Method for estimating body weight based on goose body size

Abstract

The invention relates to the technical field of machine vision, and particularly discloses a goose body size-based body weight estimation method, which comprises the following steps of: dividing a goose breeding area into n houses, constructing a house data list, and carrying out unified standard coding on goose individual information data in each house; collecting goose video data of each house, encoding the video data to obtain encoded video data, and obtaining single goose video data; key frames are extracted from the single-goose video data; detecting goose feature points of the extracted key frames by using a target detection model, and identifying and outputting a goose behavior state; and predicting the goose weight according to the goose behavior state output by the target detection model, and storing the goose weight in a database. According to the method, the limitation of single-frame isolated analysis of a traditional static image is broken through, and the correlation modeling of the goose body size and weight is upgraded to dynamic evolution modeling from instantaneous mapping; meanwhile, high unification, accuracy and consistency of data are achieved, and standard data support is provided for follow-up weight management data analysis and breeding calculation.

Description

Technical Field

[0001] The invention belongs to the technical field of machine vision and relates to a method for estimating weight based on goose body dimensions. Background Art

[0002] In the livestock industry, estimating the weight of poultry is particularly important. While this data can serve as an important basis for market prices, body size and weight are also indicators of poultry growth, reflecting physiological indicators throughout each growth cycle. This indirectly reflects the health and growth status of the poultry population, helping farmers adjust their feeding strategies in a timely manner. Currently, farmers estimate the weight of geese based on their body size, traditionally through manual measurement. However, this still presents numerous issues that lead to inaccurate measurements. Therefore, with the development of the computer vision industry, machine vision has emerged as a key advancement in the field, enabling the estimation of the weight of geese based on their body size. For example, patent publication number CN119152350A discloses a contactless cattle weight estimation device, characterized by its composition of an image data acquisition module, an image information processing module, and a neural network algorithm model platform. The device is installed on a farm to ensure proper power supply and data transmission, taking into account the farm environment, lighting conditions at different time periods, and target detection at different distances. Compared with existing technologies, this device achieves higher image acquisition efficiency and lower costs, and is adaptable to various farm environments, including nighttime monitoring and image data acquisition.

[0003] For example, patent publication number CN118781629A discloses a device and method for monitoring the size and / or weight of a flock of chickens based on posture recognition. The method comprises: collecting depth images and RGB images of the flock; identifying the flock's behavior based on the acquired depth and RGB images, obtaining an RGB image of a single chicken in a set posture and a corresponding depth image; and processing the identified RGB image of the single chicken in the set posture and the corresponding depth image to estimate the chicken's body size and / or weight. Therefore, the present invention proposes a method for identifying individual behavior in a flock of chickens, establishing a chicken body size and / or weight estimation model, and outputting changes in the chicken's body size and / or weight, providing data guidance for flock weight monitoring and optimizing feeding strategies.

[0004] The existing technology has the following problems: 1. Traditional acquisition of static RGB images and depth images may be affected by some accidental factors and can only capture the specific posture of the goose at a certain moment, such as occlusion caused by the goose's movement, resulting in incomplete image information or large errors in estimated weight.

[0005] 2. Traditional measurement methods require a large amount of human resources, but they still lack stability even under professional guidance during measurement, which will lead to reduced work efficiency during the measurement process and thus affect the resulting data. Geese are still aggressive poultry, and unexpected situations such as frightening them during measurement may cause serious injuries to workers. Goose weight is a data point with extremely high fluctuations, and simple manual measurement requires sufficient frequency to obtain a more reasonable data version. Manual measurement cannot meet the corresponding required frequency. Summary of the Invention

[0006] In view of the above problems in the prior art, the present invention provides a method for estimating weight based on goose body size, which is used to solve the above technical problems.

[0007] In order to achieve the above-mentioned and other purposes, the technical solutions adopted by the present invention are as follows: The present invention provides a method for estimating body weight based on goose body size, the method comprising the following steps: Step 1: Divide the goose breeding area into n buildings, build a building data list, and perform unified standard coding on the individual goose information data in each building; Step 2: Collect the video data of geese in each building, encode the video data to obtain the encoded video data, and obtain the video data of a single goose; Step 3: Extract key frames from single goose video data; Step 4: Use the target detection model to detect the goose feature points on the extracted key frames and identify and output the goose behavior state; Step 5: Based on the goose behavior status output by the target detection model, predict the goose weight and store it in the database; Furthermore, the building data list includes: building number, building type, building capacity; individual information data includes: species identification, variety, batch, gender, individual number, father individual code, mother individual code, generation number, date of birth, measurement date, measurement age and measurement weight; Furthermore, the unified standard coding structure is: species identification-variety-batch-sex-individual number-father's individual number-mother's individual number-generation number-date of birth-measurement date-measurement age-measurement weight.

[0008] Furthermore, the key frames include: standing and stretching frames, lying and relaxing frames, walking intermediate frames, wings-spreading frames, foraging frames, feather-grooming frames, static-rotation frames and dynamic-rotation static frames; Furthermore, a method for extracting key frames is as follows: single goose video data is input into a trained neural network model, and the neural network model predicts and outputs 1 or 0, where 1 indicates a key frame and 0 indicates a non-key frame; the neural network model includes: a temporal neural network RNN, a 3D-CNN, a LSTM, and a Transformer; the non-key frames include: repeated redundant frames and overlapping frames; Furthermore, the goose behavior states include: standing, lying down, foraging, spreading wings, and preening; the characteristic points include: the tip of the beak, the top of the head, the middle of the neck, the base of the neck, the highest point of the back, the tips of the left and right wings, the base of the left and right wings, the base of the tail, the chest, the tops of the left and right shins, and the bases of the left and right shins; Furthermore, the method for predicting goose weight includes: If the target detection model outputs that the goose behavior state is standing, the body size index parameters of the goose in the standing state are input into the weight estimation model to predict the output weight; If the target detection model outputs that the goose's behavior state is lying down, the body size index parameters of the goose in the lying down state are input into the weight estimation model to predict the output weight; If the target detection model outputs that the goose behavior state is foraging, the goose body size index parameters in the foraging state are input into the weight estimation model to predict the output weight; If the target detection model outputs that the goose behavior state is wings spread, the goose body size index parameters in the wings spread state are input into the weight estimation model to predict the output weight; If the target detection model outputs that the goose behavior state is preening, the goose body size index parameters in the preening state are input into the weight estimation model to predict the output weight; Furthermore, the goose body size index parameters include: chest depth, chest circumference, pelvic width, body length, body width, tibia length, and neck length.

[0009] As described above, the method for estimating body weight based on goose body size provided by the present invention has at least the following beneficial effects: 1. By dividing the goose breeding area into buildings and uniformly coding the individual goose information data, it can support the farm to achieve refined and compliant data management; achieve a high degree of data unification, and provide standard data support for subsequent data analysis and breeding calculations.

[0010] 2. Collect goose video data and extract key frames. Use the target detection model to detect goose feature points in the extracted key frames, identify and output the goose's behavioral status, and predict the goose's weight based on the output goose behavioral status. This breaks through the limitations of "single-frame isolated analysis" of static images and upgrades the correlation modeling of goose body size and weight from "instantaneous mapping" to "dynamic evolution modeling." This is more in line with the actual scenarios of high-frequency weight fluctuations and complex and changeable postures in poultry farming, and is an important extension of machine vision technology from static analysis to spatiotemporal joint modeling.

[0011] 3. Reduce manual data entry errors and time costs, improve data accuracy and consistency. At the same time, ensure the security of data size, storage and protection, and improve data integrity and security. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0013] Figure 1 The figure is a schematic diagram showing the connection of various steps in a method for estimating body weight based on goose body size according to the present invention. DETAILED DESCRIPTION

[0014] The above contents described below in conjunction with the implementation of the present invention are merely examples and explanations of the concept of the present invention. Those skilled in the art may make various modifications or additions to the described specific embodiments or replace them in a similar manner. As long as they do not deviate from the concept of the invention or exceed the scope defined by the claims, they shall fall within the scope of protection of the present invention.

[0015] Example 1 See also Figure 1 As shown, a method for estimating body weight based on goose body size includes the following steps: Step 1: Divide the goose breeding area into n buildings, and construct a building data list, uniformly standardize the individual information data of the geese under each building, and store it in the database system; the building data list includes: building number, building category, and building capacity; the individual information data includes: species identification, breed, batch, sex, individual number, father's individual code, mother's individual code, generation number, date of birth, measurement date, measurement age and measurement weight; the unified standard coding structure is: species identification-breed-batch-sex-individual number-father's individual number-mother's individual number-generation number-date of birth-measurement date-measurement age-measurement weight; the database system, such as MySQL or PostgreSQL, can add, modify, delete, export, search and reset the individual information data of the geese, which is convenient for subsequent data query, statistics and analysis; at the same time, the database can be backed up and stored regularly to prevent data loss.

[0016] The advantage of dividing the goose breeding area into n buildings and constructing a building data list is that the building number can be automatically verified for uniqueness, and an error will be prompted in case of duplication; building categories such as "building 01 for brooding chicks" or "building 03 for adult geese" can be used to achieve classification and screening according to the breeding stage; the building capacity can be calculated based on the breeding density standard (such as 10 geese / m2 in the brooding period and 5 geese / m2 in the adult goose period); n is an integer greater than 1.

[0017] It should be noted that the species identification coding rule uses a capital letter "G" to represent goose, "D" to represent duck, and "C" to represent chicken, which is convenient for distinguishing in a multi-species breeding environment; the breed coding rule uses two digits to represent common goose breeds, such as "01" for Lionhead goose, "02" for Taihu goose, etc., which can be pre-defined and assigned according to the actual breeding breeds; the batch coding rule uses a combination of the last two digits of the year and the batch number, such as "2503" for the third batch in 2025, to record the batch to which the goose belongs, which is convenient for unified breeding management operations, such as vaccination and growth cycle statistics; the male and female coding rule uses a single capital letter, "M" for male (Male), and "F" for female (Female), which can be used for group management and breeding plans; if the individual code of the father and the individual code of the mother are unknown, they are represented by "NULL"; the generation number coding rule uses the capital letter "P" plus a number, such as "P3" for the third generation; The coding rule for birth date is in the format of "YYYYMMDD", such as "20250315" means birth on March 15, 2025; the coding rule for measurement date is in the format of "yyyymmdd", such as "20250315" means measurement on March 15, 2025; the coding rule for measurement age is represented by three digits, such as "060" means the measurement age of the goose is 60 days.

[0018] An example explanation is that if a goose is a female Taihu goose from the third batch in 2025, the individual number is 0015, it belongs to the fourth generation, was born on March 20, 2025, and was measured on May 20, 2025, with an age of 60 days. The individual code of its father is G-02-2402-M-0008, and the individual code of its mother is G-02-2402-F-0012. The complete unified standard code of the goose is: G-02-2503-M-0015-G-02-2402-M-0008-G-02-2402-F-0012-P4-20250320-20250520-060; through the unified standard coding structure and rules, various key information of individual geese can be comprehensively and accurately recorded, providing strong support for the scientific breeding and management of the geese.

[0019] Step 2: Collect the video data of geese in each building, encode the video data to obtain the encoded video data, and obtain the video data of a single goose; It should be noted that in some possible embodiments, IP cameras or industrial cameras are selected, which have higher stability, resolution and frame rate, can provide clearer and smoother video images, and are also applicable to changing lighting conditions; a lens with a suitable focal length can be selected according to the size of the breeding site and the installation position of the camera. A wide-angle lens is suitable for large breeding areas and can cover a wider range; a telephoto lens is suitable for situations where close-up shots of specific areas or targets are required.

[0020] The interval and duration of video collection can be determined according to the needs of the research. For example, the interval for collecting videos during feeding time or activity time during the day can be shortened, such as collecting video data every 10 minutes. The collection duration can be adjusted according to actual conditions. The collection duration can be set to 10 minutes according to needs to obtain sufficient video data. For example, at night, when geese are relatively quiet, the interval for collecting videos can be shortened, such as collecting video data once an hour, and each collection duration can be 5 minutes.

[0021] Use efficient encoding methods such as H.264 or H.265 for video data. H.264 is the most widely used video encoding standard, offering high compression efficiency and image quality. H.265 further improves the compression ratio based on H.264, reducing the size of video files while maintaining the same image quality.

[0022] Single goose video data is obtained by combining the YOLOv8, DeepSORT, and Mask R-CNN algorithms to generate single goose videos. The YOLOv8 algorithm can be used to detect and locate the goose's bounding box, the DeepSORT algorithm can associate cross-frame IDs, and the Mask R-CNN algorithm can extract the single goose mask. Based on the cross-frame ID and the single goose mask, the single goose video is cropped and generated.

[0023] Step 3: extract key frames from the single goose video data, wherein the key frames include: standing and stretching frame, lying and relaxing frame, walking intermediate frame, wings-spreading frame, foraging frame, preening frame, static-rotation frame and dynamic-rotation static frame; It should be noted that key frames must meet the following core conditions to ensure that they can be effectively used for goose body size parameter measurement, such as typical goose postures (such as standing, walking, spreading wings, lowering the head to eat, lying still, etc.), especially postures with stretched bodies and no serious occlusion (for example, when standing with the head raised, the neck straight, and the side of the body fully exposed, which is convenient for measuring parameters such as body length and chest circumference). The corresponding frames can be used as key frames when the goose's motion state changes significantly (such as from walking to spreading wings, from lying still to standing). Such frames can capture the dynamic change boundaries of body size parameters. For example, frames with large differences in video content (such as the position, posture, and body contour of the goose) compared with adjacent frames, for example, frames where the back and abdomen perspectives switch due to the goose turning around, or frames where the relative positions of multiple geese change resulting in changes in occlusion, can be used as key frames. It should also be noted that for continuous frames with high repetition, such as multiple consecutive frames when the goose is stationary, only the first frame is retained. Standing stretch frame: The goose stands upright with its neck naturally extended, its feet apart, its side facing the camera, and its wings close to its body (not contracted or tense), making it easy to measure body length, leg length, chest circumference, etc. Lying and relaxed frame: The goose is in a lying position, with its body stretched out and not curled up, its head raised or placed naturally, its wings slightly spread or touching the ground, and the outline of its chest and abdomen can be clearly seen. This is used to measure body width and height (lying height); Walking intermediate frame: The goose is in the middle stage of the gait cycle (e.g., one foot is fully landed and the body center of gravity is stable), with no overlapping limbs and clear joint angles. This is used to analyze the impact of stride length and walking posture on body size. Wings-spread frame: Wings fully extended (preparing for flight or preening feathers), with the wingtips clearly positioned so the span can be measured; or wings partially extended (e.g., in a defensive stance when startled). Foraging frame: The neck is stretched forward to the lowest point, with the mouth close to the ground or the feeding trough, used to measure neck length and head pitch angle (which affects the correction of body length measurement); Feather-grooming frame: The neck is bent to the back, the beak touches the feathers, and the details of the body (such as feather texture and skin color) are clear, which can assist in judging the health status (such as feather gloss); Static-to-transition frame: the moment when you turn from lying down to standing (the body leaves the ground for the first time and the leg muscles are tense), or the starting frame when you turn from standing to walking (one foot is lifted); Moving-to-Still Frame: The frame where walking turns to standing (both feet touch the ground at the same time and the body is stable), or the frame where wings are retracted after being spread (wings are completely close to the body); The method for extracting key frames is to input the single goose video data into a trained neural network model, and the neural network model predicts and outputs 1 or 0, where 1 indicates a key frame and 0 indicates a non-key frame. Neural network models include but are not limited to: RNN, 3D-CNN, LSTM, and Transformer. The non-key frames include: repeated redundant frames and overlapping frames, etc. Repeated redundant frames: The posture, position, and behavior of the geese do not change significantly over multiple consecutive frames; Overlapping frames: The body parts of multiple geese overlap (for example, when lying side by side, their necks or wings are crossed); The method for training the neural network model is as follows: manually pre-select 500 goose video data as samples or select more samples to improve the accuracy of the model, and pre-process the goose video data samples. The pre-processing process includes resolution unification, frame rate normalization, video segmentation, deblurring and noise reduction, etc.; the pre-processed goose video data samples are annotated with key frames and non-key frames in the video by using annotation tools such as LabelStudio and CVAT; non-key frames include repeated frames (the goose posture remains unchanged for multiple consecutive frames, such as stillness), low-quality frames (blurred, too dark, most of the body is blocked by other geese), such as two consecutive frames of geese slightly turning their heads with little difference, etc.; the goose video data samples are divided into training sets and test sets, and a classifier is constructed; the goose video data in the training set is input into the neural network model, and the neural network model predicts an output of 1 or 0; the classifier is trained to obtain an initial classifier, the initial classifier is tested using the test set, and a classifier that meets the preset accuracy is output as the neural network model, and the classifier is one of the temporal neural network RNN or the long short-term memory network LSTM; Step 4: Use the target detection model to detect the goose feature points on the extracted key frames, and identify and output the goose behavior status; the goose behavior status includes: standing, lying, foraging, spreading wings, and preening; the feature points include: tip of the beak, top of the head, middle of the neck, base of the neck, highest point of the back, left and right wing tips, left and right wing bases, tail base, chest, left and right shin tops, and left and right shin bases; the target detection models include: YOLO and FasterR-CNN, etc. The target detection model is an existing technology and will not be described in detail here.

[0024] Step 5: Based on the goose behavior status output by the target detection model, predict the goose weight and store it in the database. The method for predicting goose weight includes: If the target detection model outputs that the goose behavior state is standing, the body size index parameters of the goose in the standing state are input into the weight estimation model to predict the output weight; If the target detection model outputs that the goose's behavior state is lying down, the body size index parameters of the goose in the lying down state are input into the weight estimation model to predict the output weight; If the target detection model outputs that the goose behavior state is foraging, the goose body size index parameters in the foraging state are input into the weight estimation model to predict the output weight; If the target detection model outputs that the goose behavior state is wings spread, the goose body size index parameters in the wings spread state are input into the weight estimation model to predict the output weight; If the target detection model outputs that the goose behavior state is preening, the goose body size index parameters in the preening state are input into the weight estimation model to predict the output weight; The method for training a weight estimation model involves collecting goose body measurement parameters under historical goose behavior states as a sample set, dividing the sample set into a training set and a test set; constructing a classifier, inputting the sample set into the weight estimation model, and having the model predict and output weight; training the classifier, stopping training when the output meets an accuracy rate of 98% or greater, to obtain an initial classifier, which is then tested using a test set. The classifier that meets the prediction accuracy is used as the weight estimation model; the weight estimation model includes a logistic regression model, a random forest model, a multivariate linear regression model, and the like. The goose body measurement parameters include chest depth, chest circumference, pelvic width, body length, body width, shank length, and neck length.

[0025] Example 2 This embodiment provides a method for estimating weight based on goose body measurements, and also provides a data security operation management method. The administrator is responsible for reading and writing all data, allocating role permissions, and viewing logs; the breeder is limited to entering and correcting data on geese in the area he is responsible for; and the analyst is responsible for read-only access to all data and writing analysis results. It realizes the full-link binding of "personnel-operation-data", ensuring that each piece of data has a clear responsible subject and operation track, and provides technical guarantees for the security and traceability of breeding data. The core is to build a closed loop of responsibility from data entry to management through fine-grained control of role permissions, real-time capture of operation logs, and multiple protections of security authentication, supporting farms to achieve refined and compliant data management goals.

[0026] It should be understood that in the various embodiments of the present application, the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0027] It should be understood that determining B based on A does not mean determining B based solely on A. B can also be determined based on A and / or other information.

[0028] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of this application. Therefore, the scope of protection of this application should be based on the scope of protection of the claims.

[0029] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for estimating body weight based on goose body size, characterized in that: The steps include: Step 1: Divide the goose breeding area into n buildings, build a building data list, and perform unified standard coding on the individual goose information data in each building; Step 2: Collect the video data of geese in each building, encode the video data to obtain the encoded video data, and obtain the video data of a single goose; Step 3: Extract key frames from single goose video data; Step 4: Use the target detection model to detect the goose feature points on the extracted key frames and identify and output the goose behavior state; Step 5: Based on the goose behavior status output by the target detection model, predict the goose weight and store it in the database.

2. The method for estimating body weight based on goose body size according to claim 1, characterized in that: The building data list includes: building number, building category, building capacity; individual information data includes: species identification, variety, batch, gender, individual number, father individual code, mother individual code, generation number, date of birth, measurement date, measurement age and measurement weight.

3. The method for estimating body weight based on goose body size according to claim 1, characterized in that: The unified standard coding structure is: species identification-variety-batch-sex-individual number-father individual number-mother individual number-generation number-date of birth-measurement date-measurement age-measurement weight.

4. The method for estimating body weight based on goose body size according to claim 1, characterized in that: The key frames include: standing and stretching frames, lying and relaxing frames, walking intermediate frames, wings-spreading frames, foraging frames, feather-grooming frames, static-rotation frames and dynamic-rotation static frames.

5. The method for estimating body weight based on goose body size according to claim 1, characterized in that: The method of extracting key frames is as follows: input the single goose video data into the trained neural network model, and the neural network model predicts the output 1 or 0, where 1 indicates a key frame and 0 indicates a non-key frame; The neural network model includes: temporal neural network RNN, 3D-CNN, LSTM and Transformer; the non-key frames include: repeated redundant frames and overlapping frames.

6. The method for estimating body weight based on goose body size according to claim 5, characterized in that: The goose behavior states include: standing, lying down, foraging, spreading wings, and preening feathers; the characteristic points include: tip of the beak, top of the head, middle of the neck, base of the neck, highest point of the back, tips of the left and right wings, base of the left and right wings, base of the tail, chest, tops of the left and right shins, and bases of the left and right shins.

7. The method for estimating body weight based on goose body size according to claim 1, characterized in that: Methods for predicting goose weight include: If the target detection model outputs that the goose behavior state is standing, the body size index parameters of the goose in the standing state are input into the weight estimation model to predict the output weight; If the target detection model outputs that the goose's behavior state is lying down, the body size index parameters of the goose in the lying down state are input into the weight estimation model to predict the output weight; If the target detection model outputs that the goose behavior state is foraging, the goose body size index parameters in the foraging state are input into the weight estimation model to predict the output weight; If the target detection model outputs that the goose behavior state is wings spread, the goose body size index parameters in the wings spread state are input into the weight estimation model to predict the output weight; If the target detection model outputs that the goose behavior state is preening, the goose body size index parameters in the preening state are input into the weight estimation model to predict the output weight.

8. The method for estimating body weight based on goose body size according to claim 7, characterized in that: The goose body measurement index parameters include: chest depth, chest circumference, pelvic width, body length, body width, tibia length, and neck length.

Images