Animal parasite intelligent diagnosis system and method based on artificial intelligence

Abstract

The invention discloses an animal parasite intelligent diagnosis system and method based on artificial intelligence. The system comprises a terminal device, a terminal device used for obtaining a microscopic image of an animal parasite, and a cloud server used for receiving the microscopic image. The cloud server deploys a parasite target detection model and a parasite type classification model; the parasite target detection model processes the received microscopic image and outputs bounding box coordinate information in the image; cutting out corresponding suspected area image blocks based on the bounding box coordinate information; inputting the suspected region image blocks into a parasite type classification model to obtain an authenticity discrimination result and a parasite type identification result of each suspected region; according to the authenticity judgment result and the parasite type recognition result, a diagnosis report is generated and returned to the terminal equipment for display, the problems that special equipment is expensive and terminal application computing power is insufficient are solved, and unification of high performance, low cost and high portability is achieved.

Description

Technical Field

[0001] This invention belongs to the field of artificial intelligence technology, specifically relating to an intelligent diagnostic system and method for animal parasites based on artificial intelligence. Background Technology

[0002] Animals are susceptible to a wide variety of parasites, and identifying and diagnosing these parasites requires significant time and resources. In existing technologies, Singh et al. (Singh, Naseeb et al. Development of deep learning-based mobile application for the identification of Coccidia species in pigs using microscopic images. Veterinary parasitology vol. 334 (2025): 110376. doi:10.1016 / j.vetpar.2024.110376) designed a deep learning-based mobile application for identifying coccidia in pig fecal samples. They used direct flotation (saturated brine and sucrose solution) and a modified McMaster technique to quantify the number of oocysts per gram of feces. Positive samples were sporulated, and species images were captured using an Olympus BX51 microscope at 200x and 400x magnification. A CNN model was developed using an encoder-decoder architecture for semantic segmentation of coccidia species. Resource-efficient models (EfficientNetB0, EfficientNetB1, MobileNet, and MobileNetV2) are used as encoders to extract features from the input microscopic images. The decoder uses deconvolutional layers to progressively increase the feature map size, ensuring the predicted mask is aligned with the original image resolution. Transfer learning is employed, initializing model weights with ImageNet pre-trained weights and fine-tuning them using the Adam optimizer. Dice coefficients are used as the loss function to mitigate class imbalance. The deep learning model is then translated into a mobile application with a user-friendly interface. This application allows users to upload images taken under a microscope and identify coccidia species in the images by clicking the "Predict" button; the identification results are displayed directly on the interface. Extensive testing has shown that the application runs stably on multiple Android devices. Zhu Huiyin et al. (Zhu Huiyin, Li Yuting, Zhu Daiqian, et al. Establishment and application of an AI-assisted parasite egg detection platform [J]. Chinese Journal of Schistosomiasis Control, 2024, 36(06):643-648.DOI:10.16250 / j.32.1374.2024094.) established an AI-assisted parasite egg detection platform based on the Baidu EasyDL object detection platform. This platform identifies eggs of nine parasites, including nematodes, hookworms, and flukes. The identification model can be used temporarily on smartphones by scanning QR codes, or permanently deployed as a mobile app. The mobile app can identify images stored on the phone and can also achieve "instant identification" through the phone's camera function. The average accuracy, precision, and recall of this detection platform for identifying parasite eggs were 93.42%, 92.55%, and 89.32%, respectively.

[0003] Sanchez-Montejo et al. (Sanchez-Montejo J, Marín M, Villamizar-Monsalve MA, et al. AxiWorm: a new tool using YOLOv5 to test antiparasitic drugs against Trichinella spiralis. Parasit Vectors. 2025;18(1):36. Published 2025 Feb 2. doi:10.1186 / s13071-025-06664-8) developed an artificial intelligence-based detection device: AxiWorm. This device combines a robotic arm with a Nikon Eclipse Ts2 inverted microscope to form a robotic-assisted image acquisition system. Based on model training using YOLOv5, a computer vision model combining convolutional neural networks (CNN) automatically analyzes a large number of microscope images to achieve rapid larval classification (healthy vs. damaged) and counting. Deploying the trained model to a smartphone enables accurate identification, classification, and counting of first-stage Trichinella spiralis larvae in images.

[0004] However, the existing technology has the following drawbacks: (1) Data scarcity: There is a lack of large-scale, well-annotated parasite image datasets, and existing intelligent parasite detection technologies can only detect limited parasite images; (2) Cumbersome operation: Most deep learning tools are not user-friendly for parasitologists without an artificial intelligence background; (3) Insufficient model interpretability: Deep learning models are often regarded as "black boxes," making it difficult to explain their decision-making processes. This poses a challenge for fields such as medical diagnosis that require high reliability and interpretability; (4) High hardware resource requirements: Training and inference of deep learning models usually require powerful computing resources. Summary of the Invention

[0005] To address the aforementioned technical problems, this invention provides an intelligent diagnostic system and method for animal parasites based on artificial intelligence.

[0006] The specific plan is as follows: An artificial intelligence-based intelligent diagnostic system for animal parasites includes a mobile terminal and a cloud server, wherein the mobile terminal and the cloud server are communicatively connected. The mobile terminal is equipped with an image acquisition module, an image upload module, and a detection result display module. The image acquisition module is used to acquire microscopic images of the parasite to be analyzed; The image upload module is used to encrypt and encapsulate image data and upload it to the cloud server via the HTTPS protocol; The test result display module is used to receive and display test results from the cloud server; The cloud server is equipped with trained parasite target detection models and parasite species classification models. The parasite target detection model is used to extract deep features from parasite microscopic images, output the bounding box coordinates of all suspected parasites, crop out the suspected regions based on the bounding box coordinates, and input them into the parasite species classification model. The parasite classification model is used to determine the authenticity of parasites and identify the species, and then sends the results to the mobile terminal.

[0007] The image acquisition module includes either photo capture and upload or album selection and upload. The photo capture and upload method uses the built-in camera of the mobile phone to connect to a microscopic imaging device to capture microscopic images of parasites. The album selection and upload method involves selecting microscopic images stored in the album and uploading them.

[0008] The cloud server also includes a knowledge base module and an AI assistant module. The knowledge base module is used to store parasite prevention and control knowledge data, and the AI ​​assistant module generates interpretable natural language answers based on the detection results, using a large language model and the knowledge base module.

[0009] The cloud service adopts a microservice architecture, deploying the parasite target detection model and the parasite species classification model as independent microservices, and communicating with each other through message queues.

[0010] The parasite target detection model is deployed on a cloud server in ONNX format. The parasite target detection model is a lightweight YOLO11n-SCDNet model, which includes a backbone network, a neck network, and a detection head. The backbone network uses the ShuffleNetV2 structure for feature extraction, and the neck network includes a cross-scale channel fusion module (CCFM) for multi-scale feature fusion. The detection head uses a dual convolutional structure (C3DC) for target detection.

[0011] The parasite classification model is a target detection model based on a convolutional neural network.

[0012] The cloud server supports asynchronous processing mode. For large images or batch tasks, the task ID is returned immediately after uploading, and the user is notified via push notification after processing is completed.

[0013] An artificial intelligence-based intelligent diagnostic method for animal parasites includes the following steps: S1): The mobile terminal uploads the collected microscopic images to the cloud server. S2): The cloud server calls the pre-trained parasite detection model, uses the parasite detection model to extract deep features from the input microscopic image, identifies and locates the suspected parasite regions in the image, and outputs the bounding box coordinates of each suspected region. S3): The cloud server will crop out each suspected region based on the bounding box coordinate information, and input the image patch of each suspected region into the pre-trained parasite classification model. The parasite classification model will perform true / false discrimination and species identification for each candidate region.

[0014] The pre-trained parasite detection model also includes a dataset construction process, which includes data collection and standardized labeling: The data collection includes the following steps: T1): Fecal samples were collected from cattle and sheep farms and dog and cat breeding bases. All samples were pretreated by water washing sedimentation method and saturated sucrose flotation method. Then, positive samples for parasites were screened by microscopic examination. T2: Professional veterinary technicians used optical microscopes to capture high-resolution microscopic images of parasite-positive samples at magnifications of 200x and 400x. The image size was 2464×2056 pixels, with a resolution of 300 dpi. A total of 7,120 effective microscopic images were obtained, covering eight common parasite species: Giardia, Eimeria, Isospora, Entamoeba, Toxocara, Ancylostoma, Trichuris, and Moniezia. The standardized labeling includes the following steps: T3): All microscopic images were manually annotated by two independent veterinary researchers using the LabelImg tool to ensure the accuracy and consistency of image annotation; T4): Each parasite oocyst or parasite body in each image is precisely selected and drawn with a rectangular bounding box that closely fits its morphological boundary. The images are then labeled according to the type of parasite. All labeling results are reviewed and confirmed by experienced parasitologists to ensure that the labeling quality meets high standards. T5): The annotation information is stored in the standard format supported by the YOLO object detection model. Each image corresponds to a text annotation file, which contains the bounding box information of all objects in the image. Each bounding box record includes: class index and normalized coordinate parameters, namely the x and y coordinates of the center point of the bounding box, width and height, all of which are normalized relative to the image size.

[0015] The method further includes receiving a user's AI assistant consultation request regarding the detection results; sending the detection context information to the cloud AI assistant module; the cloud AI assistant module calling a large language model and combining it with a parasite knowledge base to generate an interpretable answer; and returning the answer to the mobile terminal for display.

[0016] An AI-based intelligent diagnostic system and method for animal parasites constructs an end-cloud collaborative parasite diagnostic architecture of "terminal acquisition - cloud analysis - result return." This addresses the problems of existing integrated detection systems, which are often closed, expensive, and difficult to popularize, as well as the limitations of pure terminal applications, leading to low model performance and limited species recognition due to computational constraints. It overcomes the technical bias that "high-precision diagnosis must rely on dedicated, closed equipment." By deploying a high-performance ONNX format detection and classification model in the cloud, it enables complex and accurate multi-species identification on ordinary smartphones, combining high performance, low cost, and high portability. By setting an "AI Chat" entry point in the detection report interface and automatically carrying the current detection image and recognition results as a multimodal dialogue context, it provides users with dynamic, accurate, and traceable decision explanations and deep knowledge-based question answering, elevating the system from a static detection tool to a dynamic expert advisor. Furthermore, it independently constructs multi-species and multi-parasite image datasets as experimental data sources, solving the problem of data scarcity. Attached Figure Description

[0017] Figure 1 It is the YOLO11n-SCDNet network structure.

[0018] Figure 2 This is a diagram showing the dataset.

[0019] Figure 3 This is the login / registration page for mobile devices.

[0020] Figure 4 It refers to the mobile terminal's homepage and personal information page.

[0021] Figure 5 This is the mobile terminal detection function page.

[0022] Figure 6 It is the mobile terminal AI model page. Detailed Implementation

[0023] The technical solutions in the embodiments of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the implementation of the present invention, and not all of it. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0024] An artificial intelligence-based intelligent diagnostic system for animal parasites includes a mobile terminal and a cloud server, wherein the mobile terminal and the cloud server are communicatively connected. The mobile terminal is equipped with an image acquisition module, an image upload module, and a detection result display module. The image acquisition module is used to acquire microscopic images of the parasite to be analyzed; The image upload module is used to encrypt and encapsulate image data and upload it to the cloud server via the HTTPS protocol; The test result display module is used to receive and display test results from the cloud server; The cloud server is equipped with trained parasite target detection models and parasite species classification models. The parasite target detection model is used to extract deep features from parasite microscopic images, output the bounding box coordinates of all suspected parasites, crop out the suspected regions based on the bounding box coordinates, and input them into the parasite species classification model. The parasite classification model is used to determine the authenticity of parasites and identify the species, and then sends the results to the mobile terminal.

[0025] The image acquisition module includes either photo capture and upload or album selection and upload. The photo capture and upload method uses the built-in camera of the mobile phone to connect to a microscopic imaging device to capture microscopic images of parasites. The album selection and upload method involves selecting microscopic images stored in the album and uploading them.

[0026] The system is portable and intelligent by acquiring microscopic images of parasites through mobile terminals and uploading them to cloud servers, and receiving feedback results from the cloud servers. In this embodiment, microscopic images of parasites of the target object are acquired through two methods: one is to transmit existing microscopic images to a smartphone for processing; the other is to directly use the built-in camera of the mobile phone to connect to a microscopic imaging device to acquire microscopic images of parasites in real time.

[0027] The cloud server also includes a knowledge base module and an AI assistant module. The knowledge base module is used to store parasite prevention and control knowledge data, and the AI ​​assistant module generates interpretable natural language answers based on the detection results, using a large language model and the knowledge base module.

[0028] The collected microscopic images are uploaded to a cloud server, which has integrated a pre-trained parasite detection model. Using the parasite detection model, deep features are extracted from the input microscopic images to identify and locate suspected parasite regions in the images. The output results are preliminary detection data containing the bounding box coordinates of each suspected region and its corresponding first confidence score. The image patches of these suspected regions are further input into a pre-trained parasite classification model, which is responsible for performing true / false discrimination and species identification for each candidate region.

[0029] The system integrates the output information of the classification model to automatically generate detailed test reports. These reports clearly indicate the location of the parasite, the veracity of the test, and its species. The system also integrates professional AI large-scale model functions. Users can trigger the AI ​​assistant by clicking to access detailed information about the parasite when they have questions about the test report or wish to obtain more in-depth knowledge. This information has been reviewed by experts and includes, but is not limited to, descriptions of the parasite's morphological characteristics, epidemiological features, related clinical symptoms and pathogenic mechanisms, as well as clinical prevention and treatment recommendations.

[0030] The cloud service adopts a microservice architecture, deploying the parasite target detection model and the parasite species classification model as independent microservices and communicating through message queues, which improves the system's scalability and fault tolerance. The update or failure of a single model does not affect the other service.

[0031] like Figure 1 As shown, the parasite target detection model is deployed on a cloud server in ONNX format. The parasite target detection model is a lightweight YOLO11n-SCDNet model. To meet the comprehensive requirements of real-time performance, accuracy, and energy consumption for automatic detection of intestinal parasites in complex microscopic environments, this invention has made lightweight improvements to the YOLO11 model and proposed the YOLO11n-SCDNet lightweight model. The YOLO11n-SCDNet lightweight model includes a backbone network, a neck network, and a detection head. The backbone network uses the ShuffleNetV2 structure for feature extraction. The neck network includes a cross-scale channel fusion module (CCFM) for multi-scale feature fusion. The detection head uses a dual convolutional structure (C3DC) for target detection. Figure 1 In this code, Conv stands for Convolutional Module; ShuffleNetV2_0 is a lightweight convolutional architecture; C3DC is a feature aggregation module; SPPF is a spatial pyramid pooling module; C2PSA is a local spatial attention module; Detect represents the detection head; Concat is a feature connection module; and SCDown is a lightweight downsampling module.

[0032] The ShuffleNetV2 structure is introduced into the backbone network to improve feature extraction efficiency, the cross-scale channel fusion module CCFM is added to the neck structure to enhance multi-scale feature expression capability, and the dual convolutional structure C3DC is introduced into the detection head, thereby maintaining high detection accuracy while reducing the number of parameters and computational complexity.

[0033] By optimizing the feature extraction and information fusion process at the network structure level, the model's ability to detect tiny parasites is effectively improved while significantly reducing its computational overhead and power consumption, enabling it to run stably on resource-constrained platforms such as mobile devices and embedded microscopy equipment.

[0034] The YOLO11n-SCDNet model achieves an accuracy of 97.8% while having only 0.923M parameters, 3.0 GFLOPs of computation, and a model size of 2.15 MB, representing reductions of 64.2%, 52.4%, and 59.7% respectively compared to the baseline YOLO11n model.

[0035] The parasite classification model is a target classification model based on a dense convolutional neural network (DenseNet).

[0036] The cloud server supports asynchronous processing mode. For large images or batch tasks, the task ID is returned immediately after uploading, and the user is notified via push notification after processing is completed, thus improving the user experience.

[0037] An artificial intelligence-based intelligent diagnostic method for animal parasites includes the following steps: S1): The mobile terminal uploads the collected microscopic images to the cloud server. S2): The cloud server calls the pre-trained parasite detection model, uses the parasite detection model to extract deep features from the input microscopic image, identifies and locates the suspected parasite regions in the image, and outputs the bounding box coordinates of each suspected region. S3): The cloud server will crop out each suspected region based on the bounding box coordinate information, and input the image patch of each suspected region into the pre-trained parasite classification model. The parasite classification model will perform true / false discrimination and species identification for each candidate region.

[0038] This invention uses a self-constructed multi-species, multi-parasite image dataset as the source of experimental data. The pre-trained parasite detection model also includes a dataset construction process, which includes data collection and standardized annotation. The data collection includes the following steps: T1): A large number of fecal samples were collected from cattle and sheep farms and dog and cat breeding bases. All samples were pretreated by water washing sedimentation method and saturated sucrose flotation method. Then, positive samples for parasites were screened by microscopic examination. T2: Professional veterinary technicians used optical microscopes to capture high-resolution microscopic images of parasite-positive samples at 200x and 400x magnification. 200x magnification was used for worm detection, and 400x magnification for protozoan detection. The image size was 2464×2056 pixels, with a resolution of 300 dpi. A total of 7,120 effective microscopic images were obtained, covering eight common parasite species: Giardia, Eimeria, Isospora, Entamoeba, Toxocara, Ancylostoma, Trichuris, and Moniezia. Figure 2 As shown. Figure 2 In the diagram, A is Aegypti, B is Eimeria coccidia, C is Monizbeckia orientalis, D is Toxocara canis, E is Isospora canis, F is Giardia, G is Entamoeba histolytica, and H is Trichodina canis.

[0039] The standardized labeling includes the following steps: T3): All microscopic images were manually annotated by two independent veterinary researchers using the LabelImg tool to ensure the accuracy and consistency of image annotation; T4): Each parasite oocyst or parasite body in each image is precisely selected and drawn with a rectangular bounding box that closely fits its morphological boundary. The images are then labeled according to the type of parasite. All labeling results are reviewed and confirmed by experienced parasitologists to ensure that the labeling quality meets high standards. T5): The annotation information is stored in the standard format supported by the YOLO object detection model. Each image corresponds to a text annotation file, which contains the bounding box information of all objects in the image. Each bounding box record includes: class index and normalized coordinate parameters, namely the x and y coordinates of the center point of the bounding box, width and height, all of which are normalized relative to the image size.

[0040] Standardized annotation formats facilitate efficient integration and training with YOLO-based object detection models, enhancing the compatibility and practicality of model training.

[0041] The method further includes receiving a user's AI assistant consultation request regarding the detection results; sending the detection context information to the cloud AI assistant module; the cloud AI assistant module calling a large language model and combining it with a parasite knowledge base to generate an interpretable answer; and returning the answer to the mobile terminal for display.

[0042] This invention integrates information related to common parasitic diseases in veterinary clinical practice, improves the level of diagnosis, prevention and treatment of parasitic infections, and establishes this parasitic knowledge base.

[0043] The knowledge base will systematically summarize and structure all parasite species identified by professional veterinary technicians in clinical testing, focusing on their morphological characteristics, epidemiological features, pathogenic mechanisms, related clinical symptoms, and prevention and treatment strategies. ① Morphological characteristics: Each parasite will be described in detail according to its taxonomic position (e.g., nematodes, tapeworms, protozoa), including but not limited to: size, shape, color, body surface structure, and typical staining reactions. For protozoan parasites, the morphological characteristics of different developmental stages such as trophozoites, cysts, and schizonts will also be described. ② Epidemiological features: The seasonal dynamics, transmission routes, and susceptible animal populations of various parasites will be systematically reviewed. Risk factors for transmission will be analyzed, including husbandry practices, climate conditions, hygiene, animal density, and history of contact with wild animals. ③ Clinical symptoms and pathogenic mechanisms: The pathophysiological changes caused by parasite infection will be described in detail. Based on the parasitic site (intestines, blood, liver, lungs, skin, etc.) and invasion mode (mechanical damage, toxin release, immune-mediated damage, nutrient depletion, etc.), the pathogenic mechanism is analyzed. ④ Prevention and treatment strategies: A comprehensive prevention and control plan is proposed, covering biosafety measures, deworming procedures, and vaccine application. For treatment, highly effective, low-toxicity, and broad-spectrum antiparasitic drugs (such as ivermectin, fenbendazole, praziquantel, toltrazuril, etc.) are precisely selected according to the parasite species, clearly defining the route of administration, dosage, course of treatment, and precautions, with particular attention to the safety of pregnant animals, young animals, and the combined use of multiple drugs. Deep information queries are provided by calling the AI ​​assistant interface integrated into the system, based on a parasite knowledge base, thereby achieving one-stop, interpretable intelligent parasite diagnosis on portable devices.

[0044] This invention adopts an architecture that combines mobile terminal applications with cloud servers. It uses a mobile phone camera connected to an external microscope or reads a local image library to obtain microscope images. Then, the detection model deployed on the terminal performs image depth feature extraction and parasite target localization. The output image blocks of suspected areas are sent to a target detection model based on a convolutional neural network deployed on the same terminal for authenticity judgment and species identification. Finally, a detection report is automatically generated and sent to the mobile terminal.

[0045] The mobile terminal handles user interaction and image preprocessing, while the cloud server hosts the core AI model and expert knowledge base. The two communicate via API interfaces. The software operation flow and system interaction are as follows: 1. Terminal Applications: User Interaction and Front-End Processing The terminal application serves as the user entry point, and its operation flow is as follows: [Step 1]: Image Acquisition. The user opens the App, such as... Figures 3 to 5 As shown, log in, select the "Detection" item on the homepage, select the corresponding detection service such as "coccidiosis detection" in the "Detection" item, and obtain the microscopic image of the parasite to be analyzed by "uploading photo" or "selecting from album"; [Step Two]: Image Upload. After the user clicks the "Start Detection" button, the software automatically encrypts and encapsulates the image data and sends it to the designated cloud server analysis interface via HTTPS protocol. At this time, the interface displays a "Analyzing" loading status; [Step 3]: Receiving and Displaying Results. The terminal application waits until it receives the standardized JSON format analysis results from the cloud. The software then automatically parses the data and displays a brief summary of the detection results. Users can click "View Details" to access the detailed report interface. The report interface visually displays the results: the bounding box of the parasite is overlaid on the original image, and the species name and confidence level are labeled. It also details the specific information of each detected target in a list format.

[0046] 2. Cloud Services: Core Model Inference and Knowledge Management The cloud server is the core of the intelligent diagnostic capability of this invention. After receiving the image uploaded by the terminal, it automatically performs the following pipeline operations:

Step 1

Step 2

[0047] Centralizing computationally intensive model inference tasks in the cloud enables the application of high-performance, high-capacity AI models, while significantly reducing the performance requirements of terminal devices. Users can enjoy the performance improvements brought about by model iteration and upgrades without updating the app, achieving centralized management and efficient utilization of resources.

[0048] 3. Integration of Knowledge Base and AI Assistant This function is also achieved through the interaction between the terminal and the cloud: [Step 1]: In the detection report interface generated by the terminal app, there is an "AI Chat" button. After the user clicks this button, the application jumps to a brand new, interactive intelligent dialogue interface, such as... Figure 6As shown. This interface automatically carries the contextual information of the current detection report, including but not limited to: the parasite image currently being viewed, the parasite name, confidence level, and other key data, as well as common fixed questions, as the initial context for dialogue with the AI ​​large model; [Step Two]: After receiving the API call request, the cloud service invokes the integrated large-scale AI language model. The AI ​​model comprehensively understands the user's question, analyzes the visual features of the image, and combines this with expert knowledge to generate a targeted and interpretable natural language response, which is then returned to the terminal app. This tightly integrates AI's detection capabilities with expert domain knowledge through cloud services, providing users with a complete closed loop from "diagnosis" to "cognition," greatly improving the reliability of the results and the system's practical value.

[0049] This invention constructs an edge-cloud collaborative parasite diagnostic architecture of "terminal acquisition - cloud analysis - result return," solving the problems of existing integrated detection systems being closed, expensive, and difficult to popularize, as well as the low model performance and limited species recognition caused by computing power limitations in pure terminal applications. By deploying high-performance ONNX format detection and classification models in the cloud, complex and accurate multi-species identification can be achieved on ordinary smartphones, combining high performance, low cost, and high portability. This architecture overcomes the technical bias that "high-precision diagnosis must rely on dedicated closed equipment," and through innovative resource allocation, empowers professional-grade diagnostic capabilities to widely used mobile terminals, significantly lowering the user threshold.

[0050] An interactive AI large-scale model interpretation method deeply integrated with detection results was created, solving the key challenges of deep learning models being a "black box" in medical diagnosis, lacking transparency in the decision-making process, and resulting in low user trust. By setting up an "AI Chat" entry point in the detection report interface, and automatically carrying the current detection image and recognition results as a multimodal dialogue context, dynamic, accurate, and traceable decision explanations and deep knowledge Q&A are provided to users, elevating the system from a static detection tool to a dynamic expert advisor. This design addresses the long-standing need in this field for "explainability" of AI decision-making. Compared with traditional static knowledge base queries, it achieves a leap from "knowing what" to "knowing why," significantly improving the reliability of diagnostic results, user trust, and the practical value of the system.

[0051] A cloud-based analytics pipeline based on microservices and asynchronous tasks was adopted, resolving the performance bottlenecks and single-point-of-failure risks that may arise in a single-service architecture when facing high-concurrency user requests. By deploying the detection and classification models as independent microservices and employing asynchronous processing and message notification mechanisms for large files or batch tasks, the system ensures high reliability, high scalability, and a smooth user experience. The innovative design addressing the engineering challenges faced by core diagnostic methods in real-world application scenarios ensures that the system can stably and efficiently serve a large user base.

[0052] This invention possesses highly efficient parasite detection and analysis capabilities. At the research level, it supports teams in accurately quantifying the infection level of parasites in biological samples and enabling rapid and accurate species identification, providing reliable data support for parasite population dynamics, epidemiological surveys, and related basic research. In practical applications, the software is suitable for health management in livestock production, enabling early screening and monitoring of common parasitic diseases in livestock farms. Simultaneously, in pet clinical diagnosis and treatment, it helps improve the efficiency and accuracy of parasitic disease diagnosis, providing a technical basis for targeted prevention and treatment, and has significant practical implications for protecting animal health and promoting public health security.

[0053] This invention has been implemented in the "Research and Application of Prevention and Control Technology for Livestock and Poultry Parasitic Diseases" project under the "14th Five-Year Plan" National Key Research and Development Program (Project No.: 2023YFD1801200). The "Research and Application of Prevention and Control Technology for Livestock and Poultry Parasitic Diseases" project received financial support of RMB 24.5 million, and the project execution period is from December 2023 to December 2027.

[0054] The technical means disclosed in this invention are not limited to those disclosed in the above embodiments, but also include technical solutions composed of any combination of the above technical features. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of this invention, and these improvements and modifications are also considered within the scope of protection of this invention.

Claims

1. An intelligent diagnostic system for animal parasites based on artificial intelligence, characterized in that: This includes a mobile terminal and a cloud server, which are connected in communication. The mobile terminal is equipped with an image acquisition module, an image upload module, and a detection result display module. The image acquisition module is used to acquire microscopic images of the parasite to be analyzed; The image upload module is used to encrypt and encapsulate image data and upload it to the cloud server via the HTTPS protocol; The test result display module is used to receive and display test results from the cloud server; The cloud server is equipped with trained parasite target detection models and parasite species classification models. The parasite target detection model is used to extract deep features from parasite microscopic images, output the bounding box coordinates of all suspected parasites, crop out the suspected regions based on the bounding box coordinates, and input them into the parasite species classification model. The parasite classification model is used to determine the authenticity of parasites and identify the species, and then sends the results to the mobile terminal.

2. The intelligent diagnostic system for animal parasites based on artificial intelligence according to claim 1, characterized in that: The image acquisition module includes either photo capture and upload or album selection and upload. The photo capture and upload method uses the built-in camera of the mobile phone to connect to a microscopic imaging device to capture microscopic images of parasites. The album selection and upload method involves selecting microscopic images stored in the album and uploading them.

3. The intelligent diagnostic system for animal parasites based on artificial intelligence according to claim 1, characterized in that: The cloud server also includes a knowledge base module and an AI assistant module. The knowledge base module is used to store parasite prevention and control knowledge data, and the AI ​​assistant module generates interpretable natural language answers based on the detection results, using a large language model and the knowledge base module.

4. The intelligent diagnostic system for animal parasites based on artificial intelligence according to claim 1, characterized in that: The cloud service adopts a microservice architecture, deploying the parasite target detection model and the parasite species classification model as independent microservices, and communicating with each other through message queues.

5. The intelligent diagnostic system for animal parasites based on artificial intelligence according to claim 1, characterized in that: The parasite target detection model is deployed on a cloud server in ONNX format. The parasite target detection model is a lightweight YOLO11n-SCDNet model, which includes a backbone network, a neck network, and a detection head. The backbone network uses the ShuffleNetV2 structure for feature extraction, and the neck network includes a cross-scale channel fusion module (CCFM) for multi-scale feature fusion. The detection head uses a dual convolutional structure (C3DC) for target detection.

6. The intelligent diagnostic system for animal parasites based on artificial intelligence according to claim 1, characterized in that: The parasite classification model is a target detection model based on a convolutional neural network.

7. The intelligent diagnostic system for animal parasites based on artificial intelligence according to claim 1, characterized in that: The cloud server supports asynchronous processing mode. For batch tasks, the task ID is returned immediately after uploading, and the user is notified via push notification after processing is completed.

8. An intelligent diagnostic method for animal parasites based on artificial intelligence, the method being used in the system described in any one of claims 1 to 7, characterized in that: Includes the following steps: S1): The mobile terminal uploads the collected microscopic images to the cloud server. S2): The cloud server calls the pre-trained parasite detection model, uses the parasite detection model to extract deep features from the input microscopic image, identifies and locates the suspected parasite regions in the image, and outputs the bounding box coordinates of each suspected region. S3): The cloud server will crop out each suspected region based on the bounding box coordinate information, and input the image patch of each suspected region into the pre-trained parasite classification model. The parasite classification model will perform true / false discrimination and species identification for each candidate region.

9. The AI-based intelligent diagnostic method for animal parasites according to claim 8, characterized in that: The pre-trained parasite detection model also includes a dataset construction process, which includes data collection and standardized labeling: The data collection includes the following steps: T1): Fecal samples were collected from cattle and sheep farms and dog and cat breeding bases. All samples were pretreated by water washing sedimentation method and saturated sucrose flotation method. Then, positive samples for parasites were screened by microscopic examination. T2: Professional veterinary technicians used optical microscopes to capture high-resolution microscopic images of parasite-positive samples at magnifications of 200x and 400x. The image size was 2464×2056 pixels, with a resolution of 300 dpi. A total of 7,120 effective microscopic images were obtained, covering eight common parasite species: Giardia, Eimeria, Isospora, Entamoeba, Toxocara, Ancylostoma, Trichuris, and Moniezia. The standardized labeling includes the following steps: T3): All microscopic images were manually annotated by two independent veterinary researchers using the LabelImg tool to ensure the accuracy and consistency of image annotation; T4): Each parasite oocyst or parasite body in each image is precisely selected and drawn with a rectangular bounding box that closely fits its morphological boundary. The images are then labeled according to the type of parasite. All labeling results are reviewed and confirmed by experienced parasitologists to ensure that the labeling quality meets high standards. T5): The annotation information is stored in the standard format supported by the YOLO object detection model. Each image corresponds to a text annotation file, which contains the bounding box information of all objects in the image. Each bounding box record includes: class index and normalized coordinate parameters, namely the x and y coordinates of the center point of the bounding box, width and height, all of which are normalized relative to the image size.

10. The AI-based intelligent diagnostic method for animal parasites according to claim 8, characterized in that: The method further includes receiving a user's AI assistant consultation request regarding the detection results; sending the detection context information to the cloud AI assistant module; the cloud AI assistant module calling a large language model and combining it with a parasite knowledge base to generate an interpretable answer; and returning the answer to the mobile terminal for display.

Images