Image recognition and feature database matching-based auxiliary diagnostic system and method for pet diseases

Abstract

This invention discloses a pet disease auxiliary diagnosis system and method based on image recognition and feature database matching. The system includes a structured pet disease-symptom feature knowledge base, a multi-model collaborative symptom feature extraction engine, and a disease matching and reasoning module. This invention deeply integrates the multi-model collaborative feature extraction engine with the structured symptom feature knowledge base and employs a disease matching algorithm based on semantic similarity. This enables the system to comprehensively analyze symptom combinations from multiple parts of the pet's body, significantly improving the clinical reference value of auxiliary diagnosis. By transparently presenting the matching results and specific supporting symptom evidence, users can understand the diagnostic basis, greatly enhancing the credibility of the diagnostic results and users' willingness to accept them. Through the establishment of a closed-loop feedback mechanism for confirmed diagnosis, the system can automatically adjust medical weights and expand the high-quality training set, thereby achieving continuous iterative evolution of the recognition model and knowledge base.

Description

Technical Field

[0001] This invention relates to the interdisciplinary field of computer vision and veterinary medicine, and more specifically, to a pet disease auxiliary diagnosis system and method based on image recognition and feature database matching. Background Technology

[0002] With the development of artificial intelligence technology, some image recognition tools for pet health have emerged on the market. These tools are mostly focused on single functions. For example, some applications can identify whether a pet has a skin disease by taking a picture of its skin and using a pre-trained convolutional neural network (CNN) model, or compare its similarity with images in a database. For instance, CN120766935A discloses a pet disease medical intelligent diagnostic system. However, it lacks the ability to finely define and quantify symptom characteristics such as hair loss patterns, lesion boundaries, and secretion properties. Other initial screening systems, such as CN117672501A, disclose an AI-based pet health initial screening system. While it can analyze multi-dimensional information such as body shape, feces, and mood, its analysis results are usually isolated scores or classifications. It does not deeply correlate these discrete signs with a structured veterinary disease knowledge system, and therefore cannot form inferential conclusions from "a series of observed features" to "one or more possible diseases."

[0003] With the increasing demand for pet ownership, traditional pet health testing and management methods have many limitations and are unable to meet the needs of modern pet owners for pet health management. Specifically: 1. Pet owners often find it difficult to accurately observe and describe complex clinical symptoms (such as specific forms of hair loss, characteristics of secretions, and boundaries of skin lesions), leading to inaccurate initial judgments and inefficient online searches or consultations. 2. Existing image recognition applications mostly target single targets or isolated symptoms, lacking the ability to comprehensively analyze combinations of symptoms appearing simultaneously in multiple body parts (eyes, ears, skin, body posture, etc.) of a pet; 3. Existing technologies do not integrate a structured veterinary medicine knowledge base, making it impossible to probabilistically and interpretably correlate identified symptom features with possible diseases. Diagnostic results lack transparency and user trust is low. 4. Most existing systems are static models and lack a closed-loop mechanism to continuously optimize feature weights and recognition models using real-world diagnostic feedback, resulting in the inability of system performance to improve over time.

[0004] There is currently no effective solution to the above problems. Summary of the Invention

[0005] To address the aforementioned technical problems in related technologies, this invention proposes a pet disease auxiliary diagnosis system and method based on image recognition and feature database matching, which can overcome the above-mentioned shortcomings of the prior art.

[0006] To achieve the above-mentioned technical objectives, the technical solution of the present invention is implemented as follows: On the one hand, a pet disease auxiliary diagnosis system based on multimodal image recognition and feature library matching is provided, including: A structured pet disease-symptom feature knowledge base is used to store various diseases for different pet species and their corresponding standard symptom feature vectors. Each standard symptom feature vector consists of several quantifiable symptom feature metadata and their medical weight coefficients when diagnosing the corresponding disease. The symptom feature metadata establishes semantic relationships through a symptom ontology tree. A multi-model collaborative symptom feature extraction engine is used to receive multimodal image data of pets, and through several specially trained deep learning sub-models, automatically detect and quantify various symptom features predefined in the knowledge base from the multimodal image data, and output a confidence score for each identified symptom feature to generate a symptom feature vector to be matched. The disease matching and reasoning module is used to calculate the matching degree based on the symptom feature vector to be matched and the standard symptom feature vector of the corresponding pet type in the knowledge base, using a weighted semantic similarity algorithm based on information content, taking into account the confidence score, the medical weight coefficient and the semantic distance of the symptom features in the symptom ontology tree, and output a list of disease possibilities sorted by matching degree.

[0007] Furthermore, the multi-model collaborative symptom feature extraction engine includes a task scheduler, which distributes the multimodal image data to one or more corresponding deep learning sub-model queues according to the image type to achieve parallel inference.

[0008] Furthermore, in the weighted semantic similarity algorithm based on information content, the value of the information content is determined based on the specificity of the symptom features, specifically as follows: in, This represents the total number of diseases for the current pet species in the knowledge base. To include symptom characteristics The number of diseases.

[0009] Furthermore, when calculating the matching degree, the disease matching and inference module specifically employs an asymmetric average optimal matching algorithm, which includes: Calculate the semantic similarity between each symptom feature in the query vector and each symptom feature in the standard disease vector, and take the best match and perform a weighted sum to obtain the first match score; Calculate the semantic similarity between each symptom feature in the disease standard vector and each symptom feature in the query vector, and take the best match and perform a weighted sum to obtain the second matching score; The arithmetic mean of the first matching score and the second matching score is taken as the final matching degree; The semantic similarity is calculated based on the information content value of the deepest common ancestor node of the two symptom features in the symptom ontology tree, specifically as follows: in, To query features, For diseases in the knowledge base The standard characteristics.

[0010] Furthermore, it also includes: An interactive mobile terminal application is used to provide a standardized image acquisition guidance interface, manage pet files, upload the multimodal image data, and display the list of possible diseases, supporting evidence and auxiliary suggestions in the form of a structured report, while also providing a diagnostic feedback entry point; The feedback closed-loop optimization module is used to receive the diagnosis information from the user, automatically adjust the medical weights of the relevant symptom features in the knowledge base according to the diagnosis information, and add the diagnosis image and the corresponding symptom feature vector to the training set to update the deep learning sub-model.

[0011] On the other hand, a method for auxiliary diagnosis of pet diseases based on multimodal image recognition and feature library matching is provided, including the following steps: S1 constructs a structured pet disease-symptom feature knowledge base. The knowledge base stores multiple diseases for different pet species and their corresponding standard symptom feature vectors. Each standard symptom feature vector consists of several quantifiable symptom feature metadata and their medical weight coefficients when diagnosing the corresponding disease. The symptom feature metadata establishes semantic relationships through a symptom ontology tree. S2 acquires multimodal image data of the pet, and through several specially trained deep learning sub-models, automatically detects and quantifies various symptom features predefined in the knowledge base from the multimodal image data, outputs a confidence score for each identified symptom feature, and generates a symptom feature vector to be matched. S3 calculates the matching degree by using a weighted semantic similarity algorithm based on information content, which comprehensively considers the confidence score, the medical weight coefficient, and the semantic distance of the symptom features in the symptom ontology tree, based on the symptom feature vector to be matched and the standard symptom feature vector of the corresponding pet type in the knowledge base; S4 outputs a list of disease probabilities sorted by matching degree.

[0012] Furthermore, step S2 also includes: distributing the multimodal image data to one or more corresponding deep learning sub-model queues according to the image type to achieve parallel inference.

[0013] Furthermore, in the weighted semantic similarity algorithm based on information content, the value of the information content is determined based on the specificity of the symptom features, specifically as follows: in, This represents the total number of diseases for the current pet species in the knowledge base. To include symptom characteristics The number of diseases.

[0014] Furthermore, in step S3, calculating the matching degree specifically includes: Calculate the semantic similarity between each symptom feature in the query vector and each symptom feature in the standard disease vector, and take the best match and perform a weighted sum to obtain the first match score; Calculate the semantic similarity between each symptom feature in the disease standard vector and each symptom feature in the query vector, and take the best match and perform a weighted sum to obtain the second matching score; The arithmetic mean of the first matching score and the second matching score is taken as the final matching degree; The semantic similarity is calculated based on the information content value of the deepest common ancestor node of the two symptom features in the symptom ontology tree, specifically as follows: in, To query features, For diseases in the knowledge base The standard characteristics.

[0015] Furthermore, prior to step S2, the method includes: providing a standardized image acquisition guidance interface through an interactive mobile terminal application to manage pet files and acquire the multimodal image data.

[0016] It also includes step S5: receiving the confirmed diagnosis information from the user, automatically adjusting the medical weights of the relevant symptom features in the knowledge base according to the confirmed diagnosis information, and adding the confirmed diagnosis image and the corresponding symptom feature vector to the training set to update the deep learning sub-model.

[0017] The beneficial effects of this invention are as follows: This invention, through a standardized guided image acquisition interface, enables non-professional pet owners to easily acquire medical images with diagnostic value, thus solving the traditional acquisition problem of "not being able to interpret or explain" images. Through the deep integration of a multi-model collaborative feature extraction engine and a structured symptom feature knowledge base, and by employing a disease matching algorithm based on semantic similarity, the system can comprehensively analyze symptom combinations from multiple parts of the pet, thereby simulating the cognitive process of "differential diagnosis" in veterinarians and significantly improving the clinical reference value of auxiliary diagnosis. By transparently presenting the matching results and specific supporting symptom evidence, users can understand the diagnostic basis, greatly enhancing the credibility of the diagnostic results and users' willingness to accept them. By establishing a closed-loop feedback mechanism for confirmed diagnosis, the system can automatically adjust medical weights and expand the high-quality training set, thereby achieving continuous iterative evolution of the recognition model and knowledge base, overcoming the shortcomings of existing static models that cannot improve performance. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is an overall architecture block diagram of a pet disease auxiliary diagnosis system based on multimodal image recognition and feature library matching according to an embodiment of the present invention (showing the interaction relationship between the cloud knowledge base, multi-model engine, matching module and mobile terminal). Figure 2 This is an overall flowchart of the pet disease auxiliary diagnosis method based on multimodal image recognition and feature library matching according to an embodiment of the present invention; Figure 3 This is a schematic diagram of a standardized shooting guide interface for a mobile terminal APP according to an embodiment of the present invention (including a focus frame and body outline reference lines). Figure 4 This is a schematic diagram of the auxiliary diagnostic report generation and display interface according to an embodiment of the present invention (including probability ranking, supporting evidence, and suggestions). Figure 5 This is a flowchart of multi-model collaborative symptom feature extraction (task scheduling, parallel inference, feature assembly) according to an embodiment of the present invention. Figure 6 This is a schematic diagram of the disease matching algorithm based on Resnik semantic similarity according to an embodiment of the present invention (input vector, ontology tree mapping, MICA retrieval, IC weighting, and aggregated output). Figure 7 This is a comparison chart of experimental performance according to the embodiments of the present invention (a bar chart of the Top-1 and Top-3 accuracy rates of the present invention and the control group). Figure 8 This is a radar chart of the contribution of ablation experiments according to an embodiment of the present invention (the contribution weight of each sub-model to the overall matching degree). Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention are within the scope of protection of the present invention.

[0021] like Figure 1 As shown in the embodiment of the present invention, a pet disease auxiliary diagnosis system based on multimodal image recognition and feature library matching includes four core components: Structured Pet Disease - Symptom Feature Knowledge Base: Stored on a cloud server, this system systematically collects various common diseases for different pet species (such as dogs and cats), and constructs a "standard symptom feature vector" for each disease. This vector consists of a series of predefined, quantifiable, fine-grained appearance and behavioral feature metadata. Each feature is assigned a medical weight coefficient when diagnosing a specific disease, and semantic relationships between features are established through a symptom ontology tree.

[0022] Multi-model collaborative symptom feature extraction engine: Deployed in the cloud, it includes a series of specialized deep learning sub-models (such as eye symptom recognition models, ear symptom recognition models, skin symptom recognition models, and abnormal posture recognition models). These models can automatically detect and quantify various predefined symptom features in the knowledge base from user-uploaded images / videos of specific parts of the pet, and output a confidence score for each recognition result.

[0023] The disease matching and reasoning module based on probability and semantics receives a "symptom feature vector to be matched" with confidence scores from the feature extraction engine. Based on the pet category selected by the user, it filters the corresponding disease subset from the knowledge base. Employing a weighted semantic similarity algorithm based on information content (IC), it comprehensively considers feature recognition confidence, predefined medical weights in the knowledge base, and the semantic distance of symptom features in the ontology tree. It calculates the matching degree between the current feature vector and each standard vector for each disease in the database, ultimately outputting a list of disease probabilities sorted in descending order of matching degree, along with associated supporting symptom features.

[0024] Interactive mobile terminal application: Provides an interface to guide users in standardized image acquisition (such as shooting guidance with outline guide boxes), manages pet files, uploads data, and displays auxiliary diagnostic results in the form of structured reports, clearly marking the results as "auxiliary reference", while also providing a diagnostic feedback entry point.

[0025] Method Flow: The overall flow of the method is as follows Figure 2 As shown, it mainly includes the following steps: S1: Task Initialization Users can create examination tasks through a mobile app, fill in basic pet information such as breed, age, and gender, select the body systems they are interested in such as skin, eyes, and ears, or directly initiate a comprehensive examination.

[0026] S2: Standardized Image Acquisition and Upload The app launches a standardized shooting guidance module, which dynamically presents shooting guidance boxes according to the task type. For example, close-ups of affected areas require automatic selection of the focus area, and full-body photos require the complete body outline to be included. This guides users to shoot images / videos that meet quality requirements and automatically uploads them to the cloud server.

[0027] S3: Multi-model collaborative symptom feature extraction After receiving image data, the cloud server initiates the feature extraction engine: the task scheduler distributes images to the corresponding dedicated recognition sub-model queues based on image type (local / full-body / eye / ear, etc.). Each sub-model performs inference in parallel, outputting the probability or quantified value of the occurrence of predefined symptom features and generating a confidence score. The feature vector assembler integrates all identified features and their confidence scores into a symptom feature vector V_current for the current case to be matched.

[0028] S4: Knowledge-Based Probabilistic Matching and Reasoning The disease matching module performs the following sub-steps: Knowledge base filtering: Based on pet type, load the set of standard symptom feature vectors for the corresponding disease from the knowledge base; Semantic similarity calculation: The Resnik similarity algorithm based on MICA (deepest common ancestor) is used, combined with medical weights and recognition confidence, to calculate the matching degree between V_current and each disease standard vector; Results sorting: A list of disease probabilities is generated in descending order of matching degree.

[0029] S5: Results Output and Report Generation The cloud sends the structured results to the app, which then generates a visual, assisted diagnostic report, including: Probability ranking; list of supporting evidence (each symptom and its identification confidence level); auxiliary suggestions (medical treatment guidance, differential diagnosis tips); key evidence map (identified areas are highlighted with semi-transparent highlighting on the original image); interpretation of results and disclaimer.

[0030] S6: Feedback Closed-Loop Optimization After receiving an offline diagnosis, users can provide their actual diagnosis information via the app. Based on this, the system automatically adjusts the medical weights of relevant symptom features in the knowledge base and adds the confirmed images and feature vectors to a high-quality training set for subsequent model iterations and updates.

[0031] To facilitate understanding of the above technical solutions of the present invention, the following detailed description of the above technical solutions of the present invention will be provided through specific usage methods.

[0032] This embodiment takes the auxiliary diagnosis of canine skin diseases as an example, and the specific implementation is as follows: 1. Construction of a structured pet disease-symptom characteristic knowledge base The knowledge base constructed in this embodiment covers 32 common canine skin diseases, defining a total of 47 quantifiable symptom feature metadata items. Each feature metadata item includes: feature ID, feature name, descriptive text, quantification standard, and its position in the ontology tree. The feature ontology tree adopts a hierarchical structure; for example, hair loss is the parent node, and child nodes include patchy hair loss, symmetrical hair loss, non-inflammatory hair loss, etc.

[0033] For each disease D_j, its standard feature vector is represented as: in, Features In diagnosing diseases The medical weights at any given time were initially determined by three associate senior veterinarians using the Delphi method, with values ​​ranging from [0,1]. These weights were then dynamically adjusted through feedback learning.

[0034] 2. Definition of Symptom Feature Information (IC) To quantify the specificity of symptom features, this embodiment introduces the concept of Information Content (IC). For any symptom feature... Its IC value is defined as: Where |D| represents the total number of diseases for the current pet species in the knowledge base. To include symptom characteristics The number of diseases. The physical meaning of this formula is that the more specific the symptom, the larger the IC value, and the greater its contribution to differential diagnosis; common symptoms have lower IC values, which can prevent high-frequency symptoms from dominating the matching results.

[0035] Confidence level of feature extraction engine output Combined with IC value and medical weight, a weighted query vector is formed: in, The initial values ​​for the preset medical weights are set by experts.

[0036] 3. Disease matching algorithm based on Resnik similarity This embodiment uses an asymmetric average optimal matching algorithm to calculate the matching degree between the current case vector and the disease standard vector.

[0037] Definition 1 (Single Feature Semantic Similarity): For query features Diseases in the knowledge base Standard features The semantic similarity between the two is calculated based on their deepest common ancestor (MICA) node in the symptom ontology tree: This similarity measure measures the degree of proximity of two symptom features in medical semantics, allowing for fuzzy matching. For example, although "patchy hair loss" and "circular hair loss patches" are not exactly the same feature, they both belong to the "hair loss" subclass and can obtain a non-zero similarity through the common ancestor node "hair loss," thereby solving the problem of inaccurate descriptions by pet owners.

[0038] Definition 2 (matching score for the disease): in, To query the number of non-zero features in the vector, Features In disease The weight of medicine in China For query features The formula calculates the identification confidence level by taking the best match for each query feature in the disease standard feature set and summing them with weights, reflecting the principle that "partial matching is a contribution".

[0039] Definition 3 (Disease-to-Query Match Score): To achieve symmetry, the reverse matching score of the disease standard feature set against the query feature set is also calculated: Definition 4 (Final Match): The final match score is the arithmetic mean of the two-way scores: All candidate diseases are sorted in descending order of MatchScore, and a Top-K list is output.

[0040] 5. Example: Diagnosis of skin problems in dog "DouDou" S1: Task Initialization Users open the app and create a skin problem check task for their pet "DouDou" (a 3-year-old female Chinese rural dog). The system prompts the user to fill in basic information and select skin as the primary concern.

[0041] S2: Standardized Image Acquisition and Upload The app interface provides a shooting start guide, such as... Figure 3 As shown. First, Guide A prompts the user to take a high-resolution close-up of the affected skin area. A circular focus frame appears on the screen, automatically detecting sharpness and locking focus. Then, Guide B prompts the user to take a full-body side view photo including the complete body outline, displaying reference contour lines. After shooting, the image is automatically uploaded to the cloud.

[0042] S3: Multi-model collaborative symptom feature extraction Images are received in the cloud. The task scheduler sends close-up images of specific areas to the skin symptom recognition model queue and full-body side-view images to the overall skin lesion distribution analysis model queue. The two models infer in parallel, and the output results are as follows: The feature vector assembler generates the query vector for the current case: {(S201,0.92),(S202,0.85),(S203,0.78),(S204,0.72)}Q={(S201,0.92),(S202,0.85),(S203,0.78),(S204,0.72)} S4: Knowledge-based probabilistic matching The knowledge base contains 32 canine skin diseases, each associated with several standard feature vectors. Taking "hypothyroidism" as an example, its standard vector includes features S201 (weight 0.7), S202 (weight 0.6), S204 (weight 0.8), and S206 (symmetrical hair loss, weight 0.9), etc. The MatchScore calculation process is as follows: First, the IC values ​​(pre-calculated) of each feature are obtained from the ontology tree. For example, , , wait.

[0043] Single-item semantic similarity calculation: middle and middle Exact match ; middle and middle Exact match ; middle and There is no direct match in the text, but it matches the parent node of "scaly". (Scales) = 1.87; S204 and Matching S204 in China ; Weighted summation ; Similarly, calculate ; The final MatchScore is (2.31+2.14) / 2 = 2.225.

[0044] Similar calculations were performed for other diseases, and the results are as follows: S5: Results Output and Report Generation The app generates auxiliary diagnostic reports, such as Figure 4 As shown, the content includes: Primary possibility: Endocrine-related issues (hypothyroidism), matching score 2.225 (normalized to 76%); Supporting evidence: well-defined patchy hair loss (high confidence), bilateral symmetrical distribution trend, accompanied by non-inflammatory scaling; Additional recommendations: It is strongly recommended to take your pet to a veterinary hospital for blood tests (especially thyroid function tests); Wood's lamp examination or fungal culture can be performed to rule out tinea (41% match); Key evidence image: The original image is marked with semi-transparent, highlighted color blocks to indicate "clearly defined hair removal areas" and "symmetrically distributed areas".

[0045] S6: Feedback Closed-Loop Optimization Following the advice, the user took "DouDou" to see a doctor. After a serum T4 test and TSH stimulation test, "hypothyroidism" was diagnosed. The user marked the result as "true" on the app. The system received the feedback and then: Weight update: In the knowledge base, automatically increase the weight parameters of features S202 (clear boundaries) and S204 (symmetric distribution) in the diagnosis of "endocrine alopecia" and related diseases (increment Δ=0.05).

[0046] Training set expansion: The images of the confirmed cases and their corresponding feature vectors are used as high-quality positive samples and stored in the training pool of disease D1 for subsequent model fine-tuning.

[0047] In summary, by utilizing the technical solutions described above, the standardized guided image acquisition interface enables non-professional pet owners to easily acquire medical images with diagnostic value, thus solving the traditional acquisition problem of "not being able to interpret or explain" images. Through the deep integration of a multi-model collaborative feature extraction engine and a structured symptom feature knowledge base, and employing a disease matching algorithm based on semantic similarity, the system can comprehensively analyze symptom combinations from multiple parts of the pet, thereby simulating the cognitive process of "differential diagnosis" in veterinarians and significantly improving the clinical reference value of auxiliary diagnosis. By transparently presenting the matching results and specific supporting symptom evidence, users can understand the diagnostic basis, greatly enhancing the credibility of the diagnostic results and users' willingness to accept them. By establishing a closed-loop feedback mechanism for confirmed diagnosis, the system can automatically adjust medical weights and expand the high-quality training set, thereby achieving continuous iterative evolution of the recognition model and knowledge base, overcoming the shortcomings of existing static models in terms of performance improvement.

[0048] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A pet disease auxiliary diagnostic system based on image recognition and feature database matching, characterized in that, include: A structured pet disease-symptom feature knowledge base is used to store various diseases for different pet species and their corresponding standard symptom feature vectors. Each standard symptom feature vector consists of several quantifiable symptom feature metadata and their medical weight coefficients when diagnosing the corresponding disease. The symptom feature metadata establishes semantic relationships through a symptom ontology tree. A multi-model collaborative symptom feature extraction engine is used to receive multimodal image data of pets, and through several specially trained deep learning sub-models, automatically detect and quantify various symptom features predefined in the knowledge base from the multimodal image data, and output a confidence score for each identified symptom feature to generate a symptom feature vector to be matched. The disease matching and reasoning module is used to calculate the matching degree based on the symptom feature vector to be matched and the standard symptom feature vector of the corresponding pet type in the knowledge base, using a weighted semantic similarity algorithm based on information content, taking into account the confidence score, the medical weight coefficient and the semantic distance of the symptom features in the symptom ontology tree, and output a list of disease possibilities sorted by matching degree.

2. The pet disease auxiliary diagnosis system based on image recognition and feature database matching according to claim 1, characterized in that, The multi-model collaborative symptom feature extraction engine includes a task scheduler, which distributes the multimodal image data to one or more corresponding deep learning sub-model queues according to the image type to achieve parallel inference.

3. The pet disease auxiliary diagnosis system based on image recognition and feature database matching according to claim 1, characterized in that, The weighted semantic similarity algorithm based on information content determines the value of the information content based on the specificity of symptom features, specifically as follows: Where |D| represents the total number of diseases for the current pet species in the knowledge base. To include symptom characteristics The number of diseases.

4. The pet disease auxiliary diagnosis system based on image recognition and feature database matching according to claim 1, characterized in that, When calculating the matching degree, the disease matching and reasoning module specifically employs an asymmetric average optimal matching algorithm, which includes: Calculate the semantic similarity between each symptom feature in the query vector and each symptom feature in the standard disease vector, and take the best match and perform a weighted sum to obtain the first matching score; Calculate the semantic similarity between each symptom feature in the disease standard vector and each symptom feature in the query vector, and take the best match and perform a weighted sum to obtain the second matching score; The arithmetic mean of the first matching score and the second matching score is taken as the final matching degree; The semantic similarity is calculated based on the information content value of the deepest common ancestor node of the two symptom features in the symptom ontology tree, specifically as follows: in, To query features, For diseases in the knowledge base The standard characteristics.

5. The pet disease auxiliary diagnosis system based on image recognition and feature database matching according to claim 1, characterized in that, Also includes: An interactive mobile terminal application is used to provide a standardized image acquisition guidance interface, manage pet files, upload the multimodal image data, and display the list of possible diseases, supporting evidence and auxiliary suggestions in the form of a structured report, while also providing a diagnostic feedback entry point; The feedback closed-loop optimization module is used to receive the diagnosis information from the user, automatically adjust the medical weights of the relevant symptom features in the knowledge base according to the diagnosis information, and add the diagnosis image and the corresponding symptom feature vector to the training set to update the deep learning sub-model.

6. A method for auxiliary diagnosis of pet diseases based on image recognition and feature database matching, characterized in that, Includes the following steps: S1 constructs a structured pet disease-symptom feature knowledge base. The knowledge base stores multiple diseases for different pet species and their corresponding standard symptom feature vectors. Each standard symptom feature vector consists of several quantifiable symptom feature metadata and their medical weight coefficients when diagnosing the corresponding disease. The symptom feature metadata establishes semantic relationships through a symptom ontology tree. S2 acquires multimodal image data of the pet, and through several specially trained deep learning sub-models, automatically detects and quantifies various symptom features predefined in the knowledge base from the multimodal image data, outputs a confidence score for each identified symptom feature, and generates a symptom feature vector to be matched. S3 calculates the matching degree by using a weighted semantic similarity algorithm based on information content, which comprehensively considers the confidence score, the medical weight coefficient, and the semantic distance of the symptom features in the symptom ontology tree, based on the symptom feature vector to be matched and the standard symptom feature vector of the corresponding pet type in the knowledge base; S4 outputs a list of disease probabilities sorted by matching degree.

7. The pet disease auxiliary diagnosis method based on image recognition and feature database matching according to claim 6, characterized in that, Step S2 further includes: distributing the multimodal image data to one or more corresponding deep learning sub-model queues according to the image type to achieve parallel inference.

8. The pet disease auxiliary diagnosis method based on image recognition and feature database matching according to claim 7, characterized in that, The weighted semantic similarity algorithm based on information content determines the value of the information content based on the specificity of symptom features, specifically as follows: ,in, This represents the total number of diseases for the current pet species in the knowledge base. To include symptom characteristics The number of diseases.

9. The pet disease auxiliary diagnosis method based on image recognition and feature database matching according to claim 6, characterized in that, In step S3, calculating the matching degree specifically includes: Calculate the semantic similarity between each symptom feature in the query vector and each symptom feature in the standard disease vector, and take the best match and perform a weighted sum to obtain the first matching score; Calculate the semantic similarity between each symptom feature in the disease standard vector and each symptom feature in the query vector, and take the best match and perform a weighted sum to obtain the second matching score; The arithmetic mean of the first matching score and the second matching score is taken as the final matching degree; The semantic similarity is calculated based on the information content value of the deepest common ancestor node of the two symptom features in the symptom ontology tree, specifically as follows: ,in, To query features, For diseases in the knowledge base The standard characteristics.

10. The pet disease auxiliary diagnosis method based on image recognition and feature database matching according to claim 6, characterized in that, Before step S2, the method also includes: providing a standardized image acquisition guidance interface through an interactive mobile terminal application to manage pet files and acquire the multimodal image data; It also includes step S5: receiving the confirmed diagnosis information from the user, automatically adjusting the medical weights of the relevant symptom features in the knowledge base according to the confirmed diagnosis information, and adding the confirmed diagnosis image and the corresponding symptom feature vector to the training set to update the deep learning sub-model.

Images