direction to the right as the x-axis, and the vertical direction downward as the y-axis. Based on the acoustic signal, the time spectrum is obtained through short-time Fourier transform. The Mel-frequency cepstral coefficients and the rate of change of volume are then calculated using the following formulas: ; ; ; ; in, For time The total energy of the power spectrum, and The highest and lowest effective frequencies of a cat's meow. For time spectrum, For frequency, The integral symbol is used. For time The rate of change of volume, For time The total energy of the power spectrum, Mel power spectrum, representing the acoustic signal power in time and frequency. For the first The frequency response function (triangular window function) of a Mel filter indicates the frequency to be set on the Mel scale. For the first The Mel-frequency cepstral coefficients are the Mel-frequency cepstral coefficients. For the number of filters, For the first The center frequency of the Mel filter This is the phase correction constant; The rate of environmental change is calculated based on temperature and humidity using the following formula: ; in, For time Environmental change rate For time Temperature and humidity; The posture sequence and posture change rate are input into the VIT model to obtain the spatiotemporal feature vector. The motion feature vector, claw contact change rate, Mel frequency cepstral coefficient, volume change rate, and environmental change rate are input into the VIT model to obtain the motion feature vector. The spatiotemporal feature vector is then combined with the spatiotemporal feature vector and weighted summation to obtain the fused feature.

[0026] By calculating the time and frequency domain statistics of motion data and stacking them laterally to form a motion feature vector, this invention captures the rhythmic and periodic characteristics of pet cat movements. This is particularly crucial in identifying abnormal behaviors such as seizures or joint problems. The introduction of paw contact change rate innovatively quantifies the dynamic evolution of gait, revealing potential pain or discomfort signals and improving the sensitivity of behavioral analysis. In video frame processing, key points extracted by the ResNet-50 model are used to construct pose sequences and calculate the change rate, providing not only spatiotemporal continuity but also allowing for the analysis of complex behaviors such as hunting postures or social interactions. Fine-grained analysis and attitude modeling under this coordinate system definition enhance the geometric consistency of features. Time-spectrum analysis of acoustic signals and extraction of Mel frequency cepstral coefficients, combined with the rate of change of volume, accurately capture the spectral pattern of cat meows, helping to distinguish between normal meows and painful groans. The calculation of the rate of change of environment integrates the effects of temperature and humidity on behavior, revealing behavioral deviations under heat stress. Through weighted summation of spatiotemporal and motion features of the VIT model, high-dimensional fusion features are generated, significantly improving the discriminative power and generalization ability of features, reducing the risk of bias in single modes, and promoting multimodal collaborative decision-making.

[0027] Furthermore, by combining the dominant frequency component and periodic intensity to obtain causal enhanced rhythm features, and calculating rhythm adjustment weights, classification input features are obtained, including: The fused features are normalized, multi-scale wavelet coefficients are generated using DWT, and then enhanced to obtain fractional-order enhanced wavelet coefficients, as shown in the formula: ; in, For time scale Fractional-order enhanced wavelet coefficients, For fractional derivative operators, where For fractional order, the product of age and scaling factor is summed with the base value of the fractional order. For time scale Multiscale wavelet coefficients; The formula for calculating rhythm energy based on fractional-order enhanced wavelet coefficients is as follows: ; in, For scale The rhythmic energy; Based on rhythm energy, the main frequency component and period intensity are extracted separately, using the following formula: ; ; in, The main frequency component For scale The frequency range For scale quantity, Periodic intensity; The formulas for calculating Granger causality and grey relational degree are as follows: ; ; in, As a causal value, For the first in the fusion feature One characteristic, For behavior categories, one-hot encoding is numerical. Let covariance function be used. For time Behavioral categories, among which The lag order is... For gray relational degree, The distinguishing factor; Based on Granger causality and grey relational analysis, combined with dominant frequency components and periodic intensity, the causal enhancement rhythmic characteristics are obtained, as shown in the formula: ; in, To enhance the rhythmic features of causality, The wavelet rhythm features are obtained by horizontally stacking the main frequency component, period intensity, and rhythm energy. Based on daily habits, rhythm-adjusted weights are calculated to obtain the classification input features, using the following formula: ; ; in, Adjusting weights for rhythm, It is a day-night cycle. For mathematical constants, As a daily habit, Input features for classification, This is a feature of fusion.

[0028] By enhancing the multi-scale wavelet coefficients of DWT and introducing fractional derivative processing, a refined rhythmic energy analysis of the fused features was achieved. This not only captures the non-stationary characteristics of behavioral signals but also enhances the sensitivity of features to periodic patterns, such as identifying abnormal behaviors caused by circadian rhythm disorders. The extraction of dominant frequency components and periodic intensity further quantifies the dominant pattern and stability of the rhythm, innovatively revealing the causal relationship behind the behavior. The calculation of Granger causality value and grey relational degree provides an assessment of the causal strength between behavioral features and categories, which is rare in traditional methods. It effectively filters irrelevant noise and enhances the causal robustness of features. The causal enhanced rhythmic features formed by combining these elements significantly improve the interpretability and prediction accuracy of the classifier. The rhythmic adjustment weight calculation based on daily habits incorporates biological clock factors and dynamically adjusts feature weights, ensuring the adaptability of input features to individual differences, such as the slow rhythm of older cats and the active pattern of kittens. This constructs a highly adaptive classification input framework, potentially reducing the misclassification rate and providing a scientific basis for personalized pet behavior intervention.

[0029] S2. Input the classification input features into a random forest classifier to generate behavior category probabilities, construct an optimization objective function for iterative optimization to obtain optimized behavior category probabilities, generate predicted behavior labels, and calculate dynamic confidence. Based on fused features and causal enhanced rhythm features, use the k-nearest neighbor algorithm to establish topological neighbor relationships to obtain joint probability density. Calculate single-feature topological entropy, interaction topological entropy, and conditional entropy, apply biological rhythm modulation to generate modulated interaction topological features, and calculate entanglement weights based on daily habits to obtain the final features. Specifically, the process involves obtaining the optimized behavior category probability, generating predicted behavior labels, and calculating dynamic confidence scores, including: Based on the classification input features, the probability of behavior category is obtained through a random forest classifier; Based on the input features for classification, an optimization objective function is constructed. The optimal random forest classifier parameters (number of trees, maximum depth, and number of features) are iteratively optimized to obtain the optimized behavior category probability, as shown in the formula: ; in, To optimize the number of trees, The optimized maximum tree depth, The optimized feature number, For five-fold cross-validation, The initial number of trees, The maximum tree depth, The characteristic number; The probability of the optimized behavior category is sorted in descending order, and the category with the highest value is set as the predicted behavior label, such as walking or limping. The formula for calculating dynamic confidence level is: ; in, For dynamic confidence level, To optimize the probability of behavior categories, For rhythm intensity weighting, This is the standard deviation function.

[0030] By utilizing a random forest classifier to process the input features, an initial probability distribution of behavior categories is generated. This not only captures the diversity of behavior patterns but also lays the foundation for subsequent optimization. By constructing an optimization objective function and performing iterative optimization, the system can dynamically adjust key parameters of the classifier, such as the number of trees, maximum depth, and number of features, ensuring the robustness of the model on different subsets of data. This results in more accurate optimized behavior category probabilities, significantly improving the classifier's generalization ability, reducing the risk of overfitting, and increasing the accuracy of behavior prediction in practical applications. Sort the optimized probabilities in descending order and select the maximum value as the predicted behavior label, which not only simplifies the decision-making process but also ensures the determinism of the label. The calculation of dynamic confidence incorporates rhythm intensity weights and standard deviation evaluation, which gives the confidence context sensitivity, enabling the system to maintain high reliability in noisy environments.

[0031] Furthermore, modulated interaction topological features are generated, and entanglement weights are calculated based on everyday habits to obtain the final features, including: Based on the fusion feature combined with the causal enhanced rhythm feature, the k-nearest neighbor algorithm is used to construct topological neighbor relationships, and the joint probability density between features is calculated using the following formula: ; in, For the joint topological probability distribution, For the first One characteristic, For the first One characteristic, For feature pairs The number of times they co-occur in topological neighbor relationships. The number of features; Based on the joint probability density, the topological entropy of a single feature is calculated using the following formula: ; in, For single-feature topological entropy, For the first Features The single-feature topological probability distribution; The formulas for calculating the interaction topological entropy and conditional entropy are as follows: ; ; ; in, For interactive topological entropy, For mutual information, For conditional entropy; Based on conditional entropy, biological rhythm modulation is applied to obtain the modulation interaction topological features, as shown in the formula: ; in, To modulate interactive topological features, For biological rhythm frequency; Based on everyday habits, the entanglement weights are calculated to generate the final features, using the following formula: ; ; in, To entangle weights, The covariance between features As the final feature, for Activation function.

[0032] By fusing features with causal-enhanced rhythmic features, the k-nearest neighbor algorithm is used to establish topological neighbor relationships and calculate the joint probability density between features. This not only reveals the spatial correlation between features but also quantifies their co-occurrence patterns, thus providing fine-grained insights into complex data interactions. Calculating the topological entropy of a single feature further assesses the information content of individual features, while the introduction of interactive topological entropy, mutual information, and conditional entropy deepens the understanding of feature dependencies and enhances the model's ability to capture nonlinear relationships. Based on conditional entropy, biorhythm modulation is applied to generate modulated interactive topological features. This step incorporates time-periodic factors, such as the human diurnal rhythm, making the features more aligned with biological reality and enhancing their dynamic adaptability. Especially in health monitoring, it can better reflect individual differences. Combining daily habits to calculate entanglement weights and generating final features through activation functions, this weighting mechanism considers feature covariance, simulates strong correlations, improves the robustness of feature fusion, significantly increases the richness of feature representation and prediction accuracy, and reduces information redundancy.

[0033] S3. Based on the final features, use gradient boosting trees to calculate the health score, and combine the environmental and habit interaction feature weights to modulate the health score, calculate the dynamic adjustment threshold, trigger health warnings through screening, and upload the predicted behavior labels and dynamic confidence scores. Specifically, health alerts are triggered through screening, and predicted behavioral tags and dynamic confidence scores are uploaded, including: Based on the final features, a gradient boosting tree is used to calculate the health score, using the following formula: ; in, For health score, For the number of trees, For trees The predicted value; The weights of the interaction features between environment and habits are calculated by multiplying the rate of change of environment and daily habits using the sigmoid function, and then modulating the health score using the following formula: ; in, To adjust the health score, Weights for the interaction features between environment and habits; The formula for calculating the dynamic adjustment threshold is: ; in, To dynamically adjust the threshold, As the initial threshold, For age, This is the scaling factor; If the modulated health score is greater than the dynamic adjustment threshold, a health alert is triggered, and the predicted behavior label and dynamic confidence score are uploaded.

[0034] By using gradient boosting trees to calculate health scores based on final features, this ensemble learning method aggregates predictions from multiple decision trees, enhancing the model's robustness and accuracy. Particularly when dealing with high-dimensional health data, it effectively captures non-linear patterns, optimizes the efficiency of health assessment, reduces computational complexity, and provides scalability for multi-source data fusion. It calculates the interaction weights of environmental and habitual features, and corrects health scores by applying sigmoid modulation to the interaction between environmental change rates and daily habits. This innovatively bridges behavioral data with environmental influences. The dynamic adjustment of threshold calculations considers individual factors such as age and scaling factors, reducing false positives and improving sensitivity, thus enhancing the accuracy and user-friendliness of health monitoring.

[0035] This embodiment also provides an AI pet cat behavior monitoring system, including: The data collection and preprocessing module is used to collect behavioral data and perform preprocessing. The feature extraction and fusion module is used to extract features from behavioral data and then fuse them. The behavior analysis and classification module is used to generate predicted behavior labels by calculating rhythm energy, dominant frequency components and Granger causality values, combined with a random forest classifier. The health monitoring and early warning module is used to calculate a health score based on the final characteristics and combine environmental and habitual factors to generate modulated health warnings and dynamic confidence levels. The optimization and feedback module is used to iteratively optimize classifier parameters, improve behavior category probabilities, dynamically adjust health scores, and ultimately trigger health alerts and upload data. This embodiment also provides a computer device applicable to the AI ​​pet cat behavior monitoring method, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to realize the AI ​​pet cat behavior monitoring method proposed in the above embodiment.

[0036] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0037] This embodiment also provides a storage medium storing a computer program, which, when executed by a processor, implements the AI ​​pet cat behavior monitoring method proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0038] In summary, this invention obtains multi-scale wavelet coefficients from the fused features using DWT, generates fractional-order enhanced wavelet coefficients through fractional derivative operators, constructs causal enhanced rhythm features by combining Granger causality values ​​and grey relational degrees, calculates topological entropy and conditional entropy, and generates the final features by combining biological rhythm modulation and entanglement weights; effectively capturing changes in the behavioral patterns of pet cats at different time scales, improving the accuracy of behavioral classification and health assessment, and enhancing the precision of behavioral recognition and the reliability of health warnings.

[0039] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. An AI-powered method for monitoring the behavior of pet cats, characterized in that: include, Behavioral data is collected and preprocessed to obtain fusion features. Multi-scale wavelet coefficients are generated through DWT and enhanced to obtain fractional-order enhanced wavelet coefficients. Rhythmic energy is calculated, dominant frequency components and periodic intensity are extracted, Granger causality value and grey relational degree are calculated, and causal enhanced rhythmic features are obtained by combining dominant frequency components and periodic intensity. Rhythm adjustment weights are calculated to obtain classification input features. The classification input features are input into a random forest classifier to generate behavior category probabilities. An optimization objective function is constructed and iteratively optimized to obtain optimized behavior category probabilities. Predicted behavior labels are generated, and dynamic confidence is calculated. Based on fused features and causal enhanced rhythm features, the k-nearest neighbor algorithm is used to establish topological neighbor relationships to obtain joint probability density. Single feature topological entropy, interaction topological entropy, and conditional entropy are calculated, and biological rhythm modulation is applied to generate modulated interaction topological features. Entanglement weights are calculated in combination with daily habits to obtain the final features. Based on the final features, a health score is calculated using a gradient boosting tree. The health score is then modulated by combining the weights of environmental and habit interaction features. A dynamically adjusted threshold is calculated, and a health alert is triggered through screening. The predicted behavior label and dynamic confidence level are then uploaded.

2. The AI ​​pet cat behavior monitoring method as described in claim 1, characterized in that: The causal enhanced rhythm features are obtained by combining the dominant frequency component and the periodic intensity, and the rhythm adjustment weights are calculated to obtain the classification input features, including: The fusion features are normalized, multi-scale wavelet coefficients are generated by DWT and enhanced to obtain fractional-order enhanced wavelet coefficients. The rhythm energy is calculated and the main frequency component and period intensity are extracted. Calculate Granger causality value and grey relational degree, combine dominant frequency component and periodic intensity to obtain causal enhanced rhythm features, and combine rhythm adjustment weights to obtain classification input features.

3. The AI ​​pet cat behavior monitoring method as described in claim 2, characterized in that: The process of obtaining the optimized behavior category probability, generating predicted behavior labels, and calculating dynamic confidence includes: Based on the input features, the probability of behavior categories is obtained through a random forest classifier. An optimization objective function is constructed, and the optimal random forest classifier parameters are output through iterative optimization to obtain the optimized behavior category probability. The category with the maximum value is set as the predicted behavior label, and the dynamic confidence level is calculated.

4. The AI ​​pet cat behavior monitoring method as described in claim 3, characterized in that: The generated modulation interaction topological features, combined with everyday habits to calculate entanglement weights, yield the final features, including: Based on the fusion feature combined with the causal enhanced rhythm feature, the k-nearest neighbor algorithm is used to construct the topological neighbor relationship, calculate the joint probability density between features, calculate the topological entropy of a single feature, the interaction topological entropy and the conditional entropy, apply biological rhythm modulation to obtain the modulated interaction topological feature, calculate the entanglement weight, and generate the final feature.

5. The AI ​​pet cat behavior monitoring method as described in claim 4, characterized in that: The process of triggering health alerts through screening and uploading predicted behavioral tags and dynamic confidence scores includes: Based on the final features, a gradient boosting tree is used to calculate the health score, and the health score is modulated. Calculate the dynamic adjustment threshold. If the modulated health score is greater than the dynamic adjustment threshold, a health alert is triggered, and the predicted behavior label and dynamic confidence score are uploaded.

6. The AI ​​pet cat behavior monitoring method as described in claim 5, characterized in that: The collection and preprocessing of behavioral data includes: Behavioral data is collected and preprocessed, including high-pass filtering for denoising acceleration and angular velocity, median filtering for denoising claw contact area, band-pass filtering for denoising acoustic signals, inter-frame difference extraction for dynamic regions of video frames, and conversion of daily habits into text vectors using the BERT model.

7. The AI ​​pet cat behavior monitoring method as described in claim 6, characterized in that: The obtained fusion features include: Based on behavioral data, motion feature vectors, claw contact change rate, posture features, Mel frequency cepstral coefficients, volume change rate, and environmental change rate are calculated, input into the VIT model, and spatiotemporal feature vectors and motion feature vectors are obtained to generate fused features.

8. An AI pet cat behavior monitoring system, based on the AI ​​pet cat behavior monitoring method according to any one of claims 1 to 7, characterized in that: include, The data collection and preprocessing module is used to collect behavioral data and perform preprocessing. The feature extraction and fusion module is used to extract features from behavioral data and then fuse them. The behavior analysis and classification module is used to generate predicted behavior labels by calculating rhythm energy, dominant frequency components and Granger causality values, combined with a random forest classifier. The health monitoring and early warning module is used to calculate a health score based on the final characteristics and combine environmental and habitual factors to generate modulated health warnings and dynamic confidence levels. The optimization and feedback module is used to iteratively optimize classifier parameters, optimize behavior category probabilities, dynamically adjust health scores, and ultimately trigger health alerts and upload data.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the AI ​​pet cat behavior monitoring method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the AI ​​pet cat behavior monitoring method according to any one of claims 1 to 7.

Images