Retrieval methods and related equipment based on an intelligent case database of aquatic pathogens and diseases

By preprocessing, knowledge graph annotation, and multi-dimensional analysis of aquatic disease case data, combined with aquaculture model and day segmentation system, intelligent retrieval and precise assisted diagnosis and treatment of aquatic diseases have been achieved. This has solved the problems of insufficient standardization and intelligence in the existing system and improved the diagnosis and prevention capabilities of grassroots aquaculture.

CN123115227APending Publication Date: 2026-11-17QINGYUAN HAIBEI BIO-TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202610988133.7
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-07-03
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Existing aquatic disease diagnosis systems cannot meet the actual needs of grassroots aquaculture. The knowledge base and case database lack standardization and completeness, the standards for manual case collection are inconsistent, and special cases such as emerging diseases and mixed infections are missing. The system's intelligent reasoning ability is weak, the interaction mode is outdated, and the degree of data value mining is low, making it difficult to provide scenario-based and personalized accurate diagnostic services.

Method used

By preprocessing unstructured raw case data of aquatic pathogens, identifying, labeling, and automatically classifying them using an aquatic disease knowledge graph, summarizing the graphical and interface interaction response mechanisms based on aquaculture models, performing interval matching and multi-dimensional weight calculations, and using a large artificial intelligence model for multi-dimensional statistical analysis, a list of accurately matched cases and diagnostic and treatment assistance plans are generated.

Benefits of technology

It has achieved precise and intelligent aquatic disease retrieval, lowered the user threshold, improved the accuracy and efficiency of disease diagnosis, adapted to the operating habits of grassroots farmers, provided comprehensive technical support, and enhanced the professionalism and standardization of disease prevention and control and drug use.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123115227A_ABST
    Figure CN123115227A_ABST
Patent Text Reader

Abstract

This application relates to the field of intelligent aquaculture technology, and in particular to a retrieval method and related equipment based on an intelligent case database of aquatic pathogens and diseases. The method includes preprocessing unstructured raw case data of aquatic pathogens and diseases, completing entity annotation, classification, and tag binding based on an aquatic disease knowledge graph; the front end uses aquaculture mode summary graphical interaction to lock in the target aquaculture mode, and performs a dual joint search based on the aquaculture day interval to obtain a preliminary candidate case set; then, a six-dimensional association matching model is used for weighted calculation to output a differentiated and precise matching case list; finally, a fine-tuned artificial intelligence model in the aquaculture field, combined with the retrieval scenario and matching cases, conducts a full-domain statistical analysis of structured cases, and outputs statistical analysis results and diagnostic and treatment assistance plans. This application can lower the user threshold, improve the accuracy of case matching, and achieve intelligent database entry throughout the entire process.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This application relates to the field of intelligent aquaculture technology, and in particular to a retrieval method and related equipment based on an intelligent case database of aquatic pathogens and diseases. Background Technology

[0002] my country's aquaculture industry is dominated by decentralized, small-scale grassroots farming entities with varying levels of professional expertise. Coupled with a lack of specialized resources for aquatic diseases, insufficient coverage of scientific and technological services, and low accessibility for remote diagnosis, farmers have long faced problems such as sudden outbreaks of disease, difficulties in diagnosis, inconvenient treatment, and non-standard medication use. As a result, the industry suffers persistently high economic losses due to diseases. With the iteration of technologies such as artificial intelligence and the Internet of Things, smart fisheries have become a core direction for the transformation and upgrading of aquaculture. Intelligent aquatic disease diagnosis and decision-making systems have also become a research hotspot in the industry. Their technological development has progressed through rule-based expert systems and digital networked retrieval stages, and has now entered an AI-driven intelligent stage. Various image recognition, big data early warning, and large-model-assisted diagnostic platforms have been developed both domestically and internationally, achieving an initial shift in disease diagnosis from experience-based to data-driven approaches.

[0003] However, the current aquatic disease diagnosis system and case database still have many core shortcomings, failing to meet the actual needs of grassroots aquaculture. Firstly, the knowledge base and case database lack standardization and completeness. Case collection standards are inconsistent, coverage is limited, and special cases such as emerging diseases and mixed infections are missing. Furthermore, subjective differences in expert experience significantly reduce case matching accuracy, easily leading to misdiagnosis and missed diagnoses. Secondly, the system's intelligent reasoning ability is weak, lacking self-learning and dynamic adaptation mechanisms. It relies on fixed rules and keyword matching, failing to deeply connect disease characteristics with the complex relationships of the aquaculture environment, seasonal rhythms, and aquaculture models, making it difficult to provide scenario-based and personalized accurate diagnostic services. Thirdly, the system's interaction mode is outdated, with poor practicality at the grassroots level. It mainly relies on cumbersome text searches, lacking lightweight interactive functions. Large-scale model dialogue interaction has a high barrier to entry, making it difficult for middle-aged and elderly aquaculture farmers to operate, easily leading to model recognition errors and distorted diagnostic results, hindering the technology's widespread adoption at the grassroots level. Fourth, the level of data value mining is low. The existing case library only has data storage and query functions, lacks auxiliary decision-making capabilities such as data analysis, risk warning, and solution optimization, and cannot rely on large models to integrate industry knowledge.

[0004] In summary, the technical problems existing in the relevant technologies need to be improved. Summary of the Invention

[0005] The main objective of this application is to propose a retrieval method and related equipment based on an intelligent case database of aquatic pathogens and diseases, which can reduce the user's threshold for use, improve the accuracy of case matching, and realize intelligent database entry throughout the entire process.

[0006] To achieve the above objectives, one aspect of this application proposes a retrieval method based on an intelligent case database of aquatic pathogens and diseases, the method comprising the following steps: Preprocess the acquired unstructured raw case data of aquatic pathogens and diseases; Based on the aquatic disease knowledge graph, the pre-processed unstructured aquatic pathogen original case data are sequentially identified, labeled, automatically classified, and bidirectionally associated to obtain structured aquatic pathogen original case data. The target aquaculture model is obtained by identifying and locking user interface click operations through a summary graph of aquaculture model, a unique mapping mechanism of aquaculture model, and an interface interaction response mechanism. Based on the segmentation system of breeding days and the selection control of breeding days, the target breeding mode and breeding days are subjected to interval matching and dual joint retrieval processing to obtain a preliminary candidate case set; The preliminary candidate case set is weighted in multiple dimensions according to the six-dimensional association matching model to obtain a list of differentiated and precise matching cases. By combining the target aquaculture model, the preliminary candidate case set, and the differentiated precise matching case list with an artificial intelligence big data model, multi-dimensional statistical analysis is performed on the original case data of structured aquatic pathogens and diseases to obtain statistical analysis results and diagnostic and treatment assistance plans.

[0007] In some embodiments, after the preprocessed unstructured aquatic pathogen original case data is identified and labeled according to the aquatic disease knowledge graph, the labeled unstructured aquatic pathogen original case data is automatically classified according to a preset classification dimension. The preset classification dimensions include pathogen type classification, disease symptom classification, aquaculture species classification, aquaculture mode classification, aquaculture stage classification, and treatment effectiveness classification.

[0008] In some embodiments, the aquaculture model summary diagram includes an aquaculture process diagram, a model structure diagram, or a stage summary diagram. The interface interaction response mechanism includes displaying a summary graphic of the breeding model in a visual form on the interactive interface.

[0009] In some embodiments, the number of days of raising animals is pre-divided according to the raising mode of the animals to obtain the raising day segmentation system, wherein the raising mode includes the entire cycle of raising animals from stocking to pond clearing; The segmented breeding days system is retrieved based on the target breeding model.

[0010] In some embodiments, the matching dimensions of the six-dimensional association matching model include, in descending order of weight, breeding mode, breeding days, pathogen type, symptom similarity, environmental parameters, and drug efficacy.

[0011] In some embodiments, the differentiated precise matching case list includes multiple matching cases; The matched cases include matching score, case source, case time, and case reliability level.

[0012] In some embodiments, the multidimensional statistical analysis includes statistics on prevalent diseases under different farming models, disease risk prediction at different farming stages, and analysis of mortality and prevalence of typical pathogens.

[0013] In some embodiments, the following steps are also included: Obtain pathogen and disease images uploaded by users; the pathogen and disease images include images of disease symptoms, anatomical images, or microscopic images of pathogens; Image feature extraction is performed on the pathogen and lesion image to obtain pathogen and lesion features; The pathogen and disease characteristics are matched with the image data in the intelligent case library of aquatic pathogens and diseases to obtain the image matching results; The image matching results and the differential precise matching case list are fused and output to obtain the image-text matching case list.

[0014] To achieve the above objectives, another aspect of this application proposes a retrieval system based on an intelligent case database of aquatic pathogens and diseases, comprising: The case data processing and automatic classification module is used to preprocess the acquired unstructured aquatic pathogen original case data; based on the aquatic disease knowledge graph, the preprocessed unstructured aquatic pathogen original case data is sequentially identified, labeled, automatically classified, and bidirectionally associated to obtain structured aquatic pathogen original case data. The aquaculture mode graphical interaction module is used to identify and lock the user's interface click operations through the aquaculture mode summary graphics, the aquaculture mode unique mapping mechanism and the interface interaction response mechanism to obtain the target aquaculture mode. The breeding days stage screening module is used to perform interval matching and dual joint retrieval processing on the target breeding mode and breeding days according to the breeding days segmentation system and breeding days selection control to obtain a preliminary candidate case set. The multi-dimensional differentiated matching module is used to perform multi-dimensional weight calculation on the preliminary candidate case set according to the six-dimensional association matching model to obtain a list of differentiated and accurate matching cases. The large-scale model statistical and treatment assistance module is used to perform multi-dimensional statistical analysis on the original case data of structured aquatic pathogens and diseases by combining the target aquaculture model, the preliminary candidate case set, and the differentiated precise matching case list with an artificial intelligence large-scale model, so as to obtain statistical analysis results and treatment assistance plans.

[0015] To achieve the above objectives, another aspect of this application provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the method described above.

[0016] The embodiments of this application include at least the following beneficial effects: This application provides a retrieval method and related equipment based on an intelligent case database of aquatic pathogens and diseases. This solution achieves standardized storage of scattered unstructured data by preprocessing, intelligently labeling, classifying, and associating original disease case data, thus constructing a standardized and comprehensive intelligent case database of aquatic pathogens and diseases. This completely improves the problems of traditional case data being messy, inconsistent in standards, and having low reference value, laying a solid data foundation for accurate retrieval and intelligent analysis. At the same time, it abandons the traditional cumbersome text retrieval mode, relying on graphical interaction to lock in the aquaculture mode, and combining a segmented system of aquaculture days to complete dual joint retrieval, which is in line with the operating habits of grassroots farmers, significantly reducing the system's usage threshold, adapting to the practical needs of middle-aged and elderly farmers, and effectively improving retrieval efficiency. Furthermore, by using a six-dimensional association matching model to perform multi-dimensional weight calculations on candidate cases, it can achieve scenario-based and differentiated accurate case matching, avoiding the defects of traditional keyword retrieval matching being one-sided, inaccurate, and prone to missed or misjudgments. Finally, by integrating aquaculture scenario parameters with precisely matched cases using a domain-specific AI model, multi-dimensional statistical analysis of structured data is conducted to generate customized diagnostic and treatment support plans, overcoming the limitations of traditional systems that only offer case queries and lack decision support. This method upgrades aquatic disease retrieval from "passive querying" to "proactive intelligent analysis and precise assisted diagnosis and treatment," effectively improving the accuracy, efficiency, and professionalism of disease diagnosis. It provides comprehensive technical support for the scientific prevention and control of aquatic diseases, standardized drug use, and risk warning, and is suitable for large-scale application at the grassroots level. Attached Figure Description

[0017] Figure 1 This is a flowchart of a retrieval method based on an intelligent case database of aquatic pathogens and diseases provided in an embodiment of this application; Figure 2 This is a flowchart illustrating the construction process of aquatic disease knowledge graph; Figure 3 This is a flowchart illustrating the construction process of an intelligent case library for aquatic pathogens and diseases. Figure 4 This is a flowchart illustrating the construction process of a six-dimensional correlation matching model; Figure 5 This is a flowchart illustrating the process of creating a full case database; Figure 6 This is a flowchart illustrating the unique mapping mechanism of the aquaculture model; Figure 7 This is a flowchart illustrating the process of generating a list of differentiated and precise matching cases; Figure 8 This is a flowchart illustrating the process of multi-dimensional statistical analysis using a large artificial intelligence model; Figure 9 This is a flowchart illustrating the reasoning and computation process of a large artificial intelligence model. Figure 10 This is a schematic diagram of the functional interface of the retrieval system based on the intelligent case database of aquatic pathogens and diseases; Figure 11 This is a schematic diagram of the case search interface in the retrieval system based on the intelligent case database of aquatic pathogens and diseases; Figure 12 This is a schematic diagram of the intelligent diagnosis and treatment interface in the retrieval system based on the intelligent case database of aquatic pathogens and diseases; Figure 13 This is a schematic diagram of the differentiated precise matching case list output by an embodiment of this application; Figure 14 This is a schematic diagram of the structure of the retrieval system based on the intelligent case database of aquatic pathogens and diseases provided in the embodiments of this application; Figure 15 This is a schematic diagram of the hardware structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of this application and are not intended to limit it. In the following description, when referring to the accompanying drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with those of this application; they are merely examples of apparatuses and methods consistent with some aspects of the embodiments of this application as detailed in the appended claims.

[0019] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application only and is not intended to limit this application.

[0020] Figure 1 This is an optional flowchart of a retrieval method based on an intelligent case database of aquatic pathogens provided in this application embodiment. Figure 1The method may include, but is not limited to, steps S110 to S160.

[0021] Step S110: Preprocess the obtained unstructured raw case data of aquatic pathogens and diseases; Step S120: Based on the aquatic disease knowledge graph, the pre-processed unstructured aquatic pathogen original case data are sequentially identified, labeled, automatically classified, and bidirectionally associated to obtain structured aquatic pathogen original case data. Step S130: The user's interface click operation is identified and locked through the summary graph of the breeding mode, the unique mapping mechanism of the breeding mode and the interface interaction response mechanism to obtain the target breeding mode. Step S140: Based on the segmentation system of breeding days and the selection control of breeding days, perform interval matching and dual joint retrieval processing on the target breeding mode and breeding days to obtain a preliminary candidate case set; Step S150: Calculate the multi-dimensional weights of the preliminary candidate case set according to the six-dimensional association matching model to obtain a differentiated and precise matching case list; Step S160: By combining the target aquaculture model, the preliminary candidate case set, and the differentiated and precise matching case list with the artificial intelligence big data model, multi-dimensional statistical analysis is performed on the original case data of structured aquatic pathogens and diseases to obtain statistical analysis results and diagnosis and treatment assistance plans.

[0022] Steps S110 to S160 of this embodiment involve preprocessing the collected unstructured raw aquatic pathogen case data, and then, based on an aquatic disease knowledge graph, sequentially performing entity recognition labeling, automated classification, and bidirectional association binding of aquaculture mode and aquaculture day interval labels on the preprocessed raw case data. This generates standardized structured aquatic pathogen case data, constructing an intelligent searchable aquatic pathogen case library. During retrieval, based on the unique mapping mechanism between the aquaculture mode summary graph and the aquaculture mode, and the interface interaction response mechanism, the target aquaculture mode corresponding to the user interface click operation is identified and locked, achieving rapid location of the aquaculture scenario. On this basis, combined with a preset aquaculture day segmentation system and aquaculture day selection control, interval matching is performed based on the user-selected or input aquaculture days. The database is searched using the target aquaculture mode and the corresponding aquaculture day interval as dual joint search conditions, resulting in a preliminary candidate case set. Subsequently, a six-dimensional association matching model is used to perform multi-dimensional weight calculation and ranking of the preliminary candidate case set, outputting a differentiated and accurate matching case list tailored to the user's actual scenario. Finally, relying on the big data model of artificial intelligence, combined with the user's target aquaculture model, preliminary candidate case set and differentiated and precise matching case list, multi-dimensional statistical analysis is carried out on the structured case data in the full case database to explore data information such as disease patterns, risk characteristics and drug effects. Finally, the corresponding statistical analysis results and personalized diagnosis and treatment assistance plans including pathogen prediction, prevention and control measures, drug reference and precautions are output to realize intelligent retrieval and precise assisted diagnosis and treatment of aquatic diseases.

[0023] In some embodiments, steps S110 to S120 first involve preprocessing the unstructured aquatic pathogen disease case data collected from multiple sources. This unstructured data includes various types of frontline aquaculture data such as aquaculture logs, treatment records, and feedback from frontline farmers. This raw data generally suffers from problems such as disorganized formatting, excessive colloquialisms, redundant content, and the inclusion of user privacy information, making it unsuitable for direct retrieval and analysis. To address this, preset text cleaning rules are used to filter out invalid characters, blank content, and invalid information. Simultaneously, a similarity comparison algorithm is used for precise data deduplication, automatically removing highly duplicated cases with a similarity threshold exceeding 95%. At the same time, an intelligent privacy desensitization mechanism is activated to automatically identify, hide, and replace privacy information such as farmer names, contact information, aquaculture addresses, and aquaculture farm registration numbers contained in the case data. This optimizes data quality while ensuring the compliance of data storage and use, providing a clean and standardized raw data foundation for subsequent intelligent labeling and classification.

[0024] After data preprocessing, an intelligent identification and annotation process is performed on the preprocessed unstructured original aquatic pathogen disease case data, relying on a pre-constructed knowledge graph in the aquatic disease domain. Specifically, a dedicated entity recognition model is invoked to perform feature mining and precise extraction on the text data, uniformly annotating five core entity features: pathogen entities such as bacteria, viruses, parasites, or fungi; disease symptom entities such as surface bleeding, gill rot, ascites, swimming near the edge, or decreased feeding; entities representing various aquaculture models such as ponds, factory farming, and small sheds; environmental parameter entities such as water temperature, dissolved oxygen, pH, and ammonia nitrogen; and medication record entities such as drug type, dosage, cycle, and treatment effect. Through the intelligent annotation method empowered by the knowledge graph, the unstructured barrier of the original data is completely broken down, the core effective information of the cases is accurately extracted, and standardized feature extraction is achieved from the messy text data.

[0025] After data identification and labeling are completed, the labeled case data is automatically classified according to preset multi-dimensional classification rules. Specifically, the preset classification dimensions include six major dimensions: pathogen type classification, disease symptom classification, aquaculture species classification, aquaculture mode classification, aquaculture stage classification, and treatment effectiveness classification. This comprehensively covers the core attribute characteristics of aquatic disease cases, enabling refined classification of case data. After classification, the standardized case data undergoes bidirectional association binding, matching and binding corresponding aquaculture mode tags and aquaculture day interval tags to each structured case data, achieving precise association between case data and different aquaculture scenarios and the entire aquaculture cycle. Finally, the entire process of automated data governance is completed, generating standardized structured aquatic pathogen and disease case data and storing it in the aquatic pathogen and disease intelligent case library. The entire process requires no manual secondary processing and labeling, significantly improving the construction efficiency and data accuracy of the aquatic pathogen and disease intelligent case library.

[0026] In some embodiments, in step S130, multiple aquaculture model summary graphics are pre-stored. These graphics include aquaculture process diagrams, model structure diagrams, or stage summary diagrams, covering common aquaculture models such as intensive earthen pond culture, recirculating aquaculture systems, cage culture, elevated pond culture, and fry rearing. The corresponding graphical interfaces support online editing and real-time updates by administrators and can be dynamically expanded and optimized based on aquaculture model iterations. A unique mapping mechanism for aquaculture models is established, creating a unique binding relationship between various aquaculture model summary graphics and corresponding aquaculture model data. Through an interactive interface response mechanism, all aquaculture model summary graphics are centrally displayed in a visual form on the interactive interface, replacing the traditional cumbersome text search entry and providing users with an intuitive and convenient search operation platform.

[0027] The system identifies and locks onto user interface clicks through a combination of graphical summaries of farming patterns, a unique mapping mechanism for farming patterns, and an interactive response mechanism to determine the target farming pattern. Specifically, by combining identified key classification tags with color-block buttons on the interface, it achieves precise case matching tailored to specific farming rhythms and stages. When a user clicks the color-block button corresponding to any graphic, the system quickly locks onto the target farming pattern, automatically displays the corresponding farming day segmentation system, and guides the user to the corresponding case matching results interface. In addition to the graphical interactive search method, it also supports multiple search entry points such as field matching and key condition filtering, allowing flexible navigation to the case matching page. This significantly lowers the learning curve for grassroots farmers and adapts to real-world farming scenarios.

[0028] In some embodiments, in step S140, a pre-defined stage division is performed based on the different farming modes of farmed animals to construct a farming days segmentation system covering the entire growth cycle of farmed animals from "stocking-seedling stage-growth stage-fattening stage-pond cleaning". This segmentation system is specifically set according to the actual growth patterns of different farmed species and farming modes. For example, for the high-level pond farming mode of Litopenaeus vannamei, fixed farming days intervals are divided into seedling stage, juvenile shrimp stage, medium shrimp stage, adult shrimp stage, and fattening stage. Various fish farming modes are adapted to exclusive cycle stage division standards, which are fully in line with the rhythm of front-line farming production. Moreover, according to the target farming mode locked by the user, the exclusive farming days segmentation system that matches it can be accurately retrieved, providing a standardized basis for staged and accurate retrieval.

[0029] Based on a segmented system of aquaculture days and a selection control for aquaculture days, data matching and retrieval are completed. The front-end configuration stage features both quick selection and custom day input, adapting to the operational needs of different users. After the user selects or enters the number of aquaculture days, the system automatically completes interval matching, accurately correcting for human input bias and ensuring that the filtering conditions conform to industry standards. Subsequently, the system uses the target aquaculture mode and the matched target aquaculture day interval as dual joint search conditions to search the intelligent case library of aquatic pathogens and diseases. This effectively filters invalid cases across aquaculture scenarios and growth stages, ultimately obtaining a preliminary set of candidate cases highly adapted to the user's current aquaculture scenario, significantly improving the targeting and accuracy of disease case retrieval.

[0030] In some embodiments, in step S150, a six-dimensional association matching model is used to calculate the multi-dimensional weights of the preliminary candidate case set, thereby obtaining a differentiated and precise matching case list. The six-dimensional association matching model has a clear retrieval weight priority, and the weights of each matching dimension, from largest to smallest, are: farming mode, farming days, pathogen type, symptom similarity, environmental parameters, and drug effect. Among them, the farming stage corresponding to the farming mode and farming days is the core matching dimension, with the highest weight, which fully matches the industry characteristics of aquatic diseases being scenario-dependent and stage-specific. Feature comparison is carried out dimension by dimension for the preliminary candidate case set, and the comprehensive matching degree score of each matching case with the user's actual farming scenario is quantitatively calculated to accurately judge the case fit. High-quality cases with high fit in farming mode, farming stage, pathogen type, and disease symptoms are prioritized for selection and retention, thereby achieving refined screening and precise ranking of candidate cases.

[0031] The generated list of differentiated and precise matching cases contains multiple matching cases, each associated with a matching score, source, time, and reliability level. This provides a clear overview of case quality, allowing users to quickly identify and select high-quality reference cases. Building upon this foundation, a multimodal image matching and retrieval function is added. Users can upload images of disease symptoms, anatomical images, and microscopic images of pathogens. The system automatically extracts lesion features from the images and compares their similarity with image data within the case database. Finally, the image matching results are fused with the differentiated and precise matching case list obtained through text retrieval, effectively overcoming the limitations of single text retrieval and further improving the overall accuracy of aquatic disease case retrieval.

[0032] In some embodiments, in step S160, a large-scale artificial intelligence (AI) model specifically fine-tuned for the aquatic disease field is used. Compared to a general large-scale model, the AI ​​industry-specific large-scale model possesses stronger professional understanding of aquatic diseases, feature reasoning, and industry data summarization capabilities, enabling it to accurately adapt to the professional needs of fishery disease analysis and decision support. Based on the AI ​​industry-specific large-scale model, multi-dimensional statistical analysis of structured aquatic pathogen case data can be conducted by combining the retrieved target aquaculture model, preliminary candidate case set, and differentiated precise matching case list. The structured full-case database is input into the AI ​​large-scale model fine-tuned for the aquatic disease field. The large-scale model then completes multi-dimensional statistical analysis of the entire database. Specifically, it can achieve in-depth data analysis such as statistics of high-incidence diseases under different aquaculture models, disease risk prediction at different aquaculture day stages, and mortality and epidemic analysis of typical pathogens. Simultaneously, it can output multi-dimensional analysis results such as overall incidence rate statistics, high-incidence pathogen ranking, regional disease patterns, and disease risks for each aquaculture model, effectively completing the in-depth mining of existing case data and fully releasing the value of industry data.

[0033] Based on multi-dimensional statistical analysis, the domain-fine-tuned AI industry model can combine the user's corresponding aquaculture scenario, aquatic product growth stage, and highly matching disease case characteristics to complete intelligent reasoning and judgment, automatically generating an integrated diagnosis and treatment assistance plan. This plan covers comprehensive reference content including pathogen prediction, disease cause analysis, precise prevention and control measures, compliant medication guidance, aquaculture water environment control, and key points of refined aquaculture management. Furthermore, the output diagnosis and treatment content is a data-driven intelligent reference and does not constitute a legally binding diagnosis or treatment conclusion, effectively avoiding compliance risks in the practical application of the technology while ensuring the professionalism, practicality, and safety of aquatic disease diagnosis.

[0034] In some embodiments, this application employs a modular architecture design, encompassing six core modules: data processing, graphical interaction, staged filtering, multi-dimensional matching, large-scale artificial intelligence models, and a database. The overall architecture exhibits strong decoupling, iterative upgradeability, and support for continuous hot data updates. It enables fully automated management of aquatic disease case data, including cleaning, deduplication, anonymization, entity labeling, and classification for database entry. It unifies the structured standards for multi-source heterogeneous cases, effectively addressing the problems of fragmented, subjective, redundant, and privacy-risking traditional case data, and continuously building a standardized, high-quality intelligent aquatic disease case library. Simultaneously, this application adds an intelligent image matching and retrieval function, capable of acquiring various pathogen images uploaded by users, such as disease symptoms, anatomical images, and pathogen micrographs. By extracting core disease features from the images and performing similarity matching with the case library image data, accurate image matching results are obtained. These results are then merged with a differentiated and accurate matching case list obtained through text retrieval to form an image-text matching case list. With its lightweight operation mode featuring graphical click interaction and quick selection of breeding stages, it eliminates the need for users to input professional terms and precise instructions, completely simplifying the operation process and significantly lowering the usage threshold for middle-aged and elderly farmers and grassroots agricultural technicians. It is suitable for grassroots promotion and implementation scenarios and can be quickly and widely applied.

[0035] This application pioneers a dual-dimensional scenario-based screening mechanism combining "aquaculture model + aquaculture stage," integrating a six-dimensional weighted matching model with a text-image fusion retrieval mode. It aligns with the stage-specific and scenario-specific disease patterns of aquatic organisms, accurately quantifying case matching and completely resolving the shortcomings of traditional keyword searches, such as inaccurate matching and numerous invalid results. This significantly improves the relevance and practical reference value of disease cases. Simultaneously, relying on a fine-tuned artificial intelligence model in the aquaculture field, it breaks through the technical limitations of traditional case databases that only store data without intelligent analysis. It deeply mines the data value of the entire case database, completing multi-dimensional data analysis including disease pattern statistics, aquaculture risk prediction, and disease trend assessment. It automatically generates an integrated diagnosis and treatment assistance plan containing pathogen prediction, prevention and control measures, medication references, and precautions, achieving a technological upgrade from simple data storage and query to intelligent decision support. This effectively compensates for the industry's shortcomings, such as the shortage of grassroots aquatic disease experts, inconvenient remote diagnosis, and farmers' blind diagnosis and medication use. It can effectively reduce economic losses from aquaculture diseases and provide real and effective first-hand data support for industry disease prevention and control, technology promotion, and scientific research analysis, powerfully promoting the standardized and intelligent transformation and upgrading of smart fisheries.

[0036] In some embodiments, such as Figure 2 As shown, the pre-built aquatic disease knowledge graph serves as a domain knowledge foundation adapted to differentiated, colloquial, and regionalized diagnosis and treatment scenarios at the grassroots level. It supports standardized case management, entity alignment, multi-dimensional matching, and large-scale, multi-paradigm knowledge reasoning. The aquatic disease knowledge graph is built around aquatic regional terminology, disease characteristics, aquaculture stages, environmental factors, and pathogen diagnosis and treatment characteristics. It is constructed through six major processes: domain ontology modeling, multi-source knowledge aggregation, knowledge extraction and entity annotation, knowledge fusion and entity alignment, completion of multi-type knowledge reasoning relationships, and dynamic iteration of graph storage and online deployment. This unifies industry terminology, entity attributes, and semantic relationships, solving the problems of fragmented, inconsistent, weakly correlated, and difficult-to-uncover implicit features in traditional disease knowledge. It provides structured, correlated knowledge support for database construction, matching model computation, and artificial intelligence diagnosis and treatment reasoning. The specific construction process of the aquatic disease knowledge graph is as follows: The first step is to conduct domain ontology modeling, which involves building the core concepts, entity attributes, and entity relationships at the top level of the graph. First, the core concepts of the entire aquatic disease domain are identified, and based on the language habits of farmers in different regions, lexical units, entity categories, basic attributes, and entity interaction relationships are defined to build the overall ontology architecture. Eight primary entities are defined: aquaculture species, aquaculture region, aquaculture model, pathogen, clinical symptoms, environmental parameters, medication regimens, and aquaculture stage. Specific attributes are defined for each type of entity; for example, pathogen entities record parameters such as strain type, susceptible aquatic species, high-incidence stage, mortality rate, and transmission route, while symptom entities are labeled with information such as the affected area, morphology, associated pathogens, and inducing conditions. Finally, standardized semantic association rules between entities are defined, including fixed logic such as pathogen-induced symptoms, environmental parameters inducing disease, high-incidence specific pathogens in aquaculture stages, and matching diseases with appropriate medication regimens, forming a complete set of ontology specifications.

[0037] The second step involves aggregating multiple knowledge data sources and unifying four categories of authoritative data as map materials: industry standards and classics, front-line business data, literature and public information, and expert experience. Industry standards and classics cover national standards for aquatic diseases, aquaculture regulations, treatment manuals, and veterinary drug usage guidelines; front-line business data includes historical treatment cases, farm logs, animal health company archives, and grassroots agricultural technical service records; literature and public information include research papers, industry surveys, and official disease monitoring data; and expert experience is compiled by front-line experts in aquaculture, disease prevention and control, and disease research, including regional disease terminology, typical infection characteristics, regional disease patterns, and key prevention and control points, ensuring comprehensive and reliable knowledge coverage.

[0038] The third step involves knowledge extraction and entity annotation, employing differentiated extraction strategies for structured, semi-structured, and unstructured data sources. For structured data such as standard directories and ledgers, entities, attributes, and relationships are directly extracted through field mapping. For semi-structured data such as manual entries and web page directories, knowledge extraction is completed using rule templates and regular expressions. For unstructured text without a fixed format, such as medical records, aquaculture logs, and scientific research literature, a pre-trained entity recognition model in the aquaculture field is invoked, along with previously established ontology rules, to automatically identify all core entities, entity attributes, and semantic relationships within the text, completing batch entity annotation.

[0039] The fourth step involves knowledge fusion and entity alignment, unifying knowledge across the entire domain through three processes: ambiguity resolution, deduplication and error correction, and cross-source fusion. Ambiguity resolution integrates colloquial and written synonyms from various regions, mapping differentiated descriptions such as "stealing to death" and "drowning shrimp" to standard entity names, distinguishing similar symptoms and pathogens, and avoiding entity confusion. Deduplication and error correction remove duplicate knowledge entries and correct erroneous knowledge by comparing it with industry standards and expert experience. Cross-source fusion merges knowledge collected from different channels, supplementing missing entity attributes, improving the links between entities, and achieving the unification of multi-source knowledge.

[0040] The fifth step involves knowledge reasoning and relationship completion, relying on seven types of reasoning methods—semantic, fuzzy, rule-based, case-based, temporal logic, multimodal fusion, and multi-step chain-like—to uncover implicit knowledge and improve the graph association network. Semantic reasoning relies on entity mining to uncover deep semantic connections between symptoms, pathogens, and environmental triggers, extracting latent disease features from the text; fuzzy reasoning adapts to the colloquial and vague descriptions used by farmers, semantically normalizing non-standard symptom expressions and inferring suspected diseases based on fuzzy disease characteristics; rule-based reasoning derives hidden associations based on ontology-preset logic, supplementing the correspondence between pathogens and farming stages, and between the environment and diseases; case-based reasoning relies on existing treatment cases to mine disease patterns and the correlation between medication effects, extracting early signs of diseases; temporal logical reasoning analyzes water quality and disease cycle temporal data to predict the risk of staged disease outbreaks; multimodal fusion reasoning connects text and disease image feature vector data, combining symptom photos, anatomical microscopic images, and textual cases to jointly analyze pathogens, and outputs treatment references with a six-dimensional matching model; multi-step chain reasoning introduces a thinking chain to decompose complex aquaculture treatment problems, completing scene positioning, symptom breakdown, environmental analysis, case matching, and solution deduction step by step, improving the accuracy of answers to complex inquiries.

[0041] Finally, the graph is stored, launched, and dynamically iterated. A graph database is selected to store all entities, attributes, and relationships, supporting fast retrieval, relationship traversal, and path query. A graph version management mechanism is established to distinguish between basic and iterative versions and update content. After it is officially put into use, incremental updates are carried out based on newly added diagnosis and treatment cases, information on new diseases and pathogens, and expert revision opinions to continuously supplement and correct graph entities, attributes, and relationships, so as to achieve long-term dynamic iteration of domain knowledge.

[0042] In some embodiments, such as Figure 3 As shown, the intelligent case library for aquatic pathogens is the core structured storage carrier of this application. Using a pre-constructed aquatic disease knowledge graph as a standardized benchmark, it standardizes, transforms, and binds multi-dimensional tags to unstructured original case data of aquatic pathogens that are multi-sourced, heterogeneous, colloquial, and exhibit regional differences. This provides standardized underlying data support for screening at the aquaculture stage, six-dimensional intelligent matching, and large-scale multi-paradigm knowledge reasoning. The overall construction follows a seven-step process: defining domain data specifications, designing data table structures, defining data standardization and format specifications, initial data collection and standardization processing, developing and debugging the data entry link, optimizing index permissions and performance, and routine operation and maintenance with incremental updates. The specific construction process of the intelligent case library for aquatic pathogens is as follows: First, domain-specific data standards were defined. Based on the eight entity categories of the knowledge graph, a complete set of storage field specifications and entity mapping standards were developed, and regional language preference adaptations were completed. By unifying the storage formats of original cases, environmental parameters, image features, and other dimensions, the data inconsistencies caused by local breeding dialects and differentiated disease descriptions were specifically addressed. This ensured that the database storage logic, semantic rules, and knowledge graph ontology architecture were fully aligned, establishing a unified data entry benchmark that is compatible with regional differences.

[0043] Next, the data table structure was designed, splitting into multiple tables according to data type and setting primary and foreign key relationships. These tables include a main case table, an entity relationship table, a classification label table, aquaculture dimension relationship tables, an image data table, and a system configuration table. Each table contains basic case information, graph entity mapping, retrieval classification labels, the binding relationship between aquaculture mode and day intervals, disease images and feature vectors, system matching weights, and graph mapping configurations. This provides a data storage medium for image similarity matching, scenario-based dual retrieval, and multi-dimensional scoring matching.

[0044] The third step involves defining data standardization and format specifications, clarifying a complete set of data governance constraints, including data anonymization rules and labeling and coding rules. The anonymization rules implement a masking and replacement mechanism for private information such as farmers' contact information and farming addresses to ensure data compliance. The labeling and coding rules assign unique identifiers to farming methods, farming stages, pathogens, and symptoms. Simultaneously, standards for automatic deduplication of duplicate data and cleaning of redundant information are established, unifying the data entry, conversion, and storage formats across the entire case database.

[0045] The fourth step involves initial data collection and standardization. Multi-channel knowledge materials and historical medical records compiled from the knowledge graph are used as initial data sources. Raw data preprocessing is completed following a pipeline of text cleaning, semantic ambiguity resolution, privacy desensitization, entity annotation, and classification. Leveraging a large-scale domain entity recognition model, fully automated entity extraction and accurate semantic alignment are achieved. The knowledge graph's ambiguity resolution capabilities are reused to unify colloquial disease descriptions from various regions, completing the standardized mapping of eight major entities. Next, multi-dimensional tag binding is carried out, assigning each case multi-dimensional tags including breeding scenario, breeding stage, pathogen, symptoms, environment, and treatment effect. This transforms scattered unstructured text into standardized structured data adapted to the knowledge graph, laying a solid data foundation for multi-dimensional intelligent retrieval.

[0046] The fifth step involves developing and integrating the data import process, building an automated import interface that supports batch import and real-time incremental entry. A decoupled, partitioned storage architecture is adopted, relying on primary and foreign keys to link and correspond case, entity, tag, and image data. A built-in import verification mechanism automatically intercepts duplicate data, identifies tag mismatches, missing data, and invalid associations, issuing alerts to ensure the integrity, low redundancy, and high scalability of the imported data.

[0047] The sixth step involves implementing indexing, permission, and performance optimization. Composite indexes are created for high-frequency retrieval fields such as breeding mode, breeding days, and pathogens to adapt to graphical interactive joint retrieval scenarios, significantly improving filtering and matching efficiency. Hierarchical data access permissions are divided according to user roles such as administrators and farmers to isolate privacy-sensitive content. A hot and cold data partitioning storage scheme is adopted, with active recent cases and old historical cases stored separately to continuously optimize overall query response performance.

[0048] Finally, routine operation and maintenance and incremental updates are completed. After going online, newly collected disease images and treatment records will automatically go through the standardized management process and be incrementally added to the database. During the operation and maintenance phase, the integrity of data associations is regularly inspected and abnormal data is corrected. At the same time, entity associations and tag systems are updated in sync with the iteration rhythm of the aquatic disease knowledge graph, so as to realize the collaborative iteration of the case library and knowledge base and the long-term stable supply of standardized retrieval data.

[0049] In some embodiments, such as Figure 4 As shown, the constructed six-dimensional association matching model, which integrates aquaculture mode, aquaculture days, pathogen type, disease symptoms, environmental parameters, and drug efficacy, is developed based on the entity architecture, semantic association, and multi-paradigm reasoning system of the aquatic disease knowledge graph. It adapts to the industry characteristics of aquatic diseases, such as scenario dependence, multi-stage occurrence, and symptom ambiguity. Combined with a standardized case database, it achieves quantitative intelligent matching. It integrates semantic, fuzzy, temporal, multimodal, and multi-step chain reasoning logics to address the shortcomings of poor matching accuracy in traditional keyword retrieval. The overall construction process follows an eight-step procedure: defining six matching dimensions and matching rules, allocating dimension weights, designing single-dimensional score calculation rules, integrating multi-paradigm reasoning logic, constructing a comprehensive matching degree calculation formula, model training and testing with parameter tuning, model encapsulation and integration with output configuration, and dynamic model iteration. The specific construction process of the six-dimensional association matching model is as follows: The first step involves defining six matching dimensions and rules. Using knowledge graph entity attributes as the comparison benchmark, judgment criteria are established for six dimensions: farming mode, farming days, pathogen type, disease symptoms, environmental parameters, and medication effectiveness. Farming mode is matched precisely as an entity; farming days are scored based on temporal logic reasoning, considering interval overlap and proximity; pathogen type is categorized by knowledge graph entity attributes, further subdividing the pathogen matching level; disease symptoms are calculated for similarity after semantic and fuzzy reasoning to achieve verbal normalization; environmental parameters are compared based on disease-inducing rules, comparing interval overlap; and medication effectiveness is comprehensively evaluated by combining knowledge graph disease medication associations with case efficacy labels, unifying the matching logic standards across all dimensions.

[0050] The second step is to allocate dimensional weights, following the principle of prioritizing the weight of the breeding model and decreasing it in turn to the fixed priority of the drug effect. Combined with industry experience, a weight coefficient with a total of 1 is allocated. The parameters are stored in the system configuration table to support visual adjustment in the background. At the same time, different breeding categories can be subdivided into subdivided dimensions and sub-weights can be allocated to adapt to diverse breeding application scenarios.

[0051] The third step is to design the single-dimensional score calculation rules, unify the scoring range of all dimensions to 0-100 points, take full marks for a perfect match, convert the score for partial and adjacent matches according to the similarity ratio, and score 0 points for no corresponding features. Differentiated conversion logic is set for scenarios such as adjacent intervals of breeding days and some overlapping symptoms, and a unified single-dimensional quantitative evaluation standard is established.

[0052] The fourth step involves building a multi-paradigm reasoning fusion matching logic, linking seven types of reasoning mechanisms to optimize matching results: semantic reasoning to unify the differentiated disease descriptions from different regions, fuzzy reasoning to be compatible with incomplete and fuzzy grassroots input data, temporal reasoning to dynamically correct stage weights based on the disease incidence patterns in aquaculture, case reasoning to mine latent disease correlations in historical cases, multimodal fusion reasoning to verify text matching results by linking disease image features, and multi-step chain reasoning to decompose complex cases with mixed infections and multiple causes; ultimately forming a dual fusion matching mechanism of "six-dimensional text matching + image feature-assisted verification" to improve the accuracy of matching in complex disease scenarios.

[0053] The fifth step is to construct a comprehensive matching degree calculation formula, unify the score range of 0-100 for each dimension, and adopt a weighted summation fusion algorithm to multiply the single-dimensional score by the corresponding weight and then sum them to output a standardized comprehensive matching score of 0-100, thereby realizing quantitative scoring, horizontal comparison and orderly screening of candidate cases.

[0054] The sixth step involves model training, testing, and parameter tuning. A large number of labeled verification cases covering routine, mixed infections, regional special diseases, and newly occurring cases are used as the test set for batch computation. The accuracy of the model scores is verified by comparing them with the results of expert manual evaluation. For boundary scenarios such as mixed infections and cross-stage disease, the weights and scoring rules are fine-tuned, and the invalid information filtering logic is optimized simultaneously to eliminate matching bias.

[0055] The seventh step involves model encapsulation, integration, and output rule configuration. The six-dimensional matching algorithm is encapsulated as a separate functional module, and preliminary candidate cases output by the module are screened during the integration phase. Case sorting rules are configured in descending order of matching degree, and additional information such as matching score and case reliability are output synchronously, while image similarity score is compatible with comprehensive calculation. Interface integration with knowledge graph, case database, and large artificial intelligence model module is completed, and a visual backend configuration page is provided. Parameter fine-tuning can be completed without modifying the core algorithm, adapting to different regional and seasonal disease scenarios.

[0056] Finally, a dynamic iteration mechanism for the model is established, implementing an update plan of monthly parameter fine-tuning, quarterly overall version upgrades, and emergency optimization for sudden diseases. The matching logic is continuously corrected based on the continuously added front-line diagnosis and treatment cases and the iterated knowledge graph, thereby optimizing the model matching accuracy in the long term.

[0057] In some embodiments, such as Figure 5 As shown, the full-case database is a comprehensive data foundation upgraded and aggregated from the intelligent case library of aquatic pathogens and diseases. It is compatible with knowledge graph entity systems and multi-paradigm reasoning logic, providing complete data support for large-scale artificial intelligence models in comprehensive statistics, risk prediction, and multi-step chain reasoning. The overall construction follows an eight-step process: defining the scope of the full-domain data, designing a layered data architecture, formulating data aggregation rules, establishing a data synchronization mechanism, adapting data formats, optimizing indexes and accelerating queries, managing the data lifecycle, and ensuring coordinated online operation. The specific construction process of the full-case database is as follows: The first step is to define the scope of the full-domain data and collect three types of core data from the full case database: basic full-domain data covering all cases, entities, and tag information; statistical dimension-derived data based on the basic data to generate incidence rates and regional incidence characteristics; and AI-interactive supplementary data including model call logs and treatment plan records, fully covering all materials required for model calculation.

[0058] The second step is to complete the data layered architecture design, dividing the data into three decoupled storage layers. The raw data layer connects to the aquatic disease case library, retaining unprocessed original details to ensure data authenticity; the data aggregation layer pre-aggregates data according to dimensions such as region, breeding stage, and pathogen, reducing the real-time computing pressure on large models; the model service layer provides standardized data interfaces, completing format conversion to adapt to the input standards of large models.

[0059] The third step is to formulate data aggregation rules, benchmark against knowledge graphs and six-dimensional matching model dimensions, and set up multi-dimensional automatic aggregation logic. This can automatically collect statistics around dimensions such as breeding mode, time of year, pathogens, and drug efficacy, solidify the generation rules of core indicators such as incidence rate and disease epidemic risk, build a standardized time-series dataset, adapt to time-series logical reasoning, and support the prediction of disease trends.

[0060] The fourth step is to establish a data synchronization mechanism, so that new cases are synchronized to the original data layer in real time. The aggregation layer data is automatically refreshed through hourly and daily scheduled tasks, and periodic data consistency checks are performed to eliminate data duplication, missing data, and inconsistency issues, ensuring reliable multi-layer data synchronization.

[0061] The fifth step involves adapting the data format for large-scale artificial intelligence models. This includes semantic simplification and reorganization, completion of reasoning dimensions, and standardization of data across the entire domain. The output data undergoes desensitization and secondary verification, and the massive data is output in segments and batches. Based on the complete standardized data, the large-scale models can carry out full-dimensional intelligent reasoning, such as causal analysis and complex case deduction.

[0062] The sixth step involves index optimization and query acceleration. The disease patterns and treatment experience output by the large model are reverse-engineered into the database to expand the data dimensions. A joint aggregate index is built for high-frequency statistical fields such as breeding mode, pathogen, and time series. Real-time details and offline statistical data are stored in partitions, which greatly improves the efficiency of full database traversal and group statistics.

[0063] The seventh step is to implement data lifecycle management by adopting a hot and cold partitioning storage mechanism. Recently accessed high-frequency data is classified as hot data for high-speed reading and writing, while historical cases from the past are archived as cold data. Redundancy cleanup and data quality re-inspection are carried out regularly to balance data traceability and system inference response efficiency.

[0064] Finally, the database was put into operation and linked with other systems. The database was directly connected to the domain-specific fine-tuning aquatic disease model. The model reads data from the entire domain in layers to complete multi-dimensional statistics and risk assessment, and generates treatment assistance plans in combination with search cases. The model output content and interaction logs are stored in the database in reverse, forming a complete closed loop of data collection, processing and accumulation.

[0065] In some embodiments, such as Figure 6As shown, this method establishes a one-to-one binding relationship between aquaculture model summary graphics and various aquaculture models. The overall process follows a six-step workflow: extraction of aquaculture model feature elements, compilation of aquaculture model summary graphics, establishment of unique association fields, solidification and verification of mapping relationships, full module synchronization, and iterative updates to achieve a globally unified mapping. First, aquaculture model feature elements are extracted. For each region-aquaculture model combination, core features such as aquaculture cycle, stage division, and regional climate are extracted as the basis for graphics production, ensuring that each type of aquaculture model corresponds to a unique visualization graphic. Then, based on the aquaculture model entities within the aquatic disease knowledge graph, and combined with regional professional terminology and local dialects, unique summary graphics for each model are generated, initially establishing a one-to-one correspondence between graphic identifiers and entities and business data. Next, a unique graphic identifier association field is added to the knowledge graph, case database, full case database, and model configuration table, binding the graphic ID, storage path, hash value, and aquaculture model code to form a four-level association link: aquaculture code, entity information, graphic identifier, and graphic element. Next, a global mapping table is built, and each record is verified to avoid one-to-many and many-to-one mismatch anomalies. Duplicate and cross-matching checks are performed. Once the verification is successful, the mapping table is locked and solidified, restricting arbitrary modification of binding relationships. After the mapping is solidified, a full-module synchronization is performed, synchronizing the mapping rules to all business modules, including the knowledge graph, the graphical interactive retrieval frontend, and the six-dimensional matching model. This enables bidirectional linkage, where retrieving aquaculture modes automatically matches the corresponding graphics, and clicking on graphics reverses the location of aquaculture entities and related cases. During operation, continuous iterative updates are supported. Adding new aquaculture modes will fully reproduce the entire process of feature extraction, mapping, and binding verification. When the features of existing aquaculture modes change, only the graphic content is updated, while the original mapping primary key is retained, ensuring the continued validity of historical cases and entity data relationships.

[0066] In some embodiments, such as Figure 7As shown, the differentiated and precise matching case list is generated based on the six-dimensional association matching model, the semantic rules of the aquatic disease knowledge graph, and the full database data. It can parse non-standardized consultation information and complete precise retrieval. The overall process sequentially executes the following steps: structured parsing of user requests, calling matching dimension weights, traversing and scoring the full database data, executing differentiated filtering rules, sorting and prioritizing cases, and assembling list fields and standardizing the format. Specifically, after receiving user consultation information, the system uses a knowledge graph to extract entities and normalize semantics, parsing out key information across six search dimensions and generating standardized query vectors to complete the structured parsing of user requests. Then, it calls the weights and scoring parameters of a six-dimensional association matching model, following a progressively decreasing weight priority from aquaculture mode to medication effectiveness, and fine-tunes local weights based on seasonal and regional characteristics to determine a unified scoring standard. Next, it traverses the full case database and the intelligent case library of aquatic pathogens, performing six-dimensional quantitative scoring on each case, integrating multi-paradigm reasoning results to correct scores, and calculating the comprehensive matching degree of each case. Then, it executes multi-layered differentiated screening rules, eliminating cases with severely mismatched scenarios and low scores, and employing dedicated screening strategies for mixed infections and regionally specific diseases. Afterward, it initially sorts cases in descending order of comprehensive matching scores, then adjusts the display priority based on case timeliness, regional matching degree, and treatment effectiveness. Finally, it extracts core case fields, assembles basic case information, matching scores, treatment references, risk warnings, etc., supplements matching criteria and differentiated feature descriptions, unifies the output format, and ultimately generates and pushes a differentiated and accurately matched case list.

[0067] In some embodiments, such as Figure 8As shown, the AI-powered large-scale model relies on aquatic disease knowledge graphs, multi-dimensional tags, temporal architecture, and six-dimensional matching dimensions to perform fully automated closed-loop statistical reasoning on a comprehensive case database. The process involves, in sequence, receiving statistical tasks and loading rules, data partitioning and reading, refined cleaning and standardization correction of statistical samples, multi-dimensional cross-statistical calculation of indicators, time-series trend statistical analysis, abnormal data and risk quantification, encapsulation of pattern summarization results, and standardized output of statistical reports. After receiving various statistical instructions such as global distribution, stage risk, and medication effects, the model identifies the task type, automatically loads knowledge graph constraints, six-dimensional matching fields, and temporal granularity configurations, verifies screening boundaries, and initializes filtering and elimination rules. Subsequently, it retrieves hot and cold partition data from the entire database through a dedicated interface, generating a dataset of data to be collected by region, pathogen, and cycle. The model performs refined processing on the samples, eliminating invalid data with missing key fields, using the knowledge graph to achieve semantic normalization of colloquial content, verifying and eliminating similar redundant cases, and supplementing missing dimensions to unify statistical standards. Based on a six-dimensional matching dimension, cross-scenario cross-indicator calculations are performed to uncover the coupling correlation patterns between breeding models, breeding stages, diseases, and medication. Using time-series datasets, disease outbreak cycles and year-on-year / month-on-month growth rates are calculated, time-series curves are fitted, and short-term disease trends are predicted. Simultaneously, abnormal samples such as new diseases and regional concentrated outbreaks are identified by referring to spectral disease incidence thresholds, quantifying risk indicators such as risk levels and spread probabilities. All basic, cross-sectional, time-series, and risk data are aggregated to form a standardized indicator set. Combined with domain knowledge, explicit disease patterns and implicit triggering factors are summarized. Numerical values, rankings, and risk warnings are uniformly encapsulated into reusable structured data. Finally, data, charts, and risk interpretations are integrated according to a preset template to generate standardized statistical reports, which are then archived.

[0068] In some embodiments, such as Figure 9 As shown, the AI ​​large-scale model inference calculation is based on the ontology rules of the aquatic disease knowledge graph, the full database of case data, six-dimensional matching scores, multimodal features, and seven types of inference paradigms to carry out standardized hierarchical calculations. The entire process is traceable and reproducible. The inference results undergo multiple rounds of comparison, rule constraints, and score correction. The entire process is as follows: multimodal input data preprocessing, inference paradigm matching and strategy determination, step-by-step execution of inference calculation, inference result fusion and weight correction, conclusion verification and compliance determination, structured result assembly and storage, and inference result output. Specifically: The first step is to receive and preprocess multimodal input data, and to uniformly access multiple input sources such as user text, breeding parameters, disease images, and historical breeding records. The text is standardized by dialect and colloquialism, and the numerical units and intervals are unified. Feature vectors are extracted and compressed from the images, and the four types of information, namely text, numerical values, images and scenes, are integrated to generate a standardized input vector set.

[0069] The second step involves reasoning paradigm matching and strategy determination. Based on the knowledge graph, entity element extraction, verbal expression entity alignment, and missing feature completion are completed. The professional boundaries of reasoning are constrained by the graph attributes. Reasoning combinations are automatically matched according to the completeness of input information, the typicality of symptoms, whether disease images are uploaded, and the complexity of the case. Simple and typical cases use semantic, rule, and case reasoning. Fuzzy and incomplete descriptions are superimposed with fuzzy reasoning. Uploaded disease images are used to add multimodal fusion reasoning. Mixed infections and difficult cases with multiple causes are given priority for multi-step chain-based CoT reasoning decomposition and analysis.

[0070] The third step involves executing the entire set of reasoning calculations step by step using a paradigm, sequentially running seven types of reasoning logic: Semantic reasoning calculates the correlation scores between pathogens, symptoms, and environment based on graph triples; Fuzzy reasoning establishes membership functions for non-standard colloquial descriptions and calculates feature similarity with fault tolerance; Rule-based reasoning calls on the graph to solidify industry rules and screen out logically conflicting disease conclusions; Case-based reasoning retrieves highly similar historical cases from the entire database and derives diagnostic references based on similar cases; Temporal logical reasoning combines temporal data to quantify the risk of staged disease outbreaks; Multimodal fusion reasoning fuses image vectors to correct text matching deviations; Multi-step chain reasoning decomposes complex cases layer by layer and eliminates logical ambiguities step by step.

[0071] The fourth step involves fusing and correcting the inference results. The candidate disease set output by the seven types of inference is input into the six-dimensional association matching model to complete the full-dimensional quantitative scoring. The objective matching score is used to correct the bias caused by the text inference. The conflicting conclusions generated by different inference paradigms are arbitrated by weight, and the optimal pathogen determination result is output.

[0072] The fifth step involves verifying the conclusions and determining compliance. This includes retrieving industry-specific rules and regulations such as veterinary drug use guidelines, drug incompatibilities, and fishery disease control standards. The system automatically verifies the pathogen identification, medication regimens, and control measures generated through reasoning, eliminating non-compliant conclusions such as illegal drug use and inappropriate scenarios, and then making corrections.

[0073] Finally, the standardized and structured results are assembled by integrating various inference scores, risk assessments, causal analyses, and complete treatment plans. All intermediate inference data, paradigm scores, case and rule call logs are simultaneously retained to form a traceable log for model iteration and review, and finally, complete inference calculation results are output.

[0074] The following is a detailed description and explanation of the solutions in the embodiments of the present invention, using specific application examples: Figure 10 , Figure 11 , Figure 12 and Figure 13 As shown, based on the aquatic disease knowledge graph, a two-level case database (based on the intelligent case library of aquatic pathogens and diseases and the full case database), a six-dimensional association matching model, and multimodal image recognition capabilities, combined with a graphical user interface and a multi-paradigm artificial intelligence reasoning engine, a complete diagnosis and retrieval process is completed. Figure 10 It centrally showcases five core functions: precise case retrieval, case entry, data statistics, intelligent diagnosis and treatment, and pattern summarization. Figure 11 It provides multiple lightweight interactive entry points, including graphical filtering of aquaculture species and aquaculture models, segmented filtering of aquaculture stages across timelines, and keyword search. Figure 12 Supports variety selection, colloquial symptom input, and uploading of real-life disease photos; it completes one-click AI diagnosis and outputs disease assessment and tiered treatment plans; after the search, it outputs a standardized table with a list of differentiated and precisely matched cases, such as... Figure 13 As shown, the data displays basic case information, six-dimensional sub-scores, comprehensive matching score, treatment plan, risk warnings, and differentiated annotations.

[0075] After users input colloquial diagnostic text containing information such as region, farming mode, farming duration, abnormal feeding and mortality, and water quality indicators, and upload two photos of shrimp, the system first performs multimodal input data preprocessing. On the one hand, the colloquial text is semantically normalized by calling a knowledge graph ambiguity database, unifying common descriptions such as "white shrimp," "high-level pond," "shrimp loss," and "empty intestines and stomachs" into standardized entities such as Litopenaeus vannamei, high-level pond cultivation in South China, continuous mass mortality, and digestive system abnormalities. Information from six dimensions—farming region, mode, species, farming stage, core symptoms, and environmental parameters—is extracted to generate a text query vector. On the other hand, the uploaded disease images are denoised and feature extracted to generate image feature vectors for subsequent multimodal fusion inference. Simultaneously, the unique mapping relationship between the aquaculture mode and the summarized graph is retrieved, and the corresponding graphic identifier of the current high-level white shrimp in South China is locked. The preset weights and scoring rules of the six-dimensional association matching model are loaded at the same time. The weights of aquaculture mode, aquaculture days, pathogen type, symptom similarity, environmental parameters, and drug effect are 0.32, 0.22, 0.18, 0.14, 0.10, and 0.04, respectively. The single-dimensional score range is uniformly 0-100 points, and a comprehensive matching score of 30 points is set as the threshold for invalid cases to be removed.

[0076] After data preprocessing and model configuration loading, a full database traversal and scoring were performed on the complete case database and the aquatic pathogen and disease case database, with each case being quantitatively scored against the six dimensions. This search identified three typical candidate cases. The first case involved a 58-day case of Vibrio spp. combined with hepatopancreatic necrosis in Litopenaeus vannamei in a high-level pond in South China. All dimensions showed a high degree of match, with full marks for pattern, pathogen, and symptoms, resulting in a comprehensive matching score of 98.66. The second case involved a case of Enterocytozoon hepatocellular carcinoma infection in a high-level pond during the same period. While the farming scenario and environmental parameters showed a high degree of match, the overlap of pathogen and core symptoms was low, resulting in a comprehensive matching score of 73.02. The third case involved an early-stage bacterial gill rot in an outdoor pond. The farming model, disease stage, and pathogen were completely inconsistent, resulting in a comprehensive matching score of only 8.04. The differential screening rules are implemented to directly remove low-match cases with scores below the threshold. The remaining two cases are sorted in descending order of comprehensive score, and then the priority is finely adjusted according to the rules of the same region and the same seedling season to form the basic case ranking results. A standardized differential and accurate matching case list is output. The table fully records the basic information of each case, the six-dimensional individual scores, the comprehensive matching score, the supporting treatment plan, the risk warning and differential annotation.

[0077] Subsequently, joint verification and correction were carried out based on seven reasoning paradigms of the knowledge graph. Semantic reasoning matched the high correlation between symptoms and Vibrio infection features. Fuzzy reasoning distinguished the differences in mortality manifestations between enterocolitis hepatis and acute hepatopancreatic necrosis. Rule reasoning matched the high incidence of Vibrio in the high temperature and low dissolved oxygen of autumn in South China. Temporal logic reasoning determined the high risk of disease outbreak in November. Case reasoning statistically analyzed the distribution of historically diagnosed diseases in the same scene. Multimodal fusion reasoning compared the intestinal features of real-life images with Vibrio infection samples. Multi-step chain reasoning decomposed the cross-information of scene, symptoms, and environment layer by layer. Finally, the case sorting and matching score logic were verified to be correct, and the disease in this field was identified as acute hepatopancreatic necrosis syndrome induced by Vibrio parahaemolyticus. High-scoring cases were used as core references, and medium-scoring cases were used as exclusion controls.

[0078] Finally, the domain-fine-tuned AI model, combined with highly matched cases, knowledge graph industry rules, and veterinary drug compliance constraints, completes the inference calculation, assembles the complete structured diagnosis results in layers, and outputs a disease judgment conclusion with 95% confidence on the intelligent diagnosis interface. It also breaks down the diagnosis into six modules: emergency treatment, water quality improvement, Vibrio disinfection, internal conditioning, long-term management, and risk warning to generate personalized treatment plans. At the same time, it marks key risk warnings such as continuous water quality exceeding the standard and prohibition of irritating drugs. The entire process of consultation, retrieval, matching, inference, and plan output is completed in a closed loop. All retrieval logs, matching scores, and inference intermediate data are synchronously stored in the full case database to support the system's subsequent model iteration and data statistical analysis.

[0079] Please see Figure 14 This application also provides a retrieval system based on an intelligent case database of aquatic pathogens and diseases, which can implement the above-mentioned method. The system includes: The case data processing and automatic classification module is used to preprocess the acquired unstructured aquatic pathogen original case data; based on the aquatic disease knowledge graph, the preprocessed unstructured aquatic pathogen original case data is sequentially identified, labeled, automatically classified, and bidirectionally associated to obtain structured aquatic pathogen original case data. The aquaculture mode graphical interaction module is used to identify and lock the user's interface click operations through the aquaculture mode summary graphics, the aquaculture mode unique mapping mechanism and the interface interaction response mechanism to obtain the target aquaculture mode. The breeding days stage screening module is used to perform interval matching and dual joint retrieval processing on the target breeding mode and breeding days according to the breeding days segmentation system and breeding days selection control to obtain a preliminary candidate case set. The multi-dimensional differentiated matching module is used to perform multi-dimensional weight calculations on the preliminary candidate case set based on the six-dimensional association matching model to obtain a list of differentiated and accurate matching cases. The large-scale model statistical and treatment assistance module is used to perform multi-dimensional statistical analysis on the original case data of structured aquatic pathogens and diseases by combining the target aquaculture model, the preliminary candidate case set, and the differentiated and precise matching case list with the artificial intelligence large-scale model, so as to obtain statistical analysis results and treatment assistance plans.

[0080] It is understood that the content of the above method embodiments is applicable to this system embodiment. The specific functions implemented in this system embodiment are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those achieved in the above method embodiments.

[0081] This application also provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the above-described method. This electronic device can be any smart terminal, including tablet computers, in-vehicle computers, etc.

[0082] It is understood that the content of the above method embodiments is applicable to this device embodiment. The specific functions implemented by this device embodiment are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0083] Please see Figure 15 , Figure 15 The hardware structure of an electronic device according to another embodiment is illustrated. The electronic device includes: The processor 151 can be implemented using a general-purpose CPU (Central Processing Unit), microprocessor, application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of this application. The memory 152 can be implemented as a read-only memory (ROM), static storage device, dynamic storage device, or random access memory (RAM). The memory 152 can store the operating system and other applications. When the technical solutions provided in the embodiments of this specification are implemented through software or firmware, the relevant program code is stored in the memory 152 and is called and executed by the processor 151 using the methods described above in the embodiments of this application. Input / output interface 153 is used to implement information input and output; The communication interface 154 is used to enable communication and interaction between this device and other devices. Communication can be achieved through wired means (such as USB, network cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.). Bus 155 transmits information between various components of the device (e.g., processor 151, memory 152, input / output interface 153, and communication interface 154); The processor 151, memory 152, input / output interface 153 and communication interface 154 are connected to each other within the device via bus 155.

[0084] The embodiments described in this application are for the purpose of more clearly illustrating the technical solutions of the embodiments of this application, and do not constitute a limitation on the technical solutions provided by the embodiments of this application. As those skilled in the art will know, with the evolution of technology and the emergence of new application scenarios, the technical solutions provided by the embodiments of this application are also applicable to similar technical problems.

[0085] Those skilled in the art will understand that the technical solutions shown in the figures do not constitute a limitation on the embodiments of this application, and may include more or fewer steps than shown, or combine certain steps, or different steps.

[0086] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0087] Those skilled in the art will understand that all or some of the steps in the methods disclosed above, as well as the functional modules / units in the systems and devices, can be implemented as software, firmware, hardware, or suitable combinations thereof.

[0088] It should be understood that in this application, "at least one (item)" means one or more, and "more than" means two or more. "And / or" is used to describe the relationship between related objects, indicating that three relationships can exist. For example, "A and / or B" can represent three cases: only A exists, only B exists, and both A and B exist simultaneously, where A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one (item) of the following" or similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one (item) of a, b, or c can represent: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, and c can be single or multiple.

[0089] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of the units described above is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0090] The units described above as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0091] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0092] The preferred embodiments of the present application have been described above with reference to the accompanying drawings, but this does not limit the scope of the claims of the present application. Any modifications, equivalent substitutions, and improvements made by those skilled in the art without departing from the scope and substance of the embodiments of the present application shall be within the scope of the claims of the present application.

Claims

1. A retrieval method based on an intelligent case database of aquatic pathogens and diseases, characterized in that, The method includes the following steps: Preprocess the acquired unstructured raw case data of aquatic pathogens and diseases; Based on the aquatic disease knowledge graph, the pre-processed unstructured aquatic pathogen original case data are sequentially identified, labeled, automatically classified, and bidirectionally associated to obtain structured aquatic pathogen original case data. The target aquaculture model is obtained by identifying and locking user interface click operations through a summary graph of aquaculture model, a unique mapping mechanism of aquaculture model, and an interface interaction response mechanism. Based on the segmentation system of breeding days and the selection control of breeding days, the target breeding mode and breeding days are subjected to interval matching and dual joint retrieval processing to obtain a preliminary candidate case set; The preliminary candidate case set is weighted in multiple dimensions according to the six-dimensional association matching model to obtain a list of differentiated and precise matching cases. By combining the target aquaculture model, the preliminary candidate case set, and the differentiated precise matching case list with an artificial intelligence big data model, multi-dimensional statistical analysis is performed on the original case data of structured aquatic pathogens and diseases to obtain statistical analysis results and diagnostic and treatment assistance plans.

2. The method according to claim 1, characterized in that, After the preprocessed unstructured aquatic pathogen original case data is identified and labeled according to the aquatic disease knowledge graph, the labeled unstructured aquatic pathogen original case data is automatically classified according to the preset classification dimensions. The preset classification dimensions include pathogen type classification, disease symptom classification, aquaculture species classification, aquaculture mode classification, aquaculture stage classification, and treatment effectiveness classification.

3. The method according to claim 1, characterized in that, The summarized diagrams of the breeding model include a breeding process diagram, a model structure diagram, or a stage summary diagram. The interface interaction response mechanism includes displaying a summary graphic of the breeding model in a visual form on the interactive interface.

4. The method according to claim 1, characterized in that, The number of days for raising animals is pre-divided according to the raising mode of the animals, and the raising mode includes the entire cycle of raising animals from stocking to pond clearing. The segmented breeding days system is retrieved based on the target breeding model.

5. The method according to claim 1, characterized in that, The matching dimensions of the six-dimensional association matching model, in descending order of weight, include breeding mode, breeding days, pathogen type, symptom similarity, environmental parameters, and drug efficacy.

6. The method according to claim 1, characterized in that, The list of differentiated and precise matching cases includes multiple matching cases; The matched cases include matching score, case source, case time, and case reliability level.

7. The method according to claim 1, characterized in that, The multidimensional statistical analysis includes statistics on common diseases under different farming models, prediction of disease risks at different farming stages, and analysis of mortality and prevalence of typical pathogens.

8. The method according to claim 1, characterized in that, It also includes the following steps: Obtain pathogen and disease images uploaded by users; the pathogen and disease images include images of disease symptoms, anatomical images, or microscopic images of pathogens; Image feature extraction is performed on the pathogen and lesion image to obtain pathogen and lesion features; The pathogen and disease characteristics are matched with the image data in the intelligent case library of aquatic pathogens and diseases to obtain the image matching results; The image matching results and the differential precise matching case list are fused and output to obtain the image-text matching case list.

9. A retrieval system based on an intelligent case database of aquatic pathogens and diseases, characterized in that, include: The case data processing and automatic classification module is used to preprocess the acquired unstructured raw case data of aquatic pathogens and diseases; Based on the aquatic disease knowledge graph, the pre-processed unstructured aquatic pathogen original case data are sequentially identified, labeled, automatically classified, and bidirectionally associated to obtain structured aquatic pathogen original case data. The aquaculture mode graphical interaction module is used to identify and lock the user's interface click operations through the aquaculture mode summary graphics, the aquaculture mode unique mapping mechanism and the interface interaction response mechanism to obtain the target aquaculture mode. The breeding days stage screening module is used to perform interval matching and dual joint retrieval processing on the target breeding mode and breeding days according to the breeding days segmentation system and breeding days selection control to obtain a preliminary candidate case set. The multi-dimensional differentiated matching module is used to perform multi-dimensional weight calculation on the preliminary candidate case set according to the six-dimensional association matching model to obtain a list of differentiated and accurate matching cases. The large-scale model statistical and treatment assistance module is used to perform multi-dimensional statistical analysis on the original case data of structured aquatic pathogens and diseases by combining the target aquaculture model, the preliminary candidate case set, and the differentiated precise matching case list with an artificial intelligence large-scale model, so as to obtain statistical analysis results and treatment assistance plans.

10. An electronic device, characterized in that, The electronic device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the method according to any one of claims 1 to 8.