Livestock claim settlement data analysis method and device based on artificial intelligence, and medium

Abstract

The invention relates to the field of image processing and artificial intelligence, in particular to a livestock claim data analysis method and device based on artificial intelligence and a medium. The method comprises the following steps: acquiring a claim settlement image uploaded by a user, extracting angular points and straight lines in the image by using a feature point detection algorithm and Hough transform, and judging the normalization and shooting angle deviation of a checkerboard reference substance; and segmenting the animal image through the deep learning segmentation model, removing the shelter in combination with curvature analysis, carrying out geometric correction on the animal image, and carrying out claim settlement operation according to the corrected animal image. The automation and accuracy of livestock claim settlement data processing can be improved, and errors and fraud risks in manual operation are reduced.

Description

Technical Field

[0001] The present invention relates to the fields of artificial intelligence and finance, and in particular to an artificial intelligence-based livestock claims data analysis method, system, device and medium. Background Art

[0002] In the existing livestock insurance claims process, policyholders are usually required to submit images of dead livestock as a basis for claims. However, since the images uploaded by policyholders may contain irregular shooting angles, obstructions, or human manipulation, such as covering part of the livestock, using objects of similar colors to camouflage and increase the size of the livestock, insurance companies are unable to accurately assess the actual condition of the livestock and the amount of claims. This phenomenon not only increases the workload of insurance claims, but also provides opportunities for potential fraud, damaging the interests of insurance companies.

[0003] Traditional claims processes mostly rely on manual review, which is not only time-consuming and labor-intensive, but also subject to subjective judgment errors, and it is difficult to effectively solve the problems of occlusion, distortion and irregularity in images. Although some technologies have adopted simple image processing methods, it is still difficult to guarantee its accuracy and reliability when faced with complex scenes and diverse insurance images, which can easily lead to misjudgment or fraud risks.

[0004] Therefore, there is an urgent need for an AI-based livestock claims data analysis method that can automatically detect and process irregular behaviors in images through advanced image processing, deep learning, and intelligent analysis technologies, accurately identify potential occlusion or fraud risks, and further improve the accuracy and automation level of claims data processing, ensuring that insurance companies can make fair and accurate claims decisions. Summary of the invention

[0005] The present invention provides an analysis method, device and medium for livestock claims data based on artificial intelligence, so as to solve the problem of how to accurately identify the fraud risks caused by non-standard images, occlusion or human manipulation in livestock claims scenarios by intelligently analyzing livestock images and reference object images uploaded by users, thereby improving the accuracy and automation level of claims data processing.

[0006] In order to solve the above technical problems, the present invention provides an analysis method of livestock claims data based on artificial intelligence, comprising:

[0007] Acquire a claim image uploaded by a user, wherein the claim image includes an animal image and a checkerboard reference object;

[0008] Extracting corner points and straight lines in the chessboard reference object by using a feature point detection algorithm and Hough transform;

[0009] Determine whether the checkerboard reference object is standard according to the corner points and the straight lines, and obtain a checkerboard standardization judgment result and a shooting angle deviation parameter;

[0010] Using a deep learning image segmentation model to segment the animal image to obtain a segmentation result curve;

[0011] Performing curvature analysis on the segmentation result curve to remove occluders on the animal image;

[0012] The animal image is corrected according to the checkerboard normativeness judgment result and the shooting angle deviation parameter, and the claim settlement operation is performed based on the corrected animal image.

[0013] Further, the determining whether the checkerboard reference object is standard according to the corner points and the straight lines specifically includes:

[0014] Extracting corner points near the straight line as candidate corner points of the chessboard;

[0015] Determine the straight lines in the two axial directions of the chessboard according to the candidate corner points of the chessboard and the straight lines, and obtain small squares formed by the intersection of the straight lines;

[0016] Calculate the homography matrix corresponding to the small squares obtained by the intersection of all straight lines;

[0017] When the difference between the eigenvalue and the eigenvector of the homography matrix of any two small squares exceeds a preset threshold, it is determined that the small squares of the chessboard reference object are not standardized;

[0018] When there is no at least one corner point near any small square, it is determined that there is an obstruction on the chessboard reference object.

[0019] Furthermore, after calculating the homography matrix corresponding to the small squares obtained by the intersection of all straight lines, the method further includes:

[0020] Get the point pairs with similar homography matrices and integrate them to get the total homography matrix;

[0021] Decomposing the rotation angle of the camera posture relative to the horizontal axis and the rotation angle relative to the vertical axis according to the total homography matrix;

[0022] Obtaining the roll angle and pitch angle of the camera deflection and combining them with the rotation angle of the horizontal axis and the rotation angle of the vertical axis to calculate the deviation Δθ of the camera shooting angle;

[0023] When the shooting angle deviation Δθ is greater than a preset angle threshold, it is confirmed that the checkerboard reference object is tilted.

[0024] Further, determining whether the checkerboard reference object is standardized according to the corner points and the straight lines includes:

[0025] Obtain the four outermost corner points of the chessboard reference object;

[0026] Calculate the actual physical length represented by each point on the horizontal midline and the vertical midline of the animal image;

[0027] The geometric standardization of the entire checkerboard reference object is determined based on the physical length between two points on the edge of the checkerboard reference object and the distance ratio between the midpoints of the animal image.

[0028] Furthermore, determining the geometric standardization of the entire chessboard according to the physical length between two points on the edge of the chessboard reference object and the distance ratio between the midpoints of the animal image includes:

[0029] By comparing the deviations of the physical length and distance ratio, determining whether the checkerboard reference object is tilted or deformed;

[0030] When the deviation exceeds a preset threshold, it is confirmed that the checkerboard reference object is tilted or deformed.

[0031] Furthermore, the determining of the geometric standardization of the entire chessboard includes:

[0032] Calculate the sum of two vanishing points on both sides of the chessboard reference object;

[0033] Calculate the direction and magnitude of each point of the horizontal midline and the vertical midline according to the vanishing point;

[0034] Calculating a local Riemannian metric in the animal image through the direction and the modulus, and obtaining a deformation matrix based on the local Riemannian metric;

[0035] Comparing the local Riemann metric results of the deformation matrix and the standard rectangle, if it is found that the difference between the eigenvalue and the eigenvector exceeds a preset threshold, it is determined that the chessboard as a whole has geometric irregularities;

[0036] When the deviation value of the local Riemann metric exceeds a preset error range, the geometric regularity of the checkerboard reference object is corrected according to the deformation matrix.

[0037] Furthermore, the curvature analysis of the segmentation result curve is performed to remove the occluders on the animal image, specifically including:

[0038] Discretize the segmentation result curve to obtain discrete curve segments, and calculate the curvature of each discrete curve segment;

[0039] A splitting threshold is set according to the maximum value of the curvature, and discrete curve segments greater than the splitting threshold are removed;

[0040] Enclosing the remaining discrete curve segments into a closed curve, and calculating the centroid and major axis of the closed curve;

[0041] Based on the long axis direction, determining the animal part to which the removed discrete curve segment corresponds;

[0042] According to the tangent vector difference, determine whether the tangent vector difference of adjacent curve segments meets the merging condition;

[0043] If the tangent vector difference is less than a preset threshold, the adjacent discrete curve segments are merged;

[0044] Otherwise, the corresponding discrete curve segment is discarded.

[0045] Furthermore, the present invention also provides an artificial intelligence-based livestock claims data device, including an acquisition module, an extraction module, a specification determination module, a segmentation module, a curvature analysis module, an occlusion removal module, and a discrimination module;

[0046] The acquisition module is used to acquire the claim settlement image uploaded by the user, wherein the claim settlement image includes an animal image and a checkerboard reference object;

[0047] An extraction module, used for extracting corner points and straight lines in the chessboard reference object by using a feature point detection algorithm and Hough transform;

[0048] A standard determination module, used to determine whether the chessboard is standard according to the corner points and the straight lines, and obtain a chessboard standard determination result and a shooting angle deviation parameter;

[0049] A segmentation module, used to perform segmentation processing on the animal image using a deep learning image segmentation model to obtain a segmentation result curve;

[0050] A curvature analysis module, used to perform curvature analysis on the segmentation result curve to remove occluders on the animal image;

[0051] A discrimination module is used to correct the animal image with the obstructions removed according to the checkerboard normative discrimination result and the shooting angle deviation parameter, and perform a claim settlement operation based on the corrected animal image.

[0052] Furthermore, the present invention also provides a computer device, including a memory and a processor, wherein the memory stores computer-readable instructions, and when the processor executes the computer-readable instructions, the steps of the livestock claims data analysis method based on artificial intelligence as described above are implemented.

[0053] Furthermore, the present invention also provides a storage medium having a computer program stored thereon, wherein the computer program is called and executed by a computer to implement the above-mentioned method for analyzing livestock claims data based on artificial intelligence.

[0054] The key innovative features of the present invention include:

[0055] (1) Checkerboard reference object standardization detection and shooting angle correction: The present invention realizes automatic checkerboard standardization detection and shooting angle deviation correction through feature point detection and Hough transform algorithm, ensuring that the claims image meets the standards and reducing errors caused by improper shooting.

[0056] (2) Curvature analysis-based occlusion removal technology: Occlusions are identified and automatically removed through curvature analysis, avoiding intelligent weighing errors caused by occlusions and greatly improving the accuracy of claims image processing.

[0057] (3) Application of deep learning image segmentation technology: The system uses deep learning models to accurately segment livestock images, providing high-quality basic data for subsequent intelligent weighing and fraud risk assessment, further improving the accuracy and automation level of the claims process.

[0058] The following are its main beneficial effects:

[0059] (1) Automated image processing and correction: Through feature point detection algorithms and Hough transform, the system can accurately extract corner points and straight lines in the image. It can quickly determine the standardization of the checkerboard reference object and correct the shooting angle error, avoiding misjudgment caused by non-standard images and improving the efficiency and accuracy of claims image processing.

[0060] (2) Deep learning segmentation and occlusion removal: Using the deep learning image segmentation model, the system can accurately segment the outline of livestock and automatically detect and remove occlusions in the image through curvature analysis. This method effectively avoids errors in the claim amount caused by occlusions or human interference, ensuring the accuracy of subsequent intelligent weighing.

[0061] (3) Intelligent weighing and data consistency: Based on image segmentation and geometric correction, the present invention ensures that the physical size of the livestock image is consistent with the actual proportion, ensuring that the system obtains accurate livestock volume and weight data in the subsequent intelligent weighing process, reducing errors caused by shooting angles or image distortion. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] Figure 1 A flowchart of an artificial intelligence-based livestock claims data analysis method provided in an embodiment of the present application;

[0063] Figure 2 A structural block diagram of an analysis device for livestock claims data based on artificial intelligence provided in an embodiment of the present application;

[0064] Figure 3A structural block diagram of a computer device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0065] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by technicians in the technical field of the present application; the terms used in the specification of the application herein are only for the purpose of describing specific embodiments and are not intended to limit the present application; the terms "including" and "having" and any variations thereof in the specification and claims of the present application and the above-mentioned drawings are intended to cover non-exclusive inclusions. The terms "first", "second", etc. in the specification and claims of the present application or the above-mentioned drawings are used to distinguish different objects, not to describe a specific order.

[0066] Reference to "embodiments" herein means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present application. The appearance of the phrase in various locations in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0067] Embodiment 1:

[0068] Reference Figure 1 , is a flow chart of an analysis method for livestock claims data based on artificial intelligence provided by an embodiment of the present invention. The flow chart of the analysis method for livestock claims data based on artificial intelligence may at least include steps S100-S600:

[0069] S100: Acquire a claim image uploaded by a user, wherein the claim image includes an animal image and a checkerboard reference object.

[0070] Specifically, the system receives the claim images uploaded by the insured customer through the smart terminal device, which include livestock (such as dead pigs) and standard reference objects (chessboard reference objects). In this process, the system performs preliminary data analysis on the uploaded images and extracts the pixel resolution P of the images. x and P y , the width of the image W img and height H img As the initial input, the standard reference is a checkerboard pattern with known dimensions, with a checkerboard width of 20 cm, and the system uses this reference as the basis for image calibration.

[0071] S200, extracting corner points and straight lines in the chessboard reference object through a feature point detection algorithm and Hough transform.

[0072] Specifically, the system first receives the claim images uploaded by the user, which include animal images and checkerboard reference objects. The Harris corner detection algorithm is used to extract the corner points of each small square in the checkerboard, and the Hough transform is used to extract the straight lines on the checkerboard boundary to ensure the integrity of the geometric structure of the checkerboard. Based on the corner point and straight line data, the system establishes the geometric framework of the checkerboard as the basis for subsequent normative judgment and shooting angle correction.

[0073] Specifically, feature point detection: Harris corner point detection algorithm is used to extract the corner points of the small squares on the chessboard from the claim image (including the animal image and the chessboard reference). Each corner point represents the vertex coordinates of a small square on the chessboard, and the system ensures that the corner points of all small squares are detected.

[0074] Line detection: The Hough transform algorithm is used to extract the boundary lines of the chessboard from the claim image. Based on the detected line data, the system establishes 8 straight line frames in the two axial directions of the chessboard (4 in each direction) to ensure the integrity of the geometric structure of the chessboard.

[0075] Combination of feature points and straight lines: Combine corner points and straight line data to generate a checkerboard geometric framework for subsequent normative judgment and shooting angle correction.

[0076] In step S200, the system uses the Harris corner detection algorithm to extract the corner points of the checkerboard reference in the claim image, and uses the Hough transform to detect the boundary lines of the checkerboard. The specific process is as follows:

[0077] The Harris corner detection algorithm is used to extract the corner points in the checkerboard reference to ensure that the vertices of each small square are detected.

[0078] Use Hough transform to extract straight lines in the image, determine the straight lines in the two axial directions of the chessboard, and ensure that there are 4 straight lines in each direction.

[0079] Corner point and line data are combined to generate structural information of the chessboard in preparation for normative judgment.

[0080] S300, determining whether the chessboard is standard according to the corner points and the straight lines, and obtaining a chessboard standardization determination result and a shooting angle deviation parameter.

[0081] The system compares the geometric features of the small squares in the chessboard to determine whether there are size inconsistencies or occlusions. It then calculates the overall geometric regularity of the chessboard and corrects the geometric distortion of the image based on the shooting angle to ensure that the physical size of the animal in the claim image is consistent with the actual situation.

[0082] The feature point detection algorithm is the Harris corner detection algorithm: it is used to extract corner point information from the image to help identify the geometric features of the chessboard and the corner point positions of each small square.

[0083] Hough transform is Hough transform: It is used to extract straight line information from an image and determine the boundaries and overall regularity of the checkerboard by detecting the straight line structure of the checkerboard.

[0084] The results of “feature point detection” and “line detection” are combined to generate the overall framework of the chessboard, which is further used for normative analysis and shooting angle correction.

[0085] After acquiring the claim image, the system further extracts the camera sensor data from the insured customer’s camera device, including the camera’s focal length f, sensor size S x and S y , and the horizontal angle of the shot θ H and the vertical angle θ V These parameters, combined with the basic information of the image, become the key reference in the subsequent geometric correction.

[0086] Specifically, the system uses the following formula to preliminarily calculate the horizontal and vertical field of view angles of the claims image. The horizontal and vertical field of view angles are used to calculate the camera's shooting angle and field of view, and are parameters in the sensor data:

[0087]

[0088] Among them, θ H is the horizontal field of view; θ V is the vertical field of view; S x and S y are the horizontal and vertical dimensions of the camera sensor, respectively; f is the focal length of the camera. These data will be used to correct the shooting angle in the image to ensure the physical proportions in the image are accurate.

[0089] Specifically, checkerboard standardization judgment: the system determines whether the checkerboard is standard by calculating the geometric features of each small square in the checkerboard reference image. Specifically, if it is detected that the side lengths of the small squares in the checkerboard are inconsistent, or the angle deviation is too large, the system will prompt that there is a problem with the geometric standardization of the checkerboard.

[0090] Get the field of view parameters:

[0091] Horizontal field of view (HFOV) and vertical field of view (VFOV): After extracting the checkerboard normative information, the system further calculates the horizontal and vertical field of view angles in combination with the camera sensor parameters (focal length, sensor size, etc.);

[0092] Calculation of shooting angle correction parameters: After extracting the horizontal and vertical field of view angles, the system further corrects the shooting angle in the image in combination with the geometric norms of the chessboard. Through correction, the perspective distortion caused by the deviation of the shooting angle can be corrected, thereby ensuring that the physical size of the animal in the claim image is consistent with the actual proportion.

[0093] Generation of normative results and shooting angle deviation parameters: Finally, the system generates checkerboard normative judgment results and shooting angle deviation parameters based on the above calculation results, providing data support for subsequent geometric correction.

[0094] In step S300, the system determines whether the chessboard reference object is standardized based on the extracted corner points and straight lines, mainly through the following steps:

[0095] Extracting corner points near the straight line as candidate corner points of the chessboard;

[0096] Determine the straight lines in the two axial directions of the chessboard according to the candidate corner points of the chessboard and the straight lines, and obtain small squares formed by the intersection of the straight lines;

[0097] Calculate the homography matrix corresponding to the small squares obtained by the intersection of all straight lines;

[0098] When the difference between the eigenvalue and the eigenvector of the homography matrix of any two small squares exceeds a preset threshold, it is determined that the small squares of the chessboard reference object are not standardized;

[0099] When there is no at least one corner point near any small square, it is determined that there is an obstruction on the chessboard reference object.

[0100] Specifically, the geometric features of each small square in the chessboard are compared to determine whether there is any irregularity (such as inconsistent sizes of small squares or obstruction of the chessboard).

[0101] This application ensures that the physical size in the image is consistent with the actual size by calculating the standardization of the chessboard and the deviation of the shooting angle.

[0102] In addition, the focal length, field of view and shooting angle in the camera sensor data can be extracted to further correct the geometric distortion in the claims image.

[0103] (I) Checkerboard detection and standardization judgment

[0104] Specific steps for judging the standardization of chessboard:

[0105] Obtain the corner points and straight lines of the chessboard reference object, and determine whether the chessboard meets the specifications based on the detected corner points and straight lines.

[0106] Compare the geometric features of the small squares where the straight lines intersect to determine whether there are any irregularities on the chessboard.

[0107] Refinement of geometric regularity calculations:

[0108] The outermost corner points of the checkerboard reference are extracted to calculate the actual physical lengths on the horizontal and vertical center lines.

[0109] Based on the calculation results of the vanishing point, the geometric regularity of the image is judged by direction and modulus.

[0110] These data are used to perform comprehensive correction on the checkerboard and determine whether further correction of geometric distortion in the image is needed.

[0111] Tilt and deformation judgment of chessboard:

[0112] By calculating the deviation between the physical length of the chessboard and the distance ratio of the midpoint of the image, it is determined whether the chessboard is tilted or deformed.

[0113] If the deviation exceeds the preset threshold, the system prompts that there is a risk of tilt or deformation of the checkerboard reference.

[0114] Calibration steps:

[0115] After calculating the geometric regularity, the system corrects the geometric distortion of the image based on the shooting angle and physical feature parameters to ensure that the size ratio of the checkerboard and animal images conforms to the actual situation.

[0116] This embodiment provides a method for determining the standardization based on the checkerboard corner point and straight line information. The system first extracts the corner point and straight line information of the checkerboard reference object from the claim settlement image, and determines whether the checkerboard reference object is standard through the following steps:

[0117] The system uses the Harris corner point detection algorithm to extract the corner point information of all small squares in the chessboard reference object, and uses the corner points near the straight line as candidate points for the chessboard corner points. The candidate corner points are used to define the vertex positions of each small square in the chessboard reference object, so as to further determine whether the chessboard is standardized.

[0118] The system uses Hough transform to extract the boundary straight lines of the chessboard from the claim image, and determines the straight lines in the two axial directions of the chessboard based on the candidate corner points near the straight lines. There are 4 straight lines in each direction, and a total of 8 straight lines form the geometric framework of the chessboard. The system calculates the intersection points of all straight lines based on the positions of the corner points and straight lines, and obtains the positions and sizes of all small squares.

[0119] The system calculates the homography matrix through the vertex coordinates of each small square. The homography matrix is ​​used to describe the geometric characteristics of the small squares in the image. For each small square, the system generates a homography matrix to describe the geometric shape of the small square.

[0120] The system compares the homography matrix eigenvalues ​​and eigenvectors of all small squares. If the difference between the homography matrix eigenvalues ​​and eigenvectors of any two small squares exceeds a preset threshold, the checkerboard reference object is judged to be non-standard.

[0121] If no valid corner point is detected near any small square, the system prompts that there may be an obstruction or dirt on the chessboard reference.

[0122] Finally, the system generates a normative judgment result of the checkerboard reference object based on the homography matrix characteristics of the small squares and the distribution of the corner points, and outputs a risk warning of whether there are occlusions or checkerboard deformation.

[0123] In another embodiment, after calculating the homography matrices of all small squares, the system further integrates point pairs with similar homography matrices and calculates the deviation of the shooting angle through the following steps:

[0124] The system obtains point pairs with similar homography matrices between all adjacent small squares and integrates these point pairs. The integrated point pairs are used to construct the overall homography matrix of the chessboard, which represents the geometric form of the entire chessboard and the spatial relationship in the image.

[0125] The system decomposes the rotation angles of the camera posture in the horizontal direction and the vertical direction according to the total homography matrix. Specifically, the system obtains the rotation angles of the camera posture in the horizontal direction (horizontal axis) and the vertical direction (vertical axis) by performing SVD (singular value decomposition) on the homography matrix.

[0126] The system obtains the roll angle and pitch angle of the camera, and calculates the total deviation of the camera at the current shooting angle by combining the rotation angles of the horizontal axis and the vertical axis;

[0127] When the shooting angle deviation is greater than a preset angle threshold, the system prompts that the checkerboard reference object may be tilted or deformed during the shooting process. The system records this deviation value and marks it as "possible shooting angle deviation".

[0128] In another embodiment, the system further calculates the actual physical length represented by each point on the horizontal midline and vertical midline of the animal image based on the extracted checkerboard corner points and straight lines. The specific steps are as follows:

[0129] The system extracts the four outermost corner points of the chessboard reference object to determine the overall position and size of the chessboard in the claim image. These four corner points can define the minimum enclosing rectangle of the chessboard and serve as the basis for calculating the horizontal centerline and vertical centerline.

[0130] The system calculates the actual physical length of each point on the horizontal center line and the vertical center line according to the geometric position of the checkerboard reference in the claim image. The system calculates the actual physical length of each point in the claim image by the known actual size of the checkerboard (e.g., the side length of each small square) and the proportional relationship between the coordinates of the point in the image.

[0131] The system determines the geometric regularity of the chessboard by comparing the physical length between two points on the chessboard edge and the distance ratio of the image midpoint. If the physical length and the distance ratio of the image midpoint deviate greatly, it means that the chessboard may be tilted or deformed. The system will record this deviation and use it for subsequent tilt correction and geometric deformation judgment.

[0132] This embodiment can effectively determine whether the chessboard is tilted or deformed during the shooting process by calculating the proportional relationship between the actual physical length of the chessboard and the claim image, thereby providing data support for further image geometry correction.

[0133] In another embodiment, based on the above-mentioned geometric normative judgment, this embodiment further analyzes the inclination and deformation of the chessboard in detail, and completes it through the following steps:

[0134] The system determines whether the chessboard is tilted or deformed by comparing the physical length of the chessboard and the proportional deviation of the distance between the midpoints of the claim image. Specifically, the system calculates the physical length of the two points on the edge of the chessboard and combines it with the proportional relationship in the image. If it is found that these proportions exceed the threshold value from the preset standard deviation, the system will prompt that the chessboard is at risk of tilting or deforming.

[0135] The system presets a deviation threshold of the ratio of physical length to the distance between the midpoint of the image. When the deviation value exceeds the threshold, the system considers that the checkerboard reference object is seriously tilted or deformed, and prompts that the image may not meet the standard.

[0136] When the system determines the tilt and deformation of the chessboard, it will generate corresponding normative judgment results and record all non-standard areas. The system will use these results for subsequent image geometry correction and claim image normative verification.

[0137] This embodiment effectively determines the possible tilt and deformation of the chessboard during the shooting process by accurately calculating the physical length of the chessboard reference object and the proportion relationship of the claims image, providing an important data basis for subsequent image correction and normative analysis.

[0138] In image processing, the standardization of the chessboard is a key detection step, which mainly includes the following irregularities:

[0139] The chessboard's small squares are not of the same size: that is, the size of each small square changes.

[0140] The overall size of the chessboard does not meet the standard: the overall size of the chessboard has a large deviation (this problem will be dealt with in the subsequent steps).

[0141] The chessboard is blocked or dirty by debris: Part of the chessboard is blocked.

[0142] The system first determines the first and third situations, and the specific operations are as follows:

[0143] (1) The Harris corner point extraction algorithm is used to extract the corner point information in the image to ensure that every small square in the image can pass the corner point detection.

[0144] (2) Use Hough transform to detect straight lines in the image and determine the boundaries of the chessboard.

[0145] (3) Select suitable corner points near the straight line extracted in step (2) as candidate points.

[0146] (4) Based on the corner points and straight lines in step (3), determine eight straight lines in the two axes of the chessboard, four in each axis, to ensure that the chessboard has clear geometric constraints in both directions.

[0147] (5) Calculate the intersection points of these straight lines and generate a homography matrix Hi for each small square (where i = 1, 2, ..., 16 represents each small square). If the eigenvalues ​​or eigenvectors of the homography matrices of any two small squares are found to be too different, the system prompts that the sizes of the small squares on the chessboard are inconsistent (problem 1); if no valid corner point is detected near the corner point of any small square, the system prompts that the chessboard may be blocked or dirty by debris (problem 3).

[0148] The system extracts the precise position information of the chessboard through this step, and generates a judgment result on whether it is standardized based on the above judgment.

[0149] (II) Extracting shooting angle and reference object parameters

[0150] The system further calculates the position and angle of the chessboard, combining the sensor data with the shooting roll angle θr and pitch angle θ p , calculate the camera's shooting height H using the following formula:

[0151] H=0.5×L / (tan(θ H ×0.5+θr )+tan(θ H ×0.5-θ r ))+0.5×W / (tan(θ V ×0.5+θ p )+tan(θ V ×0.5-θ p ))

[0152] The formula is based on the actual length L and width W of the chessboard, combined with the camera's field of view and deflection angle, to calculate the camera's shooting height. If the calculated height value deviates significantly from the reasonable range (such as more than 2 meters), the system prompts that the chessboard size may not be standardized. After this step, the system generates a complete image standardization report, including whether the chessboard is standardized, whether the shooting angle is reasonable, whether the reference object is blocked, etc., for subsequent geometric correction and intelligent evaluation.

[0153] The system first uses a deep learning model to extract image features and identify the pig's body contour and the boundary features of the checkerboard reference object. The feature extraction results include the checkerboard corner point information, edge straight lines, and the preliminary outline of the pig's body area. pig and chessboard feature point P grid .

[0154] Specifically, the system first receives the image data from step S100, which includes the pre-processed uploaded livestock image and related parameters of the chessboard, including the image width W img , height H img , the size of the chessboard L real and W real , and the camera’s shooting angle θ r and θ p .

[0155] The system then performs nonlinear mapping on each feature point based on the activation function in the deep learning model to generate a segmentation map of the pig body and background. pig It will be used in the subsequent geometric correction step.

[0156] The output data generated by this step includes the pixel information P of the pig body contour pig And the pixel information P of the checkerboard area grid , these two data will be used to correct the geometric distortion of the pig body.

[0157] Based on the geometric correction algorithm, the image distortion generated during the shooting process is corrected.

[0158] S400: Segment the animal image using a deep learning image segmentation model to obtain a segmentation result curve.

[0159] Specifically, after completing the image processing and geometric correction, the system receives the corrected image data, including the pig's contour information P pig and the corrected chessboard information P grid Based on these data, the system starts to detect occluded areas.

[0160] The system first analyzes the brightness, color and texture features in the pig image to identify possible occlusion areas. By gradually analyzing the edge areas in the image and combining the high-dimensional features extracted by the convolutional neural network (CNN), the system compares the normal contour information of the pig with the original image. pig and the heterogeneous features of external occluders to generate a mask M of the area where occluders may exist occlude .

[0161] Occlusion area mask M occlude is defined as:

[0162]

[0163] Where x and y represent the pixel coordinates of the image. The generated occlusion area mask M occlude It will be used for subsequent segmentation processing.

[0164] The deep learning image segmentation model is used to accurately segment the pig body area.

[0165] After completing the occluded area detection, the system further segments the pig body area based on the deep learning image segmentation model. Specifically, the system uses a convolutional neural network (such as U-Net or MaskR-CNN) to perform global and local feature analysis on the image and extract the edge and internal structure information of the pig body.

[0166] By applying the segmentation network on the feature map F(x,y), the system accurately segments the boundaries of the pig body at the pixel level and generates the final segmentation result S pig :

[0167]

[0168] In addition, the system combines the occlusion area mask M occlude The segmentation results are optimized to remove the occluded parts and correct the incomplete pig body contour caused by occlusion. pig It will be used for occlusion detection and removal in subsequent steps.

[0169] Based on the pig body contour information P extracted in step S400 pig and chessboard pixel information P gridThe system further corrects the geometric distortion in the captured image. In particular, the system uses a geometric correction algorithm to correct the perspective distortion and angle distortion of the camera to ensure that the actual proportion of the pig in the image is consistent with its real size.

[0170] First, the system calculates the actual image perspective transformation matrix T based on the camera parameters obtained by the sensor. persp The matrix is ​​calculated based on the camera's viewing angle and the chessboard position, and is used to correct the perspective distortion in the image into a standard plane image.

[0171] The formula for perspective transformation is:

[0172]

[0173] Among them, a, b, c, d, e, f, g, h, 1 are the coefficients of perspective transformation. The system calculates the perspective transformation based on the actual physical size L of the chessboard. real ,W real and pixel size P x ,P y Calculate these parameters.

[0174] Furthermore, the system combines the pig body contour P pig and the transformation matrix T persp , the distorted pig body image is restored to a true-scale image, ensuring that the geometric size of the pig body is consistent with reality.

[0175] In step S400, the system further extracts the pig body contour information and the complete pixel information of the checkerboard reference object based on the previously extracted checkerboard corner point and straight line information combined with the deep learning image segmentation model.

[0176] The specific operations are as follows:

[0177] Feature extraction and standardization of chessboard reference objects: The system first uses the Harris corner detection algorithm to extract the corner information of the chessboard, and uses Hough transform to extract the boundary lines of the chessboard. This information is used to construct the basic geometric framework of the chessboard.

[0178] Based on the geometric framework, the system uses deep learning models to further extract the complete pixel information of the chessboard. Through deep learning models such as convolutional neural networks (CNN), the system can identify the boundary features of the chessboard, the complete shape of each small square and its pixel position relationship.

[0179] Feature extraction of pig body contour and removal of occlusions: The system segments animal images based on deep learning image segmentation models (such as U-Net and Mask R-CNN), generates segmentation result curves, and optimizes the segmentation results in combination with curvature analysis to identify and remove occlusions in pig body images.

[0180] Finally, the system generates complete pig body contour data, which includes the precise boundaries and shapes of the pig body and its pixel position relationship in the image.

[0181] Comprehensive analysis and geometric correction of chessboard and pig body information: After extracting the complete pixel information of the pig body contour and chessboard, the system combines the two parts of data together to correct the geometric distortion.

[0182] By analyzing the geometric regularity of the chessboard (such as the size consistency and shape integrity of the small squares) and its relative position to the pig's body contour, the system can accurately calculate the perspective distortion and shooting angle deviation in the image.

[0183] Finally, the system uses the chessboard as a standard reference object and performs geometric correction on the pig's body contour through the perspective transformation matrix (P′=H×P) to ensure that the physical proportions of the pig's body are consistent with its actual size.

[0184] Verification of geometric correction results: The system verifies the accuracy of the correction results by comparing the corrected pig body contour with the standard size of the chessboard. If the corrected pig body contour is consistent with the actual physical size, the system outputs the final correction image and uses it for subsequent intelligent weighing and fraud risk assessment.

[0185] Verify the standardization of the checkerboard reference. Specifically:

[0186] (I) Checkerboard detection and standardization judgment

[0187] Based on the extracted checkerboard corner points and straight lines, the system first checks the standardization of the checkerboard to ensure that it is not inconsistent in size or blocked by debris. i Test each small square:

[0188] The system extracts the corner points and straight lines of each small square, and determines whether each area of ​​the chessboard is standardized through Hough transform and Harris corner point extraction algorithm.

[0189] By calculating the homography matrix H of the chessboard i If the matrix eigenvalues ​​of different squares are quite different, it indicates that the sizes of the small squares on the chessboard are inconsistent; if some square corners are missing, it indicates that the chessboard is blocked by debris.

[0190] (II) Determination of the inclination of chessboard placement

[0191] Even if the chessboard itself is regular, it may be tilted when placed, causing the system to misjudge. The system uses the obtained homography matrix point pairs to calculate the overall homography matrix H0, and obtains the rotation angle θ of the camera posture relative to the horizontal axis and the vertical axis through matrix decomposition. x and θ y .

[0192] The formula is as follows:

[0193]

[0194] The system decomposes H0 to obtain the camera's rotation angle θ x and θ y :

[0195]

[0196] Here a, b, d, e are the decomposition coefficients of the homography matrix H0.

[0197] θ x and θ y The roll angle θ in the sensor data r and pitch angle θ p For comparison, if

[0198] max(|θ x -θ r |,|θ y -θ p |)>2°

[0199] The system prompts that the chessboard may be tilted, causing deviations in AI estimation.

[0200] In another implementation, determining the geometric regularity of the entire chessboard includes:

[0201] Calculate the two vanishing points on both sides of the chessboard reference object and

[0202] Calculate the direction and magnitude of each point of the horizontal midline and the vertical midline according to the vanishing point;

[0203] Calculating a local Riemannian metric in the animal image through the direction and the modulus, and obtaining a deformation matrix based on the local Riemannian metric;

[0204] Comparing the local Riemann metric results of the deformation matrix and the standard rectangle, if it is found that the difference between the eigenvalue and the eigenvector exceeds a preset threshold, it is determined that the chessboard as a whole has geometric irregularities;

[0205] When the deviation value of the local Riemann metric exceeds a preset error range, the geometric regularity of the checkerboard reference object is corrected according to the deformation matrix.

[0206] Specifically, according to the four outermost corner points of the chessboard detected by the feature point detection algorithm and Hough transform in S200, the corner points in the chessboard reference object and the straight line are extracted. and Calculate the actual physical length of each point on the horizontal and vertical midlines of the image. In order to accurately calculate the length, we first calculate the two vanishing points on both sides of the chessboard. and Then calculate each point on the horizontal and vertical midline of the image The local Riemann metric of and The direction and magnitude of The direction is The direction is Model Model

[0207] in It is a straight line and straight line The intersection of It is a straight line and straight line The intersection of du0 and dv0 is the lower left corner of the chessboard. exist and The physical size corresponding to the unit pixel in the direction is in cm. The actual physical size corresponding to the image width imgW and height imgH are L and W respectively.

[0208]

[0209] (2) Extract the camera's built-in FOV (field of view) parameters. For some cameras that do not provide FOV parameters, use the equivalent focal length f to calculate the FOV

[0210] (3) According to the FOV and the camera sensor size, calculate the horizontal field of view angle HFOV, which we set as θ H

[0211] And the vertical field of view angle VFOV, we let it be θ V

[0212] (4) Extract the two roll angles θ of the camera r and pitch angle θ p, use the following formula to calculate the camera shooting height H:

[0213] H=0.5*L / (tan(θ H *0.5+θ r )+tan(θ H *0.5-θ r ))+0.5*W / (tan(θ v *0.5+θ p )+tan(θ v *0.5-θ p ))

[0214] (5) We conduct statistics based on the data set and combine it with common sense in physics. If H>200, that is, the camera height is greater than 2 meters, we believe that the chessboard size is not standardized, which causes the basic measurement to be smaller, and the estimated camera height is greater than 2 meters. Therefore, we prompt whether the risk chessboard is irregular and small.

[0215] S500: Perform curvature analysis on the segmentation result curve to remove occluders on the animal image.

[0216] Furthermore, the step includes:

[0217] Discretize the segmentation result curve to obtain discrete curve segments, and calculate the curvature of each discrete curve segment;

[0218] A splitting threshold is set according to the maximum value of the curvature, and discrete curve segments greater than the splitting threshold are removed;

[0219] Enclosing the remaining discrete curve segments into a closed curve, and calculating the centroid and major axis of the closed curve;

[0220] Based on the long axis direction, determining the animal part to which the removed discrete curve segment corresponds;

[0221] According to the tangent vector difference, determine whether the tangent vector difference of adjacent curve segments meets the merging condition;

[0222] If the tangent vector difference is less than a preset threshold, the adjacent discrete curve segments are merged;

[0223] Otherwise, the corresponding discrete curve segment is discarded.

[0224] "Tangent vector difference" refers to the difference in angle or direction between the tangent vectors of adjacent curve segments. It is used to describe the geometric relationship between adjacent curve segments and determine whether these curve segments are continuous or have abrupt changes. If the difference in tangent vectors of adjacent curve segments is small, they are considered to be part of the same overall contour; if the difference is large, they may correspond to different objects or geometric discontinuities caused by occlusion.

[0225] When removing possible occluders in animal images, the shape and direction of the curve are important features for identifying occluders. Therefore, the tangent vector difference, as an important indicator for measuring the continuity between curve segments, can help the system automatically determine the location of the occluder and process it.

[0226] In the curvature analysis process, the system first discretizes the segmentation result curve. Assume that the discretization of the segmentation result curve C(t) is represented by a set of discrete points {P i}(where (i=1,2,...,n) represents the i-th discrete point on the curve), each discrete point P i =(x i ,y i ) corresponds to a position on the curve.

[0227] For each discrete point P i , the system can obtain the tangent vector of the point by calculating the difference between the two points before and after it:

[0228]

[0229] Among them, T i is the tangent vector of the i-th discrete point. The tangent vector describes the local direction of the curve at that point.

[0230] For two adjacent discrete curve segments (such as discrete curve segments A and B), the system can determine the continuity between them by calculating the difference in their corresponding tangent vectors. Suppose a point on discrete curve segment A is P i , the corresponding tangent vector is T A ; A point on discrete curve segment B is P j , the corresponding tangent vector is T B .

[0231] The system can use the following formula to calculate the angle θ between the tangent vectors to describe the tangent vector difference:

[0232] in:

[0233] T A ·T B represents the dot product of two tangent vectors; ‖T A ‖·‖·T B ‖represent the modulus of the tangent vector respectively.

[0234] If the value of the tangent vector difference θ is small (for example, less than a preset threshold), it means that the two curve segments have similar directions, and they can be considered to belong to the same coherent overall contour and should be merged; if the value of θ is large, it means that the directions of the two curve segments are quite different, and there may be discontinuities (such as occlusions or sudden changes in contours), which need to be processed separately.

[0235] In the process of removing occlusions from animal images, the specific application of tangent vector differences is as follows:

[0236] The system discretizes the segmentation result curve, generates a set of discrete points, and calculates the tangent vector of each point.

[0237] For every two adjacent discrete curve segments, the system calculates the difference in tangent vectors of their corresponding points;

[0238] If the tangent vector difference is less than a preset threshold, the adjacent discrete curve segments are merged and considered to belong to the same overall contour;

[0239] If the tangent vector difference is greater than a preset threshold, the curve segment is discarded, considering that it may correspond to an occluder or a discontinuous part.

[0240] After all adjacent tangent vector differences are processed, the system generates a closed curve surrounded by the remaining discrete curve segments and calculates the centroid and major axis of the closed curve.

[0241] The system further determines the animal part (such as ears, tail, etc.) that the discarded curve segment may correspond to based on the direction of the long axis. If it is significantly different from the main area, the curve segment is considered to be an occluder.

[0242] The threshold of the tangent vector difference is a key parameter that affects the accuracy of occlusion removal. The system can optimize the threshold according to the specific application scenario and data set. Generally, different thresholds can be set through experiments to select the best threshold that can retain the effective contour and remove interference to the greatest extent.

[0243] In this embodiment, the default value of the tangent vector difference threshold is set to 10°, that is, if the tangent vector difference angle of adjacent curve segments is less than 10°, they are considered to belong to the same overall contour; if it is greater than 10°, they are considered to be occluders or other interferences. Users can also dynamically adjust this parameter value according to actual needs to adapt to different claims image processing scenarios.

[0244] In order to solve the pig body occlusion problem that may be caused intentionally or unintentionally by the insured user in the claim scenario, the system proposes a segment splitting and re-merging algorithm to determine whether the pig body is occluded and make corrections. The specific steps are as follows:

[0245] (1) Discretization of segmentation result curve. The system discretizes the segmented pig body contour curve, selects 200 discrete points at a fixed interval, and calculates the curvature of each point. The calculation formula of curvature κ(x, y) is:

[0246]

[0247] (2) Setting the curvature threshold and splitting. Calculate the maximum curvature of the 200 points and set the curvature threshold to 0.5 times the maximum curvature. Based on all points greater than the curvature threshold, the system splits the discrete curve segments to separate the areas that may be blocked.

[0248] (3) Remove the curve segments and calculate the centroid and major axis of the closed curve. The system removes the curve segments corresponding to the points greater than the curvature threshold and calculates the centroid and major axis of the closed area enclosed by the remaining curves. This closed area represents the main part of the pig body that is not obstructed.

[0249] (4) Determine the body part corresponding to the removed curve segment. Based on the long axis direction, the system determines whether the part corresponding to the removed curve segment is a pig ear, pig tail, pig leg, etc. This step helps to distinguish the nature of the occluded part so as to proceed to the next step of processing.

[0250] (5) Merging of curve segments. Based on the roughly determined pig ears and pig tails, the system further determines whether the curve segments between these body parts need to be merged. If the difference between the tangent vectors of adjacent curve segments is very small, the two curve segments are not discarded and are merged into a continuous area; if the difference is large, the two curve segments are discarded, indicating that the part is a possible occluder.

[0251] Points with larger curvature correspond to possible occluder boundaries. The system marks and filters these points and removes them from the final pig segmentation result, thereby obtaining a complete pig body contour after removing interference. Corrected pig body region information P final It will serve as input data for the subsequent intelligent evaluation module.

[0252] S600, correcting the image according to the checkerboard normativeness judgment result and the shooting angle deviation parameter, and performing a claim settlement operation based on the corrected animal image.

[0253] In step S600, the system performs geometric correction on the claim image based on the checkerboard normativeness judgment results and shooting angle deviation parameters of the previous steps (S300, S500) to ensure that the physical size of the animal image is consistent with the actual proportion. The specific implementation steps are as follows:

[0254] To obtain the parameters required for correction, the system extracts the corner points and straight line information of the checkerboard through the feature point detection algorithm (Harris corner point detection) and Hough transform in step S300, and judges the geometric standardization of the checkerboard and the deviation of the shooting angle. The generated checkerboard standardization judgment result and shooting angle deviation parameter are used as input data for correcting the geometric distortion of the image.

[0255] Correcting the geometric distortion of the checkerboard,The first step in the correction process is to correct for the geometric distortion of the checkerboard,reference object.

[0256] Correct the geometric distortion of the animal image. After the geometric correction of the checkerboard reference object is completed, the system performs geometric correction on the animal image based on the corrected checkerboard information.

[0257] Output and verification of correction results. After the correction is completed, the system generates the final corrected image and verifies the animal outline and physical size in the image.

[0258] The system compares the corrected animal image data with the original sensor data (such as the camera's FOV, sensor size, etc.) to ensure that the accuracy of the corrected image meets the standard requirements for claims assessment.

[0259] If the correction result shows that the animal image is consistent with the actual physical size, the system outputs the corrected claims image and uses it for subsequent intelligent weighing and fraud risk assessment.

[0260] The final application of the corrected image is that the corrected image will be further used in intelligent assessment modules, such as intelligent weighing, or provide reliable technical support for subsequent claims assessment and risk detection.

[0261] The key innovative features of the present invention include:

[0262] (1) Checkerboard reference object standardization detection and shooting angle correction: The present invention realizes automatic checkerboard standardization detection and shooting angle deviation correction through feature point detection and Hough transform algorithm, ensuring that the claims image meets the standards and reducing errors caused by improper shooting.

[0263] (2) Curvature analysis-based occlusion removal technology: Occlusions are identified and automatically removed through curvature analysis, avoiding intelligent weighing errors caused by occlusions and greatly improving the accuracy of claims image processing.

[0264] (3) Application of deep learning image segmentation technology: The system uses deep learning models to accurately segment livestock images, providing high-quality basic data for subsequent intelligent weighing and fraud risk assessment, further improving the accuracy and automation level of the claims process.

[0265] The following are its main beneficial effects:

[0266] (1) Automated image processing and correction: Through feature point detection algorithms and Hough transform, the system can accurately extract corner points and straight lines in the image. It can quickly determine the standardization of the checkerboard reference object and correct the shooting angle error, avoiding misjudgment caused by non-standard images and improving the efficiency and accuracy of claims image processing.

[0267] (2) Deep learning segmentation and occlusion removal: Using the deep learning image segmentation model, the system can accurately segment the outline of livestock and automatically detect and remove occlusions in the image through curvature analysis. This method effectively avoids errors in the claim amount caused by occlusions or human interference, ensuring the accuracy of intelligent weighing.

[0268] (3) Intelligent weighing and data consistency: Based on image segmentation and geometric correction, the present invention ensures that the physical size of the livestock image is consistent with the actual proportion, ensuring that the system obtains accurate livestock volume and weight data during the intelligent weighing process, and reducing errors caused by shooting angles or image distortion.

[0269] Embodiment 2:

[0270] Figure 2 FIG. 2 is a block diagram showing a structure of an analysis device for livestock claims data based on artificial intelligence according to an embodiment of the present invention. Figure 2 As shown, the system may include:

[0271] The acquisition module 10 is used to acquire the claim images uploaded by the user, and the claim images include animal images and checkerboard reference objects; by acquiring the data uploaded by the user, the system can acquire the required livestock claim images and provide basic data for subsequent image processing and analysis.

[0272] The extraction module 20 is used to extract corner points and straight lines in the checkerboard reference object through feature point detection algorithm and Hough transform; this module is responsible for extracting feature information of the checkerboard from the image to ensure that the system can accurately identify the structure and positioning of the checkerboard.

[0273] The standard determination module 30 is used to determine whether the chessboard is standardized based on the corner points and straight lines, and obtain the chessboard standardization judgment result and the shooting angle deviation parameter; through this module, the system can judge the standardization of the chessboard to ensure that the claim image meets the preset standards and requirements.

[0274] The segmentation module 40 is used to use a deep learning image segmentation model to perform segmentation processing on the animal image to obtain a segmentation result curve; through deep learning technology, the system can segment the animal image from the background and provide accurate animal contours for subsequent analysis and processing.

[0275] The curvature analysis module 50 is used to perform curvature analysis on the segmentation result curve and remove occlusions on the animal image. Based on the curvature analysis method, this module detects abnormal points in the animal image and removes occlusions to ensure the accuracy of the image.

[0276] The judgment module 60 is used to correct the animal image with the obstructions removed according to the checkerboard normative judgment result and the shooting angle deviation parameter, and perform a claim settlement operation based on the corrected animal image.

[0277] By utilizing a deep learning image segmentation model, this application can accurately segment the outline of livestock, automatically detect and remove obstructions in the image through curvature analysis, determine the precise outline of the animal after correction, and generate a final image for intelligent weighing.

[0278] The functions of each module in the livestock claims data analysis system of the embodiment of the present invention can be referred to the above description, wherein each module can be integrated on the same device or distributed on different devices, as long as it is ensured that each module can communicate effectively, and the embodiment of the present invention does not limit this.

[0279] Embodiment 2 of the present invention provides an artificial intelligence-based livestock claims data analysis system that can efficiently acquire and process livestock claims images uploaded by users. Through the acquisition module, extraction module, specification determination module, segmentation module, curvature analysis module and discrimination module of the system, the processing and analysis of claims images can be automatically completed. The system has a high level of automation and intelligence in terms of normative detection of claims images, deep learning image segmentation, and occlusion removal, avoiding errors caused by human intervention, while improving the accuracy and efficiency of claims data processing and reducing the time cost of insurance companies in processing livestock claims data.

[0280] Embodiment three:

[0281] Figure 3 FIG. 2 shows a block diagram of a computer device according to an embodiment of the present application. Figure 3 As shown, the computer device includes:

[0282] The memory 310 is used to store program instructions for executing the livestock claims data analysis method. These program instructions may include execution codes of algorithms and models related to data acquisition, image processing, checkerboard regularity detection, deep learning segmentation, curvature analysis, etc. The memory 310 may be a random access memory (RAM), a read-only memory (ROM), a flash memory, or other storage media.

[0283] Processor 320 is responsible for loading and executing program instructions in memory 310 to implement specific steps of livestock claim data analysis, including obtaining claim images uploaded by users, performing image segmentation, normative judgment, occlusion removal, and intelligent weighing. Processor 320 can be a central processing unit (CPU), a graphics processing unit (GPU), or an application-specific integrated circuit (ASIC), which is used to process image analysis and deep learning tasks.

[0284] The communication interface 330 is used for data interaction and communication with external devices, including receiving image data uploaded by users, transmitting processing results to the server, or transmitting data with external devices. The communication interface 330 can be a wired or wireless interface, such as an Ethernet interface, a Wi-Fi interface, or a Bluetooth module.

[0285] The memory 310, the processor 320 and the communication interface 330 are connected and communicated via a bus. The bus may be an Industrial Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus or an Extended Industrial Standard Architecture (EISA) bus, and the bus is used to transmit data, control signals and address signals. To simplify the representation, Figure 3 The bus is represented as a single line, but the actual system may contain multiple buses of different types.

[0286] Optionally, in a specific implementation, the memory 310, the processor 320 and the communication interface 330 can be integrated on a chip and communicate through an internal interface. In addition, the memory can also include a persistent storage medium to store claim data and historical processing records. The computer device in the embodiment of the present invention can realize intelligent processing and analysis of livestock claim data through efficient data processing and communication.

[0287] Embodiment 3 of the present invention provides a computer device that realizes intelligent processing of livestock claims data through the synergy of memory, processor and communication interface. The device can execute complex image processing algorithms, including checkerboard standardization detection, image segmentation and occlusion removal, to ensure the accuracy of the processing results. Through the integrated communication interface, the device can also interact with external systems to realize automatic transmission and feedback of claims data, reducing manual operations and potential errors. The device also supports multiple bus structures, can flexibly adapt to different hardware environments, and has a wide range of application scenarios.

[0288] The embodiment of the present invention provides a computer-readable storage medium, in which a computer program for executing a livestock claims data analysis method is stored. When the program is run on a computer, the following steps can be performed: obtaining a claims image uploaded by a user; extracting a checkerboard reference object in the image through a feature point detection algorithm and Hough transform; performing segmentation processing on the animal image using a deep learning image segmentation model; removing obstructions based on curvature analysis and determining the effective contour of the animal, and finally generating a segmented image for intelligent weighing.

[0289] The computer readable storage medium may be a read-only memory (ROM), random access memory (RAM), flash memory, hard disk, solid state drive, or other type of storage medium, and may be used in any computer or communication device. The program instructions in the memory are executed by a processor to implement all or part of the functions of the livestock claims data analysis method described above.

[0290] The present application also provides a chip, the chip includes: a processor and a memory, the memory stores program instructions, when the processor executes these instructions, the livestock claims data analysis method based on artificial intelligence is implemented. The processor can be a central processing unit (CPU), an application-specific integrated circuit (ASIC), a digital signal processor (DSP) or a field programmable gate array (FPGA), etc.

[0291] The above-mentioned memory may include a volatile memory or a non-volatile memory, the non-volatile memory may be a read-only memory (ROM) or a flash memory, and the volatile memory may be a random access memory (RAM), which is used to temporarily store data generated during program execution. The computer-readable storage medium and its corresponding hardware system provided by the present application can realize automated and intelligent livestock claims data processing, thereby improving the efficiency and accuracy of the claims process.

[0292] The computer-readable storage medium embodiment of the present invention provides an efficient livestock claims data processing solution, and the computer program stored in the medium can realize intelligent livestock claims data analysis when running on a computer. Through the instructions in the medium, the computer can perform complex operations such as image processing, checkerboard normative detection, segmentation and occlusion removal to ensure the accuracy and processing efficiency of image data. The storage medium enables the livestock claims process to be automated, avoids errors in traditional manual operations, improves the speed and accuracy of the claims process, reduces claims costs, and enhances the reliability of claims data.

[0293] The non-Company software tools or components appearing in the embodiments of this application are merely examples and do not represent actual use.

[0294] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), and includes a number of instructions for a terminal device (which can be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in each embodiment of the present application.

[0295] Obviously, the embodiments described above are only some embodiments of the present application, rather than all embodiments. The preferred embodiments of the present application are given in the accompanying drawings, but they do not limit the patent scope of the present application. The present application can be implemented in many different forms. On the contrary, the purpose of providing these embodiments is to make the understanding of the disclosure of the present application more thorough and comprehensive. Although the present application is described in detail with reference to the aforementioned embodiments, for those skilled in the art, it is still possible to modify the technical solutions recorded in the aforementioned specific implementation methods, or to perform equivalent replacement of some of the technical features therein. Any equivalent structure made using the contents of the specification and drawings of this application, directly or indirectly used in other related technical fields, is similarly within the scope of patent protection of this application.

Claims

1. An artificial intelligence-based livestock claims data analysis method, characterized in that: include: Acquire a claim image uploaded by a user, wherein the claim image includes an animal image and a checkerboard reference object; Extracting corner points and straight lines in the chessboard reference object by using a feature point detection algorithm and Hough transform; Determine whether the checkerboard reference object is standard according to the corner points and the straight lines, and obtain a checkerboard standardization judgment result and a shooting angle deviation parameter; Using a deep learning image segmentation model to segment the animal image to obtain a segmentation result curve; Performing curvature analysis on the segmentation result curve to remove occluders on the animal image; The animal image is corrected according to the checkerboard normativeness judgment result and the shooting angle deviation parameter, and the claim settlement operation is performed based on the corrected animal image.

2. The method according to claim 1, characterized in that The determining whether the checkerboard reference object is standard according to the corner points and the straight lines specifically includes: Extracting corner points near the straight line as candidate corner points of the chessboard; Determine the straight lines in the two axial directions of the chessboard according to the candidate corner points of the chessboard and the straight lines, and obtain small squares formed by the intersection of the straight lines; Calculate the homography matrix corresponding to the small squares obtained by the intersection of all straight lines; When the difference between the eigenvalue and the eigenvector of the homography matrix of any two small squares exceeds a preset threshold, it is determined that the small squares of the chessboard reference object are not standardized; When there is no at least one corner point near any small square, it is determined that there is an obstruction on the chessboard reference object.

3. The method according to claim 2, characterized in that After calculating the homography matrix corresponding to the small squares obtained by the intersection of all straight lines, the method further includes: Get the point pairs with similar homography matrices and integrate them to get the total homography matrix H0; According to the total homography matrix, the rotation angle θ of the camera posture relative to the horizontal axis is decomposed. X and the rotation angle θ relative to the vertical axis Y ; Get the roll angle θ of the camera deflection r and the pitch angle θ p , combined with the horizontal axis rotation angle θ X and the rotation angle θ of the vertical axis Y Calculate the deviation Δθ of the camera shooting angle; When the shooting angle deviation Δθ is greater than a preset angle threshold, it is confirmed that the checkerboard reference object is tilted.

4. The method according to claim 2, characterized in that: The step of determining whether the checkerboard reference object is standard according to the corner points and the straight lines comprises: Obtain the four outermost corner points of the chessboard reference object; Calculate the actual physical length represented by each point on the horizontal midline and the vertical midline of the animal image; The geometric standardization of the entire checkerboard reference object is determined based on the physical length between two points on the edge of the checkerboard reference object and the distance ratio between the midpoints of the animal image.

5. The method according to claim 4, characterized in that Determining the geometric standardization of the entire chessboard according to the physical length between two points on the edge of the chessboard reference object and the distance ratio between the midpoints of the animal image includes: By comparing the deviations of the physical length and distance ratio, determining whether the checkerboard reference object is tilted or deformed; When the deviation exceeds a preset threshold, it is confirmed that the checkerboard reference object is tilted or deformed.

6. The method according to claim 4, characterized in that Determining the geometric standardization of the entire chessboard includes: Calculate the two vanishing points on both sides of the chessboard reference object and Calculate the direction and magnitude of each point of the horizontal midline and the vertical midline according to the vanishing point; Calculating a local Riemannian metric in the animal image through the direction and the modulus, and obtaining a deformation matrix based on the local Riemannian metric; Comparing the local Riemann metric results of the deformation matrix and the standard rectangle, if it is found that the difference between the eigenvalue and the eigenvector exceeds a preset threshold, it is determined that the chessboard as a whole has geometric irregularities; When the deviation value of the local Riemann metric exceeds a preset error range, the geometric regularity of the checkerboard reference object is corrected according to the deformation matrix.

7. The method according to claim 1, characterized in that The performing curvature analysis on the segmentation result curve to remove the occluders on the animal image specifically includes: Discretize the segmentation result curve to obtain discrete curve segments, and calculate the curvature of each discrete curve segment; A splitting threshold is set according to the maximum value of the curvature, and discrete curve segments greater than the splitting threshold are removed; Enclosing the remaining discrete curve segments into a closed curve, and calculating the centroid and major axis of the closed curve; Based on the long axis direction, determining the animal part to which the removed discrete curve segment corresponds; According to the tangent vector difference, determine whether the tangent vector difference of adjacent curve segments meets the merging condition; If the tangent vector difference is less than a preset threshold, the adjacent discrete curve segments are merged; Otherwise, the corresponding discrete curve segment is discarded.

8. An analysis device for livestock claims data based on artificial intelligence, characterized in that: include: An acquisition module, used to acquire a claim settlement image uploaded by a user, wherein the claim settlement image includes an animal image and a checkerboard reference object; An extraction module, used for extracting corner points and straight lines in the chessboard reference object by using a feature point detection algorithm and Hough transform; A standard determination module, used to determine whether the chessboard is standard according to the corner points and the straight lines, and obtain a chessboard standard determination result and a shooting angle deviation parameter; A segmentation module, used to perform segmentation processing on the animal image using a deep learning image segmentation model to obtain a segmentation result curve; A curvature analysis module, used to perform curvature analysis on the segmentation result curve to remove occluders on the animal image; The animal image is corrected according to the checkerboard normativeness judgment result and the shooting angle deviation parameter, and the claim settlement operation is performed based on the corrected animal image.

9. A computer device, characterized in that: It comprises a memory and a processor, wherein the memory stores computer-readable instructions, and when the processor executes the computer-readable instructions, it implements the steps of the livestock claims data analysis method based on artificial intelligence as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-readable instructions, which, when executed by a processor, implement the steps of the livestock claims data analysis method based on artificial intelligence as described in any one of claims 1 to 7.

Images