Cattle body condition management method and system based on computer vision scoring

Abstract

The invention relates to the technical field of agricultural technologies, in particular to a cattle body condition management method and system based on computer vision scoring, and the method comprises the following steps: obtaining cattle image data including a whole body image and a close-range image, recording key body features in the images, removing the background interference of the images, and generating cattle key morphological features; according to the invention, through the construction of the three-dimensional point cloud model, the volume and the body type of the cattle body can be expressed more comprehensively and accurately, and after the three-dimensional point cloud model is combined with the generative adversarial network, diversified composite images can be created, so that the flexibility of body condition evaluation is improved, the feature difference of different cattle individuals can be better adapted, and the accuracy of body condition evaluation is improved. The accuracy of body condition scoring is improved, a feeding strategy can be dynamically adjusted according to a health assessment result, feed adjustment measures are provided for different growth cycles and health requirements, a management scheme is optimized in real time, resource waste is reduced, and breeding benefits and sustainability are improved.

Description

Technical Field

[0001] The present invention relates to the field of agricultural technology, and in particular to a cattle body condition management method and system based on computer vision scoring. Background Art

[0002] The field of agricultural technology is a discipline that uses modern scientific and technological means to improve and optimize agricultural production. It includes the application of information technology, sensor technology, remote sensing technology, Internet of Things, big data, artificial intelligence and other technologies in agriculture, aiming to improve agricultural production efficiency, reduce resource waste, improve the quality of crop and animal husbandry output, and promote the sustainable development of agriculture. Agricultural technology has a wide range of applications, covering crop planting, pest and disease monitoring, agricultural mechanization, precision agriculture, animal husbandry management and many other aspects.

[0003] Among them, the cattle condition management method based on computer vision scoring refers to the use of computer vision technology to automatically evaluate and manage the body condition of cattle. By analyzing the appearance characteristics of cattle, combining image processing and artificial intelligence algorithms, the health status and body shape changes of cattle can be evaluated in real time and accurately, providing a scientific basis for breeding management. The purpose is to help farmers promptly identify health problems of cattle, optimize feeding management, improve breeding efficiency, reduce costs, and increase the milk production or meat quality of cattle.

[0004] In the process of evaluating the body condition of cattle, the existing technology relies on manual intervention or static image data for health checks, lacks real-time and dynamic monitoring methods, and makes it difficult to detect health problems in a timely manner. This method has a lag in dealing with rapidly changing health conditions, especially when faced with sudden health problems, it cannot respond quickly. Most body condition scoring systems rely on standardized templates, which cannot accurately reflect the individual differences of each cow, resulting in scoring errors and affecting the accuracy of breeding management. Static image processing makes it difficult to track the dynamic process of cattle's body changes in real time, resulting in the inability to identify body shape changes or health abnormalities in the first place, increasing breeding risks. The existing technology has obvious deficiencies in dynamic adjustment in combination with environmental changes, and fails to respond in real time to the impact of environmental factors on the body condition of cattle, resulting in delayed adjustment of feeding plans, and relies on manual experience, affecting the scientificity and timeliness of management decisions. Summary of the invention

[0005] The purpose of the present invention is to solve the shortcomings of the prior art and to propose a cattle body condition management method and system based on computer vision scoring.

[0006] In order to achieve the above object, the present invention adopts the following technical scheme, a cattle body condition management method based on computer vision scoring, comprising the following steps:

[0007] S1: Obtain cattle image data, including full-body images and close-up images, record key body features in the images, remove background interference from the images, and generate key morphological features of the cattle;

[0008] S2: Based on the key morphological features of the cattle, the cattle's posture and body condition are tracked using tags, the position and state changes of the cattle are identified, and the dynamic posture change information of the cattle is analyzed through continuous image data to generate dynamic body condition tracking results;

[0009] S3: Mapping key feature points in the continuous image data to a three-dimensional coordinate system based on the dynamic tracking results of the body condition, identifying the body shape and volume of the cattle, and constructing a three-dimensional point cloud model of the cattle;

[0010] S4: Based on the three-dimensional point cloud model of the cattle, a variety of cattle body condition images are obtained by generating an adversarial network, a cattle body condition label is input, and the fit between the synthetic image and the real image is evaluated to obtain a body condition synthetic image data set;

[0011] S5: using the body condition synthetic image dataset in combination with computer vision, identifying the characteristics of the cattle's body shape and muscle distribution, calculating the body condition score of each cattle, and evaluating the health status to obtain a score deviation record;

[0012] S6: Based on the score deviation record, according to the change of cattle's body condition over time and combined with feeding environment information, the trend of cattle's body condition change is predicted, and the feed amount and feeding method are dynamically adjusted to obtain a cattle body condition management plan.

[0013] As a further scheme of the present invention, the key morphological characteristics of the cattle include the head outline, trunk length, limb angles, and tail position of the cattle; the dynamic body condition tracking results include the cattle's walking speed, standing time, lying frequency, and eating behavior patterns; the cattle's three-dimensional point cloud model includes the cattle's body length, height, and width three-dimensional coordinate points, and connection relationships; the body condition synthetic image data set includes side views, front views, and back views of cattle in differentiated body conditions; the scoring deviation records include each cow's weight, muscle mass, fat ratio, and overall body shape rating; the cattle body condition management plan includes feed formula adjustment, feeding frequency, and feeding time optimization plan.

[0014] As a further solution of the present invention, the steps of obtaining cattle image data, including full-body images and close-up images, recording key body features in the images, and removing background interference from the images to generate key morphological features of the cattle are specifically as follows:

[0015] S101: Acquire cattle image data, including full-body images and close-up images, perform image cropping, identify and extract the contour of the cattle body, remove background areas unrelated to the cattle, perform boundary detection and blank area filling, and generate a cattle body image;

[0016] S102: Based on the body image of the cattle, identify key parts of the cattle body, calibrate the positions of the head, limbs, tail and back, and supplement each part to generate a body feature image;

[0017] S103: Using the body feature image, adjusting the image contrast and sharpness, performing highlight processing on the body features, optimizing the clarity of the morphological features in the image, and generating key morphological features of the cattle.

[0018] As a further solution of the present invention, according to the key morphological features of the cattle, the posture and body condition of the cattle are tracked by using tags, the position and state changes of the cattle are identified, and the dynamic posture change information of the cattle is analyzed through continuous image data to generate the dynamic tracking results of the body condition. Specifically, the steps are as follows:

[0019] S201: Based on the key morphological features of the cattle, the posture labels of the cattle are calibrated, multiple body parts and positions of the cattle are identified, the head, trunk, limbs and tail are marked, and the position of the cattle is identified through spatial mapping to generate a cattle posture label record;

[0020] S202: continuously tracking the dynamic changes of each posture of the cattle through the cattle posture label record, extracting the label information in each frame image, analyzing the position change through the difference between images, and quantifying the change of the cattle's body posture to obtain the cattle's position state change data;

[0021] S203: Performing body condition analysis based on the cattle position state change data, recording the characteristics of body condition-related changes by comparing the posture changes and position movements of continuous images, and generating body condition dynamic tracking results based on the changing trends of position and posture.

[0022] As a further solution of the present invention, the key feature points in the continuous image data are mapped to a three-dimensional coordinate system through the dynamic tracking results of the body condition, the body shape and volume of the cattle are identified, and the steps of constructing a three-dimensional point cloud model of the cattle are specifically as follows:

[0023] S301: extract key points from continuous images using a corner point detection algorithm based on the dynamic tracking results of the body condition, map the key points to a three-dimensional coordinate system through geometric transformation, perform spatial coordinate transformation based on the proportions and positional relationships of the differentiated parts of the cow's body, and generate a three-dimensional key point set of the cow;

[0024] S302: using the three-dimensional key point set of the cow, performing three-dimensional interpolation processing on the key feature points, constructing the contour line of the cow's body, and smoothing the surface of the differentiated parts of the body to generate a three-dimensional contour map of the cow;

[0025] S303: Based on the three-dimensional outline image of the cow, the characteristic points of the cow are connected, the body shape and volume of the cow are identified and estimated, and a three-dimensional point cloud model of the cow is constructed.

[0026] As a further solution of the present invention, the formula of the corner detection algorithm is as follows:

[0027]

[0028] Among them, S new is the corner response, det(M) represents the second-order matrix determinant of the local area of ​​the image, Tr(M) represents the trace of the matrix M, k is an empirical constant, Var(M) is the variance of the local area of ​​the image, Sha(M) is the sharpness of the image, is the magnitude of the image gradient, Area(M) is the area of ​​the local image region, and A1, A2, and A3 are weight coefficients.

[0029] As a further solution of the present invention, based on the three-dimensional point cloud model of the cattle, a variety of cattle body condition images are obtained by generating an adversarial network, cattle body condition labels are input, and the fit between the synthetic image and the real image is evaluated. The steps of obtaining a body condition synthetic image data set are specifically as follows:

[0030] S401: using the three-dimensional point cloud model of the cattle, adopting a generative adversarial network, using a generator to synthesize an image of the cattle's body condition, and by adjusting the random noise and cattle's body condition label input by the network, recording the differentiated body shape and body condition, and obtaining diversified cattle's body condition images;

[0031] S402: Based on the diversified cattle body condition images, input the real cattle body condition labels, use the discriminator to evaluate the fit between the synthetic image and the real image, and adjust the parameters of the generative adversarial network by comparing the texture, structure and difference of the images, optimize the quality of the synthetic image, and generate a body condition synthetic image fit evaluation record;

[0032] S403: Using the body condition composite image fitting evaluation record, screening composite image data, selecting image samples with preferred fitting, and annotating and classifying the composite images to obtain a body condition composite image data set.

[0033] As a further solution of the present invention, the steps of using the body condition synthetic image dataset in combination with computer vision to identify the characteristics of cattle body shape and muscle distribution, calculate the body condition score of each cattle, and evaluate the health status to obtain the score deviation record are specifically as follows:

[0034] S501: using the body condition synthetic image data set in combination with computer vision technology, by training a convolutional neural network model, identifying key features of the cattle's body shape and muscle distribution, extracting information on body shape, muscle density and fat distribution in the image, and generating a record of the cattle's body shape and muscle distribution features;

[0035] S502: Based on the body shape and muscle distribution feature records of the cattle, a support vector machine algorithm is used to perform data classification and quantitative evaluation of body condition scores according to set body shape, muscle distribution and fat layer thickness indicators to obtain a body condition score value of the cattle;

[0036] S503: Using the cattle body condition score value, the health status of each cattle is evaluated, and the health level of each cattle is determined by comparing the body condition score with a preset health standard, and a score deviation record is generated.

[0037] As a further solution of the present invention, the formula of the support vector machine algorithm is as follows:

[0038]

[0039] Among them, BS is the body condition score of the cattle, S1 represents the body shape score, S2 represents the muscle distribution score, S3 represents the fat layer thickness, w1, w2 and w3 are weight coefficients, w4 is the adjustment coefficient, σ1, σ2 and σ3 are the standard deviations of the body shape score, muscle distribution score and fat layer thickness data respectively.

[0040] As a further solution of the present invention, based on the score deviation record, according to the change of cattle's physical condition over time, combined with the feeding environment information, the physical condition change trend of cattle is predicted, and the feed amount and feeding method are dynamically adjusted to obtain the cattle physical condition management plan. Specifically, the steps are as follows:

[0041] S601: Analyze the trend of the cattle's body condition over time based on the score deviation record, and use a time series analysis method to predict the trend of the cattle's body condition changes in the future time period in combination with the influencing factors of the feeding environment, including temperature, humidity and activity, and obtain cattle body condition change trend data;

[0042] S602: Verify that the body condition of the cattle is within a healthy range by combining the cattle body condition change trend data with the cattle feeding environment information, including feed type, feeding frequency, and environmental temperature and humidity, and generate a feeding environment adjustment plan;

[0043] S603: Utilizing the feeding environment adjustment plan, dynamically adjusting the real-time feeding plan, and generating a cattle body condition management plan by adjusting the feed amount and feeding method through real-time monitoring of cattle body condition change data.

[0044] A cattle body condition management system based on computer vision scoring, the cattle body condition management system based on computer vision scoring is used to execute the above-mentioned cattle body condition management method based on computer vision scoring, the system comprises:

[0045] The image data recording module obtains the whole body image of the cattle and close-up images from multiple angles, automatically removes background interference in the image, extracts the body shape, muscle distribution and surface fat layer of the cattle, and generates key morphological features;

[0046] The posture analysis module tracks the posture of the cattle based on the key morphological features, analyzes the changes in posture at differentiated time points, analyzes the posture changes of the cattle through continuous images, identifies the changes in the position and state of the cattle, and generates posture dynamic tracking results;

[0047] The three-dimensional modeling module uses the posture dynamic tracking results, applies continuous image data to perform three-dimensional reconstruction, maps key feature points to a three-dimensional coordinate system, calculates the body shape and volume of the cattle, and constructs a three-dimensional point cloud model;

[0048] The image synthesis module uses the three-dimensional point cloud model, inputs the cattle body condition label, uses the generative adversarial network to synthesize diversified cattle body condition images, evaluates the fit between the synthesized image and the real image, and obtains a body condition synthesized image data set;

[0049] The body condition assessment module extracts the characteristics of the cattle's body shape and muscle distribution based on the body condition synthetic image data set, calculates the body condition score of each cattle, dynamically adjusts the feed amount and feeding method in combination with the feeding environment, and generates a cattle body condition management plan.

[0050] Compared with the prior art, the advantages and positive effects of the present invention are:

[0051] In the present invention, the timeliness and accuracy of the assessment are improved through the precise processing of image data and the dynamic tracking of the body condition of cattle. By removing background interference and extracting key morphological features, the clarity and quality of image data are significantly improved, and the reliability of the analysis results is enhanced. Continuous image data analysis can reflect the changes in the morphology and posture of cattle in real time, and effectively capture the dynamic body condition information of cattle at different growth stages. The construction of a three-dimensional point cloud model enables the volume and shape of the cattle to be expressed more comprehensively and accurately, and after combining with a generative adversarial network, a variety of synthetic images can be created, which not only increases the flexibility of body condition assessment, but also better adapts to the characteristic differences of different cattle individuals. The accuracy of the body condition score is improved, and the feeding strategy can be dynamically adjusted according to the health assessment results, and feed adjustment measures can be provided for different growth cycles and health needs. The combination of feeding environment information makes the prediction of changes in the body condition of cattle more forward-looking, and can optimize the management plan in real time, reduce resource waste, and improve breeding efficiency and sustainability. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 It is a schematic diagram of the workflow of the present invention;

[0053] Figure 2 This is a detailed flow chart of S1 of the present invention;

[0054] Figure 3 This is a detailed flow chart of S2 of the present invention;

[0055] Figure 4 This is a detailed flow chart of S3 of the present invention;

[0056] Figure 5 This is a detailed flow chart of S4 of the present invention;

[0057] Figure 6 This is a detailed flow chart of S5 of the present invention;

[0058] Figure 7 This is a detailed flow chart of S6 of the present invention;

[0059] Figure 8 It is a system flow chart of the present invention. DETAILED DESCRIPTION

[0060] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0061] In the description of the present invention, it should be understood that the terms "length", "width", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside" and the like indicate positions or positional relationships based on the positions or positional relationships shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, in the description of the present invention, "multiple" means two or more, unless otherwise clearly and specifically defined.

[0062] See also Figure 1 The present invention provides a technical solution, a cattle body condition management method based on computer vision scoring, comprising the following steps:

[0063] S1: Obtain cattle image data, including full-body images and close-up images from multiple angles, record key body features in the images, including body shape, muscle distribution, and body fat layer, remove background interference from the images, and generate key morphological features of the cattle;

[0064] S2: Based on the key morphological features of cattle, use tags to track the posture and body condition of cattle, identify the position and state changes of cattle at different time points, analyze the dynamic posture change information of cattle through continuous image data, and generate dynamic body condition tracking results;

[0065] S3: Based on the results of dynamic body tracking, the key feature points in the continuous image data are mapped to a three-dimensional coordinate system using three-dimensional reconstruction technology to identify the body shape and volume of the cattle and construct a three-dimensional point cloud model of the cattle;

[0066] S4: Based on the 3D point cloud model of cattle, a variety of cattle body condition images are obtained through generative adversarial networks, cattle body condition labels are input, and the fit between the synthetic images and the real images is evaluated to match the real cattle body condition changes and obtain a body condition synthetic image dataset;

[0067] S5: Use the synthetic body image dataset combined with computer vision to identify the characteristics of cattle body shape and muscle distribution. Calculate the body condition score of each cattle based on the training data and labels, evaluate the health status, and obtain the score deviation record;

[0068] S6: Based on the scoring deviation records, according to the changes in cattle body condition over time, combined with feeding environment information, predict the trend of cattle body condition changes in the future time period, dynamically adjust the feed amount and feeding method, and obtain the cattle body condition management plan.

[0069] The key morphological characteristics of cattle include the head outline, trunk length, limb angles, and tail position. The dynamic tracking results of body condition include the cattle's walking speed, standing time, lying frequency, and eating behavior patterns. The cattle's three-dimensional point cloud model includes the three-dimensional coordinate points of the cattle's body length, height, and width, and connection relationships. The body condition synthetic image data set includes side views, front views, and back views of cattle in differentiated body conditions. The scoring deviation records include each cow's weight, muscle mass, fat ratio, and overall body shape rating. The cattle condition management plan includes feed formula adjustment, feeding frequency, and feeding time optimization plan.

[0070] See also Figure 2 , obtain cattle image data, including full-body images and close-up images, record key body features in the images, and remove background interference from the images. The specific steps for generating key morphological features of cattle are as follows:

[0071] S101: Obtaining cattle image data, including full-body images and close-up images, performing image cropping, identifying and extracting the outline of the cattle body, removing background areas unrelated to the cattle, performing boundary detection and blank area filling, and generating the cattle body image. The execution process is as follows;

[0072] The image is processed, and the contour boundary of the cow is determined by edge detection algorithm (such as Canny algorithm), and the background area unrelated to the cow is removed by morphological operation. The foreground area of ​​the cow is extracted by image segmentation technology (such as threshold-based segmentation), and the redundant area is removed to obtain the main area related to the cow. The contour detection algorithm (such as Sobel or Laplace algorithm) is applied to enhance the clarity of the contour, and the expansion and erosion operations are used to fill the blank areas in the image to ensure the integrity and coherence of the contour, and the body image of the cow is obtained.

[0073] S102: Based on the body image of the cattle, identify the key parts of the cattle body, calibrate the positions of the head, limbs, tail and back, and supplement each part to generate the body feature image. The execution process is as follows;

[0074] The image is segmented and calibrated, and the position of the head, limbs, tail and back is automatically calibrated and located using machine learning algorithms (such as support vector machines (SVM) or convolutional neural networks (CNN) in deep learning). Each part is calibrated in detail through the labeled key points and region extraction technology, and the calibration accuracy of each part is optimized by combining the body data and actual posture of the cattle. The calibration results are post-processed to ensure that the position of each part is consistent with the actual shape of the cattle, and a body feature image is generated.

[0075] S103: Using the body feature image, adjusting the image contrast and sharpness, performing highlight processing on the body feature, optimizing the clarity of the morphological features in the image, and generating the key morphological features of the cattle. The execution process is as follows;

[0076] To highlight the body features, follow this formula: Calculate the key morphological characteristics of cattle. highli Represents the processed image, I original represents the original image, α is the contrast enhancement coefficient, It is the gradient operation of the image, which indicates the edge information of the image. Calculate the edge information of the original image to emphasize the details in the image. Edge Information It can be obtained by convolution filter (such as Sobel filter), which is used to detect the brightness change of each point in the image. Set the gradient obtained by Sobel filter to Set the contrast enhancement coefficient α = 0.3. According to the formula, the highlight processing of the image will be calculated as:

[0077] I highli =I original ×(1+0.3×0.5)=I original ×1.15

[0078] That is, the brightness of the image is increased to enhance the clarity of key morphological features in the image, especially in the joints and edges of the cattle.

[0079] See also Figure 3 According to the key morphological features of cattle, the tags are used to track the posture and body condition of cattle, identify the position and state changes of cattle, and analyze the dynamic posture change information of cattle through continuous image data. The steps to generate dynamic body condition tracking results are as follows:

[0080] S201: Based on the key morphological features of the cattle, the posture labels of the cattle are calibrated, multiple body parts and positions of the cattle are identified, the head, trunk, limbs and tail are marked, and the position of the cattle is identified through spatial mapping. The execution process of generating the cattle posture label record is as follows;

[0081] Image calibration technology is used to automatically generate posture labels, calibrate the spatial coordinates of each key part (head, trunk, limbs, tail), and determine the relative position of the cow in the image based on the spatial mapping relationship of the cow image (such as using image stitching algorithm or perspective transformation algorithm). The calculated spatial coordinates and the position information of each part correspond to the specific parts of the cow's body, including the spatial coordinates and relative position relationship of each key part of the cow, forming a cow posture label record.

[0082] S202: Through the cattle posture label record, the dynamic changes of each posture of the cattle are continuously tracked, the label information in each frame image is extracted, the position change is analyzed through the difference between images, and the change of the cattle's body posture is quantified. The execution process of obtaining the cattle position state change data is as follows;

[0083] Extract the posture label information in each frame of the image, and calculate the change in the position of the cattle through the difference analysis method between images (such as frame difference or optical flow method). The difference analysis method identifies the movement trajectory and body changes of the cattle by comparing the label position information of the current image with the previous frame. Quantify the position information in each frame of the image into specific data values, and model the dynamic changes of the cattle through data analysis (such as dynamic time warping DTR) to obtain the results of continuous tracking. Including the state data of the posture change and position change of the cattle at different time points, it is convenient to further analyze the behavioral changes of the cattle and obtain the position state change data of the cattle.

[0084] S203: Performing body condition analysis based on the cattle position state change data, recording the characteristics of body condition related changes by comparing the posture changes and position movements of continuous images, and combining the change trends of position and posture to generate the body condition dynamic tracking results. The execution process is as follows;

[0085] According to the data of cattle position status change, body condition analysis is carried out according to the formula: ΔP=P t+1 -P t , calculate the characteristics of the changes in the body condition of cattle, where ΔP represents the change in the body condition of cattle, P t+1 and P t Respectively represent the position status of the cattle in two consecutive frames of images. By comparing the position status of the cattle in two consecutive frames of images, the body condition change is calculated. It is assumed that in image t and image t+1, the position status of the cattle is P t =(x1,y1) and P t+1 =(x2, y2), where x1 and y1 are the position coordinates of the cow in image t, and x2 and y2 are the position coordinates in image t+1. The body shape change is obtained by differential calculation: ΔP = (x2-x1, y2-y1), and the positions of the cow in image t and image t+1 are set to P respectively. t =(5,3) and P t+1 =(6, 4), then the amount of posture change is: ΔP = (6-5, 4-3) = (1, 1), which means that the position of the cattle in two consecutive frames of images has changed, and the displacement is 1 unit on the x-axis and y-axis. This position change can be further combined with the posture change trend to analyze the dynamic changes in body condition.

[0086] See also Figure 4 , through the results of dynamic body tracking, the key feature points in the continuous image data are mapped to the three-dimensional coordinate system, the body shape and volume of the cattle are identified, and the steps of constructing the three-dimensional point cloud model of the cattle are as follows:

[0087] S301: Through the results of dynamic body tracking, a corner point detection algorithm is used to extract key points from continuous images, and the key points are mapped to a three-dimensional coordinate system through geometric transformation. According to the proportion and position relationship of the differentiated parts of the cow's body, spatial coordinate transformation is performed to generate a three-dimensional key point set of the cow. The execution process is as follows;

[0088] The formula of the corner detection algorithm is as follows:

[0089]

[0090] Among them, S new is the corner response, det(M) represents the second-order matrix determinant of the local area of ​​the image, Tr(M) represents the trace of the matrix M, k is an empirical constant, Var(M) is the variance of the local area of ​​the image, Sha(M) is the sharpness of the image, is the magnitude of the image gradient, Area(M) is the area of ​​the local area of ​​the image,

[0091] A1, A2 and A3 are weight coefficients.

[0092] Parameter explanation and calculation process:

[0093] det(M) is the second-order matrix determinant of the local area of ​​the image, indicating the degree of image change. In the local area of ​​a certain image, det(M) is set to 14, reflecting the intensity of pixel change in the area.

[0094] Tr(M) is the trace of the matrix M, indicating the uniformity of the image. Tr(M) is set to 9 and the empirical constant k is set to 0.04. This value is determined through batch data experiments and is suitable for most image scenes. k·(Tr(M)) 2

[0095] =0.04·9 2 =0.04·81=3.24;

[0096] Variance Var(M) indicates the degree of pixel variation in a local area of ​​the image, reflecting the complexity of the image texture. Setting μ to 3.5 and Var(M) to 1.25 indicates that the texture variation in this area is moderate;

[0097] Sharpness Sha(M) indicates the clarity of image area details, which is measured by calculating the image gradient. When Sha(M) is set to 4.5, the image details are relatively clear.

[0098] is the magnitude of the image gradient, indicating the image edge strength, and is set 5, indicating that the regional edge is more obvious;

[0099] Area(M) is the area of ​​the local region of the image, indicating the number of pixels contained in the region. Set Area(M) to 25;

[0100] Set weight coefficients A1, A2, and A3. The weight coefficients are used to adjust the influence of various parameters on the responsiveness of corner points. Adjust through experimental data or training sets to ensure that the contribution of different factors to the results is reasonable. Set A1 = 0.5, A2 = 0.3, A3 = 0.2. The values ​​indicate that: A1 = 0.5, texture changes have a greater impact on corner point recognition; A2 = 0.3, the impact of sharpness on corner point response is moderate; A3 = 0.2, the impact of edge strength on corner point recognition is small;

[0101] Substitute the calculated results into the formula:

[0102]

[0103] The result shows that the response calculated by the corner detection formula is 10.2, indicating that this area is a more significant corner area.

[0104] S302: Using the three-dimensional key point set of the cow, three-dimensional interpolation processing is performed on the key feature points to construct the contour line of the cow's body, and the surface of the differentiated parts of the body is smoothed to generate the execution process of the three-dimensional contour map of the cow as follows;

[0105] By obtaining the three-dimensional coordinate data of the cattle, the interpolation algorithm (such as cubic spline interpolation) is used to fill the gaps between the key feature points to generate a continuous and smooth three-dimensional surface. The interpolated key points are used to construct the contour line of the cattle's body to ensure the continuity and accuracy of the body contour. For the differentiated parts of the cattle's body (such as the head, limbs, tail, etc.), a smoothing algorithm (such as Gaussian smoothing or B-spline curve smoothing) is used to process the surface of the parts to remove the existing noise and irregularities and generate a three-dimensional contour map of the cattle.

[0106] S303: Based on the three-dimensional outline of the cow, the characteristic points of the cow are connected, the body shape and volume of the cow are identified and estimated, and the execution process of constructing a three-dimensional point cloud model of the cow is as follows;

[0107] By connecting the feature points of the three-dimensional contour map, the spatial relationship between the feature points is converted into a three-dimensional grid structure using a triangular mesh algorithm (such as Delaunay triangulation). Based on the three-dimensional grid, the body shape and volume of the cattle are calculated using a volume estimation method (such as the voxel method or the Monte Carlo integration method). Through calculation, each point in the model represents a feature point of the cattle's body. The density and distribution of the point cloud reflect the body shape characteristics of the cattle, which can be used to further analyze the body shape characteristics and volume estimation of the cattle, and construct a three-dimensional point cloud model of the cattle.

[0108] See also Figure 5 Based on the 3D point cloud model of cattle, a variety of cattle body condition images are obtained through generative adversarial networks. The cattle body condition labels are input, and the fit between the synthetic images and the real images is evaluated. The specific steps for obtaining the body condition synthetic image dataset are as follows:

[0109] S401: The execution process of synthesizing images of the body condition of cattle by using a generator through a three-dimensional point cloud model of cattle and a generative adversarial network, and recording differentiated body shapes and conditions by adjusting the random noise and cattle body condition labels input by the network, and obtaining diversified cattle body condition images is as follows;

[0110] The generator uses random noise and cattle body condition labels as input. The generator synthesizes cattle images that match different body conditions by adjusting the noise and changing the cattle body condition labels. Through the back propagation of the network, the generator continuously optimizes the quality of the generated images, especially in details such as body contours and body shape differences. As the training progresses, the generator is able to generate diverse and highly fitting cattle images with body condition characteristics. The images reflect the body shape differences of cattle in different body conditions. The results output by the generator show the changes in the body shape and condition of cattle, generating diverse cattle body condition images.

[0111] S402: Based on the diversified cattle body condition images, the real cattle body condition labels are input, and the discriminator is used to evaluate the fit between the synthetic image and the real image. By comparing the texture, structure and difference of the images, the parameters of the generative adversarial network are adjusted to optimize the quality of the synthetic image. The execution process of generating the body condition synthetic image fit evaluation record is as follows;

[0112] The discriminator judges the quality of the generated image by comparing the texture, structure and difference between the synthetic image and the real image. The discriminator performs layer-by-layer feature recognition based on the details of the image (such as contours, light and shadow changes in texture, body shape structure, etc.), and compares the differences between the synthetic image and the real image. Based on the evaluation results, the parameters of the generative adversarial network are adjusted so that the generator can more accurately generate images that meet the body characteristics. Through the back-propagation mechanism, the weights of the generator and the discriminator are adjusted to optimize the quality of the generated image. The image quality of the output of the generative adversarial network is improved, and the quality of the body synthetic image is getting closer and closer to the real image, generating a body synthetic image fit evaluation record.

[0113] S403: Using the body condition synthetic image fit evaluation record, screening the synthetic image data, selecting image samples with priority fit, and annotating and classifying the synthetic images, the execution process of obtaining the body condition synthetic image data set is as follows;

[0114] The synthetic image data is screened through the body condition synthetic image fit evaluation record, according to the formula: Select the image sample with the best fit, where F represents the overall fit, For synthetic images With real image The similarity measure between g is the number of image samples. By comparing each pair of synthetic images with the real image, the similarity measure between the images is calculated. Set the structural similarity index (SSIM) to measure the similarity between images:

[0115]

[0116] in, and are the average brightness of the synthetic image and the real image, and is the brightness variance between the synthetic image and the real image, is the covariance, c1 and c2 are constants. By calculating the similarity of all image pairs, the fit F of each image is obtained, and image samples with priority in fit are selected for screening. Through this screening process, high-quality and well-fitted synthetic image samples of cattle body conditions are obtained for subsequent annotation and classification.

[0117] See also Figure 6 , using the body condition synthetic image dataset, combined with computer vision, to identify the characteristics of cattle body shape and muscle distribution, calculate the body condition score of each cattle, and evaluate the health status. The specific steps to obtain the score deviation record are as follows:

[0118] S501: Using the body condition synthetic image data set, combined with computer vision technology, by training the convolutional neural network model, identifying the key features of the cattle's body shape and muscle distribution, extracting the body shape, muscle density and fat distribution information in the image, and generating the cattle's body shape and muscle distribution feature record. The execution process is as follows;

[0119] The convolutional neural network model is trained using a synthetic image dataset of cattle body condition. The network extracts features through multi-layer convolution of images and can identify key features such as the body shape, muscle density, and fat distribution of cattle. During the training process, the convolutional neural network extracts characteristic information of the body shape and muscle distribution of cattle based on the texture, shape, and color changes in the image. Through multiple training and optimization, the model can effectively extract the body shape and muscle distribution characteristics of cattle and generate records of the body shape and muscle distribution characteristics of cattle.

[0120] S502: Based on the records of the body shape and muscle distribution characteristics of the cattle, a support vector machine algorithm is used to perform data classification and quantitative evaluation of the body condition score according to the set body shape, muscle distribution and fat layer thickness indicators, and the execution process of obtaining the body condition score of the cattle is as follows;

[0121] The formula of the support vector machine algorithm is as follows:

[0122]

[0123] Among them, BS is the body condition score of the cattle, S1 represents the body shape score, S2 represents the muscle distribution score, S3 represents the fat layer thickness, w1, w2 and w3 are weight coefficients, w4 is the adjustment coefficient, σ1, σ2 and σ3 are the standard deviations of the body shape score, muscle distribution score and fat layer thickness data respectively.

[0124] Detailed explanation of the formula and the process of formula calculation and derivation:

[0125] Body Score S1: Body score is calculated based on the cow's skeleton size, height and weight. Body score is scored by comprehensive evaluation of factors using a standardized scale. The following data is obtained during monitoring: cow height: 145cm, cow weight: 600kg. Body score S1 will be calculated based on the standardized scale. The setting is based on the height and weight of the cow: S1 = 8;

[0126] Muscle distribution score S2: The muscle distribution score is evaluated by evaluating the muscle mass of various parts of the cattle (such as shoulders, waist, buttocks, etc.). The muscle thickness is measured using professional equipment (such as BIA analysis or ultrasonic measurement), and the data is summarized to obtain the muscle distribution score. The following data are set: shoulder muscle thickness: 5cm, buttock muscle thickness: 10cm. The muscle distribution score S2 combines the data of the comprehensive parts and is obtained through the relevant score standardization algorithm: S2 = 7 (the standardized result is 7, indicating that the muscle distribution is relatively uniform and the muscle development is good);

[0127] Fat layer thickness score S3: The fat layer thickness is measured by ultrasound or electrical impedance method (BIA). The fat layer thickness of different parts of the cattle is measured and scored. The fat layer data obtained by measurement are set as follows: back fat layer thickness: 1.5cm, abdominal fat layer thickness: 2.0cm. After comprehensive measurement results, the fat layer thickness score S3 is calculated as: S3 = 6 (this score reflects that the fat distribution of the cattle is reasonable and the fat layer thickness is moderate);

[0128] Standard deviations σ1, σ2 and σ3: The standard deviation is used to reflect the distribution degree of each data and the volatility of the data. The standard deviation is calculated based on the body shape, muscle distribution and fat layer data of the same type of cattle collected previously: the body shape score standard deviation σ1 = 1.2 (based on the body shape score fluctuation data of the same type of cattle), the muscle distribution score standard deviation σ2 = 1.1 (based on the muscle distribution score fluctuation data of the same type of cattle), and the fat layer thickness standard deviation σ3 = 0.9 (based on the fat layer thickness fluctuation data of the same type of cattle);

[0129] Weight coefficients w1, w2, w3 and adjustment coefficient w4: Weight coefficients w1, w2, w3 indicate the importance of each score in the body condition score. Body shape and muscle distribution are more important in the body condition score, while fat layer thickness is relatively low. According to relevant research and empirical data, the following are set: w1 = 0.4 (weight coefficient of body shape score, body shape has a greater impact on body condition score); w2 = 0.4 (weight coefficient of muscle distribution score, muscle development is also important for body condition score); w3 = 0.2 (weight coefficient of fat layer thickness, fat layer has a smaller impact); adjustment coefficient w4 is used to balance the impact of standard deviation on the score; w4 = 0.5 is set to reduce the impact of fluctuations when the standard deviation is too large;

[0130] Substitute the known data into the formula for calculation:

[0131]

[0132] Calculate the numerator part:

[0133] 8×0.4=3.2;

[0134] 7×0.4=2.8;

[0135] 6×0.2=1.2;

[0136] 3.2+2.8+1.2=7.2;

[0137] Calculate the denominator:

[0138] (1.2) 2 =1.44;

[0139] (1.1) 2 =1.21;

[0140] (0.9) 2 =0.81

[0141] 1.44+1.21+0.81=3.46

[0142]

[0143] Calculating Body Condition Score:

[0144]

[0145] The result shows that the body condition score of the cattle is 3.62, which reflects the comprehensive score of the cattle in terms of body shape, muscle distribution and fat layer thickness. Through this score, the health status, growth potential and appropriate management measures of the cattle can be further evaluated.

[0146] S503: Using the cattle body condition score value, the health status of each cattle is evaluated, and the health level of each cattle is determined by comparing the body condition score with the preset health standard. The execution process of generating the score deviation record is as follows;

[0147] Collect multiple data such as the cow's weight, food intake, body temperature, etc., apply data standardization processing, eliminate the influence of external factors such as external temperature changes, eating conditions and other unstable factors, and determine the health range and comparison value of each physiological indicator by comparing real-time data and health standard values, using regression analysis and other methods. During the comparison process, if one or more indicators exceed the preset health standards, the cow is marked as a cow with health problems. If all indicators are within the healthy range, it is a cow in normal health and a score deviation record is generated.

[0148] See also Figure 7 Based on the score deviation records, according to the changes in cattle body condition over time, combined with the feeding environment information, the trend of cattle body condition changes is predicted, and the feed amount and feeding method are dynamically adjusted. The specific steps of the cattle body condition management plan are as follows:

[0149] S601: Analyze the trend of cattle body condition changes over time based on the score deviation records, and use the time series analysis method to predict the trend of cattle body condition changes in the future time period in combination with the influencing factors of the feeding environment, including temperature, humidity and activity. The execution process of obtaining cattle body condition change trend data is as follows;

[0150] Collect scoring deviation records and related environmental data, including factors such as temperature, humidity, and cattle activity. The data will be used as input for modeling and prediction using time series analysis methods (such as ARIMA, LSTM, etc.). By analyzing the correlation between body condition changes and environmental changes in real-time data, a prediction model is constructed to predict the changing trend of cattle body condition in the future time period. The time series model is based on past data and can capture the laws of body condition changes, thereby accurately predicting future body condition trends and generating cattle body condition change trend data.

[0151] S602: Based on the trend data of cattle body condition changes and the feeding environment information of cattle, including feed type, feeding frequency, and environmental temperature and humidity, verify that the cattle body condition is within the healthy range, and generate the feeding environment adjustment plan. The execution process is as follows;

[0152] By analyzing the trend data of cattle's body condition changes and combining the cattle's feeding environment information (such as feed type, feeding frequency, ambient temperature and humidity, etc.), we can evaluate whether the cattle's body condition is within the healthy range. If the cattle's body condition data indicates that the cattle have unhealthy changes (such as a decline in body condition or deviation from the healthy range), the feed type, feeding frequency, and temperature and humidity will be adjusted. By adjusting environmental factors, we can ensure that the cattle's body condition remains at a healthy level, optimize the feeding plan, improve the cattle's production performance and health status, and generate a feeding environment adjustment plan.

[0153] S603: Using the feeding environment adjustment plan, dynamically adjusting the real-time feeding plan, and adjusting the feed amount and feeding method by real-time monitoring of cattle body condition change data, the execution process of generating a cattle body condition management plan is as follows;

[0154] Relying on the real-time monitoring system to obtain the data on changes in the cattle's physical condition, the system makes timely adjustments to the amount of feed and feeding methods based on the physical condition changes in the real-time monitoring data and the adjustment plan. If the cattle's physical condition deteriorates or their health deviates from the standard, the system will automatically increase the amount of feed or adjust the feeding method (such as increasing the amount of exercise or improving temperature and humidity) to optimize the body condition. Through continuous monitoring and feedback adjustment, the cattle's physical condition is ensured to be maintained in the best state. According to the cattle's real-time physical condition data, the feeding strategy is flexibly adjusted to improve the overall health level and production efficiency, and a cattle physical condition management plan is generated.

[0155] See also Figure 8 A cattle body condition management system based on computer vision scoring, the cattle body condition management system based on computer vision scoring is used to execute the above-mentioned cattle body condition management method based on computer vision scoring, the system comprises:

[0156] The image data recording module obtains the whole body image of the cattle and close-up images from multiple angles, automatically removes background interference in the image, extracts the body shape, muscle distribution and surface fat layer of the cattle, and generates key morphological features;

[0157] The posture analysis module tracks the posture of cattle based on key morphological features, analyzes the changes in body posture at differentiated time points, analyzes the posture changes of cattle through continuous images, identifies the changes in the position and state of cattle, and generates posture dynamic tracking results;

[0158] The 3D modeling module uses the results of dynamic posture tracking, applies continuous image data for 3D reconstruction, maps key feature points to a 3D coordinate system, calculates the body shape and volume of the cattle, and constructs a 3D point cloud model;

[0159] The image synthesis module uses a three-dimensional point cloud model, inputs the body condition labels of cattle, uses a generative adversarial network to synthesize diverse cattle body condition images, evaluates the fit between the synthesized images and the real images, and obtains a body condition synthetic image dataset;

[0160] The body condition assessment module extracts the characteristics of cattle's body shape and muscle distribution based on the body condition synthetic image dataset, calculates the body condition score of each cow, dynamically adjusts the feed amount and feeding method based on the feeding environment, and generates a cattle body condition management plan.

[0161] The above are only preferred embodiments of the present invention and are not intended to limit the present invention in other forms. Any technician familiar with the profession may use the technical contents disclosed above to change or modify them into equivalent embodiments with equivalent changes and apply them to other fields. However, any simple modification, equivalent change and modification made to the above embodiments based on the technical essence of the present invention without departing from the technical solution of the present invention still falls within the protection scope of the technical solution of the present invention.

Claims

1. A method for managing cattle body condition based on computer vision scoring, characterized in that: The following steps are involved: Obtain cattle image data, including full-body images and close-up images, record key body features in the images, remove background interference from the images, and generate key morphological features of the cattle; According to the key morphological features of the cattle, the posture and body condition of the cattle are tracked using tags, the position and state changes of the cattle are identified, and the dynamic posture change information of the cattle is analyzed through continuous image data to generate dynamic body condition tracking results; Based on the dynamic tracking results of the body condition, key feature points in the continuous image data are mapped to a three-dimensional coordinate system, the body shape and volume of the cattle are identified, and a three-dimensional point cloud model of the cattle is constructed; Based on the three-dimensional point cloud model of the cattle, a variety of cattle body condition images are obtained by generating an adversarial network, a cattle body condition label is input, and the fit between the synthetic image and the real image is evaluated to obtain a body condition synthetic image data set; Using the body condition synthetic image dataset in combination with computer vision, the characteristics of the cattle's body shape and muscle distribution are identified, the body condition score of each cattle is calculated, and the health status is evaluated to obtain a score deviation record; Based on the score deviation records, according to the changes in the cattle's body condition over time and combined with the feeding environment information, the trend of cattle's body condition changes is predicted, the feed amount and feeding method are dynamically adjusted, and a cattle body condition management plan is obtained.

2. The method for managing cattle body condition based on computer vision scoring according to claim 1, characterized in that: The key morphological features of the cattle include the head outline, trunk length, limb angles, and tail position of the cattle. The dynamic tracking results of the body condition include the cattle's walking speed, standing time, lying frequency, and eating behavior pattern. The cattle's three-dimensional point cloud model includes the three-dimensional coordinate points of the cattle's body length, height, and width, and connection relationships. The body condition synthetic image data set includes side views, front views, and back views of cattle under differentiated body conditions. The scoring deviation records include each cow's weight, muscle mass, fat ratio, and overall body shape rating. The cattle body condition management plan includes feed formula adjustment, feeding frequency, and feeding time optimization plan.

3. The method for managing cattle body condition based on computer vision scoring according to claim 1, characterized in that: The steps of obtaining cattle image data, including full-body images and close-up images, recording key body features in the images, and removing background interference from the images to generate key morphological features of cattle are as follows: Obtain cattle image data, including full-body images and close-up images, perform image cropping, identify and extract the outline of the cattle's body, remove background areas unrelated to the cattle, perform boundary detection and blank area filling, and generate cattle body images; Based on the body image of the cattle, key parts of the cattle body are identified, the positions of the head, limbs, tail and back are calibrated, and each part is supplemented to generate a body feature image; The body feature image is used to adjust the image contrast and sharpness, perform highlight processing on the body features, optimize the clarity of the morphological features in the image, and generate key morphological features of the cattle.

4. The method for managing cattle body condition based on computer vision scoring according to claim 1, characterized in that: According to the key morphological features of the cattle, the posture and body condition of the cattle are tracked by using tags, the position and state changes of the cattle are identified, and the dynamic posture change information of the cattle is analyzed through continuous image data to generate the dynamic tracking results of the body condition. Specifically, the steps are as follows: Based on the key morphological features of the cattle, the posture labels of the cattle are calibrated, multiple body parts and positions of the cattle are identified, the head, trunk, limbs and tail are marked, and the positions of the cattle are identified through spatial mapping to generate cattle posture label records; Through the cattle posture label recording, the dynamic changes of each posture of the cattle are continuously tracked, the label information in each frame of the image is extracted, the position change is analyzed through the difference between the images, and the change of the cattle's body posture is quantified to obtain the cattle's position state change data; The body condition analysis is performed based on the cattle position status change data. By comparing the posture changes and position movements of continuous images, the characteristics of the body condition-related changes are recorded, and the dynamic tracking results of the body condition are generated by combining the changing trends of the position and posture.

5. The method for managing cattle body condition based on computer vision scoring according to claim 1, characterized in that: The key feature points in the continuous image data are mapped to a three-dimensional coordinate system through the dynamic tracking results of the body condition, the body shape and volume of the cattle are identified, and the steps of constructing a three-dimensional point cloud model of the cattle are as follows: Through the dynamic tracking results of the body condition, a corner point detection algorithm is used to extract key points in continuous images, the key points are mapped to a three-dimensional coordinate system through geometric transformation, and spatial coordinate transformation is performed according to the proportion and position relationship of the differentiated parts of the cattle's body to generate a three-dimensional key point set of the cattle; Using the three-dimensional key point set of the cattle, three-dimensional interpolation processing is performed on the key feature points to construct the contour line of the cattle body, and the curved surface of the differentiated parts of the body is smoothed to generate a three-dimensional contour map of the cattle; Based on the three-dimensional outline of the cattle, the characteristic points of the cattle are connected, the body shape and volume of the cattle are identified and estimated, and a three-dimensional point cloud model of the cattle is constructed.

6. The method for managing cattle body condition based on computer vision scoring according to claim 5, characterized in that: The formula of the corner detection algorithm is as follows: Among them, S new is the corner response, det(M) represents the second-order matrix determinant of the local area of ​​the image, Tr(M) represents the trace of the matrix M, k is an empirical constant, Var(M) is the variance of the local area of ​​the image, Sha(M) is the sharpness of the image, is the magnitude of the image gradient, Area(M) is the area of ​​the local region of the image, and A1, A2 and A3 are weight coefficients.

7. The method for managing cattle body condition based on computer vision scoring according to claim 1, characterized in that: Based on the three-dimensional point cloud model of the cattle, a variety of cattle body condition images are obtained by generating adversarial networks, cattle body condition labels are input, and the fit between the synthetic image and the real image is evaluated. The steps of obtaining the body condition synthetic image data set are specifically as follows: By using the three-dimensional point cloud model of the cattle, a generative adversarial network is used to synthesize an image of the cattle's body condition using a generator, and by adjusting the random noise and cattle's body condition label input by the network, the differentiated body shape and body condition are recorded to obtain a variety of cattle's body condition images; Based on the diversified cattle body condition images, input the real cattle body condition labels, use the discriminator to evaluate the fit between the synthetic image and the real image, adjust the parameters of the generative adversarial network by comparing the texture, structure and difference of the images, optimize the quality of the synthetic image, and generate a body condition synthetic image fit evaluation record; The body condition synthetic image fitting evaluation record is used to screen the synthetic image data, select image samples with preferred fitting, and annotate and classify the synthetic images to obtain a body condition synthetic image data set.

8. The method for managing cattle body condition based on computer vision scoring according to claim 1, characterized in that: The steps of using the synthetic body image dataset and combining computer vision to identify the characteristics of cattle body shape and muscle distribution, calculate the body condition score of each cattle, and evaluate the health status to obtain the score deviation record are as follows: Using the synthetic image data set of body condition, combined with computer vision technology, by training a convolutional neural network model, the key features of the cattle's body shape and muscle distribution are identified, the information of body shape, muscle density and fat distribution in the image is extracted, and a record of the cattle's body shape and muscle distribution characteristics is generated; Based on the records of the body shape and muscle distribution characteristics of the cattle, a support vector machine algorithm is used to perform data classification and quantitative evaluation of body condition scores according to the set body shape, muscle distribution and fat layer thickness indicators to obtain a body condition score value of the cattle; The cattle body condition score is used to evaluate the health status of each cattle, and the health level of each cattle is determined by comparing the body condition score with the preset health standard, and a score deviation record is generated; The formula of the support vector machine algorithm is as follows: Among them, BS is the body condition score of the cattle, S1 represents the body shape score, S2 represents the muscle distribution score, S3 represents the fat layer thickness, w1, w2 and w3 are weight coefficients, w4 is the adjustment coefficient, σ1, σ2 and σ3 are the standard deviations of the body shape score, muscle distribution score and fat layer thickness data respectively.

9. The method for managing cattle body condition based on computer vision scoring according to claim 1, characterized in that: Based on the score deviation record, according to the change of cattle body condition over time, combined with the feeding environment information, the trend of cattle body condition change is predicted, and the feed amount and feeding method are dynamically adjusted to obtain the cattle body condition management plan. Specifically, the steps are as follows: According to the scoring deviation records, the trend of the cattle's body condition changing over time is analyzed, and combined with the influencing factors of the feeding environment, including temperature, humidity and activity, a time series analysis method is used to predict the trend of the cattle's body condition changes in the future time period, and obtain the cattle's body condition change trend data; By using the cattle body condition change trend data and combining the cattle feeding environment information, including feed type, feeding frequency, and environmental temperature and humidity, verify that the cattle body condition is within a healthy range and generate a feeding environment adjustment plan; The feeding environment adjustment plan is used to dynamically adjust the real-time feeding plan, and the feed amount and feeding method are adjusted by real-time monitoring of cattle body condition change data to generate a cattle body condition management plan.

10. A cattle body condition management system based on computer vision scoring, characterized in that: The method for managing cattle body condition based on computer vision scoring according to any one of claims 1 to 9, wherein the system comprises: The image data recording module obtains the whole body image of the cattle and close-up images from multiple angles, automatically removes background interference in the image, extracts the body shape, muscle distribution and surface fat layer of the cattle, and generates key morphological features; The posture analysis module tracks the posture of the cattle based on the key morphological features, analyzes the changes in posture at differentiated time points, analyzes the posture changes of the cattle through continuous images, identifies the changes in the position and state of the cattle, and generates posture dynamic tracking results; The three-dimensional modeling module uses the posture dynamic tracking results, applies continuous image data to perform three-dimensional reconstruction, maps key feature points to a three-dimensional coordinate system, calculates the body shape and volume of the cattle, and constructs a three-dimensional point cloud model; The image synthesis module uses the three-dimensional point cloud model, inputs the cattle body condition label, uses the generative adversarial network to synthesize diversified cattle body condition images, evaluates the fit between the synthesized image and the real image, and obtains a body condition synthesized image data set; The body condition assessment module extracts the characteristics of the cattle's body shape and muscle distribution based on the body condition synthetic image data set, calculates the body condition score of each cattle, dynamically adjusts the feed amount and feeding method in combination with the feeding environment, and generates a cattle body condition management plan.

Images