A multi-source data closed-loop fusion intelligent maintenance analysis method based on pet AI

Abstract

The application discloses a kind of multi-source data closed loop fusion intelligent maintenance analysis methods based on pet AI, belong to artificial intelligence algorithm, computer software data processing, big data heterogeneous fusion, pet intelligent maintenance technical field.This method relies on computer program scheduling and pet exclusive AI model operation, integrates pet real-time state data, self-built private database data and maintenance product actual use data, through multi-source data cross-checking, long-short cycle fusion operation and full-link closed loop iteration, realize the precise analysis of pet customized product intelligent adaptation, scene maintenance intervention.The application solves the problems of single data dimension, poor fault tolerance, large analysis deviation, easy to avoid, weak technical barriers and other problems of existing pet intelligent maintenance scheme, through multi-dimensional linkage verification architecture, open process design, double-layer determination calibration and closed loop iteration optimization, improves the precision and adaptability of maintenance analysis, and can be widely applied to various pet intelligent maintenance scenes.

Description

Technical Field

[0001] This invention relates to the fields of artificial intelligence algorithms, computer software data processing, big data heterogeneous fusion, and intelligent pet care technology, and in particular to an intelligent pet care analysis method based on multi-source data closed-loop fusion using pet AI. Background Technology

[0002] In recent years, with the improvement of people's living standards, pets have gradually become important members of families. The demand for refined and intelligent pet care has continued to increase, driving the rapid development of related technologies for intelligent pet care. Currently, various intelligent pet care analysis solutions have emerged on the market. These solutions mainly rely on pet wearable devices and environmental monitoring equipment to collect relevant data, and combine them with simple algorithm models to complete care analysis and provide suggestions, thereby improving the convenience of pet care to a certain extent.

[0003] However, existing intelligent pet care analysis solutions generally suffer from numerous technical shortcomings, making it difficult to meet the demands for high-precision, iterative, and highly protected intelligent pet care analysis. Specifically, these shortcomings are as follows: First, the data dimensions are limited, resulting in insufficient analytical precision. Most existing solutions only collect single types of pet data (such as physiological or behavioral data), lacking a collaborative verification mechanism for multi-source data. They rely solely on simple label matching for information processing, leading to poor data tolerance, susceptibility to abnormal data interference, and significant bias in analysis results. For example, some solutions determine health status solely based on pet body temperature data without cross-validating with behavioral activity and environmental data, easily resulting in misjudgments. Other solutions only collect short-term pet data, failing to incorporate long-term accumulated data, thus failing to reflect long-term pet care trends and resulting in untargeted care recommendations.

[0004] Secondly, the rigid processes are easily circumvented by competitors. Traditional smart pet care solutions have fixed business processes and timelines, with the execution order of each step being unchangeable. This makes them vulnerable to circumvention by competitors through methods such as changing steps, removing modules, or replacing simple components, resulting in insufficient technical protection. Furthermore, most existing solutions lack private, end-to-end data accumulation, relying heavily on public databases or third-party data interfaces. This not only poses data security risks but also hinders long-term data accumulation and reuse, limiting technological iteration and upgrades.

[0005] Third, the technological barriers are weak and the creativity is insufficient. Conventional solutions mostly rely on human-defined rules, meaning that collected data is judged and analyzed through pre-set manual rules. This lacks underlying computer data optimization logic and fails to incorporate the autonomous learning and iterative capabilities of AI models, resulting in insufficient technological innovation and a weak overall technological barrier. Furthermore, some solutions are deeply tied to business operation rules, posing a high risk of patent infringement examination and making it difficult to pass patent grant examination. Moreover, it is difficult to clearly define the boundaries of protection during subsequent rights protection processes, easily leading to problems in defining infringement disputes.

[0006] Fourth, poor adaptability, failing to meet personalized needs. Existing solutions mostly adopt uniform analysis standards and output modes, failing to fully consider individual differences such as different pet breeds, ages, physical conditions, and breeding environments, and also failing to dynamically adjust based on the actual usage scenarios and feedback data of pet care products. This results in the output of care suggestions and product adaptation results being not very targeted, and failing to meet the personalized care needs of different users.

[0007] In summary, existing intelligent pet care analysis solutions have significant shortcomings in data processing, process design, technological innovation, and adaptability. There is an urgent need for an intelligent pet care analysis method that can solve these problems, achieve multi-source data collaboration, flexible processes, high technical barriers, and strong adaptability, so as to promote the upgrading and development of intelligent pet care technology. Summary of the Invention

[0008] The purpose of this invention is to overcome the shortcomings of existing pet intelligent care analysis solutions, such as single data dimension, low analysis accuracy, rigid and easily circumvented processes, weak technical barriers, and poor adaptability, and to provide a multi-source data closed-loop fusion intelligent care analysis method based on pet AI.

[0009] A closed-loop fusion intelligent pet care analysis method based on multi-source data using pet AI includes the following steps: S1. Collect real-time pet status data and actual usage data of pet care products; the real-time pet status data includes pet physiological characteristic data, behavioral activity data, environmental scene data, and instantaneous health performance data; the actual usage data of pet care products includes product usage behavior data, usage habit data, adaptation feedback data, and long-term application effect data; during the collection process, short-range wireless timestamp broadcasting is used in conjunction with local time service to unify the time base and ensure the time synchronization of various real-time data. S2. Retrieve the self-built private pet database to obtain pet lifecycle feeding data; the self-built private pet database stores pet basic profiles, historical health records, long-term feeding behavior data, and structured and unstructured data accumulated throughout the feeding cycle, and uses encrypted storage with strict access permissions. S3. Using a pet-specific AI model and computer computing power scheduling, the system performs unified cleaning, abnormal data filtering, core feature extraction, and data dimension standardization on multi-source collected data and database data. For unstructured image data, the device first performs face masking, foreground separation, and skeleton key point extraction, and only uploads structured features containing key point coordinates and foreground area ratio. S4. Based on the judgment results of the pet-specific AI model, perform bidirectional cross-validation, feature association constraints, and long- and short-cycle linkage weighted fusion calculations on multiple types of data, and combine instantaneous scenarios and long-term feeding cycles to perform dual-layer linkage judgment calibration; the bidirectional cross-validation includes multi-source data consistency comparison, abnormal data interception, conflict data correction, and heterogeneous data mutual verification. S5. Based on the fusion operation calibration results, the customized maintenance analysis results are output in a hierarchical gradient. The analysis results include any one or more of the following: intelligent adaptation results of a single product, scenario-based maintenance intervention analysis results, and analysis results combining products and maintenance interventions. The output can be provided through various channels such as computer software systems, mobile terminal APPs, and smart device terminals, in the form of text, charts, voice, etc. S6. Collect feedback data and actual operation data after the implementation of the scheme and feed them back to the self-built private database. Iterate and optimize the AI ​​model parameters and data matching rules in reverse to form a continuously self-updating closed-loop intelligent computing mechanism. For unknown data that the AI ​​model cannot accurately identify, perform data error correction and convert it into effective data. Incorporate it into the iterative data system and use incremental training mode to train the AI ​​model. Furthermore, the pet's physiological characteristics data include one or more of the following: body temperature, heart rate, respiratory rate, weight, blood pressure, blood sugar, and coat condition; the behavioral activity data includes one or more of the following: pet's activity level, sleep duration, food intake, water intake frequency, excretion, and interactive behavior; the environmental scene data includes one or more of the following: temperature and humidity of the pet's living environment, air quality, light intensity, and noise level; and the instantaneous health performance data includes one or more of the following: pet's mental state, vomiting, diarrhea, coughing, and skin abnormalities.

[0010] Furthermore, the self-built private pet database stores basic pet profiles including one or more of the following: pet breed, age, gender, weight, pedigree, and medical history; the long-term accumulated unstructured data includes one or more of the following: user feeding notes, pet photos, and pet videos; the database supports real-time data updates, fast retrieval, and encrypted storage to ensure data security, privacy, and reusability.

[0011] Furthermore, in the weighted fusion calculation of short and long cycles, short-term data includes real-time status data and recent product usage data, with a weight set to 0.5-0.7; long-term data includes historical health records in the database, long-term feeding data, and long-term product usage effect data, with a weight set to 0.3-0.5; the weights can be dynamically adjusted through an AI model according to individual pet differences and actual care scenarios.

[0012] Furthermore, when performing scenario-based maintenance intervention analysis, corresponding scenario execution data and effect feedback data are collected simultaneously and incorporated into the multi-source data fusion operation and AI model iteration optimization process; the scenario execution data includes one or more of environmental adjustment data, behavior guidance data, and disease prevention data; the effect feedback data includes one or more of pet status change data and user evaluation data.

[0013] Furthermore, the pet-specific AI model is built based on deep learning algorithms and is specifically trained for the physiological characteristics, behavioral characteristics, and care needs of pets. It can achieve functions such as abnormal data detection, core feature extraction, multi-source data fusion and analysis, and parameter autonomous iterative optimization. The specific algorithm and model architecture of the AI ​​model can be replaced, but the replacement will not deviate from the protection scope of this invention.

[0014] Furthermore, the data cleaning includes removing blank values, duplicate values, and invalid values, and correcting data entry errors and format errors; the abnormal data filtering uses an AI model anomaly detection algorithm to identify data that exceeds the normal range, abnormal fluctuation data, and interference data, and after secondary verification by combining pet historical data and scene characteristics, the abnormal data is filtered.

[0015] Furthermore, the multi-dimensional standardized alignment conversion transforms multi-source data from different sources, of different types, in different formats, and in different units into a unified format and measurement standard, ensuring that multi-source data can be effectively integrated and collaboratively analyzed; the data format includes structured data format and structured format after unstructured data conversion.

[0016] The beneficial effects of this invention are as follows: 1. Improve the accuracy and adaptability of pet care analysis: Through bidirectional cross-validation of multi-source heterogeneous data, abnormal interference data is effectively filtered out and calculation deviations are corrected to ensure data reliability. By combining long- and short-cycle weighted fusion calculation and two-layer judgment calibration, collaborative analysis of short-term and long-term data, instantaneous scenarios and long-term trends is achieved. This makes the output of customized pet care analysis results and product matching suggestions more in line with the individual differences and actual care needs of pets, significantly improving the accuracy and adaptability of pet care analysis and providing more personalized and refined care support for pets.

[0017] 2. Strengthen technical protection and reduce the risk of infringement evasion: The dual protection mechanism of process architecture protection and core technology architecture as a backup is adopted, which clarifies the scope of protection and reduces the space for competitors to evade infringement through simple modification, partial replication, process adjustment and other means. The technical barrier is solid and can effectively protect the invention results and provide strong support for subsequent patent rights protection.

[0018] 3. Enhanced system scalability and adaptability: The full-link software closed-loop iterative architecture can continuously optimize AI model parameters and data matching rules based on actual operating data and user feedback, enabling autonomous updates and upgrades of technical solutions. It can adapt to the care needs of different pet breeds, ages, and physical conditions, as well as changes in different application scenarios and hardware conditions. The system has stronger scalability and adaptability and can be widely used in various intelligent pet care products and scenarios.

[0019] 4. Balancing Patent Licensing and Rights Protection Needs: The overall technical limitations are reasonable and the wording is rigorous, clearly defining the core technical features and scope of protection. There are no commercial operation rules binding it, and the subject matter is highly compliant, meeting the standards for software method patent licensing, which can effectively improve the success rate of patent licensing. At the same time, the catch-all wording is clear, clarifying the core basis for infringement determination, which facilitates the identification of infringing behavior in the subsequent rights protection process and reduces the difficulty and cost of rights protection.

[0020] 5. Ensure data security and privacy: We have built a self-owned private database for all pets, using encrypted storage and strict access control to ensure the security and privacy of pet data and avoid the risk of data leakage. At the same time, during the data collection and processing, we have implemented privacy protection measures to further protect the privacy of users and pets and improve the user experience.

[0021] 6. Reduce user maintenance costs and improve maintenance efficiency: Through precise product matching and scenario-based maintenance intervention suggestions, users can avoid blindly purchasing unsuitable pet care products, reduce resource waste, and lower maintenance costs. At the same time, the intelligent analysis and output methods simplify the user's maintenance operation process, improve the convenience and efficiency of pet maintenance, and reduce the user's maintenance burden. Attached Figure Description

[0022] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0023] Figure 1 This is a flowchart of the intelligent pet care analysis method based on multi-source data closed-loop fusion according to pet AI of the present invention; Figure 2 This is a flowchart of the data acquisition process of the present invention. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0025] The present invention relates to a multi-source data closed-loop fusion intelligent pet care analysis method based on pet AI. This method is executed by a computer software system and adopts a full-link architecture of "data acquisition - integration processing - fusion calculation - adaptation output - closed-loop iteration." Each link is interconnected and works synergistically to form a complete intelligent pet care analysis system. The specific steps are as follows: 1. Data Collection A multi-channel data collection terminal is built through a backend software program to achieve real-time and comprehensive collection of various data types, ensuring the diversity and integrity of data sources and providing reliable support for subsequent data processing and fusion operations. The specific collection content includes two main parts: The first part involves real-time pet status data collection. This includes collecting data on pet physiological characteristics, behavioral activities, environmental scenarios, and instantaneous health conditions through smart wearable devices (such as smart collars and harnesses), environmental monitoring devices (such as temperature and humidity sensors and air quality sensors for the pet's living environment), image acquisition devices (such as smart cameras), and user interaction terminals (such as a pet care app). Physiological characteristic data includes the pet's body temperature, heart rate, respiratory rate, weight, blood pressure, blood sugar, and coat condition; behavioral activity data includes the pet's activity level, sleep duration, food intake, water intake frequency, excretion, and interactive behaviors; environmental scenario data includes the temperature and humidity of the pet's living environment, air quality, light intensity, and noise levels; and instantaneous health condition data includes the pet's mental state, vomiting, diarrhea, coughing, skin abnormalities, and other sudden health-related symptoms. During the collection process, short-range wireless timestamp broadcasting, combined with a local time service, ensures the time synchronization of various real-time data, avoiding data conflicts caused by time discrepancies.

[0026] The second part involves collecting data on the actual use of pet care products. This includes simultaneously collecting data on the actual usage behavior, habits, adaptation feedback, and long-term application effects of pet care products (such as pet food, supplements, grooming products, toys, and medical supplies). Usage behavior data includes product usage frequency, duration, dosage, and usage scenarios; usage habit data includes user feeding habits, product replacement cycles, and usage preferences; adaptation feedback data includes pet acceptance of the product, changes in pet's condition after use, and user evaluation of the product's adaptation effect; and long-term application effect data includes changes in the pet's physiological state, behavior, and health status after a period of product use. All collected data is uploaded to the computer software system via encrypted transmission to ensure the security and integrity of data transmission.

[0027] 2. Data integration and processing First, the system accesses a self-built, dedicated private database for pets. This database stores data accumulated throughout the entire pet lifecycle, including basic pet profiles (breed, age, sex, weight, pedigree, medical history, etc.), historical health records (past diagnoses, treatments, and health check reports), long-term feeding behavior data (e.g., feeding patterns, sleep patterns, activity patterns), and structured and unstructured data accumulated throughout the lifecycle (e.g., user feeding notes, pet photos, videos, etc.). The self-built private database employs encrypted storage and strict access permissions to ensure data security and privacy, while supporting real-time updates and rapid retrieval, providing data support for multi-source data fusion and computation.

[0028] Secondly, a pet-specific AI model is used in conjunction with computing power to uniformly clean, filter out abnormal data, extract core features, and perform multi-dimensional standardized alignment and conversion processing on multi-source collected data (including real-time status data and care product usage data) and database inventory data. This ensures the accuracy, consistency, and standardization of the data, laying the foundation for subsequent fusion calculations. The specific processing steps are as follows: (1) Data cleaning: The collected raw data is preprocessed by computer software programs to remove blank values, duplicate values ​​and invalid values, and to correct data entry errors, format errors and other problems to ensure the integrity and accuracy of the data; (2) Abnormal data filtering: Based on the abnormal detection algorithm of the pet-specific AI model, the cleaned data is subjected to abnormal identification and filtering. Data that exceeds the normal range (such as abnormally high / low pet body temperature, sudden increase / decrease in activity level, etc.), abnormal fluctuation data and interference data are identified. The data is then verified in combination with pet historical data and scene characteristics. After confirming the abnormality, the data is filtered to avoid abnormal data from affecting the analysis results. (3) Core feature extraction: Through the feature extraction algorithm of the pet-specific AI model, core features related to pet care analysis are extracted from the processed data, such as the trend of body temperature change in physiological features, sleep quality in behavioral features, and compatibility-related features in product usage features. Irrelevant features are eliminated, data dimensions are reduced, and computational efficiency is improved. (4) Multi-dimensional standardized alignment and conversion: For data from different sources, of different types, and in different formats (such as structured physiological data and unstructured image data, and measurement data in different units), a standardized algorithm is used for unified conversion and alignment to convert various types of data into a unified format and measurement standard, ensuring that multi-source data can be effectively integrated and collaboratively analyzed. For unstructured image data, the device first performs face masking, foreground separation, and skeleton key point extraction, and only uploads structured features containing key point coordinates and foreground area ratio to reduce data transmission volume and protect privacy.

[0029] 3. Core Fusion Computing Based on the analysis results of a pet-specific AI model, bidirectional cross-validation, feature association constraints, and weighted fusion calculations involving long and short cycles are performed on the integrated real-time status data, product usage data, and database data. This is combined with dual-layer calibration based on both instantaneous feeding scenarios and long-term feeding cycles to correct data conflicts and analysis errors, thereby improving the accuracy and reliability of the fusion calculation results. The specific calculation process is as follows: (1) Two-way cross-validation: This involves cross-validating three types of core data, including multi-source data consistency comparison, abnormal data interception, conflict data correction, and heterogeneous data cross-validation. For example, real-time physiological data of pets is compared with historical health data to verify the rationality of real-time data; product usage feedback data is compared with real-time pet status data to verify the authenticity of product adaptation effects; when data from different sources conflict, the conflicting data is corrected by combining the AI ​​model's judgment results with scene characteristics to ensure data consistency. Through two-way cross-validation, abnormal interference data is effectively filtered out, improving data reliability and providing accurate data support for subsequent analysis. This differs from the traditional single data validation mode and further enhances data fault tolerance.

[0030] (2) Feature Association Constraints: Through a pet-specific AI model, the correlation between core features of various data is explored, and a feature association constraint mechanism is established. For example, the physiological characteristics of pets (such as body temperature and weight) are associated with product usage characteristics (such as pet food type and feeding amount) and environmental scene characteristics (such as temperature and humidity). The mutual influence between different characteristics is analyzed to ensure the synergy and rationality of each feature during the fusion calculation process. At the same time, through the constraint mechanism, irrelevant or conflicting features are avoided from interfering with the calculation results, thereby improving the efficiency and accuracy of the calculation. For pet behavioral characteristics, frequency domain analysis can be performed on the tail trajectory, and features such as ear pitch angle can be associated with physiological data to improve the comprehensiveness of feature association.

[0031] (3) Weighted Fusion Calculation Based on Long and Short Cycles: A weighted fusion algorithm based on long and short cycles is used to perform weighted fusion calculations on short-term data (such as real-time status data and recent product usage data) and long-term data (such as historical health records and long-term feeding data in the database). Among them, short-term data focuses on reflecting the current status and needs of pets and is given a higher weight; long-term data focuses on reflecting the long-term feeding trends and individual differences of pets and is given a reasonable weight. Through weighted calculation, the synergistic fusion of short-term and long-term data is achieved, which not only ensures the real-time nature of the analysis results but also takes into account the long-term nature and specificity. During the fusion calculation process, the weight allocation is dynamically adjusted by combining the autonomous learning ability of the AI ​​model to ensure the accuracy of the calculation results.

[0032] (4) Two-layer judgment calibration: A two-layer linkage judgment calibration is performed by combining instantaneous feeding scenarios and long-term feeding cycles to make secondary corrections to the fusion calculation results. Instantaneous feeding scenario judgment mainly targets the current feeding environment, the pet's real-time status, and the product usage scenario to calibrate the calculation results in real time; long-term feeding cycle judgment mainly combines the pet's long-term feeding patterns, individual differences, and historical care effects to calibrate the calculation results for long-term trends. Through two-layer judgment calibration, data conflicts and analysis errors are effectively corrected to ensure that the fusion calculation results can accurately reflect the pet's actual care needs and improve the accuracy of the analysis results.

[0033] 4. Intelligent Adaptive Output Based on the data fusion and calibration results, a hierarchical gradient output mode is adopted to output customized maintenance analysis results, meeting the personalized needs of different scenarios and users. The analysis results include any one or more of the following: intelligent adaptation results for single products, scenario-based maintenance intervention analysis results, and combined analysis results of products and maintenance interventions, as detailed below: (1) Intelligent Adaptation Results for Single Products: For various pet care products, based on the pet's real-time status, historical data, product usage feedback, and fusion calculation results, product adaptation suggestions are output, including whether the product is suitable, the degree of suitability, usage suggestions (such as dosage, frequency of use, and usage scenarios), and replacement suggestions. For example, based on the pet's weight, age, physiological state, and historical feeding data, suitable pet food types and feeding amounts are recommended for the pet; based on the pet's coat condition and skin condition, suitable grooming products and usage frequencies are recommended.

[0034] (2) Scenario-based care intervention analysis results: Combining the pet's real-time feeding scenario (such as environmental temperature and humidity, activity scenario), real-time status and long-term feeding data, scenario-based care intervention suggestions are output, including health monitoring, environmental adjustment, behavior guidance, and disease prevention. For example, when an abnormal body temperature is detected in the pet and the environmental temperature and humidity are unsuitable, environmental adjustment suggestions and health monitoring reminders are output; when insufficient pet activity is detected, behavior guidance suggestions and interaction plans are output.

[0035] (3) Analysis results of combining supplies and maintenance interventions: Combine supply matching with maintenance interventions to output comprehensive maintenance analysis results. For example, based on the pet's health status and product matching, recommend suitable health products and corresponding administration plans, and provide supporting maintenance intervention suggestions such as diet and exercise, so as to achieve the coordinated promotion of supply matching and maintenance interventions and improve the pet maintenance effect.

[0036] The output method can be customized according to user needs, and can be delivered in various forms such as text, charts, and voice through computer software systems, mobile terminal apps, and smart device terminals, ensuring that users can obtain maintenance analysis results clearly and conveniently.

[0037] 5. Closed-loop iterative optimization The system collects feedback data and actual application effect data after the method is implemented, and feeds this data back into a self-built private database covering the entire pet domain, forming a data repository. Simultaneously, it iteratively optimizes the parameters of the pet-specific AI model and the data matching rules, creating a continuously updated, fully closed-loop software operation mechanism to achieve continuous upgrading and improvement of the technical solution. The specific iteration process is as follows: (1) Feedback and Effect Data Collection: Feedback from users on maintenance analysis results, product matching suggestions, and maintenance intervention suggestions is collected through user interaction terminals, smart device terminals, etc., including satisfaction and modification suggestions; at the same time, actual application effect data after the method is implemented is collected, including data on changes in pet physiological state, behavior, and health status, and continuous data on product use effects, to ensure the comprehensiveness and authenticity of the data. For unknown data that the AI ​​model cannot accurately identify, data error correction is performed and the data is transformed into effective data and incorporated into the iterative data system. Data expansion can be achieved without manual intervention, thereby increasing the amount and diversity of data for model training.

[0038] (2) Data Feedback and Accumulation: The collected feedback data and actual application effect data are cleaned, filtered, and standardized before being fed back into a self-built private pet database to enrich the database's existing data, enabling long-term data accumulation and reuse, and providing richer data support for subsequent fusion operations and model optimization. At the same time, the labeled valid data is reviewed and updated to ensure data quality before being uploaded to the private database for storage.

[0039] (3) Iterative optimization of models and rules: Relying on the computing power of the computer software system and combining the returned data, the parameters of the pet-specific AI model are dynamically adjusted and optimized to improve the model's judgment accuracy, feature extraction capability, and fusion computing efficiency. At the same time, the data matching rules are optimized, and the data verification standards, weight allocation ratios, and adaptation judgment standards are adjusted based on actual application effects and user feedback to ensure that the technical solution can continuously adapt to changes in pet care needs and improve the system's adaptability and overall technological progress. During the iteration process, an incremental training mode is adopted, which is based on the existing model and new data for training to improve iteration efficiency and reduce resource consumption.

[0040] The data processing and computation steps in this solution have no fixed execution order and support asynchronous calls, parallel operation, and step splitting or merging. Any adjustment, splitting or merging of the step order, as long as it does not deviate from the core technical architecture of this invention, does not deviate from the protection scope of this invention. Based on this solution, adding conventional auxiliary technical steps such as data encryption, operation log recording, data caching and scheduling, and data format conversion, replacing different AI implementation forms (such as replacing the specific algorithm or model architecture of the AI ​​model), or changing the computer running platform (such as changing from a cloud server to a local server or mobile terminal) will not depart from the protection scope of this invention. Any computer technology solution that achieves intelligent pet care and analysis based on the unique multi-source data cross-validation, long and short cycle fusion, and closed-loop iteration core technology architecture of this invention, regardless of how its specific implementation method or technical details are adjusted, falls within the protection scope of this invention. Example

[0041] This embodiment provides a multi-source data closed-loop fusion intelligent pet care analysis method based on pet AI, which is applied to a pet intelligent care APP. Relying on computer software system, pet smart collar, environmental monitoring sensor, user interaction terminal and self-built pet full-domain private database, it realizes customized product adaptation and scenario-based care intervention analysis for pets (taking pet dogs as an example in this embodiment). The specific execution process is as follows.

[0042] 1. Data Collection 1.1. Real-time Pet Status Data Collection: Physiological data (body temperature 38.5℃, heart rate 120 bpm, respiratory rate 25 breaths / min, weight 10kg) and behavioral activity data (daily activity 8000 steps, sleep duration 10 hours, food intake 500g, water intake 6 times) are collected via a smart pet collar. Environmental monitoring sensors collect data on the temperature and humidity of the dog's living environment (temperature 25℃, humidity 50%) and air quality (PM2.5 concentration 30μg / m³). Users input the dog's instantaneous health status data (good mental state, no vomiting, diarrhea, or other abnormalities) via a user interaction terminal (care APP). During data collection, short-range wireless timestamp broadcasting, combined with a local time service, ensures the time synchronization of all data types.

[0043] 1.2. Pet Care Product Usage Data Collection: Data on the pet dog's current pet food (Brand A, adult dog food) and supplements (calcium tablets) were collected via a pet care app. This included usage behavior data (pet food fed twice daily, 250g each time; calcium tablets given once daily before bedtime), usage habit data (users fed the dog regularly in the morning and evening, changing the pet food flavor weekly), compatibility feedback data (the dog showed high acceptance of the pet food, ate actively, and had no allergic reactions; after taking calcium tablets, its bone condition was good), and long-term application effect data (after 3 months of continuous use of the pet food, the dog's weight stabilized at around 10kg, and its coat was smooth; after 2 months of calcium tablet use, joint mobility improved). All collected data was uploaded to the computer software system using AES encryption.

[0044] 2. Data integration and processing 2.1. Access the self-built private pet database to obtain the pet dog's full-cycle feeding data, including basic profile (breed: Labrador, age: 2 years, sex: male, weight history: stable at 9.5-10.5kg in the past 6 months, past medical history: none), historical health records (normal physical examination reports in the past 6 months, no disease records), long-term feeding behavior data (average daily activity of 7500-8500 steps in the past 6 months, average sleep duration of 9.5-10.5 hours, average food intake of 480-520g), and unstructured data (daily photos of the pet dog and feeding notes uploaded by the user).

[0045] 2.2. A pet-specific AI model (built based on deep learning algorithms and specifically trained for the physiological and behavioral characteristics of dogs) is used in conjunction with computer computing power to uniformly process the collected real-time status data, product usage data, and database inventory data. (1) Data cleaning: Remove blank and duplicate values ​​from the data, and correct a feeding amount input error by the user (originally entered as 50g, corrected to 500g). (2) Abnormal data filtering: Through the AI ​​model anomaly detection algorithm, abnormal data is identified and all collected data are within the normal range; (3) Core feature extraction: Extract core features, including body temperature change trend (stable at 38.3-38.6℃ in the past week), activity trend (stable at 7800-8200 steps in the past week), pet food suitability features (stable food intake, no allergic reaction), and environmental suitability features (suitable temperature and humidity, good air quality); For pet dog photos uploaded by users, the front end first performs foreground separation and skeleton key point extraction, and only uploads structured features such as key point coordinates to protect pet privacy; (4) Multi-dimensional standardization and alignment conversion: Convert data of different units (such as weight kg, body temperature ℃, activity steps) into a unified standardized format, and convert unstructured feeding notes into structured text features to ensure that multi-source data can be effectively integrated.

[0046] 3 Core Fusion Computing 3.1. Two-way cross-validation: The real-time physiological data of the pet dog (body temperature 38.5℃) is compared with the historical health data (average body temperature of 38.4℃ in the past week) to verify the rationality of the real-time data; the pet food usage feedback data (positive eating, no allergies) is cross-validated with the real-time physiological data (stable weight, smooth coat) to confirm the authenticity of the pet food's compatibility; there are no data conflicts and no correction is required.

[0047] 3.2. Feature Association Constraints: The AI ​​model is used to mine the association between core features. For example, the environmental temperature and humidity (25℃, 50%) are associated with the pet's activity level (8000 steps), and the analysis shows that the current environment is suitable and the pet's activity level is normal. The pet food feeding amount is associated with the weight, and the analysis shows that the feeding amount is reasonable and the weight is stable.

[0048] 3.3. Weighted Fusion Calculation Based on Long and Short Cycles: The weight of short-term data (real-time status data and product usage data for the day) is set to 0.6, and the weight of long-term data (historical data for the past 6 months and long-term product usage effect data) is set to 0.4. Through weighted fusion calculation, it is concluded that the pet dog is currently in good condition and the pet food and health products are highly compatible.

[0049] 3.4. Dual-layer judgment calibration: Instantaneous scene judgment (current environmental temperature and humidity are suitable, and the pet's real-time condition is good) to perform real-time calibration of the calculation results; Long-term feeding cycle judgment (the pet's condition has been stable and feeding is regular over the past 6 months) to perform long-term trend calibration of the calculation results. The final corrected fusion calculation result is: the pet dog is currently in good health, the pet food and health products are highly compatible, and the feeding environment is suitable.

[0050] 4. Intelligent Adaptive Output Based on the calibration results of the fusion operation, customized maintenance analysis results are output through the hierarchical gradient output of the maintenance APP: 4.1. Results of intelligent matching of single products: Pet food (Brand A adult dog food) has a compatibility rate of 95%, and it is recommended to continue using it, maintaining a feeding amount of 250g twice a day; health supplement (calcium tablets) has a compatibility rate of 90%, and it is recommended to continue taking it. The frequency of administration can be adjusted appropriately according to the dog's activity level (e.g., if the activity level increases, it can be increased to 2 tablets per day). 4.2. Scenario-based maintenance intervention analysis results: The current ambient temperature and humidity are suitable, and the air quality is good, so no adjustment is needed; the pet dog's daily activity level and sleep duration are normal, and it is recommended to continue the routine of walking the dog twice a day for 30 minutes each time; 4.3. Analysis of the combination of supplies and care interventions: Based on the current condition of the pet dog, it is recommended to continue using the existing pet food and calcium supplements, while providing weekly hair care, feeding appropriate amounts of vegetables to supplement vitamins, and improving the quality of the pet dog's coat and immunity.

[0051] 5. Closed-loop iterative optimization 5.1. Feedback and Effect Data Collection: User feedback was collected through the pet care app (satisfaction with the pet care analysis results was 95%, with suggestions to add specific methods for pet hair care); actual application effect data was collected after one month (the pet dog's weight stabilized at 10kg, the coat was smoother, the joints were more flexible, and there were no health abnormalities); at the same time, a small amount of behavioral data that the AI ​​model could not accurately identify was collected, and after error correction, it was converted into valid data.

[0052] 5.2. Data Feedback and Consolidation: After cleaning and standardizing user feedback and actual application effect data, the data is fed back into the self-built private database covering the entire pet domain to enrich the historical data of the pet dog and update the product adaptation effect data in the database. After reviewing and updating the labeled valid data, it is also uploaded to the private database.

[0053] 5.3. Model and Rule Iteration Optimization: Based on the returned data, the parameters of the pet-specific AI model were adjusted to improve the model's analysis accuracy of pet fur condition; data matching rules were optimized, and adaptation criteria and suggested templates related to fur care were added; at the same time, an incremental training mode was adopted to train the model based on the existing model and newly added effective data to improve model performance and achieve iterative optimization of the technical solution. The subsequent output of maintenance analysis results will include specific methods for fur care, further enhancing the user experience.

[0054] This embodiment, through the above steps, achieves intelligent care analysis of pet dogs through multi-source data fusion, effectively solving the problems of existing solutions such as single data dimensions, low analysis accuracy, and poor adaptability. After one month of practical application, the pet dogs' health status remained stable, and users were satisfied with the care analysis results and product matching suggestions. The workflow design of this embodiment is flexible and can be adjusted according to the different pets' conditions, exhibiting strong adaptability and wide applicability. The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A multi-source data closed-loop fusion intelligent pet care analysis method based on pet AI, characterized in that, Includes the following steps: S1. Collect real-time pet status data, and simultaneously collect actual usage data of pet care products; The real-time pet status data includes pet physiological characteristics data, behavioral activity data, environmental scene data, and instantaneous health performance data; the actual usage data of the pet care products includes product usage behavior data, usage habit data, adaptation feedback data, and long-term application effect data; during the collection process, short-range wireless timestamp broadcasting is used in conjunction with local time service to unify the time base and ensure the time synchronization of various real-time data. S2. Retrieve the self-built private pet database to obtain pet lifecycle feeding data; the self-built private pet database stores pet basic profiles, historical health records, long-term feeding behavior data, and structured and unstructured data accumulated throughout the feeding cycle, and uses encrypted storage with strict access permissions. S3. Using a pet-specific AI model and computer computing power scheduling, the system performs unified cleaning, abnormal data filtering, core feature extraction, and data dimension standardization on multi-source collected data and database data. For unstructured image data, the device first performs face masking, foreground separation, and skeleton key point extraction, and only uploads structured features containing key point coordinates and foreground area ratio. S4. Based on the judgment results of the pet-specific AI model, perform bidirectional cross-validation, feature association constraints, and long- and short-cycle linkage weighted fusion calculations on multiple types of data, and combine instantaneous scenarios and long-term feeding cycles to perform dual-layer linkage judgment calibration; the bidirectional cross-validation includes multi-source data consistency comparison, abnormal data interception, conflict data correction, and heterogeneous data mutual verification. S5. Based on the fusion operation calibration results, the customized maintenance analysis results are output in a hierarchical gradient. The analysis results include any one or more of the following: intelligent adaptation results of a single product, scenario-based maintenance intervention analysis results, and analysis results combining products and maintenance interventions. The output can be provided through various channels such as computer software systems, mobile terminal APPs, and smart device terminals, in the form of text, charts, voice, etc. S6. Collect feedback data and actual operation data after the implementation of the scheme and feed them back to the self-built private database. Iterate and optimize the AI ​​model parameters and data matching rules in reverse to form a continuously self-updating closed-loop intelligent computing mechanism. For unknown data that the AI ​​model cannot accurately identify, perform data error correction and transform it into effective data, incorporate it into the iterative data system, and use incremental training mode to train the AI ​​model.

2. The method according to claim 1, characterized in that: The pet's physiological characteristics data include one or more of the following: body temperature, heart rate, respiratory rate, weight, blood pressure, blood sugar, and coat condition; the behavioral activity data includes one or more of the following: pet's activity level, sleep duration, food intake, water intake frequency, excretion, and interactive behavior; the environmental scene data includes one or more of the following: temperature and humidity of the pet's living environment, air quality, light intensity, and noise level; and the instantaneous health performance data includes one or more of the following: pet's mental state, vomiting, diarrhea, coughing, and skin abnormalities.

3. The method according to claim 1, characterized in that: The self-built private pet database stores basic pet profiles including one or more of the following: pet breed, age, gender, weight, pedigree, and medical history. The long-term unstructured data accumulated from long-term pet care includes one or more of the following: user pet care notes, pet photos, and pet videos. The database supports real-time updates, fast retrieval, and encrypted storage of data to ensure data security, privacy, and reusability.

4. The method according to claim 1, characterized in that: In the weighted fusion calculation of short and long cycles, short-term data includes real-time status data and recent product usage data, with a weight of 0.5-0.7; long-term data includes historical health records in the database, long-term feeding data, and long-term product usage effect data, with a weight of 0.3-0.5; the weights can be dynamically adjusted by the AI ​​model according to individual pet differences and actual care scenarios.

5. The method according to claim 1, characterized in that: When performing scenario-based maintenance intervention analysis, corresponding scenario execution data and effect feedback data are collected simultaneously and incorporated into the multi-source data fusion operation and AI model iteration optimization process. The scenario execution data includes one or more of environmental adjustment data, behavior guidance data, and disease prevention data. The effect feedback data includes one or more of pet status change data and user evaluation data.

6. The method according to claim 1, characterized in that: The pet-specific AI model is built on a deep learning algorithm and is specifically trained to meet the physiological characteristics, behavioral characteristics, and care needs of pets. It can achieve functions such as abnormal data detection, core feature extraction, multi-source data fusion and analysis, and parameter autonomous iterative optimization. The specific algorithm and model architecture of the AI ​​model can be replaced, but the replacement will not deviate from the protection scope of this invention.

7. The method according to claim 1, characterized in that: The data cleaning includes removing blank values, duplicate values, and invalid values, and correcting data entry errors and format errors. The abnormal data filtering uses an AI model anomaly detection algorithm to identify data that exceeds the normal range, abnormal fluctuation data, and interference data. After secondary verification by combining pet historical data and scene characteristics, the abnormal data is filtered.

8. The method according to claim 1, characterized in that: The multi-dimensional standardized alignment conversion transforms multi-source data of different origins, types, formats, and units into a unified format and measurement standard, ensuring that multi-source data can be effectively integrated and collaboratively analyzed; the data format includes structured data format and structured format after unstructured data conversion.

Images