A method and system for correcting the error of captive animal weight estimation based on geometric parameters

Abstract

This invention relates to the field of intelligent monitoring and weight estimation technology in livestock farming, specifically providing a method and system for correcting weight estimation errors in captive animals based on geometric parameters. The method includes: collecting image data of target animals within a preset area; extracting pixel coordinates of body measurement points and calculating geometric parameters using reference objects; extracting spatial distribution features of body posture, including trunk axis offset, body curvature, and standing tilt; fusing three types of posture deviations to obtain a comprehensive deviation value, which is then used to correct the geometric parameters to obtain geometric correction parameters; inputting these parameters into a weight estimation model to output a preliminary estimated weight; and calculating a weight correction coefficient by combining animal breed characteristics and dynamic physiological parameters to finally correct the estimated weight and error range. This invention enables non-contact continuous monitoring and improves the accuracy and stability of weight estimation through multi-dimensional error correction, providing data support for the management of large-scale captive farming.

Description

Technical Field

[0001] This invention relates to the field of intelligent monitoring and weight estimation technology in livestock farming, specifically to a method and system for correcting errors in weight estimation of captive animals based on geometric parameters. Background Technology

[0002] In modern, large-scale livestock production, animal weight is a core basis for feeding control, fattening management, and slaughter decisions. Traditional manual weighing methods are stressful and inefficient, and cannot meet the needs of continuous dynamic monitoring. Therefore, non-contact weight estimation technology based on machine vision is gradually becoming the mainstream technology in the industry. Most existing visual estimation methods extract geometric parameters such as body length and chest circumference from animal images and then substitute them into a weight model to obtain the estimation result. However, in actual captive breeding scenarios, they generally suffer from poor estimation accuracy and insufficient stability of results. Their core technical defects are concentrated in the following aspects: Existing methods generally use body posture features extracted from images directly for calculations without verifying the rationality of these features based on the natural physiological structure of captive animals. This makes it easy to include abnormal offsets, bends, or tilts that exceed the normal physiological range into the geometric parameter calculation process, directly causing distortion of key parameters such as body length and chest circumference. At the same time, existing technologies only process one type of posture deviation in isolation, without considering the physiological linkage and constraint relationship between trunk axis offset, body bending, and standing tilt. It is impossible to determine whether the combination of multiple posture features conforms to the animal's true standing posture. Even if local deviation corrections are made, it is difficult to eliminate the systematic errors caused by posture as a whole.

[0003] Geometric parameter measurement errors caused by body posture deviations will be directly transmitted to the subsequent weight estimation process. As geometric parameters such as body length and chest circumference are the core input items of the weight estimation model, even small deviations in these parameters will be amplified by the model, ultimately leading to a significant deviation between the weight estimation result and the actual value.

[0004] Therefore, the present invention provides a method and system for correcting errors in the estimation of the weight of captive animals based on geometric parameters, in order to solve the above-mentioned problems existing in the prior art. Summary of the Invention

[0005] To overcome the shortcomings of the prior art, the present invention provides a method and system for correcting errors in the estimation of the weight of captive animals based on geometric parameters, so as to solve the problems in the prior art.

[0006] One embodiment of the present invention provides a method for correcting errors in estimating the weight of captive animals based on geometric parameters, comprising the following steps: Collect image data of the target animal within a preset area, extract the pixel coordinates of the target animal's body measurement points using a preset image recognition algorithm, and calculate the geometric parameters of the target animal by combining the size parameters of a preset reference object; Extract the body spatial distribution features of body measurement points, including trunk axis offset features, body bending features, and standing tilt features; Based on the aforementioned body spatial distribution characteristics, the trunk axis offset deviation value, body bending deviation value, and standing tilt deviation value are calculated and fused to obtain a comprehensive posture deviation value. The geometric parameters are then corrected using the comprehensive posture deviation value to obtain the geometric correction parameters. The geometric correction parameters are input into a preset weight estimation model, and the preliminary estimated weight of the target animal is output. Calculate the weight correction factor based on the breed characteristics and dynamic physiological parameters of the target animal; The initial estimated weight is corrected using the weight correction factor to obtain the estimated weight of the target animal and its error range.

[0007] This application also relates to a system for correcting errors in estimating the weight of captive animals based on geometric parameters, including: The image acquisition module is used to acquire image data of the target animal within a preset area, extract the pixel coordinates of the target animal's body measurement points using a preset image recognition algorithm, and calculate the geometric parameters of the target animal by combining the size parameters of a preset reference object. The feature extraction module is used to extract the body spatial distribution features of body measurement points, including trunk axis offset features, body bending features, and standing tilt features; The first correction module is used to calculate and fuse the trunk axis offset deviation value, the body bending deviation value and the standing tilt deviation value based on the body spatial distribution characteristics to obtain a comprehensive posture deviation value, and use the comprehensive posture deviation value to correct the geometric parameters to obtain geometric correction parameters; The preliminary estimation module is used to input the geometric correction parameters into a preset weight estimation model and output the preliminary estimated weight of the target animal. The coefficient calculation module is used to calculate the weight correction coefficient based on the breed characteristics and dynamic physiological parameters of the target animal. The second correction module is used to correct the preliminary estimated weight using the weight correction coefficient to obtain the estimated weight and error range of the target animal.

[0008] This application also relates to a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the above-described method for correcting errors in estimating the weight of captive animals based on geometric parameters.

[0009] This application also relates to a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the above-described method for correcting errors in estimating the weight of captive animals based on geometric parameters.

[0010] The above embodiments provide a method and system for correcting errors in the estimation of captive animal weight based on geometric parameters, which has the following beneficial effects: 1. This invention effectively solves the problems of poor accuracy, insufficient stability and limited practicality of existing visual weight estimation technology through multi-dimensional error correction logic, and realizes non-contact and efficient monitoring. It avoids the stress interference to animals caused by traditional manual weighing, ensures the normal growth and development of animals, and meets the needs of batch and continuous dynamic monitoring in large-scale farms, greatly improving the efficiency of weight monitoring and the convenience of breeding management. 2. This invention extracts three core body posture spatial distribution features: trunk axis offset, body bending, and standing tilt. After comprehensive calculation and fusion, a quantitative deviation value is formed, which specifically corrects the geometric parameter distortion caused by body posture deviation. This reduces the systematic error caused by body posture interference from the source and improves the measurement accuracy of core geometric parameters such as body length and chest circumference. 3. This invention constructs a dual error correction mechanism of geometric parameter correction + weight coefficient correction. Based on the correction of body shape deviation, it calculates a specific weight correction coefficient by combining the breed characteristics of the target animal with dynamic physiological parameters such as food intake, water intake, and activity level. This makes the weight estimation model adaptable to captive animals of different breeds, growth stages and physiological states, further improving the accuracy and adaptability of the estimation results. Attached Figure Description

[0011] Figure 1 A flowchart illustrating a method for correcting errors in estimating the weight of captive animals based on geometric parameters, provided in an embodiment of the present invention; Figure 2 This is a schematic block diagram of a computer device provided in an embodiment of the present invention. Detailed Implementation

[0012] The technical solutions in the embodiments of the present invention will now be clearly and completely described in conjunction with the accompanying drawings.

[0013] Reference Figure 1 One embodiment of the present invention provides a method for correcting errors in estimating the weight of captive animals based on geometric parameters, comprising the following steps: S10. Collect image data of the target animal within a preset area, extract the pixel coordinates of the target animal's body measurement points using a preset image recognition algorithm, and calculate the geometric parameters of the target animal by combining the size parameters of a preset reference object. S20. Extract the body spatial distribution features of body measurement points, including trunk axis offset features, body bending features, and standing tilt features; S30. Based on the aforementioned body posture spatial distribution characteristics, calculate and fuse the trunk axis offset deviation value, body bending deviation value, and standing tilt deviation value to obtain a comprehensive posture deviation value. Use the comprehensive posture deviation value to correct the geometric parameters to obtain geometric correction parameters. S40. Input the geometric correction parameters into the preset weight estimation model and output the preliminary estimated weight of the target animal; S50. Calculate the weight correction factor based on the breed characteristics and dynamic physiological parameters of the target animal; S60. The preliminary estimated weight is corrected using the weight correction coefficient to obtain the estimated weight and error range of the target animal.

[0014] In this embodiment, the present invention is applicable to the weight estimation of domesticated livestock such as pigs, cattle, and sheep. This embodiment takes domesticated pigs as an example.

[0015] As described in step S10 above, this step specifically involves the basic data acquisition and geometric parameter calculation for the overall weight estimation method. The preset area is a pre-defined activity area for pigs within a penned environment, adapted for image acquisition. The target animal is the pig within the penned area whose weight needs to be estimated. Complete image data of the target pig within this area is acquired using conventional image acquisition equipment. A preset image recognition algorithm is used to process the acquired images, extracting the pixel coordinates of key measurement points on the target pig's body used for calculating core indicators such as body length and chest circumference. Then, using a pre-set reference object within the scene with known actual size parameters as a benchmark, a conversion relationship between image pixel coordinates and actual physical dimensions is established. This allows the calculation of the target pig's geometric parameters, including body length and chest circumference, providing basic data for body shape deviation correction and weight estimation.

[0016] As described in step S20 above, specifically, this step is the body posture feature extraction stage. Based on the pixel coordinates of the target pig's body measurement points obtained in step S10, further extract body posture spatial distribution features that can characterize the pig's actual standing posture. Among them, the trunk axis offset feature is used to reflect the deviation of the pig's overall trunk from the standard trunk axis; the trunk curvature feature is used to reflect the degree of curvature deformation of the pig's trunk itself; and the standing tilt feature is used to characterize the tilt state of the pig's body relative to the shooting reference plane. The above three types of features together constitute a complete body posture spatial distribution feature, comprehensively describing the posture deviation of the pig when it stands naturally, and providing a basis for calculating the posture deviation value.

[0017] As described in step S30 above, this step specifically involves the quantification of posture deviation and correction of geometric parameters. It follows the three types of spatial distribution features of body posture extracted in step S20 and the geometric parameters calculated in step S10, achieving the quantitative transformation of posture deviation and the precise correction of geometric parameters. First, for the trunk axis offset feature, body bending feature, and standing tilt feature, corresponding quantification algorithms are used to calculate the trunk axis offset deviation value, body bending deviation value, and standing tilt deviation value, respectively. The calculated deviation values ​​are all quantitative indicators that can reflect the difference between the actual posture and the standard posture. Then, the three types of deviation values ​​are integrated through a preset fusion rule to obtain a comprehensive posture deviation value that can comprehensively characterize the overall posture deviation of the pig. Finally, based on this comprehensive posture deviation value, the geometric parameters obtained in step S10, including body length and chest circumference, are corrected to offset the parameter distortion caused by posture deviation, obtaining accurate geometric correction parameters and providing reliable input data for weight estimation.

[0018] As described in step S40 above, it should be noted that the preset weight estimation model used in this embodiment is a dedicated body size-weight mapping model constructed for the growth and development characteristics of penned pigs (such as the correlation between body size and weight of different breeds of pigs, the impact of changes in body fat percentage on weight during the growth stage, etc.). The training process of this model is as follows: In a large-scale breeding scenario, actual measured data of pigs covering mainstream breeds such as Large White, Landrace, and Duroc, as well as piglets, fattening pigs, and sows at different growth stages are collected, with a cumulative sample size of no less than several thousand or even tens of thousands of groups; each sample includes accurately measured body length (the straight-line distance from the anterior edge of the shoulder blade to the posterior edge of the ischial tuberosity) and chest circumference (the circumference around the widest part of the chest) as input feature samples, and the actual weight data obtained by actual weighing on a standard electronic scale is used as the output calibration label. Subsequently, conventional mathematical modeling methods in this field, such as multiple linear regression and nonlinear fitting (e.g., multinomial fitting, exponential fitting), were used to train the model by substituting the input feature samples and calibration labels. The model parameters were adjusted using conventional optimization algorithms such as least squares and gradient descent to ensure that the error between the model's predicted value and the measured weight reached a preset threshold (e.g., average relative error ≤ 5%). After training, the model was validated by dividing the training set and the test set (e.g., a 7:3 ratio) to ensure that the model still has stable predictive performance on unseen samples. Finally, a mature model that can stably reflect the objective correspondence between pig body length, chest circumference, and weight was formed.

[0019] This model uses only geometric parameters related to body size as input variables, eliminating the need for additional collection of complex data such as physiological indicators and movement status of pigs. It can quickly output weight values, featuring high computational efficiency, convenient operation, and adaptability to large-scale pen farming scenarios. The construction principles, algorithm selection, training methods, optimization logic, and application scenarios of the above model are all conventional techniques well-known to those skilled in the art. Those skilled in the art can flexibly adjust the model's sample source, input feature weights, or select suitable modeling algorithms according to actual farming needs, all to achieve the same weight estimation purpose. This embodiment does not impose a unique limitation on this.

[0020] Specifically, this step is the preliminary weight estimation stage. Based on the geometric correction parameters obtained in step S30 after eliminating posture interference, the corrected geometric correction parameters (body length, chest circumference) are input into the pre-built weight estimation model. The weight estimation model first performs standardization preprocessing on the input body length and chest circumference parameters (e.g., converting the parameters to the feature scale range during model training to eliminate the influence of dimensional differences on the calculation results); then, based on the body size-weight mapping relationship fixed after model training (if it is a multiple linear regression model, then substitute it into the formula W=a×L+b×C+c, where W is the predicted weight value, L is the corrected body length, C is the corrected chest circumference, a and b are the weight coefficients of body length and chest circumference, and c is a constant term, which are fixed values ​​determined during model training; if it is a nonlinear fitting model, then substitute it into the corresponding fitting formula to complete the calculation); after the model completes the parameter weighting, numerical fitting and other calculation processes through its built-in calculation logic, it directly outputs the corresponding weight value, which is the preliminary estimated weight after eliminating posture deviation interference.

[0021] As described in step S50 above, this step specifically involves calculating the weight correction coefficient. The aim is to combine the breed specificity of the target pigs with their real-time physiological state to construct a correction coefficient that can offset individual differences, providing a basis for the accurate optimization of the initial weight estimate. Breed characteristics refer to the inherent attributes of the target pig breed, such as the high lean meat percentage of Large White pigs, the growth rate advantage of Landrace pigs, and the body fat distribution characteristics of Duroc pigs. Different breeds have subtle differences in their body size-weight mapping relationship, requiring targeted adjustments based on breed characteristics. Dynamic physiological parameters refer to the pigs' recent real-time physiological state data, including daily feed intake, activity level, health status (e.g., whether in the recovery period from disease), and physiological stage (e.g., mid-fat feeding, gestation). These parameters directly affect the short-term weight change trend and body composition ratio of the pigs, and are key dynamic factors for correcting the initial weight estimate.

[0022] In the calculation process, the breed characteristics and dynamic physiological parameters are first quantified: the breed characteristics are mapped to fixed weight coefficients (e.g., based on actual breed data, the breed coefficient is preset to 1.02 for Large White pigs, 1.01 for Landrace pigs, and 0.99 for Duroc pigs), and the dynamic physiological parameters are converted into quantitative indicators (e.g., daily feed intake achievement rate, activity level score, and health status quantitative value). Then, through preset weighted calculation rules, the breed quantitative coefficients and dynamic physiological quantitative indicators are integrated and calculated to finally obtain a weight correction coefficient with a value range within a preset interval (e.g., 0.95~1.05). This coefficient can reflect the degree of influence of individual differences on the weight estimation results and provide a quantitative basis for the final weight correction.

[0023] As described in step S60 above, this step specifically involves the final weight calculation and error calibration. Based on the preliminary estimated weight obtained in step S40 and the weight correction coefficient determined in step S50, the final optimization of the overall estimation process is completed. The correction process uses a weighted calculation method, directly multiplying the preliminary estimated weight by the weight correction coefficient to obtain the final estimated weight after eliminating the interference of individual factors such as breed differences and dynamic physiological states. This result is closer to the true weight of the target pig.

[0024] This step outputs the corresponding error range, which is determined by the inherent verification error of the preset weight estimation model, the accuracy of body shape deviation correction, and the quantitative accuracy of the weight correction coefficient. It is usually controlled within ±3%, which can intuitively reflect the accuracy of the weight estimation result and provide reliable and referential complete data for aquaculture management work such as feeding adjustment and slaughter determination in captivity scenarios.

[0025] In one embodiment, step S10 specifically includes: S101. Several image acquisition devices are set up at fixed acquisition positions in the pre-set captive breeding area, and fixed and known actual size preset reference objects are set up in the complete acquisition field of view of each image acquisition device. The preset reference objects and the body measurement parts of the target animal are on the same shooting reference plane. S102. The image acquisition device acquires image data of a single target animal in a natural standing position within a pre-defined captive area, and the image data includes the target animal's body and a pre-defined reference object. S103. Using a preset image recognition algorithm, the image data is used to detect and locate key body measurement points and preset reference objects, and the pixel coordinates of the target animal's body measurement points and the pixel size of the preset reference objects in the image are extracted. S104. Establish a size conversion ratio based on the actual size and pixel size of the preset reference object, and calculate the geometric parameters of the target animal based on the size conversion ratio and the pixel coordinates of the target animal's body measurement points.

[0026] In this embodiment, to further improve the accuracy of geometric parameter calculation, step S10 is implemented through the following steps. This embodiment still uses penned pigs as the application object, and the specific explanations of each step are as follows: As described in step S101 above, this step specifically involves the deployment of the image acquisition system, which is fundamental to ensuring the accuracy of subsequent size conversions. The fixed acquisition positions need to be pre-selected based on the spatial layout of the designated pen area, prioritizing unobstructed locations with a shooting angle perpendicular to the side of the pig's body (such as above the pen passageway or on a fixed bracket outside the fence). The image acquisition equipment can be conventional devices such as high-definition industrial cameras or infrared cameras (resolution not less than 1920×1080, frame rate ≥30fps). This invention does not impose a single limitation on the specific number and precise locations of the image acquisition equipment. In practical applications, adjustments can be made flexibly based on differences in the spatial shape of different pigpens, breeding scale, and pen layout. It is sufficient to ensure that the acquisition field of the deployed equipment covers the core area of ​​pig activity, has no blind spots, and can clearly capture the target pig's body and the pre-set reference object. This deployment adjustment logic is a technical choice that can be routinely implemented by those skilled in the art based on the conditions of the breeding site, and does not affect the normal implementation of this technical solution.

[0027] The preset reference object should be a standard part made of hard, non-deformable material with high contrast (such as a black and white checkerboard calibration plate or a graduated metal ruler). Its actual dimensions (such as ruler length and checkerboard side length) must be accurately measured and recorded to ensure an error ≤ 0.1. This invention does not impose a single limitation on the specific number and precise fixed positions of the preset reference objects. They can be flexibly arranged in conjunction with the field of view of the image acquisition equipment and the pig's typical standing area. The only requirement is that the reference objects are always within the complete field of view of each image acquisition device and are on the same shooting reference plane as the pig's body measurement area. The key placement requirement is that the reference objects and the pig's body measurement area (body length corresponding to the front and rear ends of the torso, chest circumference corresponding to the chest area) are on the same shooting reference plane, meaning that the vertical distance from both to the lens of the image acquisition equipment is the same. This eliminates pixel size distortion caused by differences in shooting angles, providing a flawless reference for subsequent proportional conversion.

[0028] As described in step S102 above, specifically, this step is the image acquisition stage, the core of which is to acquire valid image data that meets the measurement requirements. During acquisition, conventional methods such as pen isolation devices or timed herding are required to ensure that only a single pig exists in the acquisition area, avoiding multiple animals obscuring or overlapping, which could lead to the failure of measurement point positioning; the pigs must be in a natural standing position, that is, with all four limbs on the ground, the torso extended, and no curling up or lying on their side, to ensure that the key measurement points of the body can be fully presented.

[0029] The collected image data must clearly include both the complete body of the pig and the preset reference object, and neither of them should have obvious occlusion, blurring or reflection. The pig's body should occupy no less than 60% of the image to ensure that the subsequent recognition algorithm can accurately detect and locate key measurement points and reference objects, providing high-quality data support for pixel coordinate extraction.

[0030] As described in step S103 above, specifically, this step is the image recognition and coordinate extraction stage. Following the valid image data obtained in step S102, image recognition processing simultaneously locates and extracts data from key measurement points of the pig's body and preset reference objects, providing accurate pixel-level data for subsequent size conversion. The preset image recognition algorithm can employ conventional target detection and key point localization algorithms in the field (such as YOLO series algorithms, CNN convolutional neural networks). This algorithm needs to be pre-trained and optimized based on a large number of pig image samples with different postures and shooting angles, as well as image samples of various types of preset reference objects, to possess the ability to stably identify key measurement points of the body and features of reference objects.

[0031] During the recognition process, the image data is first preprocessed using conventional methods (such as noise reduction, grayscale enhancement, and distortion correction) to improve the clarity of image features. Then, the trained recognition algorithm is used to detect and locate key measurement points on the pig's body, including the anterior edge of the scapula corresponding to body length, the posterior edge of the ischial tuberosity, and the left and right endpoints of the widest part of the chest corresponding to chest circumference. The two-dimensional coordinates of each measurement point in the image pixel coordinate system are then output. Simultaneously, the algorithm performs contour detection or feature identification and localization on preset reference objects in the image, extracts the pixel size of the reference objects in the image (such as the pixel length of the ruler, the pixel side length of the checkerboard), and strictly ensures the extraction accuracy of the pixel coordinates of the body measurement points and the pixel size of the reference objects throughout the process, so that they match the image resolution, and the pixel error is controlled within ≤1 pixel unit.

[0032] The algorithms and processing procedures for image recognition, key point localization, and pixel data extraction described above are all conventional techniques well-known to those skilled in the art. Those skilled in the art can flexibly select suitable algorithm models or adjust image preprocessing parameters according to actual recognition accuracy requirements, and all can achieve the technical objectives of this step. This embodiment does not limit this to a single method.

[0033] As described in step S104 above, specifically, this step is the precise calculation of geometric parameters. It establishes a relationship between pixel scale and actual physical scale based on a preset reference object, realizing the conversion of pixel coordinates of body measurement points to true geometric parameters, thus fundamentally avoiding dimensional deviations caused by relying solely on pixel calculations. First, the size conversion ratio is calculated: Let the actual size of the preset reference object be... (unit: Its pixel size in the image is (Unit: pixels), then the conversion ratio (unit: ( / pixel), this ratio intuitively reflects the actual physical length corresponding to each image pixel, and is the core basis for realizing pixel-to-actual size conversion.

[0034] Then, geometric parameters are calculated: for body length, the straight-line pixel distance between the anterior edge of the scapula and the posterior edge of the ischial tuberosity, extracted in step S103, is calculated using the Euclidean distance formula, based on the two-dimensional pixel coordinates of the anterior edge of the scapula and the posterior edge of the ischial tuberosity. Then multiply the pixel distance by the conversion ratio. The actual body length of the target pig was obtained. For the bust circumference, first calculate the corresponding pixel distance using the pixel coordinates of the left and right endpoints of the widest part of the bust, then verify it using the conventional geometric relationship between circumference and diameter, or directly collect the surrounding pixel length of the bust contour and multiply it by the conversion ratio. The final actual bust measurement is obtained. The calculations are performed with values ​​rounded to two decimal places, ensuring that the final output of body length and chest circumference geometric parameters has an error margin of ±0.5%. This improves the accuracy of basic data, providing reliable support for body shape deviation correction and weight estimation.

[0035] The above-mentioned methods for calculating size conversion ratios and geometric calculations of body length and chest circumference are all conventional mathematical calculation methods in the field of livestock and poultry body size measurement. Those skilled in the art can flexibly adjust the calculation formulas or calculation logic according to the actual measurement accuracy requirements and the type of reference object, and all can achieve the same geometric parameter calculation effect. This adjustment method is a conventional technical choice for those skilled in the art and does not affect the complete implementation of this technical solution.

[0036] In one embodiment, prior to step S20, the method further includes alignment correction of the pixel coordinates of the target animal body measurement points, specifically including the following steps: S2001. Using the shooting reference plane where the preset reference object is located as a reference, define the origin, horizontal axis and vertical axis of the two-dimensional reference coordinate system, and set the mapping relationship between the coordinate scale and the actual size. S2002. Extract the pixel coordinates of the key corner points of the preset reference object and substitute them into the two-dimensional reference coordinate system to calibrate and form at least three non-collinear reference coordinate anchor points. S2003. Calculate the relative positional deviation between each pixel coordinate of the target animal body measurement point and the reference coordinate anchor point, and perform coordinate translation and scaling correction on the body measurement point according to the relative positional deviation. S2004. After calibration, output the pixel coordinates of the target animal body measurement points that match the shooting reference plane.

[0037] In this embodiment, to further eliminate pixel coordinate offsets caused by factors such as slight deviations in shooting angle and equipment calibration errors, and to improve the accuracy of subsequent body feature extraction, an alignment and correction step for the pixel coordinates of the target animal's body measurement points is added before extracting the body spatial distribution features in step S20. This embodiment still uses penned pigs as the application object, and the specific explanations of each step are as follows: As described in step S2001 above, this step specifically involves constructing a reference coordinate system. The core of this step is to establish a unified and accurate coordinate reference framework to provide a standard basis for subsequent calibration. Specifically, the two-dimensional reference coordinate system is established using the shooting reference plane where the preset reference object is located in step S101 as the sole reference, avoiding calibration deviations caused by inconsistent references. The origin of the coordinate system can be flexibly selected (such as the lower left corner vertex of the preset reference object, the geometric center, etc.). The horizontal axis is set along the long side of the reference object, and the vertical axis is perpendicular to the horizontal axis and parallel to the shooting reference plane. This invention does not impose a unique limitation on the specific selection of the origin position and axis direction; it only needs to ensure that the system is stable and repeatable.

[0038] The mapping relationship between coordinate scales and actual dimensions directly adopts the dimension conversion ratio established in step S104. (unit: / pixel), meaning that each unit of measurement in the coordinate system corresponds to the actual size. This achieves a precise correlation between pixel coordinates and actual physical space. The two-dimensional reference coordinate system constructed in this step can unify the pixel coordinates of different image acquisition devices and shooting times within the same reference frame, laying the foundation for alignment correction.

[0039] As described in step S2002 above, specifically, this step is the benchmark coordinate anchor point calibration stage, which establishes a calibration reference benchmark through the known features of the reference object. First, the pixel coordinates of the key corner points of the preset reference object are extracted. The key corner points must be feature points on the reference object with clear physical locations, easy identification, and resistance to deformation (such as the intersection corners of a black and white checkerboard calibration board, the endpoints and midpoints of a metal ruler, etc.). Then, the pixel coordinates of these corner points are substituted into the two-dimensional benchmark coordinate system constructed in step S2001, combined with the actual size of the reference object and the conversion ratio. The theoretical coordinates of each corner point in the reference system are calculated, and the corresponding calibration of pixel coordinates and reference theoretical coordinates is completed, forming at least three non-collinear reference coordinate anchor points.

[0040] It should be noted that selecting at least three non-collinear anchor points is based on the fundamental principles of plane geometry. Three points determine a unique plane, which can achieve complete calibration of image perspective distortion and scale deviation. This calibration logic is a conventional technique in coordinate correction in this field. This invention does not impose a unique limitation on the specific number of anchor points (≥3 is sufficient) or the selection position of corner points. Those skilled in the art can flexibly adjust according to the type of reference object, and all can achieve the same calibration effect.

[0041] As described in step S2003 above, specifically, this step is the coordinate precision correction stage, which relies on the reference anchor point to correct the deviation of the measurement point coordinates. First, the relative position deviation is calculated: using conventional deviation calculation methods in the field (such as the least squares method and the iterative nearest point algorithm), the original pixel coordinates of each measurement point on the target pig's body (extracted in step S103) are compared with the theoretical coordinates after the reference coordinate anchor point is calibrated, to obtain the translation deviation of each measurement point in the horizontal and vertical axes, as well as the scale deviation caused by the slight change in shooting distance; then, targeted correction is performed based on these deviation values: the overall position offset is eliminated by coordinate translation (making the position of the measurement point relative to the anchor point consistent with the actual physical space), and the scale deviation is eliminated by proportional scaling (ensuring that the pixel spacing of different measurement points is consistent with the actual size ratio). The correction process can be implemented using conventional coordinate correction algorithms in the field, such as affine transformation and perspective transformation.

[0042] This invention does not impose a single limitation on the specific deviation calculation method or correction algorithm. Those skilled in the art can choose the appropriate technical means according to the actual deviation type (such as translation deviation only, deviation including scale and distortion), and all can achieve the purpose of accurate coordinate alignment.

[0043] As described in step S2004 above, specifically, this step is the post-calibration coordinate output stage. Following the calibration result of step S2003, it outputs the pixel coordinates of the target pig's body measurement points that precisely match the shooting reference plane. These coordinates have completely eliminated offset errors caused by factors such as shooting angle, equipment calibration, and scale changes. The relative positions and spacing between each measurement point are consistent with the actual physical space height, providing unbiased coordinate data support for step S20 to extract body posture features such as trunk axis offset and body curvature, further improving the accuracy of subsequent posture deviation value calculations.

[0044] The principles and techniques of the above-mentioned reference coordinate system construction, anchor point calibration, and coordinate correction are all conventional technical solutions in image coordinate processing in this field. Those skilled in the art can flexibly adjust the system parameters, anchor point selection, or correction algorithm according to the characteristics of the reference object and the needs of the aquaculture scenario, which will not affect the complete implementation of this technical solution. This embodiment does not limit it to a single one.

[0045] In one embodiment, step S20 specifically includes: S201. Based on the pixel coordinates of the target animal's body measurement points, fit and construct the target animal's trunk reference axis; S202. Based on the relative positional relationship between the trunk reference axis and the target animal's standard trunk axis, calculate the trunk axis offset characteristics; S203. Based on the distribution deviation of the target animal's body measurement points on both sides of the trunk reference axis, the body bending characteristics are identified; S204. Based on the difference in vertical pixel coordinates of the target animal's body measurement points, the standing tilt feature is determined; S205. Integrate the trunk axis offset features, body bending features, and standing tilt features to form the spatial distribution features of the target animal's body posture. This specifically includes the following steps: S2051. Calibrate the physiologically reasonable ranges of the target animal corresponding to the trunk axis offset characteristics, body bending characteristics, and standing tilt characteristics, respectively. S2052. Compare the actual values ​​of the trunk axis offset feature, body curvature feature, and standing tilt feature with the corresponding physiologically reasonable ranges respectively; if the actual value of any of the trunk axis offset feature, body curvature feature, and standing tilt feature exceeds the corresponding physiologically reasonable range, the single feature verification is deemed unqualified and the process does not proceed to the next step. S2053. For the trunk axis offset feature, body bending feature and standing tilt feature that pass the single feature verification, establish linkage constraint relationship according to the body posture law of the natural standing of captive animals, and verify the rationality of the combination of trunk axis offset feature, body bending feature and standing tilt feature based on the linkage constraint relationship. S2054. The trunk axis offset features, body bending features and standing tilt features that have passed the combination rationality verification are coupled and integrated to form the body spatial distribution features of the target animal.

[0046] In this embodiment, to achieve accurate quantification, effective screening, and reasonable integration of body shape spatial distribution features, step S20 unfolds through a logic of single feature extraction → dual verification → coupled integration. This ensures the accuracy of feature extraction while eliminating abnormal data through a verification mechanism. This embodiment still uses penned pigs as the application object, and the specific explanations of each step are as follows: As described in step S201 above, specifically, this step is the trunk baseline fitting process, which is the core reference basis for all subsequent body feature calculations. Key longitudinal measurement points of the pig's body are selected (including feature points distributed along the trunk midline such as the anterior edge of the shoulder blade, the midpoint of the neck and back, the midpoint of the chest and back, the midpoint of the lumbar back, and the posterior edge of the ischial tuberosity). A conventional least-squares linear fitting algorithm is used to substitute these discrete measurement points into the fitting formula (e.g., ...). ,in , To measure the pixel coordinates of the point, The slope (where the intercept is used) to solve for the trunk reference axis that runs through the longitudinal extension trend of the pig's trunk. This invention does not limit the fitting algorithm to a single one. Those skilled in the art can also choose conventional methods such as polynomial fitting or B-spline curve fitting according to the complexity of the body morphology, as long as the axis can stably represent the overall extension direction of the trunk.

[0047] As described in step S202 above, specifically, this step is the trunk axis offset feature calculation stage, which is quantified by comparing the "actual axis - standard axis". The target animal standard trunk axis is determined based on the measured data of thousands of healthy pigs standing naturally in a large-scale breeding scenario, and is an ideal trunk axis (i.e., a standard straight line extending horizontally along the shooting reference plane without lateral offset, with an axis slope of...). =0). The relative positional relationship between the torso reference axis and the standard torso axis is calculated through geometric operations: First, the horizontal and vertical distances between the two axes are calculated (unit: pixels, which can be converted to scale by size). Convert to actual distance The first step is to calculate the offset magnitude; the second step is to calculate the angular deviation between the two axes (unit: °) to represent the offset direction; the distance and angle values ​​are normalized (e.g., mapped to the 0~1 interval) and integrated to form the trunk axis offset feature, which can accurately quantify the degree of lateral offset of the pig's trunk relative to the standard posture.

[0048] As described in step S203 above, specifically, this step is the body bending feature recognition stage, which determines the bending state based on the distribution deviation of measurement points relative to the reference axis. Using the trunk reference axis constructed in step S201 as the central reference line, key measurement points on both sides of the pig's trunk (such as the left side of the chest, the right side of the chest, the left side of the abdomen, and the right side of the abdomen) are extracted. The vertical distance (pixel value) from each lateral measurement point to the reference axis is calculated, and the distance dispersion (such as variance and standard deviation) and curvature change (calculated by the distance change rate between adjacent measurement points) of all lateral measurement points are statistically analyzed. A preset bending judgment threshold is set (this threshold is determined based on the measured data of the natural bending of the body of healthy pigs, such as dispersion ≤ 5 pixels and curvature change rate ≤ 0.1). If the measurement points are concentrated on one side of the axis, and the dispersion or curvature change rate exceeds the preset threshold, then the body is determined to be bent. Combining the dispersion value (representing the bending amplitude) and the distribution orientation (representing the bending direction, such as left-side bending or right-side bending), the body bending feature is quantified.

[0049] As described in step S204 above, specifically, this step is the standing tilt feature determination step, which identifies the tilt state based on the difference in vertical pixel coordinates. The vertical pixel coordinates of key measurement points at the upper and lower ends of the pig's body (such as the highest point of the shoulder, the highest point of the rump, the end point of the forelimb hoof, and the end point of the hindlimb hoof) are extracted, and the vertical coordinate difference between the end points of the forelimb and hindlimb on the same side is calculated. ), the vertical coordinate difference between the highest points of the shoulder and hip ( ), by converting the size ratio Convert pixel differences into actual height differences ( ). Calculate the body tilt angle based on the height difference (e.g., tanθ = ,in, The horizontal pixel distance between the shoulder and hip is converted into the actual horizontal distance and then calculated. If θ is positive and exceeds the preset tilt threshold (e.g., θ≤3°), it is determined to be forward tilt; if it is negative and exceeds the threshold, it is determined to be backward tilt. The tilt angle and tilt direction are integrated to form the standing tilt feature.

[0050] As described in step S205 above, this step involves single feature verification, combined verification, and coupling integration of body posture features. Dual verification ensures the validity and rationality of the features, preventing abnormal body posture data from affecting subsequent deviation calculations. The specific steps are as follows: As described in step S2051 above, specifically, this step is the physiologically reasonable range calibration stage, the core of which is to establish the validity boundary of a single feature. The physiologically reasonable range is the allowable range for each physical characteristic determined through statistical analysis (e.g., using a 95% confidence interval) based on measured data from healthy pigs of different breeds (Large White, Landrace, Duroc, etc.) and different growth stages (piglet, finishing pig, sow, etc.). The physiologically reasonable range for trunk axis deviation characteristics: actual lateral deviation distance ≤ 2 The deviation of the included angle of the axis is ≤5°; The physiologically reasonable range for body curvature characteristics: the actual distance corresponding to the curvature amplitude is ≤3. The rate of change of curvature is ≤0.15; The physiologically reasonable range for standing tilt characteristics: tilt angle ≤ 5° (forward or backward tilt).

[0051] This invention does not impose a unique limitation on the specific values ​​of the interval. Those skilled in the art can dynamically adjust it based on measured data of the breeding species and growth stage, as long as the interval can cover the natural body size range of healthy pigs.

[0052] As described in step S2052 above, specifically, this step is a single-feature validity verification step, used to exclude abnormal body posture data (such as pigs falling down, limb deformities, stress curling, and other unnatural standing states). The actual values ​​of each feature calculated in steps S202~S204 are compared with the corresponding physiologically reasonable intervals marked in step S2051: if any feature's actual value exceeds the corresponding interval (e.g., a lateral deviation of 3 mm from the torso), the verification is performed. If the tilt angle is 8°, the body posture data is directly determined to be invalid and will not proceed to the subsequent combination verification and integration steps. The image acquisition device can be triggered to re-acquire data. If all feature actual values ​​are within the corresponding range, the single feature verification is determined to be qualified and proceed to the next step.

[0053] As described in step S2053 above, this step specifically involves verifying the rationality of feature combinations. It establishes linkage constraints based on the natural standing posture of pigs to avoid situations where a single feature is acceptable but the combination is contradictory (e.g., significant lateral displacement of the torso accompanied by severe backward leaning, which does not conform to the biomechanical balance of natural standing). The linkage constraint relationship is established based on the biomechanical laws of healthy pigs naturally standing. An example constraint relationship is as follows: If the lateral offset of the torso axis is ≥1 If the angle of inclination is ≤3°, then the standing angle should be ≤3° (to avoid the superposition of offset and inclination leading to imbalance). If the body is bent (bending angle ≥ 1.5) If the torso axis offset is less than or equal to 1, then the torso axis offset must be less than or equal to 1. And the tilt angle is ≤2° (it is difficult to maintain a large offset and tilt at the same time when the body is bent).

[0054] During verification, the actual values ​​of the three types of features that are qualified for a single feature are substituted into the linkage constraint relationship to determine whether all constraint conditions are met: if they are met, the combination rationality verification is deemed passed; if not (e.g., offset 1.2), the verification is deemed successful. If the tilt is 4°, the data is deemed invalid and will not proceed to the integration step. This invention does not impose a unique limitation on the specific number and conditions of the linkage constraint relationships; those skilled in the art can supplement or adjust the constraint logic according to the body's dynamic patterns to achieve the purpose of verifying the rationality of the combination.

[0055] As described in step S2054 above, specifically, this step is the body posture feature coupling and integration stage, forming a complete and reasonable body posture spatial distribution feature. First, the three types of features that have passed the combination verification are normalized preprocessed (the feature values ​​are mapped to the 0~1 interval) to eliminate the dimensional differences between different features; then, based on the influence weight of each feature on the overall body posture (the weights are determined through regression analysis of healthy pig body posture samples, such as trunk axis offset weight 0.35, body bending weight 0.3, and standing tilt weight 0.35), the three types of features are coupled into a set of multidimensional feature data (such as [0.25, 0.18, 0.12]) using conventional integration methods such as weighted summation or feature vector concatenation. This data is the body posture spatial distribution feature of the target pig, which can comprehensively, quantitatively, and reasonably characterize the overall body posture deviation in its natural standing state, providing reliable feature input for the subsequent step S30 to calculate the comprehensive posture deviation value.

[0056] The technical principles and implementation methods of the above-mentioned feature extraction, interval calibration, linkage constraint establishment, and coupling integration are all conventional technical solutions in the fields of image pose analysis, data verification, and feature fusion. Those skilled in the art can flexibly adjust the algorithm model, threshold parameters, constraint conditions, or integration methods according to the actual breeding scenario and accuracy requirements, which will not affect the complete implementation of this technical solution. This embodiment does not limit it to a single method.

[0057] In one embodiment, step S30 specifically includes: S301. Based on the preset offset quantization model, the torso axis offset features are converted into offset distances and offset angles along the transverse and longitudinal directions of the torso to obtain the torso axis offset deviation value. S302. Based on the mapping relationship between the curvature of the body and the geometric parameter error, the body curvature feature is quantified into the curvature length deviation and the curvature section deformation coefficient to obtain the body curvature deviation value. S303. Calculate the projection deviation angle of the target animal's body on the acquisition plane based on the standing tilt characteristics to obtain the standing tilt deviation value; S304. The torso axis offset deviation value, torso bending deviation value and standing tilt deviation value are weighted and fused according to the preset weights to generate a comprehensive posture deviation value. S305. Based on the comprehensive posture deviation value, the geometric parameters of body length and chest circumference of the target animal are corrected respectively to obtain geometric correction parameters.

[0058] In this embodiment, to convert the quantified body features into deviation values ​​that can be directly corrected for geometric parameters, step S30 achieves correction through the logic of deviation quantization, weighted fusion, and parameter correction. This embodiment still uses penned pigs as the application object, and the specific explanations of each step are as follows: As described in step S301 above, specifically, this step is the precise quantification of trunk axis offset deviation values. The core of this step is to use a pre-designed offset quantification model specifically designed for the body shape characteristics of penned pigs to transform the abstract trunk axis offset features into concrete deviation values ​​that can be directly correlated with the measurement errors of body length and chest circumference. The offset quantification model is as follows: This preset offset quantization model is a specialized geometric quantization model adapted to the structural characteristics of pigs (the trunk is approximately cuboid, and the dimensions of body length and chest circumference are separated), rather than a general geometric model. Its core architecture adopts a three-layer structured design of input layer - computation layer - output layer, and the functions and parameters of each layer are as follows: Input layer: explicitly receives two types of data that have been quantized in step S202 and have passed the size conversion ratio. The core parameters that can be converted into actual physical quantities are: firstly, the lateral and perpendicular distance between the trunk reference axis and the standard axis. (unit: The first factor is the degree of lateral displacement of the torso, which is directly related to the error in chest circumference measurement; the second factor is the angle between the torso's baseline axis and the standard axis. (Unit: °, corresponding to the direction of axis tilt, directly related to the error in body length measurement).

[0059] Computational layer: Employs dual logic operations of geometric component decomposition and measured error calibration, closely matching the dimensional specificity of pig body size measurements. The first step is the decomposition of the lateral / longitudinal components. Based on the principle of orthogonal decomposition in plane geometry, the axial offset is transformed into independent components directly related to the body size measurement: the lateral offset component. = × (unit: (theoretical deviation corresponding to the direction of chest circumference measurement), longitudinal offset component. = × (unit: (the theoretical deviation corresponding to the body length measurement direction); The second step is to calibrate the measured error coefficient. To eliminate the deviation between the theoretical decomposition and the actual breeding scenario, a correction coefficient based on pig-specific samples is introduced. , By collecting several sets of experimental samples from major breeds such as Large White, Landrace, and Duroc, covering all growth stages from piglets to fattening pigs, and artificially controlling... and Simultaneously measure the body size under offset conditions and the actual body size under no offset conditions, calculate the ratio of actual error to theoretical error, and obtain the result through linear regression fitting. , (fit) ≥0.95), the final correction formula is: Actual lateral offset deviation value = × Actual longitudinal offset deviation value = × .

[0060] Output layer: Clearly outputs three types of deviation values ​​that directly serve subsequent parameter correction: the first is the lateral offset deviation value of the torso axis. (unit: (1) the error range corresponding to chest circumference measurement; and 2) the longitudinal offset deviation value of the trunk axis. (unit: (corresponding to the error range in body length measurement); and thirdly, the contribution value θ of the axial angle deviation (unit: %, θ= Since the maximum angle within the physiologically reasonable range is 5°, it is mapped to 0~100% for subsequent weighted fusion adaptation.

[0061] The training and validation process of this model is fully reproducible: the sample set must contain three-dimensional samples of "different breeds × different growth stages × different offset states", with a sample size of ≥1000 groups; a linear regression algorithm (optimized for pig body parameters) is used. , The independent variable is the actual measured error, and the iteration optimization is performed. , Continue until the model prediction error is ≤0.1cm; divide the training set and test set into a 7:3 ratio, and the model is considered qualified if the average relative error of the test set is ≤3%.

[0062] The model integrates the above logic of "input feature quantification → geometric component decomposition → measured coefficient calibration → deviation value output" to finally form a quantified trunk axis offset deviation value. This deviation value directly corresponds to the measurement error of body length and chest circumference, providing an accurate quantitative basis for geometric parameter correction.

[0063] As described in step S302 above, specifically, this step is the body curvature deviation value quantification step. Through a preset mapping relationship, the abstract body curvature characteristics are transformed into a concrete deviation quantity that can be directly associated with the measurement errors of body length and chest circumference, ensuring the accuracy and operability of deviation quantification. The mapping relationship between the body curvature and geometric parameter errors is a conventional technical logic for correcting livestock and poultry body size deviations in this field. Its specific construction process is as follows: Select mainstream penned breeds such as Large White, Landrace, and Duroc pigs, covering different growth stages including piglets, fattening pigs, and multiparous sows. Accumulate and statistically analyze several groups of measured samples of healthy pigs. For each sample group, artificial intervention is used to control the pigs to exhibit different degrees of left-side bending, right-side bending, and slight arching of the back, etc., in a natural bending state. Simultaneously collect two types of core data: first, measured body size data (body length, chest circumference) obtained through image recognition in the bending state; second, actual body size data of the same pig in a naturally straightened state, measured precisely by hand. Calculate the difference between the two sets of data (i.e., body size measurement error). Then, the body curvature (characterized by the vertical distance between the bending apex and the trunk reference axis, unit: ...) is used as the basis for the error. Using the independent variable and the body size measurement error as the dependent variable, a conventional quadratic polynomial nonlinear regression analysis was employed for fitting, yielding a fixed correspondence formula (goodness of fit). (≥0.95), forming a stable and reliable preset mapping relationship.

[0064] Based on this pre-defined mapping relationship, the body bending characteristics are precisely converted into two targeted quantitative indicators: one is the bending length deviation (unit: The first indicator is the bending deviation. When a pig's body bends, image measurement can only capture the straight chord length of the front and rear ends of the torso, while the true body length is the arc length after bending. The difference between the two is the deviation, which directly corresponds to the actual error in body length measurement. The second indicator is the bending cross-sectional deformation coefficient (dimensionless, ranging from 0 to 0.05). This coefficient represents the deviation in chest circumference measurement caused by the compression of the body during bending, changing the chest cross-section from an approximately circular shape in its natural state to an ellipse. This coefficient is the ratio of the deviation to the true chest circumference, directly reflecting the degree of interference of cross-sectional deformation on chest circumference measurement. Using a conventional linear coupling method (such as weighted summation, with a weight of 0.6 for bending length deviation and 0.4 for the weighted cross-sectional deformation coefficient), the two indicators are integrated into a single body bending deviation value. This deviation value comprehensively and quantitatively reflects the combined interference of the bending posture on the two geometric parameters of body length and chest circumference, providing standardized deviation data support for the weighted fusion in the subsequent step S304.

[0065] As described in step S303 above, specifically, this step is the calculation of the standing tilt deviation value. The core is to convert the standing tilt characteristics (tilt angle, height difference) obtained in step S204 into a quantitative deviation value that characterizes the body size measurement error, based on the conventional planar projection geometric operation logic in this field. The details are as follows: The calculation process uses the standing tilt feature data determined in step S204 as the core input: first, the standing tilt angle θ (unit: °, including forward / backward tilt angle, calculated by the difference in vertical pixel coordinates); second, the height difference of key points. (unit: This refers to the actual height difference between the highest point of the shoulder and the highest point of the hip, as well as the endpoints of the fore and hind limbs on the same side, calculated by combining the vertical pixel coordinate difference with the size conversion ratio. (Originated from conversion).

[0066] The calculation logic follows the conventional principles of planar projection geometry: when a pig's body is tilted while standing, its projection onto the image acquisition plane (shooting reference plane) undergoes stretching or compression deformation, resulting in a deviation between the pixel measurement of body size and the actual physical size. In specific calculations, the shooting reference plane is used as the projection plane, and the pig's body is treated as a rigid cuboid structure. The conventional sine and cosine theorems are used for calculation: first, using the tilt angle θ as the core parameter, the ratio of the projected length of the body's longitudinal direction (body length direction) on the projection plane to the actual length is calculated (projection ratio = cosθ; this ratio is less than 1 when tilting forward / backward, characterizing the degree of projection compression); then, the height difference is considered... Verify the accuracy of the projection scale (e.g., by...) θ is derived by inversely calculating the tangent of the horizontal projection distance of the body to ensure parameter consistency.

[0067] Finally, the projection ratio is converted into a quantified standing tilt deviation value: Standing tilt deviation value = (1 - projection ratio) × baseline body length / chest circumference value. This deviation value directly represents the measurement error of body length / chest circumference caused by standing tilt (a positive deviation value indicates that the projection compression causes the measurement value to be smaller; the larger the deviation value, the more significant the error).

[0068] Those skilled in the art can calculate the tilt deviation value based on the above-disclosed input data, calculation principles and conversion logic. This deviation value can independently reflect the degree of interference of standing tilt on the geometric parameters of body length and chest circumference, and provide standardized data for the subsequent weighted fusion of comprehensive posture deviation values.

[0069] As described in step S304 above, specifically, this step is the comprehensive posture deviation value generation stage. By weighted fusion and integrating three types of single posture deviations—torso axis offset deviation value, body bending deviation value, and standing tilt deviation value—a comprehensive quantitative index that can be uniformly used for geometric parameter correction is obtained, as follows: The predetermined weights are determined based on the actual impact of the three types of deviations on the measurement of geometric parameters. The calibration process involves collecting several sets of measured samples from penned pigs, covering different types of postural abnormalities (axis offset, body curvature, standing tilt), different breeds (Large White, Landrace, Duroc), and different growth stages. Correlation analysis is performed on the three types of single postural deviations and the total error in body size measurement in each sample set. Analysis of variance (ANOVA) is used to clarify the significance of each type of deviation on the measurement error. Then, analytic hierarchy process (AHP) is used to determine the relative importance weight of each deviation. Finally, the optimal weight allocation scheme is determined: the weight for trunk axis offset deviation is 0.4 (because axis offset interferes with body length and chest circumference measurements across two dimensions, its impact is most significant), the weight for body curvature deviation is 0.3, and the weight for standing tilt deviation is 0.3. This weight allocation is merely an example; the present invention does not impose a unique limitation on the specific values ​​of the weights. Those skilled in the art can adjust the weight ratios using the same sample analysis method according to the body structure characteristics of the breed and the actual measurement accuracy requirements to achieve the same fusion effect.

[0070] The fusion process employs a conventional weighted summation method. The specific calculation logic is as follows: Comprehensive posture deviation value = Trunk axis offset deviation value × preset weight + Body bending deviation value × preset weight + Standing tilt deviation value × preset weight (calculated according to the weights in the example above: Comprehensive posture deviation value = Trunk axis offset deviation value × 0.4 + Body bending deviation value × 0.3 + Standing tilt deviation value × 0.3). Before fusion, it is ensured that the dimensions of the three types of deviation values ​​are consistent (all are quantitative deviations related to body size), and no additional normalization processing is required for direct calculation. The final generated unique comprehensive posture deviation value comprehensively integrates the geometric parameter measurement deviations caused by all postural factors such as trunk axis offset, body bending, and standing tilt, avoiding the one-sidedness of single deviation correction, and providing a core and unified quantitative basis for the accurate correction of geometric parameters in the subsequent step S305.

[0071] As described in step S305 above, specifically, this step is a precise correction step for geometric parameters, eliminating the interference of body posture factors on body size measurement based on the comprehensive posture deviation value. Based on the magnitude of the comprehensive posture deviation value, the original body length and chest circumference geometric parameters obtained in step S104 are linearly compensated and corrected, with the correction formula being: Corrected body length = Original body length × (1 − Overall posture deviation value) Corrected chest circumference = Original chest circumference × (1 − Overall posture deviation value) Through the above corrections, measurement errors caused by various body shape deviations are eliminated, and finally geometric correction parameters that fit the actual body size of the pigs are obtained, providing accurate input data without body shape interference for subsequent weight estimation.

[0072] The aforementioned techniques for constructing the offset quantization model, fitting the bending error mapping relationship, weighted fusion of multiple deviations, and linear correction of geometric parameters are all implemented around the scenario of measuring the body size of penned pigs. Those skilled in the art can flexibly adjust the model parameters, weight ratios, and correction formulas according to the actual breeding scenario, and all of these can achieve the technical objectives of this step without affecting the complete implementation of this technical solution. This embodiment does not impose a unique limitation on this.

[0073] In one embodiment, step S50 specifically includes: S501. Extract the breed characteristics of the target animal, including the breed, current growth stage and sex of the target animal; S502. Collect dynamic physiological parameters of the target animal within a preset time period, wherein the dynamic physiological parameters include food intake, water intake and activity level; S503. Retrieve a preset parameter comparison database to obtain a basic correction coefficient that matches the characteristics of the variety; S504. The food intake, water intake, and activity level are weighted according to preset weights to obtain dynamic correction increment coefficients; S505. Based on the aforementioned basic correction coefficient and dynamic correction increment coefficient, the initial weight correction coefficient is calculated. S506. According to the preset threshold range of the weight correction coefficient for animals of the same species, the initial weight correction coefficient is verified, and abnormal coefficients that exceed the threshold range are removed to obtain the weight correction coefficient.

[0074] In this embodiment, to further eliminate the interference of breed, growth stage, sex, and dynamic physiological state on weight estimation and to achieve accurate generation of weight correction coefficients, step S50 is implemented through the logic of breed feature extraction → dynamic physiological parameter collection → baseline coefficient matching → dynamic coefficient calculation → coefficient coupling → anomaly verification. This embodiment still uses penned pigs as the application object, and the specific explanations of each step are as follows: As described in step S501 above, this step is specifically the breed feature extraction stage, used to determine the static basic attributes of the target animal, providing a benchmark for subsequent coefficient matching. The breed features are the core static attributes affecting the correspondence between livestock weight and body size, specifically including: the breed of the target animal (such as mainstream domesticated breeds like Large White, Landrace, Duroc, and three-way crossbred pigs), its current growth stage (such as piglets, nursery pigs, fattening pigs, replacement gilts, and multiparous sows), and its sex (such as boars, castrated boars, and sows). These breed features can be obtained through conventional methods in the field, such as manual input into the livestock management system, electronic ear tag recognition, and image-based breed recognition. This invention does not impose a unique limitation on the specific method of feature extraction, as long as the feature information is accurate.

[0075] As described in step S502 above, this step specifically involves the dynamic physiological parameter acquisition process. It is used to obtain real-time physiological data of the target animal, reflecting its short-term growth and activity. The preset duration is a standard monitoring cycle for animal husbandry (e.g., 24 hours, 72 hours, which can be flexibly adjusted according to the required precision in animal husbandry). The dynamic physiological parameters are real-time indicators directly related to changes in body weight, specifically including: feed intake (unit: kg, the total feed weight within the preset duration, automatically collected by intelligent feeding equipment), water intake (unit: L, the total water volume within the preset duration, collected by an intelligent water flow meter), and activity level (activity duration, number of steps, or percentage of activity range within the preset duration, monitored by a behavior recognition algorithm of the image acquisition device). All of the above parameters are collected using conventional monitoring equipment in captivity settings. The data is continuous and reproducible, accurately reflecting the real-time physiological state of the target animal.

[0076] As described in step S503 above, specifically, this step is the basic correction coefficient matching step, which retrieves the static benchmark correction coefficient through a preset database. The parameter comparison database is a pre-constructed livestock and poultry weight correction benchmark database. Its construction method involves collecting measured data of healthy livestock and poultry of various breeds, growth stages, and sexes, statistically analyzing the correspondence between body size and weight, and calibrating the basic correction coefficients corresponding to different attribute combinations. The database can be updated periodically to adapt to different breeding environments. This step directly retrieves the uniquely corresponding basic correction coefficient by matching the breed characteristics extracted in step S501 with the attribute entries in the database. This coefficient is a static benchmark value used to characterize the weight correction benchmark of the target animal under standard physiological conditions.

[0077] As described in step S504 above, specifically, this step is the dynamic correction increment coefficient calculation stage. By weighted fusion of dynamic physiological parameters, an increment adjustment value reflecting the real-time state is obtained. Preset weights are determined based on the degree of influence of each dynamic parameter on weight change (example: food intake weight 0.5, water intake weight 0.2, activity level weight 0.3). These weights are for illustrative purposes only, and the specific values ​​of the weights are not uniquely limited in this invention. The dynamic correction increment coefficient is calculated using a conventional weighted summation method in the art: Dynamic correction increment coefficient = Food intake × corresponding weight + Water intake × corresponding weight + Activity level × corresponding weight. This coefficient is a dynamic increment value used to characterize the fluctuation adjustment range of the real-time physiological state to the standard weight.

[0078] As described in step S505 above, specifically, this step is the initial weight correction coefficient calculation stage, which couples the static baseline coefficient and the dynamic incremental coefficient to obtain the preliminary correction coefficient. The conventional coefficient coupling logic used in this field is employed for calculation: Initial weight correction coefficient = Base correction coefficient × (1 + Dynamic correction incremental coefficient). By combining the static baseline and the dynamic incremental coefficient, the correction coefficient simultaneously adapts to the fixed breed attributes and real-time physiological state of the target animal, thus better reflecting the actual weight situation.

[0079] As described in step S506 above, this step specifically involves verifying the weight correction coefficient, eliminating abnormal coefficients to ensure the accuracy of subsequent weight estimation. The preset threshold range for weight correction coefficients for the same breed of animal is calibrated based on extensive measured data from healthy livestock and poultry of the same breed, growth stage, and sex, representing the effective range of coefficients under normal physiological conditions. During verification, the initial weight correction coefficient is compared with the corresponding threshold range. If the initial coefficient is within the threshold range, it is retained; if it exceeds the threshold range (e.g., abnormal coefficients caused by disease, stress, or abnormal feeding), it is deemed invalid and eliminated. This process ultimately yields a valid and stable weight correction coefficient, providing a reliable basis for weight estimation.

[0080] The above-mentioned techniques for extracting breed characteristics, collecting dynamic physiological parameters, constructing parameter comparison databases, weighted calculations, and coefficient verification are all conventional technical solutions for estimating and correcting livestock and poultry weight in this field. Those skilled in the art can flexibly adjust the collection methods, weight ratios, and threshold ranges according to the actual breeding scenario and monitoring equipment conditions, and all of these can achieve the technical objectives of this step without affecting the complete implementation of this technical solution. This embodiment does not limit this to a single method.

[0081] In one embodiment, step S60 specifically includes: S601. Multiply the preliminary estimated weight with the weight correction coefficient to obtain the initial corrected weight; S602. Retrieve a preset parameter comparison database to obtain the actual weight measurement benchmark error range corresponding to the characteristics of the variety; S603. Obtain the acquisition fluctuation amplitude of the dynamic physiological parameters and the verification deviation of the weight correction coefficient, and adjust the benchmark error range of the actual weight measurement using the acquisition fluctuation amplitude and the verification deviation to obtain the target error range. S604. The rationality of the initial corrected weight is verified based on the preset physiological threshold of weight for animals of the same breed and growth stage. S605. After the verification is passed, the initial corrected weight is determined as the estimated weight of the target animal, and the target error range is integrated to obtain the estimated weight and error range of the target animal.

[0082] In this embodiment, to further improve the accuracy and reliability of weight estimation, and to complete accurate weight correction, error adaptive calibration, and rationality verification, step S60 is implemented through the logic of preliminary weight correction → benchmark error retrieval → error range adjustment → weight rationality verification → final result output. This embodiment still uses penned pigs as the application object, and the specific explanations of each step are as follows: As described in step S601 above, specifically, this step is the initial corrected weight calculation stage, completing the basic correction of the weight. The preliminary estimated weight obtained in the previous step is directly multiplied by the weight correction coefficient. The calculation formula is: Initial corrected weight = Preliminary estimated weight × Weight correction coefficient. This calculation is a standard mathematical operation for livestock and poultry weight correction in this field. Through multiplication, the correction effects of breed, growth stage, sex, and dynamic physiological state can be directly incorporated into the weight value, eliminating the weight estimation deviation caused by static attributes and real-time physiological state, and obtaining an initial corrected weight that closely matches the actual physiological state.

[0083] As described in step S602 above, specifically, this step involves retrieving the baseline error range for actual weight measurement to obtain a standard error reference range. The parameter comparison database pre-constructed in step S503 is then retrieved. Based on the target animal breed characteristics (breed, growth stage, sex) extracted in step S501, matching is performed to retrieve the pre-calibrated baseline error range for the corresponding breed in the database. This baseline error range is obtained by collecting and statistically analyzing body size-weight measurement data from several groups of healthy pigs of the same type. It serves as the standard reference error range for weight estimation in this field, used to characterize the inherent error of weight estimation under ideal acquisition conditions.

[0084] As described in step S603 above, specifically, this step is the adaptive adjustment step for the target error range, optimizing the error interval based on the actual data acquisition status. Based on the baseline error interval for actual weight measurement obtained in step S602, two types of actual deviation parameters are introduced for dynamic adjustment: first, the fluctuation range of dynamic physiological parameters (i.e., the percentage fluctuation of data on food intake, water intake, and activity levels within a preset time period, reflecting the stability of the monitoring equipment and the real-time status); second, the verification deviation of the weight correction coefficient (i.e., the deviation between the initial weight correction coefficient and the threshold center value in step S506, reflecting the reliability of the coefficient itself). Through a conventional linear superposition adjustment method in this field, these two types of deviation parameters are integrated into the baseline error interval to obtain a target error range that adapts to the current data acquisition and calculation status, making the error range more consistent with the actual measurement scenario and improving the accuracy of the error labeling.

[0085] As described in step S604 above, this step specifically involves verifying the rationality of the initial corrected weight, eliminating abnormal weight data. The preset physiological weight thresholds for animals of the same breed and growth stage are derived from standard growth data from large-scale farming, representing the upper and lower limits of reasonable weight for the corresponding breed of pigs at the current growth stage. During verification, the initial corrected weight is compared with these physiological thresholds: if the initial corrected weight is within the threshold range, it is considered reasonable and valid; if it exceeds the threshold range (e.g., due to abnormal data collection or calculation errors leading to excessively high or low weight), it is considered abnormal data, triggering a recalculation process. This verification step effectively eliminates invalid data, ensuring the physiological rationality of the final estimated weight.

[0086] As described in step S605 above, this step specifically involves outputting the final estimated weight and error range, forming a complete calculation result. After passing the rationality verification, the initial corrected weight is determined as the final estimated weight of the target animal, and simultaneously linked to the target error range obtained in step S603, ultimately outputting a complete result of "estimated weight + error range". This result includes both accurate weight values ​​and a clearly defined reliable measurement range, providing standardized and highly reliable data support for feeding management, growth monitoring, and disease early warning during the breeding process.

[0087] The aforementioned technical means of product correction, error range retrieval, adaptive adjustment, and weight rationality verification are all conventional technical solutions for livestock and poultry weight estimation and data quality control in this field. Those skilled in the art can flexibly adjust the calculation logic, error parameters, and physiological thresholds according to the actual breeding scenario and monitoring accuracy requirements, and all of these can achieve the technical purpose of this step without affecting the complete implementation of this technical solution. This embodiment does not limit this to a single method.

[0088] In one embodiment, a geometric parameter-based error correction system for estimating the weight of captive animals is provided, which corresponds to the geometric parameter-based error correction method for estimating the weight of captive animals described in the above embodiments. This geometric parameter-based error correction system for estimating the weight of captive animals includes: The image acquisition module is used to acquire image data of the target animal within a preset area, extract the pixel coordinates of the target animal's body measurement points using a preset image recognition algorithm, and calculate the geometric parameters of the target animal by combining the size parameters of a preset reference object. The feature extraction module is used to extract the body spatial distribution features of body measurement points, including trunk axis offset features, body bending features, and standing tilt features; The first correction module is used to calculate and fuse the trunk axis offset deviation value, the body bending deviation value and the standing tilt deviation value based on the body spatial distribution characteristics to obtain a comprehensive posture deviation value, and use the comprehensive posture deviation value to correct the geometric parameters to obtain geometric correction parameters; The preliminary estimation module is used to input the geometric correction parameters into a preset weight estimation model and output the preliminary estimated weight of the target animal. The coefficient calculation module is used to calculate the weight correction coefficient based on the breed characteristics and dynamic physiological parameters of the target animal. The second correction module is used to correct the preliminary estimated weight using the weight correction coefficient to obtain the estimated weight and error range of the target animal.

[0089] Specific limitations regarding the geometric parameter-based error correction system for estimating the weight of captive animals can be found in the above description of the method for correcting errors in estimating the weight of captive animals based on geometric parameters, and will not be repeated here. Each module in the aforementioned geometric parameter-based error correction system for estimating the weight of captive animals can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the corresponding operations of each module.

[0090] In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows. Figure 2 As shown, the computer device includes a processor, memory, network interface, and database connected via a system bus. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and database. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The database is used for data storage, data processing, and data analysis. The network interface is used for communication with external terminals via a network connection. When executed by the processor, the computer program implements a method for correcting errors in the estimation of captive animal weight based on geometric parameters.

[0091] In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements a method for correcting errors in the estimation of the weight of captive animals based on geometric parameters.

[0092] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements a method for correcting errors in the estimation of the weight of captive animals based on geometric parameters.

[0093] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0094] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is used as an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0095] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A method for correcting errors in estimating the weight of captive animals based on geometric parameters, characterized in that, Includes the following steps: Collect image data of the target animal within a preset area, extract the pixel coordinates of the target animal's body measurement points using a preset image recognition algorithm, and calculate the geometric parameters of the target animal by combining the size parameters of a preset reference object; Extract the body spatial distribution features of body measurement points, including trunk axis offset features, body bending features, and standing tilt features; Based on the aforementioned body spatial distribution characteristics, the trunk axis offset deviation value, body bending deviation value, and standing tilt deviation value are calculated and fused to obtain a comprehensive posture deviation value. The geometric parameters are then corrected using the comprehensive posture deviation value to obtain the geometric correction parameters. The geometric correction parameters are input into a preset weight estimation model, and the preliminary estimated weight of the target animal is output. Calculate the weight correction factor based on the breed characteristics and dynamic physiological parameters of the target animal; The initial estimated weight is corrected using the weight correction factor to obtain the estimated weight of the target animal and its error range.

2. The method for correcting errors in estimating the weight of captive animals based on geometric parameters as described in claim 1, characterized in that, The steps of acquiring image data of the target animal within a preset area, extracting the pixel coordinates of the target animal's body measurement points using a preset image recognition algorithm, and calculating the geometric parameters of the target animal in conjunction with the size parameters of a preset reference object specifically include: Several image acquisition devices are set up at fixed acquisition positions in the pre-set captive area, and fixed, pre-set reference objects with known actual sizes are set up within the complete acquisition field of view of each image acquisition device. The pre-set reference objects and the body measurement parts of the target animal are on the same shooting reference plane. The image acquisition device acquires image data of a single target animal in a natural standing position within a pre-defined captive area, and the image data includes the target animal's body and a pre-defined reference object; A preset image recognition algorithm is used to detect and locate key body measurement points and preset reference objects in the image data, and the pixel coordinates of the target animal's body measurement points and the pixel size of the preset reference objects in the image are extracted. A size conversion ratio is established based on the actual size and pixel size of a preset reference object. Based on the size conversion ratio and the pixel coordinates of the target animal's body measurement points, the geometric parameters of the target animal are calculated.

3. The method for correcting errors in estimating the weight of captive animals based on geometric parameters as described in claim 1, characterized in that, Before the step of extracting the body spatial distribution features of the body measurement points, which include trunk axis offset features, body bending features, and standing tilt features, the method also includes the alignment and correction of the pixel coordinates of the target animal's body measurement points, specifically including the following steps: Using the shooting reference plane where the preset reference object is located as a reference, the origin, horizontal axis and vertical axis of the two-dimensional reference coordinate system are defined, and the mapping relationship between the coordinate scale and the actual size is set. Extract the pixel coordinates of the key corner points of the preset reference object and substitute them into the two-dimensional reference coordinate system to calibrate and form at least three non-collinear reference coordinate anchor points; Calculate the relative positional deviation between each pixel coordinate of the target animal's body measurement point and the reference coordinate anchor point, and perform coordinate translation and scaling correction on the body measurement point based on the relative positional deviation; After calibration, the pixel coordinates of the target animal body measurement points that match the shooting reference plane are output.

4. The method for correcting errors in estimating the weight of captive animals based on geometric parameters as described in claim 1, characterized in that, The step of extracting the spatial distribution features of body measurement points, including trunk axis offset features, body curvature features, and standing tilt features, specifically includes: Based on the pixel coordinates of the target animal's body measurement points, a trunk reference axis of the target animal is fitted and constructed; Based on the relative positional relationship between the trunk reference axis and the target animal's standard trunk axis, the trunk axis offset characteristics are calculated. Based on the distribution deviation of the target animal's body measurement points on both sides of the trunk reference axis, the body bending characteristics are identified; The standing tilt feature is determined based on the difference in vertical pixel coordinates of the target animal's body measurement points; By integrating the trunk axis offset features, body bending features, and standing tilt features, the spatial distribution features of the target animal's body shape are formed.

5. The method for correcting errors in estimating the weight of captive animals based on geometric parameters as described in claim 1, characterized in that, The step of calculating and fusing the trunk axis offset deviation value, body bending deviation value, and standing tilt deviation value based on the spatial distribution characteristics of the body posture to obtain a comprehensive posture deviation value, and using the comprehensive posture deviation value to correct the geometric parameters to obtain the geometric correction parameters, specifically includes: Based on the preset offset quantization model, the torso axis offset features are converted into offset distances and offset angles along the transverse and longitudinal directions of the torso to obtain the torso axis offset deviation value. Based on the mapping relationship between the curvature of the body and the geometric parameter error, the body curvature characteristics are quantified into the curvature length deviation and the curvature section deformation coefficient to obtain the body curvature deviation value. The projection deviation angle of the target animal's body on the acquisition plane is calculated based on the standing tilt characteristics to obtain the standing tilt deviation value; The torso axis offset deviation value, torso bending deviation value and standing tilt deviation value are weighted and fused according to preset weights to generate a comprehensive posture deviation value. Based on the comprehensive posture deviation value, the geometric parameters of body length and chest circumference of the target animal are corrected to obtain geometric correction parameters.

6. The method for correcting errors in estimating the weight of captive animals based on geometric parameters as described in claim 1, characterized in that, The step of calculating the weight correction factor based on the breed characteristics and dynamic physiological parameters of the target animal specifically includes: Extract the breed characteristics of the target animal, including the breed, current growth stage and sex of the target animal; Collect dynamic physiological parameters of the target animal within a preset time period. The dynamic physiological parameters include food intake, water intake, and activity level. Retrieve a preset parameter comparison database to obtain a basic correction coefficient that matches the characteristics of the variety; The food intake, water intake, and activity level are weighted according to preset weights to obtain a dynamic correction increment coefficient; Based on the aforementioned basic correction coefficient and dynamic correction increment coefficient, the initial weight correction coefficient is calculated. The initial weight correction coefficient is verified based on the preset threshold range of the weight correction coefficient for animals of the same species. Abnormal coefficients that exceed the threshold range are removed to obtain the weight correction coefficient.

7. The method for correcting errors in estimating the weight of captive animals based on geometric parameters as described in claim 1, characterized in that, The step of correcting the preliminary estimated weight using the weight correction coefficient to obtain the estimated weight and error range of the target animal specifically includes: The initial corrected weight is obtained by multiplying the preliminary estimated weight with the weight correction factor. Retrieve a preset parameter comparison database to obtain the actual weight measurement benchmark error range corresponding to the characteristics of the variety; The acquisition fluctuation amplitude of the dynamic physiological parameters and the verification deviation of the weight correction coefficient are obtained. The acquisition fluctuation amplitude and the verification deviation are used to adjust the benchmark error range of the actual weight measurement to obtain the target error range. The initial corrected weight is verified for rationality based on the preset physiological threshold of weight for animals of the same breed and growth stage. After verification, the initial corrected weight is determined as the estimated weight of the target animal, and the target error range is integrated to obtain the estimated weight and error range of the target animal.

8. A system for correcting errors in estimating the weight of captive animals based on geometric parameters, used to implement the steps of the method for correcting errors in estimating the weight of captive animals based on geometric parameters as described in any one of claims 1-7, characterized in that, include: The image acquisition module is used to acquire image data of the target animal within a preset area, extract the pixel coordinates of the target animal's body measurement points using a preset image recognition algorithm, and calculate the geometric parameters of the target animal by combining the size parameters of a preset reference object. The feature extraction module is used to extract the body spatial distribution features of body measurement points, including trunk axis offset features, body bending features, and standing tilt features; The first correction module is used to calculate and fuse the trunk axis offset deviation value, the body bending deviation value and the standing tilt deviation value based on the body spatial distribution characteristics to obtain a comprehensive posture deviation value, and use the comprehensive posture deviation value to correct the geometric parameters to obtain geometric correction parameters; The preliminary estimation module is used to input the geometric correction parameters into a preset weight estimation model and output the preliminary estimated weight of the target animal. The coefficient calculation module is used to calculate the weight correction coefficient based on the breed characteristics and dynamic physiological parameters of the target animal. The second correction module is used to correct the preliminary estimated weight using the weight correction coefficient to obtain the estimated weight and error range of the target animal.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method for correcting errors in estimating the weight of captive animals based on geometric parameters as described in any one of claims 1-7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the method for correcting errors in estimating the weight of captive animals based on geometric parameters as described in any one of claims 1-7.

Images