A high-yield five-dragon goose screening method based on dynamic ultrasonic body shape recognition
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- POULTRY INSTITUTE SHANDONG ACADEMY OF AGRICULTURAL SCIENCE (SHANDONG SPECIFIC PATHOGEN FREE CHICKS RESEARCH CENTER)
- Filing Date
- 2026-05-07
- Publication Date
- 2026-07-14
AI Technical Summary
Traditional high-yield screening methods for Wulong geese rely on manual observation and measurement, which are subject to strong subjectivity, low efficiency, poor accuracy, and are prone to triggering stress responses. Existing technologies have failed to effectively utilize dynamic ultrasonic body posture recognition technology, resulting in insufficient screening accuracy.
A multi-view ultrasonic sensor array was used to collect dynamic body posture data of Wulong geese. Combined with an improved feature extraction algorithm and AI recognition model, a high-yield trait feature library was constructed to achieve non-contact and accurate body posture feature extraction and matching.
It achieves a comprehensive and objective reflection of the physical condition of Wulong geese, avoids the subjectivity and stress interference of manual measurement, has the ability to resist environmental interference, improves the accuracy and efficiency of screening, and adapts to complex breeding scenarios.
Smart Images
Figure CN122388511A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the fields of artificial intelligence and ultrasonic detection technology, specifically to a method for screening high-yield Wulong geese based on dynamic ultrasonic body posture recognition. Background Technology
[0002] Wulong geese are a high-quality egg-laying goose breed, and their egg production performance directly affects the breeding efficiency. Therefore, the accurate selection of high-yielding Wulong geese is a key link in improving the quality and efficiency of the Wulong goose breeding industry. Traditional high-yielding screening methods for Wulong geese mainly rely on manual observation and manual measurement. This involves manually judging the appearance, vitality, body proportions, and other characteristics of the geese, or using tools such as calipers to measure body size parameters, and then combining this with experience to judge the egg production potential.
[0003] While traditional screening methods are widely used, they have several drawbacks: First, they are highly subjective, relying on human experience, and differing judgment criteria among individuals can easily lead to screening errors. Second, they are inefficient, requiring individual observation and measurement, making rapid screening of large groups impossible. Third, they are inaccurate, with limited precision in manual measurement and difficulty in capturing the dynamic physical characteristics of geese in their natural activity state, which often better reflect their growth, development, reproductive health, and egg production performance. Fourth, they can easily trigger stress responses, as manual contact measurement can disturb the geese, affecting their normal growth and physiological state, thus interfering with the objectivity of the screening results.
[0004] With the development of artificial intelligence and ultrasonic detection technology, their application in the Wulong goose breeding field has gradually become widespread. Dynamic ultrasonic body posture data acquisition technology can penetrate environmental interference such as dust and water mist to achieve non-contact and accurate capture of Wulong goose body posture characteristics. Wulong goose feature AI recognition algorithm can realize intelligent analysis and matching of features, providing a new technical approach for the selection of high-yield Wulong geese.
[0005] Currently, there is no mature technical solution for dynamic ultrasonic body shape recognition and high-yield trait matching specifically for Wulong geese. Existing technologies are mostly applicable to large animal breeds and do not fully consider the breed-specific body shape characteristics and feather thickness of Wulong geese, resulting in insufficient screening accuracy and failing to meet the actual needs of screening high-yield goose eggs from Wulong geese. Summary of the Invention
[0006] This invention provides a method for screening high-yield Wulong geese based on dynamic ultrasonic body posture recognition. The method is rationally designed and employs dynamic ultrasonic body posture data acquisition technology. A multi-view ultrasonic sensor array captures the dynamic body posture characteristics of Wulong geese in their natural activity state, generating body contour data and performing precise preprocessing. Combined with an improved feature extraction algorithm, it can accurately extract core body posture characteristic parameters such as body length, chest width, and pubic bone distance. This not only comprehensively and objectively reflects the body posture of Wulong geese, avoiding the subjectivity and stress interference of manual observation and contact measurement, but also has the advantage of resisting environmental interference such as dust, water mist, and light changes, adapting to complex breeding scenarios. It provides stable and accurate basic data for high-yield screening. Furthermore, a high-yield trait feature library is constructed specifically for the Wulong goose breed, improving the targeting and accuracy of screening and solving the problems existing in the prior art.
[0007] The technical solution adopted by the present invention to solve the above-mentioned technical problems is as follows: A method for screening high-yield Wulong geese based on dynamic ultrasonic body posture recognition, the screening method comprising the following steps: S1, the screening system collects dynamic distance data of the five dragon geese in their natural activity state through a multi-view ultrasonic sensor array, generates body contour data of the five dragon geese, and performs noise reduction, splicing and calibration preprocessing on the body contour data. S2. Based on the body contour data after denoising, splicing and calibration preprocessing, the contour segmentation algorithm is used to separate the individual Wulong geese from the background. Then, through the feature point localization and fitting algorithm, multiple core body contour parameters of the Wulong geese are extracted. The core body contour parameters include at least body oblique length, chest width, chest depth, hip width and neck length. S3. Construct a Wulong Goose Feature AI Recognition Model. Input the extracted core body shape feature parameters into the Wulong Goose Feature AI Recognition Model and perform similarity matching with the preset Wulong Goose high-yield trait feature library. When the similarity reaches the preset threshold, it is determined to be a high-yield candidate Wulong Goose and the screening result is output.
[0008] The multi-view ultrasonic sensor array includes nine ultrasonic sensors, which are evenly distributed around and on top of the breeding area. The detection range overlap between adjacent ultrasonic sensors is no less than 25%, ensuring full coverage of the activity area of the five dragon geese and capturing dynamic data of their various natural activity postures such as walking, feeding, and standing. The ultrasonic sensor's acquisition frame rate is set to 10~20 fps to balance data acquisition integrity and efficiency; the ultrasonic sensor's detection frequency is set to 40~80 kHz to balance detection accuracy and anti-interference capability.
[0009] During preprocessing, a median filtering algorithm is used to remove noise points from the body contour data to avoid the impact of environmental interference on feature extraction; a coordinate calibration algorithm is used to stitch together multi-view data to ensure that data collected from different perspectives can be accurately aligned; a temperature compensation unit is used to collect ambient temperature, and a distance correction algorithm is used to compensate for the influence of temperature on the propagation speed of ultrasonic waves, generating a complete and clear body contour model of the five dragon geese.
[0010] The contour segmentation algorithm adopts a threshold-based segmentation method, which distinguishes the individual geese from the background area based on the gray-level difference of the distance data, ensuring the accuracy of the segmentation. The feature point localization and fitting algorithm adopts an improved Canny edge detection algorithm combined with the Hough transform algorithm. It accurately locates the key feature points of the head, neck, chest, abdomen and pubic area of the five dragon geese through contour feature matching, and then fits the key feature points through the least squares method to calculate the corresponding core body feature parameters.
[0011] The steps to construct the Five Dragon Goose Feature AI Recognition Model are as follows: S3.1 Collect dynamic 3D body shape data of Wulong geese with known egg production performance and corresponding egg production records, label the body shape characteristic parameters of high-producing Wulong geese, and construct training dataset and test dataset; S3.2, a network structure integrating CNN and LSTM is used as the basic model. The CNN network is used to extract deep features of body posture parameters, and the LSTM network is used to capture the temporal variation of dynamic body posture features. The training dataset is input into the model for training, and the model parameters are optimized by an adaptive learning rate adjustment algorithm so that the model converges to the preset accuracy. S3.3 introduces an attention mechanism to strengthen the weight of core body posture feature parameters and improve the model's ability to recognize key features; the dropout algorithm is used to prevent model overfitting and improve the model's generalization ability, resulting in an optimized AI recognition model for the five dragon geese features.
[0012] The construction method of the Wulong Goose high-yield trait feature library is as follows: collect a large number of body feature parameters of high-yield Wulong geese, use K-means clustering algorithm to analyze the parameters, determine the high-yield threshold range of each body feature parameter, form the Wulong Goose high-yield trait feature library, and provide a standard basis for similarity matching.
[0013] The screening system includes: A dynamic ultrasonic data acquisition module is used to acquire dynamic distance data of the five dragon geese in their natural activity state through a multi-view ultrasonic sensor array, generate body contour data and complete preprocessing. The body feature extraction module is used to separate the individual geese from the background using a contour segmentation algorithm based on the denoised, stitched and calibrated body contour data, and then extract multiple core body feature parameters of the geese using a feature point localization and fitting algorithm. The AI recognition and matching module is used to construct a Wulong goose feature AI recognition model, match the extracted body feature parameters with the Wulong goose high-yield trait feature library, and output the screening results. The data storage module is used to store the collected 3D body shape data, extracted feature parameters, high-yield trait feature library and screening results, and supports data query, export and backup; The display interaction module is used to display the body contour model of the five dragon geese, the extracted feature parameters and the filtering results in real time. It supports users to set parameters and query filtering results, thereby improving the ease of use of the system.
[0014] The dynamic ultrasonic data acquisition module includes a synchronization control unit and a temperature compensation unit. The synchronization control unit is used to control the synchronous acquisition of the multi-view ultrasonic sensor array to ensure the time consistency of the data acquired by each sensor and improve the accuracy of data stitching. The AI recognition and matching module includes a model update unit, which is used to periodically update the AI recognition model of Wulong geese features based on newly collected body shape data and egg production performance data, thereby improving the matching accuracy of the model.
[0015] This invention employs the aforementioned structure and method, using a multi-view ultrasonic sensor array to capture the dynamic physical characteristics of Wulong geese during their natural activity state. It generates body contour data and performs precise preprocessing, comprehensively and objectively reflecting the physical condition of Wulong geese, avoiding the subjectivity and stress interference of manual observation and contact measurement. Furthermore, it possesses advantages such as resistance to environmental interference from dust, water mist, and light changes, adapting to complex breeding scenarios and providing stable and accurate basic data for high-yield selection. Through an improved feature extraction algorithm, it can accurately extract core physical characteristic parameters of Wulong geese, such as body length, chest width, and pubic bone distance. These parameters are closely related to the egg-laying performance of Wulong geese, providing a basis for high-yield selection. Trait matching provides crucial evidence; the constructed AI recognition model for Wulong geese, which integrates CNN and LSTM, enables intelligent matching of physical characteristics with high-yield traits, significantly improving screening efficiency and allowing for rapid screening of large-scale Wulong goose populations. Furthermore, the model incorporates attention mechanisms and dropout algorithms, resulting in significantly higher matching accuracy than traditional manual screening methods. By setting up data storage, model updates, and interactive display functions, it not only achieves efficient screening of high-yield Wulong geese but also accumulates breeding data, providing data support for the genetic improvement of superior Wulong geese. This model possesses significant practical application value and promising prospects, offering advantages such as comprehensive accuracy, practicality, and reliability. Attached Figure Description
[0016] Figure 1 This is a schematic diagram of the process of the present invention.
[0017] Figure 2 This is a schematic diagram of the screening system of the present invention.
[0018] Figure 3 This is a schematic diagram of the structure of the multi-view ultrasonic sensor array of the present invention.
[0019] Figure 4 This is a schematic diagram of the feature point positioning of the five dragon goose body shape wheel of the present invention. Detailed Implementation
[0020] To clearly illustrate the technical features of this solution, the invention will be described in detail below through specific implementation methods and in conjunction with the accompanying drawings.
[0021] like Figures 1-4 As shown, a method for screening high-yield Wulong geese based on dynamic ultrasonic body posture recognition includes the following steps: S1, the screening system collects dynamic distance data of the five dragon geese in their natural activity state through a multi-view ultrasonic sensor array, generates body contour data of the five dragon geese, and performs noise reduction, splicing and calibration preprocessing on the body contour data. S2. Based on the body contour data after denoising, splicing and calibration preprocessing, the contour segmentation algorithm is used to separate the individual Wulong geese from the background. Then, through the feature point localization and fitting algorithm, multiple core body contour parameters of the Wulong geese are extracted. The core body contour parameters include at least body oblique length, chest width, chest depth, hip width and neck length. S3. Construct a Wulong Goose Feature AI Recognition Model. Input the extracted core body shape feature parameters into the Wulong Goose Feature AI Recognition Model and perform similarity matching with the preset Wulong Goose high-yield trait feature library. When the similarity reaches the preset threshold, it is determined to be a high-yield candidate Wulong Goose and the screening result is output.
[0022] The multi-view ultrasonic sensor array includes nine ultrasonic sensors, which are evenly distributed around and on top of the breeding area. The detection range overlap between adjacent ultrasonic sensors is no less than 25%, ensuring full coverage of the activity area of the five dragon geese and capturing dynamic data of their various natural activity postures such as walking, feeding, and standing. The ultrasonic sensor's acquisition frame rate is set to 10~20 fps to balance data acquisition integrity and efficiency; the ultrasonic sensor's detection frequency is set to 40~80 kHz to balance detection accuracy and anti-interference capability.
[0023] During preprocessing, a median filtering algorithm is used to remove noise points from the body contour data to avoid the impact of environmental interference on feature extraction; a coordinate calibration algorithm is used to stitch together multi-view data to ensure that data collected from different perspectives can be accurately aligned; a temperature compensation unit is used to collect ambient temperature, and a distance correction algorithm is used to compensate for the influence of temperature on the propagation speed of ultrasonic waves, generating a complete and clear body contour model of the five dragon geese.
[0024] The contour segmentation algorithm adopts a threshold-based segmentation method, which distinguishes the individual geese from the background area based on the gray-level difference of the distance data, ensuring the accuracy of the segmentation. The feature point localization and fitting algorithm adopts an improved Canny edge detection algorithm combined with the Hough transform algorithm. It accurately locates the key feature points of the head, neck, chest, abdomen and pubic area of the five dragon geese through contour feature matching, and then fits the key feature points through the least squares method to calculate the corresponding core body feature parameters.
[0025] The steps to construct the Five Dragon Goose Feature AI Recognition Model are as follows: S3.1 Collect dynamic 3D body shape data of Wulong geese with known egg production performance and corresponding egg production records, label the body shape characteristic parameters of high-producing Wulong geese, and construct training dataset and test dataset; S3.2, a network structure integrating CNN and LSTM is used as the basic model. The CNN network is used to extract deep features of body posture parameters, and the LSTM network is used to capture the temporal variation of dynamic body posture features. The training dataset is input into the model for training, and the model parameters are optimized by an adaptive learning rate adjustment algorithm so that the model converges to the preset accuracy. S3.3 introduces an attention mechanism to strengthen the weight of core body posture feature parameters and improve the model's ability to recognize key features; the dropout algorithm is used to prevent model overfitting and improve the model's generalization ability, resulting in an optimized AI recognition model for the five dragon geese features.
[0026] The construction method of the Wulong Goose high-yield trait feature library is as follows: collect a large number of body feature parameters of high-yield Wulong geese, use K-means clustering algorithm to analyze the parameters, determine the high-yield threshold range of each body feature parameter, form the Wulong Goose high-yield trait feature library, and provide a standard basis for similarity matching.
[0027] The screening system includes: A dynamic ultrasonic data acquisition module is used to acquire dynamic distance data of the five dragon geese in their natural activity state through a multi-view ultrasonic sensor array, generate body contour data and complete preprocessing. The body feature extraction module is used to separate the individual geese from the background using a contour segmentation algorithm based on the denoised, stitched and calibrated body contour data, and then extract multiple core body feature parameters of the geese using a feature point localization and fitting algorithm. The AI recognition and matching module is used to construct a Wulong goose feature AI recognition model, match the extracted body feature parameters with the Wulong goose high-yield trait feature library, and output the screening results. The data storage module is used to store the collected 3D body shape data, extracted feature parameters, high-yield trait feature library and screening results, and supports data query, export and backup; The display interaction module is used to display the body contour model of the five dragon geese, the extracted feature parameters and the filtering results in real time. It supports users to set parameters and query filtering results, thereby improving the ease of use of the system.
[0028] The dynamic ultrasonic data acquisition module includes a synchronization control unit and a temperature compensation unit. The synchronization control unit is used to control the synchronous acquisition of the multi-view ultrasonic sensor array to ensure the time consistency of the data acquired by each sensor and improve the accuracy of data stitching. The AI recognition and matching module includes a model update unit, which is used to periodically update the AI recognition model of Wulong geese features based on newly collected body shape data and egg production performance data, thereby improving the matching accuracy of the model.
[0029] The working principle of a high-yield Wulong goose screening method based on dynamic ultrasonic body posture recognition in this invention is as follows: Dynamic ultrasonic body posture data acquisition technology is used to capture the dynamic body posture characteristics of Wulong geese in their natural activity state via a multi-view ultrasonic sensor array. Body posture contour data is generated and accurately preprocessed. Combined with an improved feature extraction algorithm, core body posture characteristic parameters such as body length, chest width, and pubic bone distance of the Wulong goose can be accurately extracted. This not only comprehensively and objectively reflects the body posture of the Wulong goose, avoiding the subjectivity and stress interference of manual observation and contact measurement, but also has the advantage of resisting environmental interference such as dust, water mist, and light changes. It can adapt to complex breeding scenarios and provides stable and accurate basic data for high-yield screening. Simultaneously, a high-yield trait feature library is constructed based on the breed specificity of Wulong geese, improving the targeting and accuracy of the screening.
[0030] The screening method in the overall scheme specifically includes the following steps: S1. Dynamic Ultrasonic Body Shape Data Acquisition: A multi-view ultrasonic sensor array consisting of 9 ultrasonic sensors was constructed in the breeding shed during the rearing and egg-laying periods of the Wulong geese. The array was evenly distributed around the perimeter and top of the shed, with an overlap of 30% between adjacent sensors. The detection frequency was preset to 60 kHz, and the acquisition frame rate was set to 15 fps. Dynamic distance data of the Wulong geese's natural activities were continuously collected for 2 hours using the sensor array to generate body shape contour data for each goose. Furthermore, a median filtering algorithm was used to remove noise points from the body shape contour data, and a coordinate calibration algorithm was used to stitch together the multi-view data. Combined with the ambient temperature collected by the temperature compensation unit, a distance correction algorithm was used to compensate for the temperature effect, resulting in a complete body shape contour model of the Wulong geese.
[0031] S2. Body Feature Extraction: A threshold-based contour segmentation algorithm was used to separate the individual Wulong geese from the background region. Then, an improved Canny edge detection algorithm combined with the Hough transform algorithm was used to locate key feature points of each Wulong goose, such as the front of the head, the end of the tail, both sides of the chest, the highest point of the back, the lowest point of the abdomen, and both sides of the hip bones. The key feature points were fitted by the least squares method to calculate body feature parameters such as body length, neck length, chest width, chest depth, and hip bone width. The specific parameters are shown in Table 1 below. Table 1 Examples of physical characteristics of the Five Dragon Goose
[0032] S3, High-yield trait matching: Collect the physical characteristics and egg production records of 1000 Wulong geese with known egg production performance, label the characteristic parameters of 500 high-yield Wulong geese, and construct training and test datasets; A basic model is constructed using a fusion of CNN and LSTM networks. The CNN network employs 6 convolutional layers and 3 pooling layers, while the LSTM network uses 2 hidden layers. The model parameters are optimized using the Adam algorithm, with 100 training iterations and an initial learning rate of 0.001. An attention mechanism is introduced to strengthen the weights of pubic interventricular distance and abdominal circumference, and the dropout algorithm is used to prevent overfitting, resulting in an optimized AI recognition model for the five dragon geese features.
[0033] Furthermore, a high-yield trait feature library for Wulong geese was constructed, and the high-yield thresholds for each feature parameter were determined: body length ≥ 23.83 cm, neck length ≥ 24.01 cm, chest width ≥ 9.11 cm, hip width ≥ 7.52 cm, and chest depth ≥ 8.12 cm. The body feature parameters extracted in step S2 were input into the AI recognition model and matched with the high-yield trait feature library for similarity. When the similarity was ≥ 85%, the goose was determined to be a high-yield candidate Wulong goose.
[0034] In this embodiment, the similarity of the five-dragon goose numbered G01 is 91%, and it is determined to be a high-yield candidate five-dragon goose; the similarity of the five-dragon goose numbered G02 is 78%, and it is determined to be a non-high-yield five-dragon goose.
[0035] The screening system of this application includes a dynamic ultrasonic data acquisition module, a body feature extraction module, an AI recognition and matching module, a data storage module, and a display and interaction module. The specific configuration of each functional module is as follows: Dynamic ultrasonic data acquisition module: It adopts a multi-view sensor array composed of 9 ultrasonic sensors (model optional TDK ICU-30201), equipped with an official original factory synchronization control unit to realize synchronous acquisition of sensors; it is equipped with a temperature sensor (model: TMP112) as a temperature compensation unit to collect the ambient temperature in real time; the preprocessing unit adopts an Intel Core i7-12700K processor to run median filtering, coordinate calibration and distance correction algorithms to complete the preprocessing of body contour data.
[0036] Body posture feature extraction module: The NVIDIA Jetson AGX Xavier edge computing module is used to run a contour segmentation algorithm based on threshold segmentation and an improved Canny+Hough transform feature extraction algorithm to achieve rapid extraction of body posture feature parameters.
[0037] AI recognition and matching module: It uses NVIDIA Tesla V100 GPU as the computing core to build an AI recognition model that integrates CNN and LSTM. It is equipped with a model update unit and supports monthly iterative updates of the model based on newly collected data. The high-yield trait feature library is stored in a MySQL database and supports dynamic adjustment of parameters.
[0038] Data storage module: Uses a 1TB SSD to store 3D body shape data, feature parameters, filtering results and other data, and supports data query, export and automatic backup.
[0039] Display and interaction module: It adopts a 15.6-inch touch screen to display the body contour model, feature parameters and filtering results of the five dragon geese in real time; it supports users to set parameters such as similarity threshold (range: 80%~95%) and acquisition frame rate (10~20 fps) through touch operation, and supports querying and exporting filtering results by conditions.
[0040] In the appendix Figure 4 The system can clearly identify the specific locations of multiple feature points of the Five Dragon Goose body shape wheel, and can obtain various body shape feature parameters of the Five Dragon Goose to be screened based on the distance between different feature point locations, so as to compare and evaluate them with the corresponding thresholds.
[0041] It is worth noting that, compared with the existing technology, the innovation of this invention lies in breaking through the bottleneck of traditional static manual measurement by leveraging the different penetrating power characteristics of ultrasonic waves at different frequencies to different media, reducing the error in the thickness of the feathers on the body surface of the five dragon geese, realizing fully automatic, high-precision, non-contact direct measurement and screening, while avoiding stress reactions to the five dragon geese.
[0042] Furthermore, this invention employs an array-type ultrasonic sensor to overcome the problem of poor performance with a single probe, generating a complete and coherent three-dimensional body shape point cloud and contour model. It integrates median filtering and temperature compensation correction algorithms to offset the interference of ambient temperature on the propagation speed of ultrasonic waves in real time, while filtering out noise data from farm dust and water mist, ensuring image clarity and contour accuracy. Combined with an improved Canny edge detection and Hough transform contour refinement algorithm, it is specifically optimized for the body shape characteristics of the Wulong goose, accurately distinguishing the goose body from the farming background, avoiding blurred contours and missing features, achieving sub-pixel level accurate body shape contour imaging, perfectly adapting to the subsequent high-yield trait feature extraction requirements.
[0043] In summary, the high-yield Wulong goose screening method based on dynamic ultrasonic body posture recognition in this embodiment of the invention employs dynamic ultrasonic body posture data acquisition technology. It captures the dynamic body posture characteristics of Wulong geese in their natural activity state via a multi-view ultrasonic sensor array, generates body contour data, and completes precise preprocessing. Combined with an improved feature extraction algorithm, it can accurately extract core body posture characteristic parameters such as body length, chest width, and pubic bone distance of Wulong geese. This not only comprehensively and objectively reflects the body posture of Wulong geese, avoiding the subjectivity and stress interference of manual observation and contact measurement, but also has the advantage of resisting environmental interference such as dust, water mist, and light changes. It can adapt to complex breeding scenarios and provides stable and accurate basic data for high-yield screening. Furthermore, a high-yield trait feature library is constructed based on the breed specificity of Wulong geese, improving the targeting and accuracy of the screening.
[0044] The above specific embodiments should not be construed as limiting the scope of protection of the present invention. For those skilled in the art, any alternative improvements or modifications made to the embodiments of the present invention shall fall within the scope of protection of the present invention.
[0045] Any aspects of this invention not described in detail are well-known to those skilled in the art.
Claims
1. A method for screening high-yield Wulong geese based on dynamic ultrasonic body posture recognition, characterized in that, The screening method includes the following steps: S1, the screening system collects dynamic distance data of the five dragon geese in their natural activity state through a multi-view ultrasonic sensor array, generates body contour data of the five dragon geese, and performs noise reduction, splicing and calibration preprocessing on the body contour data. S2. Based on the body contour data after denoising, splicing and calibration preprocessing, the contour segmentation algorithm is used to separate the individual Wulong geese from the background. Then, through the feature point localization and fitting algorithm, multiple core body contour parameters of the Wulong geese are extracted. The core body contour parameters include at least body oblique length, chest width, chest depth, hip width and neck length. S3. Construct a Wulong Goose Feature AI Recognition Model. Input the extracted core body shape feature parameters into the Wulong Goose Feature AI Recognition Model and perform similarity matching with the preset Wulong Goose high-yield trait feature library. When the similarity reaches the preset threshold, it is determined to be a high-yield candidate Wulong Goose and the screening result is output.
2. The method for screening high-yield Wulong geese based on dynamic ultrasonic body posture recognition according to claim 1, characterized in that: The multi-view ultrasonic sensor array includes nine ultrasonic sensors, which are evenly distributed around and on top of the breeding area. The detection range overlap between adjacent ultrasonic sensors is no less than 25%, ensuring full coverage of the activity area of the five dragon geese and capturing dynamic data of their various natural activity postures such as walking, feeding, and standing. The ultrasonic sensor's acquisition frame rate is set to 10~20 fps to balance data acquisition integrity and efficiency; the ultrasonic sensor's detection frequency is set to 40~80 kHz to balance detection accuracy and anti-interference capability.
3. The method for screening high-yield Wulong geese based on dynamic ultrasonic body posture recognition according to claim 1, characterized in that: During preprocessing, a median filtering algorithm is used to remove noise points from the body contour data to avoid the impact of environmental interference on feature extraction; a coordinate calibration algorithm is used to stitch together multi-view data to ensure that data collected from different perspectives can be accurately aligned; a temperature compensation unit is used to collect ambient temperature, and a distance correction algorithm is used to compensate for the influence of temperature on the propagation speed of ultrasonic waves, generating a complete and clear body contour model of the five dragon geese.
4. The method for screening high-yield Wulong geese based on dynamic ultrasonic body posture recognition according to claim 1, characterized in that: The contour segmentation algorithm adopts a threshold-based segmentation method, which distinguishes the individual geese from the background area based on the gray-level difference of the distance data, ensuring the accuracy of the segmentation. The feature point localization and fitting algorithm adopts an improved Canny edge detection algorithm combined with the Hough transform algorithm. It accurately locates the key feature points of the head, neck, chest, abdomen and pubic area of the five dragon geese through contour feature matching, and then fits the key feature points through the least squares method to calculate the corresponding core body feature parameters.
5. The method for screening high-yield Wulong geese based on dynamic ultrasonic body posture recognition according to claim 1, characterized in that, The steps to construct the Five Dragon Goose Feature AI Recognition Model are as follows: S3.1 Collect dynamic 3D body shape data of Wulong geese with known egg production performance and corresponding egg production records, label the body shape characteristic parameters of high-producing Wulong geese, and construct training dataset and test dataset; S3.2, a network structure integrating CNN and LSTM is used as the basic model. The CNN network is used to extract deep features of body posture parameters, and the LSTM network is used to capture the temporal variation of dynamic body posture features. The training dataset is input into the model for training, and the model parameters are optimized by an adaptive learning rate adjustment algorithm so that the model converges to the preset accuracy. S3.3 introduces an attention mechanism to strengthen the weight of core body posture feature parameters and improve the model's ability to recognize key features; the dropout algorithm is used to prevent model overfitting and improve the model's generalization ability, resulting in an optimized AI recognition model for the five dragon geese features.
6. The method for screening high-yield Wulong geese based on dynamic ultrasonic body posture recognition according to claim 1, characterized in that, The construction method of the Wulong Goose high-yield trait feature library is as follows: collect a large number of body feature parameters of high-yield Wulong geese, use K-means clustering algorithm to analyze the parameters, determine the high-yield threshold range of each body feature parameter, form the Wulong Goose high-yield trait feature library, and provide a standard basis for similarity matching.
7. The method for screening high-yield Wulong geese based on dynamic ultrasonic body posture recognition according to claim 1, characterized in that, The screening system includes: A dynamic ultrasonic data acquisition module is used to acquire dynamic distance data of the five dragon geese in their natural activity state through a multi-view ultrasonic sensor array, generate body contour data and complete preprocessing. The body feature extraction module is used to separate the individual geese from the background using a contour segmentation algorithm based on the denoised, stitched and calibrated body contour data, and then extract multiple core body feature parameters of the geese using a feature point localization and fitting algorithm. The AI recognition and matching module is used to construct a Wulong goose feature AI recognition model, match the extracted body feature parameters with the Wulong goose high-yield trait feature library, and output the screening results. The data storage module is used to store the collected 3D body shape data, extracted feature parameters, high-yield trait feature library and screening results, and supports data query, export and backup; The display interaction module is used to display the body contour model of the five dragon geese, the extracted feature parameters and the filtering results in real time. It supports users to set parameters and query filtering results, thereby improving the ease of use of the system.
8. The method for screening high-yield Wulong geese based on dynamic ultrasonic body posture recognition according to claim 7, characterized in that: The dynamic ultrasonic data acquisition module includes a synchronization control unit and a temperature compensation unit. The synchronization control unit is used to control the synchronous acquisition of the multi-view ultrasonic sensor array to ensure the time consistency of the data acquired by each sensor and improve the accuracy of data stitching. The AI recognition and matching module includes a model update unit, which is used to periodically update the AI recognition model of Wulong geese features based on newly collected body shape data and egg production performance data, thereby improving the matching accuracy of the model.