Pet disease detection method and system based on near-infrared image time sequence change analysis
By analyzing the temporal changes in near-infrared images, the problems of lag in pet disease detection and hair occlusion were solved, enabling early and objective disease warning and personalized health monitoring, thus assisting veterinarians in diagnosis.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- SUZHOU JIDITAI BIOTECHNOLOGY CO LTD
- Filing Date
- 2026-01-29
- Publication Date
- 2026-04-24
AI Technical Summary
Existing methods for detecting pet diseases suffer from problems such as lag, subjectivity, susceptibility to obstruction by fur, and high costs, making it difficult to achieve early and objective detection of physiological changes.
A method based on near-infrared image temporal change analysis is adopted. Through near-infrared image acquisition, image preprocessing, hair interference removal, temporal feature extraction and disease identification, combined with deep learning and thermal imaging, change maps are generated and disease early warning is provided.
It enables non-contact, non-invasive early disease warning, improves the objectivity and sensitivity of detection, reduces subjective misjudgment, and provides personalized health tracking and assisted veterinary diagnosis.
Smart Images
Figure CN121922401A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of pet health monitoring and intelligent diagnostic technology, and more specifically, to a pet disease detection method and system based on near-infrared image temporal change analysis. Background Technology
[0002] As pets gain more status in households, the demand for their health monitoring is increasing. Currently, the detection of pet diseases mainly relies on pet owners' visual observation (such as noticing lethargy or loss of appetite) and regular veterinary checkups. These methods have significant shortcomings:
[0003] 1. Delay: Owners usually only notice when their pets show obvious clinical symptoms, which may cause them to miss the best opportunity for early intervention.
[0004] 2. Subjectivity: It relies on personal experience and makes it difficult to detect subtle changes.
[0005] 3. Many interfering factors: A pet's thick fur can obscure key physical features such as changes in skin color, swelling, and rashes, making visual diagnosis extremely difficult.
[0006] 4. High cost: Frequent veterinary clinic check-ups are time-consuming and costly.
[0007] In existing technologies, there are some solutions for monitoring pets using cameras, but most of them are based on visible light images, which are greatly affected by ambient light and hair occlusion, and are mostly used for behavioral analysis (such as activity level), making it difficult to directly use them for the detection of physiological indicators and deep tissue changes.
[0008] Therefore, there is an urgent need for an intelligent detection method that is non-contact, stress-free, penetrates hair interference, and can provide early disease warnings through subtle changes. Summary of the Invention
[0009] The purpose of this invention is to overcome the shortcomings of existing technologies and to propose a method and system for detecting pet diseases based on the temporal variation analysis of near-infrared images.
[0010] To address the aforementioned technical problems, the present invention adopts the following technical solution:
[0011] A pet disease detection method based on near-infrared image temporal variation analysis includes the following steps:
[0012] S1: Image Acquisition: Using a near-infrared image acquisition device, take regular or continuous photos of specific parts of the pet while it is in a quiet or standard posture to obtain the original near-infrared image sequence.
[0013] S2: Image preprocessing and hair interference removal: Each original image is standardized and image segmentation methods are used to reduce or remove interference caused by pet hair in the image, and the tissue background layer reflecting subcutaneous tissue information is extracted.
[0014] S3: Temporal change feature extraction: Register the processed temporal images, calculate the feature value changes of each pixel or region of interest in the image sequence at different time points, and generate a change map;
[0015] S4: Disease Identification and Early Warning: The extracted temporal change features are input into a pre-trained classification model, the identification results are output, and the detection report and early warning information are pushed to the pet owner through the user terminal.
[0016] As a further aspect of the present invention: in step S2, the image segmentation method includes a method based on frequency domain filtering or a deep learning segmentation network.
[0017] As a further aspect of the present invention: the deep learning segmentation network is a U-Net network with an attention mechanism, used to separate the hair layer and tissue layer from the original near-infrared image.
[0018] As a further aspect of the present invention: before first use or during the pet's health status confirmation stage, a personalized baseline establishment step is performed: when the pet is in good health, multiple sets of near-infrared images are collected, and statistical analysis is used to establish the pet's individual health baseline image and normal fluctuation range of characteristics, which serves as a comparison benchmark for subsequent time-series change analysis.
[0019] As a further aspect of the present invention: the near-infrared image acquisition device also integrates a thermal imaging module to fuse and analyze near-infrared images and thermal imaging images as part of the temporal variation characteristics.
[0020] As a further aspect of the present invention: the feature value change includes at least one of grayscale value, texture feature, and spectral feature.
[0021] As a further aspect of the present invention, the classification model is trained based on a large amount of labeled near-infrared time-series images of pets in healthy and diseased states, and is able to identify feature change patterns associated with specific diseases.
[0022] The present invention also provides a pet disease detection system for implementing the above method, comprising:
[0023] Near-infrared image acquisition terminal, used to capture and upload near-infrared images of specific parts of a pet;
[0024] The processing unit, located in a cloud server or local computing device, includes an image preprocessing module, a hair removal module, a temporal analysis module, and a disease identification module. The image preprocessing module performs image normalization and hair interference removal; the temporal analysis module performs image registration, feature extraction, and change map generation; and the disease identification module has a built-in pre-trained classification model for disease identification based on temporal features.
[0025] The user interaction terminal is used to receive and display test reports and early warning information, and to manage pet health records.
[0026] As a further aspect of the present invention: the near-infrared image acquisition terminal is an intelligent device or a near-infrared camera equipped with a near-infrared light source and sensor.
[0027] As a further aspect of the present invention: the user interaction terminal is a mobile APP or a web-based health management platform.
[0028] Compared with the prior art, the advantages of this invention are:
[0029] 1. Deep information acquisition: By utilizing the penetrating power of near-infrared light, bypassing hair obstruction, it directly acquires physiological information of subcutaneous tissue and microcirculation, breaking through the bottleneck of visible light detection.
[0030] 2. High-sensitivity early warning: Focusing on "change analysis", it can capture subtle physiological deviations that are difficult to detect in a single examination and develop slowly, achieving true early detection and early warning.
[0031] 3. Objective quantitative assessment: Transforming health status into calculable and traceable image features and time series data greatly reduces subjective experience-based misjudgments.
[0032] 4. Pet-friendly monitoring: The entire process is non-contact, non-invasive, and stress-free, making it easy to integrate into a pet's daily life and achieve routine family health monitoring.
[0033] 5. Combining intelligence and personalization: By establishing individual baselines, the system can adapt to the huge differences in pets of different breeds, sizes, and coat colors, and provide accurate personalized health tracking and risk warnings.
[0034] 6. Diagnostic Value: The generated quantitative reports and change atlases can provide veterinarians with valuable objective imaging evidence, assisting them in making faster and more accurate clinical diagnoses. Attached Figure Description
[0035] Figure 1 This is a flowchart illustrating the overall workflow of the method of the present invention.
[0036] Figure 2The following diagram illustrates the comparison between image preprocessing and hair removal: (a) original near-infrared image, (b) tissue image after hair removal.
[0037] Figure 3 The process of extracting temporal variation features is illustrated, and images at different time points and their synthesized variation maps are displayed.
[0038] Figure 4 This is a block diagram of the overall structure of the system of the present invention. Detailed Implementation
[0039] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. 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 without creative effort are within the scope of protection of the present invention.
[0040] Example 1
[0041] Please see Figure 1 A method for detecting pet diseases based on temporal variation analysis of near-infrared images includes the following steps:
[0042] S1: Image Acquisition: Using a near-infrared image acquisition device equipped with a near-infrared light source and a near-infrared sensor, take regular (e.g., daily or weekly) or continuous photos of specific parts of the pet (such as the abdomen, limb joints, ear flaps, gums, etc.) while the pet is in a quiet or standard posture, and obtain an original time-series image sequence consisting of near-infrared images from at least two time points.
[0043] S2: Image preprocessing and hair removal:
[0044] S2.1: Perform standardized preprocessing on each raw near-infrared image, including grayscale conversion, noise suppression, and illumination uniformity correction;
[0045] S2.2: Image segmentation methods are used to reduce or remove texture interference caused by pet hair in the image, and to extract the tissue background layer reflecting the subcutaneous tissue structure and physiological information. The image segmentation method is either a frequency-domain high-pass filtering method or a deep learning-based semantic segmentation method. The deep learning segmentation network is a U-Net network with an attention mechanism, used to separate the hair layer and tissue layer from the original near-infrared image.
[0046] S3: Extraction of temporal variation features:
[0047] S3.1: Perform pixel-level registration on the processed temporal images to ensure that images of the same body part at different time points are spatially aligned;
[0048] S3.2: Calculate the change of at least one feature value for each pixel or preset region of interest in the image sequence at different time points. The feature value changes include gray intensity (reflecting changes in tissue density and blood volume), texture features (such as local binary patterns, reflecting changes in tissue structure), and spectral features (used to estimate parameters such as blood oxygen saturation).
[0049] S3.3: Generate a visual change map based on feature changes to identify regions and trends (intensification or weakening) of significant physiological changes within a selected time window (e.g., the most recent week). Figure 3 As shown.
[0050] S4: Disease Identification and Early Warning
[0051] S4.1: The extracted temporal variation features are input into a pre-trained classification model. This model is trained on a large amount of labeled near-infrared temporal image data of pets in healthy and diseased states, and is capable of identifying feature change patterns associated with specific diseases. The classification model is one or more combinations of convolutional neural networks, recurrent neural networks, temporal convolutional networks, or Transformer architectures, and transfer learning or incremental learning strategies are used for model updates and optimization.
[0052] S4.2: Based on the output of the classification model, generate a test report that includes health status assessment, abnormal site indication, suspected disease type, and recommended medical treatment level;
[0053] S4.3: Push the detection report and warning information to the pet owner through the user interaction terminal.
[0054] In addition, before first use or during the pet's health status confirmation phase, a personalized baseline establishment step is performed: While the pet is in good health, multiple sets of near-infrared images are acquired, and statistical analysis is used to establish the individual pet's health baseline image and normal fluctuation range of characteristics, serving as a comparison benchmark for subsequent time-series change analysis. The near-infrared image acquisition device also integrates a thermal imaging module, fusing and analyzing near-infrared images with thermal images as part of the time-series change characteristics.
[0055] Example 2
[0056] like Figure 4 As shown, this embodiment provides a pet disease detection system that implements the above method, including:
[0057] Near-infrared image acquisition terminal, used to capture and upload near-infrared images of specific parts of a pet;
[0058] The processing unit, located in a cloud server or local computing device, includes an image preprocessing module, a hair removal module, a temporal analysis module, and a disease identification module. The image preprocessing module performs image normalization and hair interference removal; the temporal analysis module performs image registration, feature extraction, and change map generation; and the disease identification module has a built-in pre-trained classification model for disease identification based on temporal features.
[0059] The user interaction terminal is used to receive and display test reports and early warning information, and to manage pet health records.
[0060] The near-infrared image acquisition terminal is a smart device or near-infrared camera equipped with a near-infrared light source and sensor. The user interaction terminal is a mobile APP or a web-based health management platform.
[0061] System Applications and Examples of Clinical Auxiliary Diagnosis
[0062] This embodiment uses the monitoring of arthritis in a pet dog in a home environment as an example, and applies the above-mentioned methods and systems to specifically illustrate the application process:
[0063] 1. Equipment Preparation and Data Acquisition: Pet owners use a smart stand equipped with a near-infrared camera of a specific wavelength (e.g., 850nm). Every day, when the dog is relatively quiet before sleeping, place the stand in front of its limb joints (e.g., the right elbow joint) to automatically take pictures and acquire near-infrared images of that area.
[0064] 2. Data Upload: The captured images are automatically uploaded to the cloud server via the home network.
[0065] 3. Processing and Analysis: The cloud processing unit immediately initiates the analysis process;
[0066] Image preprocessing and hair removal: The system first performs standardization preprocessing (grayscale conversion, noise reduction, etc.) on the near-infrared image of the right elbow joint uploaded that day. Then, it calls a pre-trained U-Net hair removal model with an attention mechanism to process the image, ultimately obtaining a clear "tissue background layer" image reflecting the subcutaneous joint capsule and surrounding soft tissue, such as... Figure 2 As shown.
[0067] Time-series comparison and analysis: The system performs high-precision pixel-level registration of the processed image of the current day with a personalized health baseline image library established for the dog over the previous 30 consecutive days. The time-series analysis module then performs feature calculations on the registered image sequence and finds that two key indicators of a specific joint capsule region show significant trend changes in the past 5 days: first, the near-infrared absorption rate, which reflects blood volume and edema, continues to rise; second, the texture features characterizing tissue homogeneity become blurred, which is consistent with the pathological changes of tissue exudation caused by inflammation.
[0068] 4. Output Results: The extracted temporal variation features are input into a pre-trained early arthritis identification and classification model (e.g., built based on a temporal convolutional network). After comparing feature patterns, the model determines that the current change matches the pattern of early arthritis lesions with 85% accuracy. The system immediately generates a structured detection report and pushes a warning message to the pet owner via the linked mobile app: "Tip: Persistent inflammatory signal changes have been detected in the right elbow joint area of your pet. It is recommended to reduce strenuous exercise and pay attention to its gait. If limping occurs, please seek medical attention promptly."
[0069] 5. Veterinary Reference: Pet owners can bring the detailed reports (including "change graphs", characteristic trend curves and quantitative data) provided in the mobile app to the veterinarian as an important objective diagnostic reference to help the veterinarian quickly locate the problem.
[0070] Through the above process, this invention enables early, objective, convenient, and home-based intelligent monitoring and early warning of diseases such as arthritis in pets, providing innovative technical means for pet health management.
[0071] The above description is merely a preferred embodiment of the present invention; however, the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and its improved concepts, should be covered within the scope of protection of the present invention.
Claims
1. A method for detecting pet diseases based on temporal variation analysis of near-infrared images, characterized in that, Includes the following steps: S1: Image Acquisition: Using a near-infrared image acquisition device, take regular or continuous photos of specific parts of the pet while it is in a quiet or standard posture to obtain the original near-infrared image sequence. S2: Image preprocessing and hair interference removal: Each original image is standardized and image segmentation methods are used to reduce or remove interference caused by pet hair in the image, and the tissue background layer reflecting subcutaneous tissue information is extracted. S3: Temporal change feature extraction: Register the processed temporal images, calculate the feature value changes of each pixel or region of interest in the image sequence at different time points, and generate a change map; S4: Disease Identification and Early Warning: The extracted temporal change features are input into a pre-trained classification model, the identification results are output, and the detection report and early warning information are pushed to the pet owner through the user terminal.
2. The pet disease detection method based on near-infrared image temporal change analysis according to claim 1, characterized in that: In step S2, the image segmentation method includes methods based on frequency domain filtering or deep learning segmentation networks.
3. The pet disease detection method and system based on near-infrared image temporal change analysis according to claim 2, characterized in that: The deep learning segmentation network is a U-Net network that incorporates an attention mechanism, used to separate the hair layer and tissue layer from the raw near-infrared image.
4. The pet disease detection method based on near-infrared image temporal change analysis according to claim 1, characterized in that: Before first use or during the pet's health status confirmation phase, a personalized baseline establishment step is performed: under the pet's health status, multiple sets of near-infrared images are collected, and statistical analysis is used to establish the pet's individual health baseline image and normal fluctuation range of characteristics, which serves as a comparison benchmark for subsequent time-series change analysis.
5. The pet disease detection method based on near-infrared image temporal change analysis according to claim 1, characterized in that: The near-infrared image acquisition device also integrates a thermal imaging module to fuse and analyze near-infrared images and thermal images as part of the temporal variation characteristics.
6. The pet disease detection method based on near-infrared image temporal change analysis according to claim 1, characterized in that: The feature value changes include at least one of grayscale value, texture feature, and spectral feature.
7. The pet disease detection method and system based on near-infrared image temporal change analysis according to claim 1, characterized in that: The classification model is trained on a large amount of labeled near-infrared time-series images of pets in healthy and diseased states, and is able to identify feature change patterns associated with specific diseases.
8. A pet disease detection system implementing the method as described in any one of claims 1 to 7, characterized in that, include: Near-infrared image acquisition terminal, used to capture and upload near-infrared images of specific parts of a pet; The processing unit, located in a cloud server or local computing device, includes an image preprocessing module, a hair removal module, a temporal analysis module, and a disease identification module. The image preprocessing module performs image normalization and hair interference removal; the temporal analysis module performs image registration, feature extraction, and change map generation; and the disease identification module has a built-in pre-trained classification model for disease identification based on temporal features. The user interaction terminal is used to receive and display test reports and early warning information, and to manage pet health records.
9. The pet disease detection system according to claim 8, characterized in that: The near-infrared image acquisition terminal is an intelligent device or a near-infrared camera equipped with a near-infrared light source and sensor.
10. The pet disease detection system according to claim 8, characterized in that: The user interaction terminal is a mobile APP or a web-based health management platform.