An artificial intelligence-based aquatic pet cloud primary screening and on-site treatment full-cycle management system and method

Abstract

The application discloses an aquatic pet cloud primary screening and on-site treatment whole-cycle management system and method based on artificial intelligence, which comprises an offline service terminal used for collecting multi-modal data of aquatic pets and performing on-site service operations; a cloud hub system is in communication connection with the offline service terminal through a mobile communication network, is used for receiving offline service terminal data and performing data processing, decision-making and resource scheduling processing, and comprises an AI computing cluster, which is used for analyzing the multi-modal data to generate diagnosis suggestion information; an intelligent service scheduling module is used for acquiring the diagnosis suggestion information and on-site resource states fed back by the offline service terminal, generating and issuing a service execution scheme to the offline service terminal; and the offline service terminal receives the service execution scheme and performs corresponding on-site service operations. The application constructs an aquatic pet multi-modal AI primary screening diagnosis module based on data resources accumulated by a national aquatic animal pathogen library, and significantly improves disease identification accuracy and diagnosis efficiency.

Description

Technical Field

[0001] This invention relates to the field of smart healthcare technology, and in particular to a cloud-based initial screening and home care management system and method for aquatic pets based on artificial intelligence. Background Technology

[0002] With the explosive growth of China's pet economy, aquatic pets have become the fastest-growing segment. By the end of 2025, the total number of aquatic pets in China exceeded 120 million, with a compound annual growth rate of 35%, and the total industry output value exceeded 80 billion yuan. Shanghai, as the city with the highest aquatic pet consumption in China, has a total number of over 8 million aquatic pets, and its high-end koi market accounts for 22% of the national market, with some collectible koi reaching a value of several million yuan per animal. However, in stark contrast to the booming market, the domestic aquatic pet veterinary industry is almost non-existent. Aquatic pets are extremely sensitive to environmental changes; even slight fluctuations in temperature, water quality, and dissolved oxygen levels can trigger severe stress reactions. Data shows that the mortality rate of aquatic pets during transportation for veterinary care is as high as 30%, with many pets dying not from the disease itself, but from the stress during transport. For high-end koi worth hundreds of thousands of yuan, the risks of transporting them for veterinary care are extremely high.

[0003] Currently, pet medical technology solutions on the market mainly fall into two categories: First, purely online AI consultation platforms, all trained on cat and dog data, lacking AI models specifically for aquatic pets. Their disease matching accuracy is less than 55%, and they can only provide text-based consultations, failing to address transportation stress issues. Second, offline pet hospital management systems, which only serve sick pets arriving at the hospital, unable to provide home visits. Their data is isolated, and they lack technological empowerment and training systems, failing to alleviate the fundamental contradiction of a shortage of medical personnel. It is evident that existing solutions all face a common bottleneck: the lack of systematic and authoritative aquatic pathogen data. Aquatic pet diagnosis heavily relies on pathogen data, but no commercial company currently possesses a systematic aquatic pathogen database and clinical case database.

[0004] In summary, existing technologies cannot simultaneously address the core issues of scarce aquatic pet medical resources, high transportation mortality rates, lack of industry standards, and limited service coverage. Furthermore, there is no dedicated cloud-based integrated management system for initial screening and home-based treatment of aquatic pets. Therefore, developing an AI-based cloud-based, full-cycle management system for initial screening and home-based treatment of aquatic pets is one of the most pressing issues that needs to be addressed. Summary of the Invention

[0005] In view of the above-mentioned problems in the existing technology, the purpose of this invention is to provide a cloud-based initial screening and home treatment full-cycle management system for aquatic pets based on artificial intelligence, so as to solve the fundamental problems of the lack of a dedicated diagnostic technology system for aquatic pets and the inability of the diagnostic accuracy to meet clinical needs in the existing technology.

[0006] Another objective of this invention is to provide a cloud-based initial screening and home-based treatment management method for aquatic pets based on artificial intelligence.

[0007] To address the above issues, this application proposes an AI-based cloud-based initial screening and home care management system for aquatic pets. The system includes at least one offline service terminal deployed at on-site service nodes to collect multimodal data of aquatic pets and perform on-site service operations.

[0008] This includes a cloud-based central system that communicates with the offline service terminals via a mobile communication network. This system receives data from the offline service terminals and performs data processing, decision-making, and resource scheduling.

[0009] An AI computing cluster is configured to receive multimodal data of aquatic pets from the offline service terminal, and to analyze the multimodal data to generate diagnostic suggestion information.

[0010] The intelligent service scheduling module is connected to the AI ​​computing cluster and is configured to obtain the diagnostic suggestion information and the on-site resource status fed back by the offline service terminal, generate and send the service execution plan to the offline service terminal;

[0011] The offline service terminal receives the service execution plan and performs the corresponding on-site service operations.

[0012] Furthermore, the cloud-based central system also includes a full-cycle health management module, which is communicatively connected to the AI ​​computing cluster and the intelligent service scheduling module. This module is used to establish and store electronic health records for each aquatic pet, including the pet's historical medical data and diagnostic recommendations.

[0013] Furthermore, the AI ​​computing cluster includes a multimodal AI initial screening and diagnosis module, whose input end is connected to the offline service terminal application, and whose output end is connected to the intelligent service scheduling module and the full-cycle health management module respectively.

[0014] The multimodal AI initial screening and diagnosis module integrates a large vertical model of aquatic pet diseases and a disease image recognition unit, supporting text symptom description, image upload, and video consultation and diagnosis modes. It identifies diseases based on the large vertical model of aquatic pet diseases and outputs diagnostic results, disease classification, and standardized treatment plans. Specifically, the multimodal AI initial screening and diagnosis module receives image and video data collected by the terminal through a data preprocessing unit and performs noise reduction, enhancement, cropping, and standardization processing.

[0015] The disease image recognition unit extracts and recognizes features from preprocessed images using a fusion model built on a target detection network and an image classification network. It also identifies visually distinctive diseases by combining a large vertical model of aquatic pet diseases and generates image recognition results.

[0016] The multimodal AI initial screening and diagnosis module also includes a natural language understanding unit, which is configured to perform semantic parsing on the natural language symptom descriptions collected by the terminal based on a pre-trained language model, extract symptom information, and output structured symptom labels and symptom severity scores.

[0017] Furthermore, the input end of the intelligent service scheduling module is communicatively connected to the multimodal AI initial screening and diagnosis module, and the output end is communicatively connected to the veterinary mobile work terminal. The intelligent service scheduling module adopts a multi-objective optimization scheduling algorithm based on genetic algorithm, which integrates spatial factors, clinical factors, and quality factors to prioritize users and certified veterinarians.

[0018] Furthermore, the system also includes a standardized training and certification module, which specifically involves constructing a three-in-one training system that combines online AI simulation training with offline hands-on instruction and unified assessment and certification. This system is used to provide standardized training and assessment for certified veterinarians in their diagnostic and treatment capabilities.

[0019] Furthermore, the offline service terminal includes at least one of a veterinary mobile terminal, a user terminal, and a standardized treatment package; the input end of the full-cycle health management module is communicatively connected to the multimodal AI initial screening and diagnosis module and the veterinary mobile work terminal, respectively, and the output end is communicatively connected to the offline service terminal application.

[0020] Furthermore, the full-cycle health management module establishes a unique digital health record for each aquatic pet, recording its basic information, growth and development, medical history, medication records, vaccination status, and water quality testing data. Based on the record information, it provides personalized prevention suggestions, rehabilitation tracking, and follow-up visit reminders.

[0021] Furthermore, when a user uses the system for the first time, they add information about their aquatic pet in the mobile application. The system then generates a unique health record number based on this information. After each treatment, the veterinarian fills in the treatment record and medication details on the mobile terminal and synchronizes this information to the corresponding health record. Based on the pet's breed, age, health status, and historical treatment records, the system regularly pushes personalized prevention suggestions to the user.

[0022] Furthermore, the cloud-based central system also includes an intelligent optimization module, which is connected to the AI ​​computing cluster and the intelligent scheduling module. The intelligent optimization module adopts federated learning technology to collect and integrate clinical diagnosis and treatment data from across the country while protecting user privacy and institutional data security. It regularly iterates and optimizes the AI ​​computing cluster and updates the corresponding training course content and diagnosis and treatment guidelines.

[0023] This invention also provides a cloud-based initial screening and home-based treatment management method for aquatic pets based on artificial intelligence. The method is applied to the system and includes the following steps:

[0024] Step 1: Multidimensional data collection and uploading. The IoT node, acting as the on-site execution node, is responsible for collecting pet and environmental data. Users collect and upload pet health information through their user terminal devices. The IoT node collects water quality data through water quality environmental sensors and collects pet swimming posture and body surface image data through visual monitoring devices. The above data is then uploaded to the cloud central system via mobile network.

[0025] Step 2: Cloud data processing and multimodal AI preliminary screening. The platform layer server receives data uploaded by user terminal devices and IoT nodes. The multimodal AI preliminary screening and diagnosis module in the platform layer server analyzes the data and combines it with environmental data and pet health data to conduct a preliminary health status assessment and anomaly identification, forming the AI ​​preliminary screening result.

[0026] Step 3: Service Triage and Dispatch. Based on the AI ​​preliminary screening results from the multimodal AI preliminary screening and diagnosis module, the system triggers the corresponding service process, specifically:

[0027] Triggering online medical consultations:

[0028] In the consultation distribution process, if the multimodal AI preliminary screening and diagnosis module determines that expert intervention is required, the platform server will distribute the data and consultation request described in step one to online experts and diagnosis and treatment terminals.

[0029] Expert diagnosis: Experts receive and distribute data and consultation requests through online expert and diagnosis terminals, conduct remote video communication with user terminal devices through expert video consultation workstations, and diagnose aquatic pets by combining the data from the AI ​​preliminary screening results.

[0030] For follow-up services, experts can issue prescriptions online through online consultation terminals, and users can purchase related supplies through the online pharmacy / shopping mall module;

[0031] Triggering offline services:

[0032] Order generation: If the multimodal AI initial screening and diagnosis module determines that the situation is urgent or requires on-site handling, the platform layer server will generate an on-site sampling / treatment order request from the data described in step one and distribute it to the offline service scheduling gateway of the intelligent service scheduling module.

[0033] Service scheduling: The offline service scheduling gateway's built-in home medical dispatch system receives the dispatch request and dispatches experts according to priority based on geographical location and veterinary expertise.

[0034] For door-to-door service, the offline expert's veterinary mobile terminal receives the dispatch request and goes to the user's location to provide door-to-door service;

[0035] Step 4: The online diagnosis and treatment results or offline service data are recorded and synchronized to the platform server, and users can view the diagnosis report, treatment suggestions and follow-up care guidelines through their user terminals.

[0036] Furthermore, step two, the multimodal AI initial screening, includes the following steps:

[0037] S21. Multi-source heterogeneous data input: The multi-modal AI initial screening and diagnosis module receives pet health information collected and uploaded by the user terminal; water quality data uploaded by the water quality environment sensor in the IoT node; and pet swimming posture and body surface image data collected and uploaded by the visual monitoring equipment.

[0038] S22. Data preprocessing: The data preprocessing module performs noise reduction, enhancement, cropping and standardization on the data described in step S21, and performs feature extraction and recognition on the preprocessed data.

[0039] S23. Dual-path parallel analysis: Performing visual and linguistic analysis on the preprocessed data, including:

[0040] The visual analysis module analyzes images and time-series data; the body surface lesion identification module identifies physical lesions on the body surface of aquatic pets and extracts visual lesion features; the behavior trajectory analysis module analyzes the swimming posture of aquatic pets and extracts abnormal behavior features, generating visual analysis results.

[0041] The language and data understanding module receives and reads aquatic pet health information and water quality data, and uses pathogen characteristics and pharmacokinetic parameters from the aquatic pet pathology knowledge base for auxiliary analysis. The RAG retrieval enhancement module constructs a vector database, i.e., the aquatic pet pathology knowledge base, through document segmentation and vectorization. It then rewrites and expands user queries, and finally recalls fragments through a hybrid retrieval of vector retrieval and keywords, and performs rearrangement and context compression based on the number of recalled fragments. The symptom semantic analysis module analyzes the textual descriptions of pet health provided by users, summarizes semantic symptoms, and generates language analysis results.

[0042] S24. Multimodal fusion decision-making: The multimodal fusion decision-making unit integrates visual lesion features, behavioral abnormality features, reference treatment guidelines, and semantic symptom summaries, performs multidimensional feature cross-validation, and calculates the probability and confidence of possible diseases.

[0043] S25. Output diagnostic report: After fusion decision-making, the system generates and outputs an AI preliminary diagnosis and intervention report to provide a scientific basis for subsequent expert diagnosis or treatment.

[0044] Compared with the prior art, the beneficial technical effects of the present invention are as follows:

[0045] 1. Existing general-purpose pet AI models are all trained on cat and dog data, lacking support from aquatic pathogen data. This results in a very high misdiagnosis rate when faced with complex symptoms in aquatic pets, involving surface lesions, swimming behavior, and water quality. This invention leverages decades of data resources accumulated by the National Aquatic Animal Pathogen Bank of Shanghai Ocean University to construct a multimodal AI initial screening and diagnostic module specifically for aquatic pets. Through the collaborative work of a data preprocessing unit, a disease image recognition unit, a natural language understanding unit, and a multimodal fusion decision unit, it achieves the fusion analysis of multi-source heterogeneous data, including textual symptoms, surface images, behavioral videos, and water quality parameters. By fusing a target detection network for precise lesion localization, an image classification network for probabilistic disease classification, and a cross-modal attention mechanism for weighted fusion of visual features and semantic symptoms, the system can capture early, subtle symptoms that are difficult to identify using a single modality. Test results show that, under the same test set conditions, the diagnostic accuracy of the system in this application is significantly improved compared with the existing model trained on a general dataset. The response time for a single initial screening is reduced from tens of minutes in traditional manual consultation to seconds, providing efficient scientific pre-screening for subsequent expert diagnosis.

[0046] 2. Aquatic pets are extremely sensitive to environmental changes. Existing technologies require pet owners to transport their pets to offline hospitals. Fluctuations in temperature, water quality, and dissolved oxygen levels during transportation can easily trigger severe stress reactions, leading to rapid deterioration of the condition or even death. This invention provides an intelligent service scheduling module that integrates cloud-based AI initial screening with certified veterinarian home visits into a unified process design. When the multimodal AI initial screening diagnosis module determines that the condition requires on-site treatment, the system automatically generates and issues a home visit dispatch plan based on a multi-objective optimization scheduling algorithm, taking into account spatial distance, veterinarian expertise matching degree, real-time workload, and urgency of the condition. The method described in this application allows aquatic pets to receive professional treatment without leaving the water during transportation, effectively avoiding stress risks and significantly reducing mortality during the medical process.

[0047] 3. The standardized training and certification module provided by this invention constructs a standardized certification system through online AI simulation training, offline hands-on teaching, and nationwide unified assessment and certification, which effectively reduces the fluctuation of diagnosis and treatment quality caused by regional differences. The online simulation training system builds a typical case database based on the multimodal AI initial screening and diagnosis module of this application, allowing trainees to conduct diagnostic decision training in a virtual environment and obtain real-time error correction feedback; offline hands-on teaching is conducted by qualified senior veterinarians who provide standardized operation demonstrations.

[0048] The system described in this application enables primary healthcare personnel to quickly master standardized diagnosis and treatment procedures for common diseases in aquatic pets, significantly shortening the talent training cycle and reducing fluctuations in diagnosis and treatment quality caused by regional differences. At the same time, the training content is updated synchronously with the AI ​​model iteration to ensure that the output standards are consistent with the latest clinical understanding, effectively alleviating the contradiction of the shortage of medical personnel in the industry.

[0049] 4. The full-cycle health management module provided by this invention realizes the full life cycle management of aquatic pets. By establishing a unique digital health file for each aquatic pet, integrating data such as basic information of aquatic pets, medical records, and water quality records, it generates prevention or rehabilitation plans, effectively managing aquatic pets throughout their entire life cycle.

[0050] 5. This application brings the previously scattered and disordered aquatic pet medical services into a standardized and digital management system by unifying cloud-based initial screening standards, on-site medical service specifications, and veterinary certification systems. Through cloud technology output, high-quality medical resources can overcome geographical limitations and cover second-tier and lower-tier cities and remote areas, alleviating the problem of uneven geographical distribution of medical resources. At the same time, this application promotes the transformation and application of scientific research resources such as the National Aquatic Animal Pathogen Bank from the traditional field of edible aquatic products to the field of pet medical care, providing a new implementation path for the industrialization of university scientific research results. Attached Figure Description

[0051] Figure 1 This is a system architecture diagram as described in the embodiments of this application;

[0052] Figure 2 This is a flowchart illustrating the workflow of the multimodal AI initial screening and diagnosis module described in the embodiments of this application.

[0053] Figure 3 This is a flowchart of the diagnostic and treatment events described in the embodiments of this application;

[0054] The system comprises the following components: 100. User terminal equipment; 200. Platform layer server; 300. IoT node; 400. Online expert and diagnosis terminal; 500. Offline service scheduling gateway; 600. Multimodal AI initial screening and diagnosis module; 301. Water quality environment sensor; 302. Visual monitoring equipment; 401. Expert video consultation workstation; 402. Online pharmacy / e-commerce module; 403. Pet owner communication community module; 501. Home visit dispatch system; 610. Aquatic pet pathology knowledge base; 620. Data preprocessing module; 630. Visual large model module; 631. Body surface lesion identification submodule; 632. Behavioral trajectory analysis submodule; 640. Language large model module; 641. RAG retrieval enhancement module; 642. Symptom semantic analysis submodule; and 650. Multimodal fusion decision unit. Detailed Implementation

[0055] 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. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0056] Example 1

[0057] like Figure 1 As shown in the embodiment of this application, a cloud-based initial screening and home-based treatment full-cycle management system for aquatic pets based on artificial intelligence is provided. It includes at least one offline service terminal deployed at the field service node for collecting multimodal data of aquatic pets and performing on-site service operations. The multimodal data includes at least two of the following: image data, video data, water quality parameter datasets, and behavioral data of aquatic pets.

[0058] This includes a cloud-based central system that communicates with the offline service terminals via a mobile communication network. This system receives data from the offline service terminals and performs data processing, decision-making, and resource scheduling.

[0059] An AI computing cluster is configured to receive multimodal data of aquatic pets from the offline service terminal, and to analyze the multimodal data to generate diagnostic suggestion information.

[0060] The intelligent service scheduling module is connected to the AI ​​computing cluster and is configured to obtain the diagnostic suggestion information and the on-site resource status fed back by the offline service terminal, generate and send the service execution plan to the offline service terminal;

[0061] The service terminal now receives the service execution plan and executes the corresponding on-site service operation.

[0062] like Figure 1 As shown in the embodiment of this application, the system adopts a distributed architecture, consisting of a terminal layer, a network layer, a platform layer, and an application layer from bottom to top. The terminal layer includes offline service terminals, such as veterinary mobile terminals, IoT nodes, standardized diagnostic kits, and regional training base terminals, which are used to collect on-site data and execute commands issued by the application layer. The network layer includes the mobile communication network, such as a 5G / 4G mobile communication network, to realize data transmission between the terminal layer, the platform layer, and the application layer.

[0063] The platform layer includes cloud server clusters, database clusters, and AI computing clusters, which provide computing, storage, and AI inference capabilities.

[0064] The application layer includes the multimodal AI initial screening and diagnosis module and the intelligent service scheduling module. The multimodal AI initial screening and diagnosis module is used to receive and analyze multi-source data collected by the terminal layer and output diagnostic suggestions. The intelligent service scheduling module is used to generate service execution plans based on the diagnostic suggestions and actual on-site resources. In some embodiments, the application layer also includes a standardized training and certification module and a full-cycle health management module. The standardized training and certification module includes a standardized treatment plan library, which is used to build a unified standard for the service process of each execution node. The full-cycle health management module is used to establish a health record for each aquatic pet.

[0065] In some preferred embodiments, the application layer also includes a cooperative structure management module and a data intelligence optimization module. The cooperative institution management module is used to manage all cooperative medical institutions throughout their entire lifecycle. The cooperative institution management module is communicatively connected to the standardized training and certification module. After veterinary personnel receive the corresponding training and assessment certification, they obtain system certification and authorization to carry out door-to-door medical services.

[0066] The data intelligence optimization module is connected to the multimodal AI initial screening and diagnosis module, the standardized training and certification module, and the full-cycle health management module. The data intelligence optimization module adopts federated learning technology to collect and integrate clinical diagnosis and treatment data from across the country while protecting user privacy and institutional data security. It regularly iterates and optimizes the AI ​​model and updates training course content and diagnosis and treatment guidelines.

[0067] The multimodal AI initial screening and diagnosis module has its input end connected to the offline service terminal application, and its output end connected to the intelligent service scheduling module and the full-cycle health management module, respectively. This module integrates a large-scale vertical model of aquatic pet diseases and a CV image recognition system, supporting text symptom descriptions, image uploads, and video consultation diagnostic modes. It identifies diseases and outputs diagnostic results, disease grading, and standardized treatment plans. In this embodiment, the large-scale vertical model of aquatic pet diseases is constructed based on pathogen data accumulated in the National Aquatic Animal Pathogen Bank of Shanghai Ocean University. This data includes 12,000 samples of common turtle diseases and 15,000 samples of common koi diseases, covering surface images, pathological sections, pathogen microscopic images, and corresponding environmental parameters (water temperature, pH, dissolved oxygen, ammonia nitrogen concentration) for 22 common diseases. The model training adopts a two-stage strategy:

[0068] In the first stage, pre-trained weights are initialized based on a general-purpose large vision model (such as ResNet-50);

[0069] In the second stage, the model's full parameters were fine-tuned using the pathogen database data. During the fine-tuning, the Focal Loss function was used to handle the class imbalance problem. The sample weights for high-frequency diseases such as turtle white spot disease and koi gill rot were 1.0, while the sample weights for rare diseases were 2.5.

[0070] The multimodal AI initial screening and diagnosis module receives image and video data collected by the terminal through the data preprocessing unit and performs noise reduction, enhancement, cropping and standardization processing.

[0071] The disease image recognition unit extracts and recognizes features from preprocessed images based on a fusion model constructed from a target detection network and an image classification network. In this embodiment, a fusion model based on YOLOv8 and ResNet50 is used to extract and recognize features from preprocessed images, identify diseases with visual characteristics, and generate image recognition results. Meanwhile, the multimodal AI initial screening and diagnosis module also includes a natural language understanding unit (e.g., LoRA fine-tuning technology), configured to perform semantic parsing on natural language symptom descriptions collected by the terminal based on a pre-trained language model, extract symptom information, and generate structured natural language understanding results.

[0072] The diagnosis and treatment plan generation unit uses an attention mechanism to fuse the image recognition results and natural language understanding results to make a final diagnosis recommendation and classify the condition into three levels: mild, severe, and critical. It then matches the corresponding treatment plan from a standardized diagnosis and treatment plan library.

[0073] When the multimodal AI initial screening and diagnosis module determines that a user's pet is severely or critically ill and requires on-site treatment, it automatically sends the diagnosis results, disease severity level, and user location information to the intelligent service scheduling module. The scheduling module first determines service priority based on the severity of the illness, with critically ill cases having the highest priority, followed by severely ill cases, and ordinary cases having the lowest. Then, it filters out partner institutions and veterinarians with the appropriate qualifications and service capabilities within a 3-kilometer radius of the user, calculates the estimated arrival time of each candidate veterinarian at the user's location, and performs a comprehensive evaluation based on their current work status and expertise, selecting the veterinarian with the highest score for order allocation. After order allocation, the system automatically sends a notification to both the user and the veterinarian, including their contact information, address, and estimated arrival time, and tracks the veterinarian's location and service progress in real time. As an alternative technical solution, the intelligent service scheduling module can also use other heuristic optimization algorithms such as simulated annealing and particle swarm optimization for order matching. For remote areas, when there is no certified veterinarian within a 3-kilometer radius, the system can expand the matching range to 5 kilometers and prioritize matching veterinarians with experience in providing on-site services.

[0074] In some embodiments, the standardized training and certification module's output is communicatively connected to the cooperative institution management module and the data intelligence optimization module. This standardized training and certification module constructs a three-in-one training system integrating an online AI simulation training platform, offline practical instruction, and unified assessment and certification, including a virtual diagnosis and treatment training system for typical cases. It should be noted that the online AI simulation training system is based on the multimodal AI initial screening and diagnosis module. Trainees can upload images and symptom descriptions of simulated cases to the system, which will provide real-time diagnostic results and treatment plans, and compare them with standard answers for error correction and scoring, helping trainees quickly master the diagnosis and treatment methods for common diseases. Offline practical instruction is conducted by experts from Shanghai Ocean University and senior veterinarians from core cooperative hospitals, primarily teaching basic examination methods for aquatic pets, trauma treatment techniques, common surgical procedures, and laboratory testing techniques. The nationally unified assessment consists of two parts: a theoretical examination and a practical examination. Upon passing the assessment, a nationally unified aquatic pet diagnosis and treatment skills level certificate is issued, divided into three levels: primary, intermediate, and advanced, with different levels corresponding to different service permissions. As an alternative technical solution, online training can take various forms such as live streaming and recorded courses, while offline practical training can be conducted at regional training bases, with certified senior veterinarians serving as instructors.

[0075] The input end of the full-cycle health management module described in this application embodiment is communicatively connected to the multimodal AI initial screening and diagnosis module and the user terminal device, respectively, and the output end is communicatively connected to the offline service terminal application. The full-cycle health management module establishes a unique digital health record for each aquatic pet, recording its basic information, growth and development, treatment history, medication records, vaccination status, and water quality testing data. Based on the record information, it provides personalized prevention suggestions, rehabilitation tracking, and follow-up visit reminders. When a user uses the system described in this application for the first time, the user can add aquatic pet information in the application on the user terminal device. The system generates a unique health record number based on the aquatic pet information. After each treatment, the veterinarian fills in the treatment record and medication information on the mobile work terminal and synchronizes the information to the corresponding health record. Based on the pet's breed, age, health status, and historical treatment records, personalized prevention suggestions are periodically pushed to the user.

[0076] Example 2

[0077] A cloud-based initial screening and home-based treatment management method for aquatic pets based on artificial intelligence, the method being applied to the system described in this application, the method comprising the following steps:

[0078] Step 1: Multidimensional data collection and uploading. The IoT node 300, as the execution node, is responsible for collecting pet and environmental data. The user collects and uploads pet health information through the user terminal 100. The IoT node 300 collects water quality data through the water quality environmental sensor 301 and collects pet swimming posture and body surface image data through the visual monitoring device 302. The above data is uploaded to the cloud central system through the mobile network.

[0079] Step 2: Cloud data processing and multimodal AI preliminary screening. The platform layer server 200 receives data uploaded by the user terminal 100 and the IoT node 300. The multimodal AI preliminary screening and diagnosis module 600 in the platform layer server 200 analyzes the data and, in combination with environmental data and pet health data, performs a preliminary health status assessment and anomaly identification.

[0080] Step 3: Service Triage and Dispatch. Based on the AI ​​preliminary screening results from the multimodal AI preliminary screening and diagnosis module 600, the system triggers the corresponding service process, specifically:

[0081] Triggering online medical consultations:

[0082] In the consultation distribution process, if the multimodal AI preliminary screening and diagnosis module 600 determines that expert intervention is required, the platform layer server 200 will distribute the data and consultation request described in step one to the online experts and diagnosis and treatment terminals 400.

[0083] Expert diagnosis: Experts receive and distribute data and consultation requests through online expert and diagnosis terminal 400, and conduct remote video communication with user terminal 100 through expert video consultation workstation 401, and diagnose aquatic pets by combining the data of AI preliminary screening results.

[0084] For follow-up services, experts can issue prescriptions online through the online expert and diagnosis terminal 400, and users can purchase related supplies through the online pharmacy / shopping mall module 402; in some preferred embodiments, it also includes a pet owner communication community module 403, through which users can share experiences or seek support.

[0085] Triggering offline services:

[0086] If the multimodal AI initial screening and diagnosis module 600 determines that the situation is urgent or requires on-site handling, the platform layer server 200 will generate an on-site sampling / treatment dispatch request based on the data and distribute it to the offline service scheduling gateway 500 of the intelligent service scheduling module.

[0087] Service scheduling: The offline service scheduling gateway 500 has a built-in home medical dispatch system 501 that receives the dispatch request and dispatches experts according to priority based on geographical location and veterinary expertise.

[0088] For door-to-door service, the offline expert's veterinary mobile terminal receives the dispatch request and goes to the user's location to provide door-to-door service;

[0089] Step 4: The online diagnosis and treatment results or offline service data are recorded and synchronized to the platform server 200. Users can view the diagnosis report, treatment suggestions and follow-up care guidelines through their user terminal 100.

[0090] Step two, the multimodal AI initial screening, includes the following steps:

[0091] S21. Multi-source heterogeneous data input: The multi-modal AI initial screening and diagnosis module 600 receives pet health information collected and uploaded by the user terminal 100; water quality data collected and uploaded by the water quality environment sensor 301 in the IoT node 300, and pet swimming posture and body surface image data collected and uploaded by the visual monitoring device 302.

[0092] S22. Data preprocessing: The data preprocessing module 620 performs noise reduction, enhancement, cropping and standardization on the data described in step S21, and performs feature extraction and recognition on the preprocessed data.

[0093] S23. Dual-path parallel analysis: Performing visual and linguistic analysis on the preprocessed data, including:

[0094] Visual analysis: The visual large model module 630 analyzes images and time-series data; the body surface lesion identification submodule 631 identifies physical lesions on the body surface of aquatic pets and extracts visual lesion features; the behavior trajectory analysis submodule 632 analyzes the swimming posture of aquatic pets and extracts abnormal behavior features, generating visual analysis results.

[0095] Language and data understanding: The language model module 640 receives and reads health information and water quality data of aquatic pets, and calls upon pathogen characteristics and pharmacokinetic parameters from the aquatic pet pathology knowledge base 610 for auxiliary analysis; the RAG retrieval enhancement module 641 uses retrieval enhancement generation technology to search for and refer to the latest diagnosis and treatment guidelines from professional documents; the symptom semantic analysis submodule 642 analyzes the textual description of pet health provided by the user, summarizes semantic symptoms, and generates language analysis results.

[0096] S24. Multimodal fusion decision-making: The multimodal fusion decision-making unit 650 integrates visual lesion features, behavioral abnormality features, reference treatment guidelines, and semantic symptom summaries, performs multidimensional feature cross-validation, and calculates the probability and confidence of possible diseases.

[0097] S25. Output diagnostic report: After fusion decision-making, the system generates and outputs an AI preliminary diagnosis and intervention report to provide a scientific basis for subsequent expert diagnosis or treatment.

[0098] Using the visual large model module 630 as the trigger point, combined with basic water temperature data, a lightweight family closed loop (S31 path) of "early detection, simple consultation, and online medicine purchase" is realized, including the following steps:

[0099] S10. Scene triggering and data collection. For example, a pet owner keeps several high-value koi in his family aquarium. One day, during a routine patrol and shooting, the visual monitoring device 302 deployed on the side wall of the aquarium captured multiple tiny white spots on the body surface and fins of one of the koi. At the same time, the water quality environment sensor 301 reported that the current water temperature was 21℃.

[0100] S20, Multimodal AI Cloud Preliminary Screening: After the image and video stream uploaded by S10 is uploaded to the cloud, the body surface lesion identification submodule 631 under the visual large model module 630 performs image segmentation and feature extraction, accurately delineating the location and density of white spots;

[0101] The multimodal fusion decision unit 650 combines water temperature data collected by the water quality environment sensor 301 (where 21℃ is the temperature range where this pathogen is prone to occur), and calls the proprietary aquatic pet pathology knowledge base 610 for cross-comparison. The system determines that it is Ichthyophthirius multifiliis infection (white spot disease), with a diagnostic confidence level of up to 95%.

[0102] S30. Tiered response and intervention decision-making: Given that Ichthyophthirius multifiliis is a common disease and is in the early stage of infection (mild warning range), the system automatically switches to the online light consultation path (S31). The system pushes the "AI Preliminary Diagnosis and Intervention Report" to the user terminal 100, which recommends: physical therapy, specifically, gradually raising the water temperature to above 28°C to disrupt the life cycle of Ichthyophthirius multifiliis, and supplementing with drug treatment.

[0103] S50, a closed-loop online service: pet owners can directly access the online pharmacy / store module 402 via links within the report to purchase specialized white spot treatment (e.g., compliant medications containing malachite green substitutes); the system accurately calculates the milliliter-level dosage based on the aquarium water volume (e.g., 500L) pre-entered by the pet owner. After recovery, the image features of the koi before and after recovery are fed back as positive samples for fine-tuning the visual model.

[0104] In some embodiments, the water quality environment sensor 301 and the pet owner's text input serve as dual trigger points to address complex illnesses with high risk of infection and high mortality, and to facilitate offline dispatching (S40 path). The specific steps include:

[0105] S10, Scene Triggering and Data Acquisition: In a shrimp farm or large live-holding pond, the water quality sensor 301 issues an alarm, indicating that the dissolved oxygen at the bottom has dropped sharply within 2 hours, and the nitrite level is approaching the red line. The farmer then anxiously inputs voice / text into the user terminal 100: "The shrimp have empty intestines and stomachs, swim to the surface, and their body color is white and translucent."

[0106] S20, Multimodal Analysis and Risk Assessment: The data preprocessing module 620 performs trend fitting on the water quality fluctuation curve; the symptom semantic analysis submodule 642 under the language large model module 640 converts user speech into structured tags: [empty intestine and stomach, floating head, muscle turbidity]; the multimodal fusion decision unit 650, with the assistance of the RAG retrieval enhancement module 641, determines that there is an extremely high risk of an outbreak of acute hepatopancreatic necrosis disease (AHPND, early death syndrome) or severe Vibrio infection. Due to the lack of microscopic pathological data, the confidence level for diagnosis is only 60%, but the risk rating is "extremely high".

[0107] S40, Expert Intervention and Offline Dispatch: The system forcibly switches to the expert video consultation workstation 401, where the aquatic pathology expert takes over and reviews the AI-generated emergency report and water quality curve; the expert determines that PCR nucleic acid testing is needed immediately to confirm the pathogenic strain, and the system instruction is issued to the door-to-door medical dispatch system 501.

[0108] The platform matches and dispatches nearby aquatic veterinarians (or grid-based technicians) with portable microscopes and sampling tubes to arrive at the scene within 2 hours to carry out water quality control and rescue and extract pathogens;

[0109] S50, data feedback, and finally offline laboratory PCR confirmation results are entered into the system. The system records the characteristics of drastic changes in water quality and subjective symptom descriptions before the disease on this batch of shrimp, and optimizes the AI's early prediction weight for sudden major diseases.

[0110] In some preferred embodiments, the visual behavior trajectory analysis submodule 632 and the language large model module 640 interact dynamically through multiple rounds of questioning to investigate "environmental stress" problems caused by non-pathogenic bacteria, as detailed below:

[0111] (1) Scene triggering and behavior trajectory analysis (S10 & S20 pre-processing), for example, a pet owner keeps a red arowana. The fish has no lesions on its body surface, but the visual monitoring device 302 continuously records the fish's abnormal swimming trajectory (such as: frequent rubbing of the tank, nervous darting, or even attempts to jump out of the tank). The behavior trajectory analysis submodule 632 extracts the kinematic abnormal vector and triggers the system to actively query;

[0112] (2) Dynamic multi-turn interaction dominated by the large language model (S20): The system initiates anthropomorphic follow-up questions on the user terminal 100 through the language large model module 640: "The system has detected that your arowana is severely agitated and rubbing the tank. Have you changed the filter cotton, changed a large amount of water, or added new equipment in the last 24 hours?"

[0113] The user replied: "Yesterday I changed half of the water, and I added the tap water directly without letting it sit (aerate)."

[0114] (3) Feature fusion and drug-free decision-making S30, symptom semantic analysis submodule 642 extracts the core reason: [large-scale replacement of unaerated tap water].

[0115] The multimodal fusion decision unit 650, combining the abnormal fish trajectory with subjective operational errors, immediately ruled out parasitic infection and diagnosed it as a severe stress response caused by residual chlorine poisoning.

[0116] (4) Emergency intervention guidance and closed loop (S31 & S50): The system intercepts the behavior of pet owners who may blindly use insecticides or antibiotics and immediately generates the "First Aid Guide for Residual Chlorine Poisoning": It is recommended to immediately add a large dose of water quality stabilizer (sodium thiosulfate / sodium sodium thiosulfate) and turn on the maximum power of oxygenation.

[0117] 24 hours later, the visual monitoring device 302 continued to track the arowana's swimming trajectory. After confirming that it had returned to a stable swimming state, the system successfully associated this "abnormal trajectory + no lesions + water change semantics" as a benchmark case of "acute residual chlorine stress" and stored it in the underlying database.

[0118] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A cloud-based, full-cycle management system for cloud-based initial screening and home-based treatment of aquatic pets based on artificial intelligence, characterized in that: The system includes at least one offline service terminal, deployed at the field service node, used to collect multimodal data of aquatic pets and perform field service operations; This includes a cloud-based central system that communicates with the offline service terminals via a mobile communication network. This system receives data from the offline service terminals and performs data processing, decision-making, and resource scheduling. An AI computing cluster is configured to receive multimodal data of aquatic pets from the offline service terminal, and to analyze the multimodal data to generate diagnostic suggestion information. The intelligent service scheduling module is connected to the AI ​​computing cluster and is configured to obtain the diagnostic suggestion information and the on-site resource status fed back by the offline service terminal, generate and send the service execution plan to the offline service terminal; The offline service terminal receives the service execution plan and performs the corresponding on-site service operations.

2. The cloud-based initial screening and home-based treatment management system for aquatic pets based on artificial intelligence as described in claim 1, characterized in that, The cloud-based central system also includes a full-cycle health management module, which is communicatively connected to the AI ​​computing cluster and the intelligent service scheduling module. This module is used to establish and store electronic health records for each aquatic pet. The electronic health records contain the aquatic pet's historical medical data and diagnostic recommendations.

3. A cloud-based initial screening and home-based treatment management system for aquatic pets based on artificial intelligence, as described in claim 1 or 2, is characterized in that... The AI ​​computing cluster includes a multimodal AI initial screening and diagnosis module, whose input end is connected to the offline service terminal application, and whose output end is connected to the intelligent service scheduling module and the full-cycle health management module respectively. The multimodal AI initial screening and diagnosis module integrates a large vertical model of aquatic pet diseases and a disease image recognition unit. It supports text symptom description, image upload, and video consultation and diagnosis modes. Based on the large vertical model of aquatic pet diseases, it identifies diseases and outputs diagnostic results, disease classification, and standardized treatment plans. The disease image recognition unit extracts and recognizes features from preprocessed images using a fusion model built on a target detection network and an image classification network. It also identifies visually distinctive diseases by combining a large vertical model of aquatic pet diseases and generates image recognition results. The multimodal AI initial screening and diagnosis module also includes a natural language understanding unit, which is configured to perform semantic parsing on the natural language symptom descriptions collected by the terminal based on a pre-trained language model, extract symptom information, and output structured symptom labels and symptom severity scores.

4. A cloud-based initial screening and home visit management system for aquatic pets based on artificial intelligence, as described in claim 1 or 2, is characterized in that... The input end of the intelligent service scheduling module is connected to the multimodal AI initial screening and diagnosis module, and the output end is connected to the veterinary mobile work terminal. The intelligent service scheduling module adopts a multi-objective optimization scheduling algorithm based on genetic algorithm, which integrates spatial factors, clinical factors, and quality factors to prioritize users and certified veterinarians.

5. The cloud-based initial screening and home-based treatment management system for aquatic pets based on artificial intelligence as described in claim 1, characterized in that, The system also includes a standardized training and certification module, which specifically involves building a three-in-one training system that combines online AI simulation training with offline hands-on instruction and unified assessment and certification. This system is used to provide standardized training and assessment for certified veterinarians in their diagnostic and treatment capabilities.

6. The cloud-based initial screening and home-based treatment full-cycle management system for aquatic pets based on artificial intelligence as described in claim 2, characterized in that, The offline service terminal includes at least one of a veterinary mobile terminal, a user terminal, and a standardized diagnosis and treatment package; the input end of the full-cycle health management module is communicatively connected to the multimodal AI initial screening and diagnosis module and the veterinary mobile work terminal, respectively, and the output end is communicatively connected to the offline service terminal application.

7. The cloud-based initial screening and home-based treatment full-cycle management system for aquatic pets based on artificial intelligence as described in claim 2, characterized in that, The full-cycle health management module establishes a unique digital health record for each aquatic pet, recording its basic information, growth and development, medical history, medication records, vaccination status, and water quality testing data. Based on the record information, it provides personalized prevention suggestions, rehabilitation tracking, and follow-up visit reminders.

8. The cloud-based initial screening and home visit management system for aquatic pets based on artificial intelligence as described in claim 1, characterized in that, The cloud-based central system also includes an intelligent optimization module, which communicates with the AI ​​computing cluster and the intelligent scheduling module. The intelligent optimization module uses federated learning technology to collect and integrate clinical diagnosis and treatment data from across the country while protecting user privacy and institutional data security. It regularly iterates and optimizes the AI ​​computing cluster and updates relevant training course content and treatment guidelines.

9. A method for cloud-based initial screening and home-based treatment management of aquatic pets based on artificial intelligence, wherein the method is applied to the system described in any one of claims 1-8, characterized in that, The method includes the following steps: Step 1: Multidimensional data collection and uploading. The IoT node (300) is responsible for collecting pet and environmental data as the on-site execution node. The user collects and uploads pet health information through the user terminal device (100). The IoT node (300) collects water quality data through the water quality environmental sensor (301) and collects pet swimming posture and body surface image data through the visual monitoring device (302). The above data is uploaded to the cloud central system through the mobile network. Step 2: Cloud data processing and multimodal AI preliminary screening. The platform layer server (200) receives data uploaded by user terminal devices (100) and IoT nodes (300). The multimodal AI preliminary screening diagnostic module (600) in the platform layer server (200) analyzes the data and combines environmental data and pet health data to conduct a preliminary health status assessment and anomaly identification, forming the AI ​​preliminary screening result. Step 3: Service Triage and Dispatch. Based on the AI ​​preliminary screening results from the multimodal AI preliminary screening and diagnosis module (600), the system triggers the corresponding service process, specifically: Triggering online medical consultations: In the consultation distribution, if the multimodal AI preliminary screening diagnostic module (600) determines that expert intervention is required, the platform layer server (200) will distribute the data and consultation request described in step one to online experts and diagnosis and treatment terminals (400). Expert diagnosis: Experts receive and distribute data and consultation requests through online expert and diagnosis terminal (400), and conduct remote video communication with user terminal device (100) through expert video consultation workstation (401) to diagnose aquatic pets in combination with the data of AI preliminary screening results; For follow-up services, experts can issue prescriptions online through the online expert and diagnosis terminal (400), and users can purchase related supplies through the online pharmacy / shopping mall module (402); Triggering offline services: When the dispatch order is generated, if the multimodal AI screening and diagnosis module (600) determines that the situation is urgent or requires on-site handling, the platform layer server (200) will generate an on-site sampling / treatment dispatch request from the data described in step one and distribute it to the offline service scheduling gateway (500) of the intelligent service scheduling module. Service scheduling: The offline service scheduling gateway (500) has a built-in home medical dispatch system (501) that receives the dispatch request and dispatches experts according to priority based on geographical location and veterinary expertise. For door-to-door service, the offline expert's veterinary mobile terminal receives the dispatch request and goes to the user's location to provide door-to-door service; Step 4: The online diagnosis and treatment results or offline service data are recorded and synchronized to the platform layer server (200), and the user can view the diagnosis report, treatment suggestions and follow-up care guidelines through their user terminal device (100).

10. A method for the full-cycle management of cloud-based initial screening and home-based treatment of aquatic pets based on artificial intelligence, as described in claim 9, is characterized in that... Step two, the multimodal AI initial screening, includes the following steps: S21. Multi-source heterogeneous data input: The multimodal AI screening and diagnosis module (600) receives pet health information collected and uploaded by the user terminal device (100); water quality data collected and uploaded by the water quality environment sensor (301) in the Internet of Things node (300), and pet swimming posture and body surface image data collected and uploaded by the visual monitoring device (302); S22. Data preprocessing: The data preprocessing module (620) performs noise reduction, enhancement, cropping and standardization on the data described in step S21, and performs feature extraction and recognition on the preprocessed data. S23. Dual-path parallel analysis: Performs visual and linguistic analysis on the preprocessed data, including: Visual analysis: The visual large model module (630) analyzes images and time series data; the body surface lesion identification submodule (631) identifies physical lesions on the body surface of aquatic pets and extracts visual lesion features; the behavior trajectory analysis submodule (632) analyzes the swimming posture of aquatic pets and extracts abnormal behavior features, generating visual analysis results. Language and data understanding, the language big model module (640) receives and reads aquatic pet health information and water quality data, and calls the pathogen characteristics and pharmacokinetic parameters in the aquatic pet pathology knowledge base (610) for auxiliary analysis; the RAG retrieval enhancement module (641) constructs a vector database, namely the aquatic pet pathology knowledge base, through document segmentation and vectorization, and then performs query rewriting and query expansion on user queries, and finally recalls fragments through a hybrid retrieval of vector retrieval and keywords, and performs rearrangement and context compression according to the number of recalled fragments; the symptom semantic analysis submodule (642) analyzes the pet health text description provided by the user, performs semantic symptom summarization, and generates language analysis results; S24, Multimodal fusion decision-making, the multimodal fusion decision-making unit (650) integrates visual lesion features, behavioral abnormality features, reference treatment guidelines, and semantic symptom summarization, performs multidimensional feature cross-validation, and calculates the probability and confidence of possible diseases; S25. Output diagnostic report: After fusion decision-making, the system generates and outputs an AI preliminary diagnosis and intervention report to provide a scientific basis for subsequent expert diagnosis or treatment.

Images