Pig living body B-ultrasonic image carcass character determination method and system based on deep learning

Through the analysis method of live B-ultrasound imaging data of pigs based on deep learning, the YOLOv8 model is used for key point detection, and the problems of low efficiency and insufficient accuracy of B-ultrasound imaging data analysis in the existing technology are solved, and efficient and accurate pig backs are achieved Fat thickness and eye muscle depth measurements support scientific breeding decisions and reduce production costs.

CN118968556APending Publication Date: 2024-11-15HUAZHONG AGRI UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411044325.X
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-31
Publication Date
2024-11-15

AI Technical Summary

Technical Problem

In the prior art, pig production and breeding lack efficient and accurate B-ultrasound image data analysis software and corresponding database support, resulting in the measurement results relying on manual operations, low efficiency and insufficient accuracy, and cannot meet the needs of the breeding industry.

Method used

Using the data analysis method of live B-ultrasound imaging of pigs based on deep learning, key point detection is carried out through the YOLOv8 model, combined with data enhancement technology and data set construction, a configuration includes image upload, model management, image analysis and result display, data management and Abnormal processing system enables accurate measurement of backfat thickness and eye muscle depth.

Benefits of technology

It significantly improves the accuracy and efficiency of measurement, reduces manual errors, realizes automated and intelligent measurements, supports scientific breeding decisions, reduces production costs, and improves production management level.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118968556A_ABST
    Figure CN118968556A_ABST
Patent Text Reader

Abstract

The invention belongs to the technical field of computer vision and artificial intelligence, discloses a living pig B-ultrasonic image data analysis method and system based on deep learning, and aims to solve the problems of low efficiency, large error and the like in automatic measurement in the pig industry. The system utilizes a YOLOv8Pose model to carry out key point detection so as to realize accurate and rapid measurement of backfat thickness and eye muscle depth. The method comprises the specific steps that a B-ultrasonic image data set containing diversity and comprehensiveness is constructed, detection frames and key points are marked, a deep learning model is trained by applying a data enhancement strategy, and a corresponding B-ultrasonic image analysis and calculation algorithm is developed. The system integrates an image uploading module, a model management module, an image analysis module, a result display and export module, a data storage and management module and an exception handling and log recording module. The stability and reliability of the system are verified through model performance evaluation and multiple measurement jump interval analysis.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the field of computer vision and artificial intelligence technology, and in particular relates to a method and system for analyzing pig live B-ultrasound image data based on deep learning. Background Art

[0002] With the rapid development of global agriculture and animal husbandry, the pig farming industry is gradually transforming towards intensification, automation and intelligence. In this context, achieving efficient measurement and management of pig production and breeding has become a key factor in improving production efficiency and product quality. At present, there is no intelligent solution specifically for the analysis of live pig ultrasound imaging data on the market. Although traditional manual measurement methods are widely used, they are cumbersome, time-consuming and labor-intensive to operate, and the measurement results rely on the experience and skills of the operator, which is prone to errors and difficult to meet the needs of large-scale production.

[0003] Traditional evaluation methods, such as manual measurement of backfat thickness and loin depth, can provide evaluation information to a certain extent, but they are limited by low efficiency, strong subjectivity, and insufficient precision. These measurements rely on physical measurements and anatomical evaluations after slaughter, which are not only cumbersome and cruel, but also cannot be measured on live pigs. When using B-ultrasound equipment to measure backfat thickness and loin depth, the operator needs to manually select the measurement points, which is not only time-consuming, but also easily affected by subjective factors, resulting in inaccurate measurement results.

[0004] In pig production and breeding, fast and accurate measurement of backfat thickness and eye muscle depth is of great significance for integrating production data, optimizing breeding strategies and improving meat quality. Through B-ultrasound imaging, the internal structure of pigs can be observed intuitively, so as to more accurately evaluate their growth performance and meat quality. However, there is a lack of efficient and accurate B-ultrasound image analysis software and corresponding database support on the market, which limits the widespread application of B-ultrasound technology in the breeding industry.

[0005] Technical problems existing in the prior art and their analysis.

[0006] 1. Complicated operation, time-consuming and labor-intensive: The traditional manual measurement method is complicated to operate and requires manual selection of measurement points and recording of data. The process is time-consuming and labor-intensive, especially in large-scale farms, which increases labor intensity, reduces production efficiency, and cannot meet the needs of efficient production.

[0007] 2. The measurement results are dependent on the operator's experience and skills: The precision and accuracy of manual measurement methods are highly dependent on the operator's experience and skills. Differences in technical levels between different operators can lead to errors and inconsistencies in measurement results. Since the measurement results rely on human factors, it is difficult to ensure the consistency and accuracy of the measurement, which is a significant drawback for production and breeding decisions that require precise data.

[0008] 3. Strong subjectivity and insufficient precision: Traditional evaluation methods, such as manual measurement of backfat thickness and eye muscle depth, are highly subjective and easily affected by the operator's personal judgment, resulting in insufficient measurement precision. The problems of subjectivity and insufficient precision are particularly serious in pig production and breeding. Insufficient precision will affect the accuracy of production data, which in turn affects the optimization of breeding strategies and the improvement of meat quality.

[0009] 4. Lack of efficient and accurate B-ultrasound image analysis software: Currently, there is a lack of intelligent solutions specifically for live pig B-ultrasound image data analysis on the market. These software are insufficient in accuracy and efficiency and cannot meet the needs of the breeding industry. Existing B-ultrasound image analysis software not only limits the application of B-ultrasound technology in the breeding industry, but also affects the intuitive observation of the pig's internal structure and the evaluation of growth performance and meat quality. Due to the lack of efficient and accurate software support, the advantages of B-ultrasound technology cannot be fully utilized, further hindering the breeding industry's demand for accurate data and production optimization.

[0010] 5. Lack of corresponding database support: B-ultrasound image analysis requires powerful database support to store and manage large amounts of image data for data analysis and model training. However, there is currently a lack of such specialized database support on the market. Without powerful database support, B-ultrasound image data cannot be effectively managed and utilized, limiting the effectiveness of data analysis and model training, and further limiting the application of B-ultrasound technology in pig production and breeding.

[0011] There are many technical problems and defects in the application of existing technologies in pig production and breeding, mainly including cumbersome operation, time-consuming and labor-intensive, measurement results relying on the experience and skills of the operator, strong subjectivity, insufficient precision, lack of efficient and accurate B-ultrasound image analysis software, and lack of corresponding database support. These problems limit the widespread application of B-ultrasound technology in the breeding industry and cannot meet the needs of intensive, automated and intelligent transformation. To solve these problems, it is urgent to develop efficient and accurate B-ultrasound image analysis software and provide corresponding database support to improve the efficiency and quality of pig production and breeding. Summary of the invention

[0012] In response to the problems existing in the prior art, the present invention provides a method and system for analyzing pig live B-ultrasound image data based on deep learning.

[0013] The present invention is implemented as follows: a method for analyzing pig in vivo B-ultrasound image data based on deep learning includes:

[0014] S1, unique dataset construction:

[0015] 404 sets of B-ultrasound image data from different breeds of pigs were collected at the pig farm of Zhule Tianyuan Ecological Agriculture Co., Ltd. using a portable veterinary B-ultrasound imager. After data sorting, a total of 404 sets of B-ultrasound image data were obtained, each of which was equipped with the true value information of the pig's breed, weight, age, back fat thickness and eye muscle depth.

[0016] S2, detection box and key point annotation:

[0017] All images were annotated using the Labelme tool, and all detection boxes and key points were annotated by the same worker. The annotation results were saved in json format and then converted into YOLO (You Only Look Once) format. During the image annotation process, the conventional backfat thickness and eye muscle depth were annotated separately.

[0018] S3, data enhancement strategy:

[0019] The training data was expanded through data augmentation methods such as color transformation, exposure adjustment, saturation adjustment, brightness change, and simulated image noise and blur;

[0020] S4, selection and training of deep learning models:

[0021] Select the YOLOv8pose (You Only Look Once) model for training, set the appropriate training cycle and early stopping strategy, and use data enhancement methods to optimize model performance to ensure its accuracy and robustness.

[0022] S5, Develop B-ultrasound image analysis and calculation algorithms:

[0023] In terms of image processing and key point detection, a deep learning model is used to analyze B-ultrasound images. By detecting and obtaining key point coordinate information, the pixel measurements of eye muscle depth and back fat thickness are calculated and converted into actual physical dimensions.

[0024] S6, building a pig in vivo B-ultrasound image data analysis system based on deep learning:

[0025] A system including image uploading, model management, image analysis and result display, data export, data management, exception handling and log recording has been built. Users can upload image files, change models, perform key point detection and calculate backfat thickness and eye muscle depth, intuitively display and export analysis results, maintain data storage, capture and record exceptions and system logs, and ensure stable operation of the system.

[0026] S7, model performance and B-ultrasound image analysis evaluation: Based on the conventional model performance analysis such as precision, recall rate, F1 score, confusion matrix and normalized confusion matrix, the “multiple measurement jump intervals” were introduced to evaluate the stability of the results, and the matching degree between the predicted value and the true value was verified through correlation analysis and deviation analysis.

[0027] Furthermore, in S1, 100 data sets are randomly selected as test sets and do not participate in training and verification; after image data enhancement is performed on the 304 data sets other than the test set, 304 images are extracted as verification sets, and the remaining 3040 images are used as training sets.

[0028] Furthermore, the mechanism based on the YOLOv8-pose model learning in S2 needs to first mark the detection box and then mark the key points in the image;

[0029] S21, box-point relationship design: In order to determine the detection box design that optimizes the model performance, four box-point relationship experiments were designed, ranging from small detection boxes that are close to key points to large detection boxes that cover a wider image area, to evaluate the impact of different detection box configurations on key point recognition accuracy;

[0030] S22, detection box annotation: All images are annotated by the Labelme tool. All detection boxes and key points are annotated by the same measurement personnel who measure the true value to ensure uniformity and reduce subjective errors. The annotation results are saved in json format and then converted to YOLO format by the label conversion algorithm;

[0031] S23, key point marking: In B-ultrasound image analysis, the measurement definition of eye muscle depth and back fat thickness is based on the vertical distance between the third and fourth ribs of the pig; because the upper point of the eye muscle and the lower point of the back fat are the same point, three key points are marked: the lower point of the eye muscle, the upper point of the eye muscle and the upper point of the back fat, and the three should be arranged vertically in space;

[0032] S24, key point annotation correction: The accurate annotation of key points is crucial to the accuracy of measurement; therefore, a key point annotation correction process was implemented, which includes adjusting the x-coordinates of other key points based on the x-coordinate of the subocular point.

[0033] Further, S3 is to construct a pig B-ultrasound image dataset;

[0034] S31, key point image data enhancement: its core is to change the visual characteristics of the image without affecting the labels of the key points to simulate changes in the real world, such as color transformation, exposure adjustment, saturation adjustment, brightness change, and simulated image noise and blur;

[0035] S32, key point rotation data enhancement: further considered rotation data enhancement, this method not only rotates the image, but also adapts the label of the rotated image accordingly;

[0036] In S4, the selected YOLOv8pose model is used for key point detection;

[0037] S41, training parameter setting: YOLOv8 model is the core of this study. Ultralytics library is used to train YOLOv8 model. YOLOv8npose is used as the main model. The training cycle is set to 500 epochs, which is medium. As a control, a fully trained model is made, that is, 100 epochs of early stopping patience is set. In these 100 epochs, the training is terminated without any prompt.

[0038] S42, YOLOv8pose model training: The input image size is 640*640 resolution, and RTX3090 GPU is used for training. Other parameters including learning rate and batch size use the default weights of Ultralytics for YOLOv8 training. During the training process, data enhancement is also used, including hue, saturation, brightness adjustment intensity, as well as translation, zoom, and mosaic data enhancement.

[0039] Furthermore, the S5 step of analyzing the B-ultrasound image includes using the OpenCV software library to convert the image from the original BGR color space to the RGB color space required by the YOLO model; using the support provided by the Ultralytics library for YOLO detection to perform deep learning analysis on the image, thereby identifying key points in the image and obtaining the coordinate information of these points;

[0040] S51, secondary drawing and calculation: In order to ensure the accuracy of the calculation of eye muscle depth and back fat thickness, secondary drawing and calculation were implemented on the analysis results; first, the coordinates of the lower point of the eye muscle were determined as a reference, and then a vertical line was drawn upward from this point to intersect with the y-axis coordinate of the upper point of the eye muscle to determine a new upper point position of the eye muscle;

[0041] S52, pixel value conversion to real value method: Since the model outputs the pixel position of the key point, and these pixel values ​​need to be converted to the actual physical size, a conversion factor is introduced to achieve this conversion; considering the resolution of B-ultrasound images and the needs of actual measurement, the height of 540 pixels of each image corresponds to the actual length of 150 mm as a benchmark to obtain the conversion factor;

[0042]

[0043] S53, Two-point algorithm for B-ultrasound analysis: A two-point algorithm was proposed based on the improvement and innovation of the three-point algorithm.

[0044] Furthermore, the S6 constructs a pig live B-ultrasound image data analysis system based on deep learning, including the following main modules: image uploading, model management, image analysis and result display, data export, exception handling and log recording, and model performance evaluation; specifically:

[0045] S61, Image upload module: The image upload module supports users to select a single B-ultrasound image file by clicking the "Upload Image" button, or upload image files in batches by clicking the "Select Folder" button; this module is compatible with common image formats such as JPG and PNG, ensuring that users can successfully upload images in various application scenarios; after the upload is successful or failed, the system will feedback the upload status to the user through the interface prompt or message window;

[0046] S62, Model Management Module: The model management module allows users to load and replace different YOLO model files to meet different analysis needs; users can select and load new model files from the local disk through the "Change Model" button on the interface; the system supports dynamic loading technology, and the model can be replaced without restarting the software; through this module, users can choose the most suitable model according to the specific analysis goals;

[0047] S63, Image Analysis Module: The image analysis module is the core of the system, responsible for processing images uploaded by users and performing deep learning model analysis; this module uses the YOLOv8Pose model to detect key points, automatically identify and calculate important biometric indicators such as backfat thickness and eye muscle depth; during the analysis process, the system will pre-process the image, use the deep learning model to detect key points and output coordinate information; the analysis results will be displayed in real time in the specified area of ​​the user interface, including key points and measurement lines marked on the image;

[0048] S64, result display and export module: The result display and export module is responsible for intuitively displaying the analysis results in the user interface in the form of charts or texts; users can clearly see the specific values ​​of backfat thickness and eye muscle depth, and mark the measurement lines and key points on the processed images; this module also provides data export function, and users can choose to export the analysis results into TXT format files to facilitate subsequent records, report writing and further data analysis;

[0049] S65, Data Management Module: The data management module processes and maintains the storage and management of image files and analysis results to ensure the integrity and reliability of the data; the system automatically stores the analyzed key indicators and processed image files in the local file system, following certain naming conventions to prevent file conflicts;

[0050] S66, Exception handling and logging module: The exception handling and logging module uses a global exception capture mechanism to timely capture and record runtime errors, ensuring that the system can provide timely feedback to users when encountering problems; for foreseeable exceptions, such as file read and write permission restrictions or unsupported image formats, the system will provide friendly error prompts to guide users on how to solve the problem; the logging function includes operation logs and system logs, which record the user's operation history and detailed information on system operation;

[0051] S7 introduces the innovative model evaluation index of multiple measurement jump intervals to evaluate the stability and repeatability of the results based on conventional model performance analysis such as precision, recall rate, F1 score, confusion matrix and normalized confusion matrix. In addition, the correlation analysis and deviation analysis of B-ultrasound analysis results and true values ​​are also included in the evaluation system, specifically:

[0052] S71, Multiple Measurement Jump Interval: To evaluate the consistency and repeatability of measurements, an innovative metric called “Multiple Measurement Jump Interval” is introduced. This metric measures the maximum difference between the results when multiple measurements are made on the same image. Ideally, this difference should be zero, indicating that the model gives completely consistent results across multiple measurements.

[0053] S72, B-ultrasound image analysis and evaluation: To further verify the effectiveness of the model, correlation analysis and deviation analysis were performed between the predicted values ​​and the true values; the correlation coefficient between the predicted values ​​and the true values ​​was calculated.

[0054] Another object of the present invention is to provide a pig in vivo B-ultrasound image data analysis system based on deep learning, comprising:

[0055] Image upload module: The image upload module supports users to select a single B-ultrasound image file by clicking the "Upload Image" button, or to upload image files in batches by clicking the "Select Folder" button; this module is compatible with common image formats such as JPG and PNG, ensuring that users can successfully upload images in various application scenarios; after the upload succeeds or fails, the system will feedback the upload status to the user through the interface prompt or message window;

[0056] Model management module: The model management module allows users to load and replace different YOLO model files to meet different analysis needs; through the "Change Model" button on the interface, select and load a new model file from the local disk; the system supports dynamic loading technology, and the model can be replaced without restarting the software, thereby improving the flexibility and convenience of operation;

[0057] Image analysis module: The image analysis module is the core of the system, responsible for processing images uploaded by users and performing deep learning model analysis;

[0058] Result display and export module: The result display and export module is responsible for intuitively displaying the analysis results in the user interface in the form of charts or text;

[0059] Data management module: The data management module processes and maintains the storage and management of image files and analysis results;

[0060] Exception handling and logging module: The exception handling and logging module uses a global exception capture mechanism to promptly capture and record runtime errors, ensuring that the system can provide timely feedback to users when encountering problems; for foreseeable exceptions.

[0061] Another object of the present invention is to provide a computer device, comprising a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the method for analyzing pig live B-ultrasound image data based on deep learning.

[0062] Another object of the present invention is to provide a computer-readable storage medium storing a computer program, which, when executed by a processor, enables the processor to perform the steps of the method for analyzing pig live B-ultrasound image data based on deep learning.

[0063] Another object of the present invention is to provide an information data processing terminal, which is used to implement the pig living B-ultrasound image data analysis system based on deep learning.

[0064] In combination with the above technical solutions and the technical problems solved, the advantages and positive effects of the technical solutions to be protected by the present invention are as follows:

[0065] First, the present invention provides a method and system for analyzing pig live B-ultrasound image data based on deep learning, which adopts the YOLOv8 model for key point detection to achieve accurate measurement of back fat thickness and eye muscle depth, overcomes many difficulties of the existing technology in automated measurement, provides efficient and accurate measurement means, comprehensively improves measurement efficiency and supports scientific breeding.

[0066] (1) Improve recognition accuracy:

[0067] By using the YOLOv8Pose model and high-quality data sets, the present invention significantly improves the accuracy of backfat thickness and eye muscle depth measurement. The model can automatically identify and remove interference factors during key point detection to ensure the reliability of each measurement result.

[0068] (2) Enhance model generalization ability:

[0069] By adopting various data enhancement techniques, such as color transformation, exposure adjustment, saturation adjustment, brightness change and simulated image noise, the model can accurately measure under different lighting, angles and backgrounds, adapt to various environmental conditions, and enhance the generalization ability of the model.

[0070] (3) Adaptability and flexibility:

[0071] The YOLOv8Pose model not only performs well in key point detection, but also has high adaptability. It can be optimized according to different computing capabilities and application requirements to meet the needs of diverse operating environments.

[0072] (4) Real-time and high efficiency:

[0073] The system meets the requirements for real-time measurement of backfat thickness and eye muscle depth. Deep learning technology and efficient system integration ensure stable and efficient accurate measurement even in complex environments, significantly improving measurement efficiency and reducing labor costs.

[0074] (5) Automation and intelligence:

[0075] The system is fully automated and intelligent, requiring no human intervention. It can automatically analyze and store results and data, provide a scientific basis for breeding, reduce the labor intensity of operators, and improve production management.

[0076] The present invention uses advanced deep learning technology to achieve intelligent real-time measurement of backfat thickness and eye muscle depth of live pigs, significantly improving the accuracy and efficiency of measurement. The system does not require expensive mechanical equipment and a large amount of manual operation, which can significantly reduce production costs and improve economic benefits. The high-frequency real-time measurement capability ensures the efficient operation of the production line and avoids production stagnation caused by measurement delays. By analyzing the measurement data, a scientific basis is provided for breeding, breeding strategies are optimized, and breeding efficiency and effects are improved.

[0077] The system records the data of each measurement, helping enterprises meet strict food safety and quality standards and improve market competitiveness. This technology not only improves production efficiency, but also ensures the transparency and traceability of the breeding process, which is of great significance to the development of modern pig farming. Through precise data analysis and prediction, the system can optimize production processes and management decisions, improve the overall production management level, promote the automation and intelligentization of the pig farming industry, and bring significant economic benefits and technological progress.

[0078] Second, as auxiliary evidence of the inventiveness of the claims of the present invention, it is also reflected in the following important aspects:

[0079] (1) The expected benefits and commercial value of the technical solution of the present invention after transformation are:

[0080] The pig live B-ultrasound imaging data analysis system of the present invention has significant commercial value and expected benefits, especially in improving the production efficiency and management level of pig farms. Through the accurate back fat thickness and eye muscle depth measurement technology, the pig management efficiency of the farm is greatly improved, allowing operators to perform more accurate production monitoring, pig breeding and breeding pig determination, thereby improving overall production efficiency and breeding quality. In addition, this automated measurement method reduces dependence on manual inspection, reduces labor costs, and significantly reduces problems caused by human errors, such as subjective errors of measurement personnel and inconsistent measurement standards.

[0081] The system's precise measurement capabilities can optimize farm management processes, and provide a scientific basis for pig assessment through accurate thickness and depth measurements combined with other data. The application of the system can not only optimize supply chain management, but also improve the transparency and traceability of the entire chain, and enhance consumer trust in the product.

[0082] The system can record and analyze the data of each measurement, provide detailed feedback information for breeders, help optimize breeding strategies, improve breeding efficiency and results, and promote the development of scientific breeding. Through automated and intelligent management methods, the present invention promotes the technical upgrading and management optimization of the pig farming industry. The system can continuously monitor and record the measured data, provide data support for enterprises, optimize breeding processes and management decisions, and improve the overall breeding management level.

[0083] This technology not only improves production efficiency, but also ensures transparency and traceability of the breeding process, helping to meet increasingly stringent food safety and quality standards. The system provides a scientific basis for meat quality assessment. Through accurate backfat thickness and eye muscle depth measurement, combined with other indicators, meat quality assessment can be performed more accurately, thereby increasing the added value and market recognition of products. The application of the system enables breeding companies to better respond to changes in market demand and improve product consistency and market competitiveness through precise data analysis and prediction.

[0084] Finally, the large amount of data collected by the system can be used for in-depth analysis to provide data-driven decision support, such as optimizing breeding strategies and improving breeding efficiency and effectiveness. Through data analysis, breeders can better understand the key factors in the production process, make more scientific and effective breeding decisions, further improve the quality and effectiveness of breeding work, promote the development of scientific breeding, and enhance the competitiveness and market performance of enterprises.

[0085] In summary, the present invention is not only technically innovative and practical, but also shows great potential and value in commercial applications. Through the application of the present invention, enterprises can achieve more efficient and accurate production management, reduce costs, improve product quality and market competitiveness, while ensuring the hygiene and safety of products, and open up new markets and business opportunities.

[0086] (2) The technical solution of the present invention fills the technical gap in the industry at home and abroad:

[0087] The deep learning-based pig live B-ultrasound image data analysis system proposed in this invention represents an important breakthrough in the field of pig farming technology at home and abroad, filling the technical gap in the measurement of live pig back fat thickness and eye muscle depth. This system uses advanced deep learning technology, combined with a unique key point detection method, to provide an innovative solution for accurate measurement and management. Compared with traditional manual measurement or other physical measurement methods, this invention shows obvious advantages and innovations in several key aspects:

[0088] High precision and reliability:

[0089] Using the deep learning model YOLOv8Pose and farm-specific image datasets, this system can achieve high measurement accuracy. Through subsequent continuous learning and optimization, the system can adapt to various breeding environments and different breeds of pigs, ensuring accurate measurement of backfat thickness and eye muscle depth under various conditions.

[0090] Powerful data processing capabilities:

[0091] The present invention can not only process complex B-ultrasound image data, but also extract key features from it for precise measurement. At the same time, the data provided by the system supports decision-making and helps optimize production processes and management strategies. The system integrates key point detection and data processing in deep learning, realizes synchronous processing of image data, and ensures the accuracy and real-time nature of the data.

[0092] Feedback Breeding Support:

[0093] The system can record and analyze the data of each measurement, provide detailed feedback information to breeders, help optimize breeding strategies, and improve breeding efficiency and effectiveness. Through data-driven decision support, companies can better understand the key factors in the breeding process and promote the development of scientific breeding.

[0094] Wide range of application potential:

[0095] Although this system is mainly used for measuring backfat thickness and eye muscle depth of live pigs, its technical framework and methods have broad application prospects. The modular design and flexibility of the system enable it to adapt to different application scenarios and needs, providing versatility for automated management of animal husbandry.

[0096] Through the above advantages and innovations, the present invention fills the gap in the industry at home and abroad in terms of technology, provides an efficient, accurate and reliable method for measuring backfat thickness and eye muscle depth, and promotes the technological progress and management optimization of the pig farming industry. The wide application of the system can not only improve production efficiency and product quality, but also bring significant economic benefits and market competitiveness to breeding enterprises, and at the same time provide valuable technical support and inspiration for research and application in related fields.

[0097] (3) Whether the technical solution of the present invention solves the technical problems that people have been eager to solve but have not been able to solve successfully:

[0098] The pig live B-ultrasound image data analysis system of the present invention solves many long-standing technical problems in the automated measurement of the pig farming industry and meets the urgent demand for efficient, accurate and non-invasive measurement technology in the industry. Specifically, the present invention overcomes the following major problems:

[0099] Accuracy of automated measurements:

[0100] Traditional manual measurement methods rely on the operator's experience and skills, which is not only inefficient but also prone to errors and inconsistencies. This paper adopts the deep learning model YOLOv8Pose, through high-precision image recognition and key point detection technology, to achieve accurate measurement of backfat thickness and eye muscle depth, completely solving the problems of errors and inefficiency in manual measurement.

[0101] Measurement stability in complex environments:

[0102] The farm environment is complex, and factors such as lighting conditions and operating speed will affect the measurement results. Traditional methods are difficult to maintain stability in various complex environments. This invention improves the generalization ability of the model in different environments and conditions through the application of data enhancement and deep learning technology, ensuring stable and accurate measurement in various complex environments.

[0103] Real-time and high efficiency:

[0104] Traditional measurement methods require manual operation, which makes it difficult to achieve real-time measurement and data processing. This invention uses high-speed cameras and deep learning models, combined with efficient image processing algorithms, to achieve real-time data acquisition, processing and display. The measurement results can be displayed in real time, and automatically stored and analyzed, greatly improving the measurement efficiency and the timeliness of data processing. The high efficiency of the system enables the production line to run seamlessly, avoiding production stagnation caused by measurement delays.

[0105] Data-driven decision support:

[0106] Traditional measurement methods are difficult to provide sufficient data to support decision-making. The present invention provides rich decision-making support information by automatically collecting and analyzing a large amount of measurement data. Through data analysis, breeding companies can optimize production processes, improve quality control, predict market demand, and achieve scientific and data-based management.

[0107] Feedback Breeding Support:

[0108] The system can record and analyze the data of each measurement, provide detailed feedback information to breeders, help optimize breeding strategies, and improve breeding efficiency and effectiveness. Through data-driven decision support, companies can better understand the key factors in the breeding process and promote the development of scientific breeding. The system not only helps breeders select high-quality breeding individuals, but also allows breeders to improve and perfect breeding plans through historical data analysis and improve the overall breeding level.

[0109] In summary, the present invention solves many long-standing technical problems in automated measurement in the pig farming industry, and meets the industry's urgent need for efficient, accurate, and non-invasive measurement technology. Through the application of the present invention, breeding companies can not only achieve efficient and accurate backfat thickness and eye muscle depth measurement, improve production efficiency and product quality, but also improve management level and market competitiveness through data-driven decision support. The breakthrough of this technology provides strong technical support for the modern management of the pig farming industry and even the entire animal husbandry industry, and has broad application prospects and significant economic benefits.

[0110] (4) Whether the technical solution of the present invention overcomes technical prejudice:

[0111] The pig live B-ultrasound image data analysis system of the present invention overcomes various technical biases in traditional pig farming measurement methods, breaks through the long-standing mindset and technical limitations in the industry, and is specifically embodied in the following aspects:

[0112] The transition from mechanical measurement to intelligent measurement:

[0113] Traditional pig farming measurement methods rely on mechanical equipment and manual operation, which have problems such as low efficiency, poor accuracy, and easy errors. Mechanical measurement methods are considered to be the only reliable solution, thus forming a technical bias. This invention successfully realizes intelligent and non-invasive backfat thickness and eye muscle depth measurement by introducing deep learning models and image processing technology, greatly improving measurement accuracy and efficiency, and changing the industry's inherent dependence on mechanical measurement methods.

[0114] Break through the reliance on manual measurements:

[0115] Traditional pig farming measurement methods are highly dependent on manual operation. Although manual measurement has errors and low efficiency, it has always been considered indispensable due to the limitations of technical means. This invention successfully realizes full-process automated measurement through automated image recognition and data processing technology, reduces the impact of human factors, significantly improves the accuracy and efficiency of measurement, and breaks through the technical prejudice against manual measurement.

[0116] Change the question of adaptability to complex environments:

[0117] Traditional mechanical and manual measurement methods are difficult to maintain stability and efficiency in complex farming environments. The industry generally believes that automated measurement technology is difficult to adapt to complex and changeable actual operating environments. This invention successfully achieves stable and accurate measurement in various complex environments through data enhancement technology and the high generalization ability of deep learning models, changing the industry's doubts and prejudices about the adaptability of automated measurement technology.

[0118] Innovation in data-driven decision making:

[0119] Traditional measurement methods often lack systematic data recording and analysis, and cannot provide effective support for management decisions. Data-driven decision-making has not yet been fully valued and applied in the pig farming industry. The present invention provides rich data support for management decisions through automated data collection and analysis, promotes the popularization and application of data-driven decision-making, and breaks the limitations of traditional management models.

[0120] Feedback Breeding Support:

[0121] The system can record and analyze the data of each measurement, provide detailed feedback information to breeders, help optimize breeding strategies, and improve breeding efficiency and effectiveness. Through data-driven decision support, companies can better understand the key factors in the breeding process and promote the development of scientific breeding. This data analysis capability has changed the reliance on experience and intuition in the traditional breeding process, and significantly improved the scientificity and efficiency of breeding work.

[0122] In summary, the present invention overcomes a variety of technical prejudices and breaks through the long-standing technical limitations in the industry through an intelligent, non-invasive, efficient and accurate backfat thickness and eye muscle depth measurement method. Through the application of the present invention, breeding companies can not only achieve more efficient and accurate production management, improve product quality and market competitiveness, but also achieve scientific and modern management through data-driven decision support, providing strong technical support and innovation power for the development of the pig farming industry and even the entire animal husbandry industry. BRIEF DESCRIPTION OF THE DRAWINGS

[0123] Figure 1 This is a flow chart of a method for analyzing pig in vivo B-ultrasound image data based on deep learning provided in an embodiment of the present invention.

[0124] Figure 2 This is a structural block diagram of a pig live B-ultrasound image data analysis system based on deep learning provided in an embodiment of the present invention.

[0125] Figure 3 It is the construction of a B-ultrasound image analysis dataset, in which 304 pig data were enhanced to form 3040 training sets and 304 validation sets, and 100 pigs were used as a test set.

[0126] Figure 4 are the annotations of the detection boxes, (A) is the interspersed detection box, (B) is the small detection box, (C) is the medium detection box, and (D) is the large detection box.

[0127] Figure 5 It is the marking and correction of key points. (A) is three-point marking, (B) is two-point marking, (C) is the initial marking map, and (D) is the corrected marking map, all on a straight line.

[0128] Figure 6 It is a data enhancement method, (A) is image data enhancement, and (B) is rotation data enhancement.

[0129] Figure 7 It is the flow chart of the model, which inputs the image into the deep neural network, obtains the structure of key point prediction, and finally draws it on the original image.

[0130] Figure 8 It is a comparison of the effects of two methods. (A) is a comparison of the effects of the three-point method, and (B) is a comparison of the effects of the two-point method.

[0131] Fig. 9 It is a method to convert pixel values ​​into real values, showing the process of converting pixel values ​​(540px) into actual length (150mm) through conversion factors.

[0132] Fig.10It is a flow chart of the pig live B-ultrasound image analysis software, showing the complete process from image uploading, model reasoning, result display to log recording and exporting.

[0133] Fig.11 This is a real screen shot of the pig live B-ultrasound image analysis software, showing the actual operation screen of the software interface, image uploading and analysis, key point measurement and result display.

[0134] Fig.12 is a visualization effect diagram of the feature map, (A) is a visualization effect diagram of the interspersed detection box, (B) is a visualization effect diagram of the small detection box, (C) is a visualization effect diagram of the medium detection box, and (D) is a visualization effect diagram of the large detection box.

[0135] Fig.13 is the evaluation of the model, (A) is the F1-confidence curve, (B) is the precision-recall curve, (C) is the normalized confusion matrix, (D) is the precision-confidence curve, (E) is the recall-confidence curve, and (F) is the normalized confusion matrix.

[0136] Fig.14 It is a multiple measurement jump interval diagram, showing the changes in eye muscle depth and back fat thickness in multiple measurements.

[0137] Fig.15 : are correlation analysis diagrams, (A) is the correlation analysis of backfat thickness prediction, and (B) is the correlation analysis of eye muscle depth prediction.

[0138] Fig.16 These are deviation analysis diagrams. (A) is the back fat thickness deviation analysis, and (B) is the eye muscle depth deviation analysis. DETAILED DESCRIPTION

[0139] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0140] The implementation scheme of the embodiment of the present invention is as follows:

[0141] First, the user needs to prepare the ultrasound image data for analysis. The user can collect ultrasound images of pigs using the ultrasound imager. Then, upload the image data to the analysis system. The user can select a single image file through the "Upload Image" button, or upload image files in batches through the "Select Folder" button. The system will automatically identify and load all image files in the folder in supported formats.

[0142] Once the image file is uploaded successfully, the user can select or change the deep learning model for analysis. By clicking the "Change Model" button, the user can select and load different YOLO model files from the local disk. This allows the user to select the most suitable model to optimize the analysis effect according to the specific analysis needs or goals.

[0143] Next, the user clicks the "Start Analysis" button and the system automatically processes the uploaded image. The system uses the pre-trained YOLOv8Pose model to analyze the image, automatically detect key points and calculate biometric indicators such as backfat thickness and eye muscle depth. The analysis process includes image preprocessing, key point detection, and result calculation. After the analysis is completed, the system will display the analysis results in real time in the designated area of ​​the user interface. The user can clearly see the key points and measurement lines marked on the image, and view the specific values ​​of backfat thickness and eye muscle depth.

[0144] In order to record and export the analysis results, the system provides a data export function. Users can choose to export the analysis results as a TXT file to facilitate further data processing and report writing. The export content includes key indicators such as image file name, backfat thickness, eye muscle depth, etc. In addition, the system automatically records the user's operation history and analysis log. Users can save the log information as a file through the "Export Log" function to facilitate problem diagnosis and user support.

[0145] During use, the system also has exception handling and logging functions. Through the global exception capture mechanism, the system can capture and record runtime errors in a timely manner, and provide friendly error prompts to guide users to solve problems. The logging function includes operation logs and system logs, which record the user's operation history and system operation information in detail to ensure the stability and traceability of the system.

[0146] Through the above steps, users can efficiently utilize the deep learning-based pig live B-ultrasound image data analysis system to achieve fast and accurate measurement of pig back fat thickness and eye muscle depth, thereby supporting pig health assessment and breeding decisions.

[0147] like Figure 1 As shown, an embodiment of the present invention provides a method and system for analyzing pig live B-ultrasound image data based on deep learning, including:

[0148] In view of the problems existing in the prior art, the present invention provides a method and system for analyzing pig live B-ultrasound image data based on deep learning, which uses the YOLOv8 model to detect key points and achieve accurate measurement of backfat thickness and eye muscle depth. This method aims to overcome the many difficulties of the prior art in automated measurement, provide an efficient and accurate measurement method, comprehensively improve measurement efficiency and support scientific breeding.

[0149] The present invention is implemented as follows: a method and system for analyzing pig live B-ultrasound image data based on deep learning, comprising the following steps:

[0150] S1, Construction of a unique data set: This study collected 404 ultrasound image data from different breeds of pigs (Duroc, Yorkshire, Landrace and Baksha) at the pig farm of Zhule Tianyuan Ecological Agriculture Co., Ltd., using a portable veterinary ultrasound imager. After data sorting, a total of 404 ultrasound image data were obtained, each of which was equipped with real value information of the pig's breed, weight, age, back fat thickness and eye muscle depth. The diversity and comprehensiveness of these data ensure the stability and reliability of the model in various complex environments;

[0151] S2, detection box and key point annotation: All images are annotated by the Labelme tool, and all detection boxes and key point annotations are annotated by the same worker to ensure uniformity and reduce subjective errors. The annotation results are saved in json format and then converted into YOLO format. During the image annotation process, the conventional backfat thickness and eye muscle depth are annotated separately to ensure that the model can accurately identify and locate key points and improve the accuracy of the prediction.

[0152] S3, Data enhancement strategy: In order to construct a B-ultrasound image dataset that has both practical application value and can accurately reflect the real situation, this study expanded the training data through data enhancement methods such as color transformation, exposure adjustment, saturation adjustment, brightness change, and simulated image noise and blur. These enhancement methods effectively improve the generalization ability of the model, enabling it to adapt to a variety of different image quality and environmental conditions;

[0153] S4, selection and training of deep learning models: select and apply the YOLOv8Pose model, which has efficient and fast reasoning capabilities and is suitable for key point detection and pose estimation. The wide range of applications of YOLOv8 makes it not limited to target detection. Through the key point detection function, YOLOv8Pose can play an important role in the analysis of live pig B-ultrasound images. During the training process, determine the appropriate hyperparameters, such as epochs, batch size, etc., to ensure the efficiency and accuracy of the model;

[0154] S5, Development of B-ultrasound image analysis and calculation algorithms: In terms of image processing and key point detection, a deep learning model is used to analyze B-ultrasound images. By detecting and obtaining key point coordinate information, the pixel measurements of eye muscle depth and back fat thickness are calculated and converted into actual physical dimensions. This process uses efficient image processing technology and deep learning algorithms to ensure the accuracy and consistency of the measurement results;

[0155] S6. The present invention provides a pig live B-ultrasound image data analysis system based on deep learning, which aims to overcome the many difficulties of the existing technology in automated measurement, provide an efficient and accurate measurement method, comprehensively improve measurement efficiency and support scientific breeding.

[0156] S7, Model performance and B-ultrasound image analysis evaluation

[0157] In order to comprehensively evaluate the model performance and the accuracy of B-ultrasound image analysis, the present invention adopts a series of scientific evaluation indicators and methods. These evaluation methods not only cover traditional indicators such as precision, recall rate, F1 score, etc., but also introduce innovative multiple measurement jump interval evaluation methods to ensure the stability and reliability of the model in various environments.

[0158] The present invention achieves accurate and rapid measurement of backfat thickness and eye muscle depth of live pigs by constructing a diverse and comprehensive B-ultrasound image data set and using advanced deep learning technology. The system integrates modules such as image uploading, model management, image analysis, result display and export, data storage and management, and visualization to form a complete automated B-ultrasound image analysis platform. Through efficient image processing technology and deep learning algorithms, the accuracy and consistency of the measurement results are ensured, providing reliable data support for scientific breeding. The system not only improves measurement efficiency and reduces manual errors, but also has broad application prospects and significant economic benefits, and promotes the automation and intelligentization of animal husbandry.

[0159] In S1 provided in the embodiment of the present invention, 100 data are randomly selected as the test set and do not participate in training and verification. After image data enhancement is performed on the 304 data other than the test set, 304 images are extracted as the verification set, and the remaining 3040 images are used as the training set to improve the diversity of the data and the generalization ability of the model.

[0160] The mechanism based on YOLOv8-pose model learning in S2 provided in the embodiment of the present invention requires first marking the detection box and then marking the key points in the image.

[0161] S21, frame point relationship design: In order to determine the detection frame design that optimizes the model performance, the present invention designs four frame point relationship experiments, ranging from small detection frames that are close to key points to large detection frames that cover a wider image area, to evaluate the impact of different detection frame configurations on key point recognition accuracy. The purpose is to find the detection frame configuration that is most conducive to accurate key point recognition, so as to improve the accuracy of key point detection.

[0162] S22, detection box annotation: All images are annotated by the Labelme tool. All detection boxes and key points are annotated by the same measurement personnel who measure the true value to ensure uniformity and reduce subjective errors. The annotation results are saved in json format and then converted to YOLO format by the label conversion algorithm.

[0163] S23, key point marking: In B-ultrasound image analysis, the measurement definition of eye muscle depth and back fat thickness is based on the vertical distance between the third and fourth ribs of the pig. Since the upper point of the eye muscle and the lower point of the back fat are the same point, the present invention marks three key points: the lower point of the eye muscle (the most depressed point), the upper point of the eye muscle and the upper point of the back fat, and the three should be arranged vertically in space.

[0164] S24, key point labeling correction: accurate labeling of key points is crucial to the accuracy of measurement. Therefore, the present invention implements a key point labeling correction process, which includes adjusting the x coordinates of other key points based on the x coordinate of the eye muscle subpoint to ensure that all key points are on the same vertical line to reduce human labeling errors.

[0165] Through this series of carefully designed annotations and experimental steps, we aim to optimize the application effect of the Yolov8 model in B-ultrasound image analysis and improve the accuracy and efficiency of the model in measuring key traits in complex imaging data.

[0166] The embodiment of the present invention provides a pig B-ultrasound image dataset in S3 that has both practical application value and can accurately reflect the actual situation. The present invention adopts two main data enhancement methods to achieve this goal.

[0167] S31, key point image data augmentation: Its core is to change the visual characteristics of the image without affecting the labels of the key points to simulate changes in the real world, such as color transformation, exposure adjustment, saturation adjustment, brightness change, and simulated image noise and blur. This method aims to train a robust model that can adapt to different image quality conditions and help the model learn to maintain the stability of key point detection under visual changes.

[0168] S32, key point rotation data enhancement: further considers rotation data enhancement, which not only rotates the image, but also adapts the label of the rotated image accordingly. Usually this comprehensive data enhancement strategy will enable the model to better adapt to the image changes caused by the change of object posture in the real world, thereby improving the accuracy of key point detection and the application flexibility of the model.

[0169] In S4 provided by the embodiment of the present invention, the selected YOLOv8pose model can not only be used for key point detection, but also has the characteristics of high speed and high precision, so that it has wide application potential in real-time application scenarios.

[0170] S41, training parameter setting: YOLOv8 model is the core of this study, and the Ultralytics library is used to train the YOLOv8 model. YOLOv8npose is used as the main model, and the training cycle is set to 500 epochs, which is medium. As a control, the present invention makes a fully trained model, that is, the early stopping patience is set to 100 epochs. In these 100 epochs, the training is terminated without any prompt.

[0171] S42, YOLOv8pose model training: The input image size is 640*640 resolution, and the RTX3090 GPU is used for training. Other parameters including learning rate and batch size use the default weights of Ultralytics for YOLOv8 training. During the training process, data enhancement is also used, including hue, saturation, brightness adjustment intensity, as well as translation, zoom, mosaic data enhancement, etc.

[0172] The S5 analysis step of the B-ultrasound image provided by the embodiment of the present invention includes using the OpenCV software library to convert the image from the original BGR color space to the RGB color space required by the YOLO model. The support provided by the Ultralytics library for YOLO detection is used to perform deep learning analysis on the image, thereby identifying the key points in the image and obtaining the coordinate information of these points.

[0173] S51, secondary drawing and calculation: In order to ensure the accuracy of the calculation of eye muscle depth and back fat thickness, the present invention implements secondary drawing and calculation on the analysis results. The present invention first determines the coordinates of the lower point of the eye muscle as a reference, and then draws a vertical line upward along this point, intersecting with the y-axis coordinates of the upper point of the eye muscle, and determines a new upper point position of the eye muscle. This correction scheme effectively prevents the situation in which the upper point of the eye muscle and the lower point of the eye muscle are not on the same vertical line in the prediction result, thereby improving the accuracy of the model prediction. By calculating the Euclidean distance between the newly defined upper and lower points of the eye muscle, the present invention obtains the pixel measurement value of the eye muscle depth. Similarly, the pixel measurement value of the back fat thickness can be obtained.

[0174] S52, pixel value conversion method: Since the model outputs the pixel position of the key point, and the present invention needs to convert these pixel values ​​into actual physical sizes, a conversion factor is introduced to achieve this conversion. Considering the resolution of B-ultrasound images and the needs of actual measurement, the present invention uses the height of 540 pixels per image corresponding to the actual length of 150 mm as a benchmark, thereby obtaining an accurate conversion factor.

[0175]

[0176] At the same time, in order to enhance the clarity of visualization and the comprehensiveness of information, the present invention adds small horizontal lines (end caps) at both ends of the lines between key points to clearly indicate the start and end of the measurement interval, and draws text representing the actual physical length (in millimeters) and the corresponding pixel value (in px) next to these measurement lines, so that the entire analysis result is both intuitive and easy to understand.

[0177] S53, B-ultrasound analysis two-point algorithm: The present invention improves and innovates on the three-point algorithm and proposes a two-point algorithm. It aims to improve the accuracy of backfat thickness by fixing the upper backfat point and reduce the problem of fluctuation in the positioning of the upper backfat point in the prediction. This innovation utilizes a key position feature of the upper backfat point in the B-ultrasound image: the upper backfat point is always located at the top of the image area.

[0178] The main advantage of the two-point algorithm is that it uses the prior knowledge about the position of the backfat points, making the prediction process more accurate with the inherent position characteristics of the backfat points. Compared with the three-point algorithm, this method simplifies the analysis process by reducing the number of points considered, and can achieve more accurate positioning due to the specific position attributes of the backfat points.

[0179] The S6 provided in the embodiment of the present invention builds a pig live B-ultrasound image data analysis system based on deep learning, including the following main modules: image upload, model management, image analysis and result display, data export, exception handling and log recording, and model performance evaluation. The design of each module is centered on improving the automation, intelligence and accuracy of the system to ensure that the system can operate stably in a complex production environment. Specifically:

[0180] S61, Image Upload Module: The image upload module supports users to select a single B-ultrasound image file by clicking the "Upload Image" button, or to upload image files in batches by clicking the "Select Folder" button. This module is compatible with common image formats such as JPG and PNG, ensuring that users can successfully upload images in various application scenarios. After the upload succeeds or fails, the system will feedback the upload status to the user through the interface prompt or message window to improve the user experience.

[0181] S62, Model Management Module: The model management module allows users to load and replace different YOLO model files to meet different analysis needs. Users can select and load new model files from the local disk through the "Change Model" button on the interface. The system supports dynamic loading technology, and the model can be replaced without restarting the software, thereby improving the flexibility and convenience of operation. Through this module, users can select the most suitable model according to specific analysis goals and improve analysis results.

[0182] S63, Image Analysis Module: The Image Analysis Module is the core of the system, responsible for processing images uploaded by users and performing deep learning model analysis. This module uses the YOLOv8Pose model for key point detection, automatically identifying and calculating important biometric indicators such as backfat thickness and eye muscle depth. During the analysis process, the system pre-processes the image, uses the deep learning model to detect key points and outputs coordinate information. The analysis results will be displayed in real time in the designated area of ​​the user interface, including key points and measurement lines marked on the image, to ensure the accuracy and visualization of the results.

[0183] S64, Result Display and Export Module: The Result Display and Export Module is responsible for displaying the analysis results in the form of charts or text in the user interface. Users can clearly see the specific values ​​of backfat thickness and eye muscle depth, and mark the measurement lines and key points on the processed images. This module also provides data export function, and users can choose to export the analysis results as TXT format files for subsequent recording, report writing and further data analysis.

[0184] S65, Data Management Module: The data management module processes and maintains the storage and management of image files and analysis results to ensure data integrity and reliability. The system automatically stores the analyzed key indicators and processed image files in the local file system, following certain naming conventions to prevent file conflicts. The module also supports batch processing, which can automatically filter non-image format files to ensure the smooth progress of the entire analysis process. In addition, users can save the analysis results as TXT files through the export function to facilitate further data processing and analysis.

[0185] S66, Exception handling and logging module: The exception handling and logging module uses a global exception capture mechanism to capture and record runtime errors in a timely manner, ensuring that the system can provide timely feedback to users when encountering problems. For foreseeable exceptions, such as file read and write permission restrictions or unsupported image formats, the system will provide friendly error prompts to guide users on how to solve the problem. The logging function includes operation logs and system logs, which record the user's operation history and detailed information on system operation to facilitate problem diagnosis and user support. Users can use the log export function to save operation logs and system logs as TXT files to facilitate problem tracking and technical support from the support team.

[0186] The S7 provided in the embodiment of the present invention introduces an innovative model evaluation index of multiple measurement jump intervals to evaluate the stability and repeatability of the results on the basis of conventional model performance analysis such as precision, recall rate, F1 score, confusion matrix and normalized confusion matrix. In addition, the correlation analysis and deviation analysis of B-ultrasound analysis results and true values ​​are also included in the evaluation system, specifically:

[0187] S71, Multiple Measurement Jump Interval: In order to evaluate the consistency and repeatability of the measurements, the present invention introduces the innovative indicator of "Multiple Measurement Jump Interval". This indicator measures the maximum difference between the results when multiple measurements are made on the same image. Ideally, this difference should be zero, indicating that the model gives completely consistent results in multiple measurements. The specific value of the jump interval reflects the model's sensitivity to subtle changes in the image, as well as its ability to handle random errors and systematic errors.

[0188] S72, B-ultrasound image analysis and evaluation: To further verify the effect of the model, the present invention performs correlation analysis and deviation analysis on the predicted value and the true value. The accuracy and consistency of the model prediction are evaluated by calculating the correlation coefficient between the predicted value and the true value. At the same time, the systematic error and random error of the predicted value are determined by deviation analysis to ensure the reliability of the model's prediction results in practical applications. These analytical methods are combined to fully verify the stability and reliability of the model of the present invention in various complex environments, ensure the accuracy and consistency of the measurement results, provide users with scientific breeding data support, and significantly improve measurement efficiency and accuracy.

[0189] like Figure 2 As shown, a system for applying the deep learning-based pig in vivo B-ultrasound image data analysis method provided by an embodiment of the present invention includes:

[0190] Image upload module: The image upload module supports users to select a single B-ultrasound image file by clicking the "Upload Image" button, or to upload image files in batches by clicking the "Select Folder" button. This module is compatible with common image formats such as JPG and PNG, ensuring that users can successfully upload images in various application scenarios. After the upload succeeds or fails, the system will feedback the upload status to the user through the interface prompt or message window to improve the user experience.

[0191] Model Management Module: The model management module allows users to load and replace different YOLO model files to meet different analysis needs. Users can select and load new model files from the local disk through the "Change Model" button on the interface. The system supports dynamic loading technology, and the model can be replaced without restarting the software, thereby improving the flexibility and convenience of operation. Through this module, users can select the most suitable model according to specific analysis goals and improve analysis results.

[0192] Image Analysis Module: The image analysis module is the core of the system, responsible for processing images uploaded by users and performing deep learning model analysis. The module uses the YOLOv8Pose model for key point detection, automatically identifying and calculating important biometric indicators such as backfat thickness and eye muscle depth. During the analysis process, the system pre-processes the image, uses the deep learning model to detect key points and outputs coordinate information. The analysis results will be displayed in real time in the designated area of ​​the user interface, including key points and measurement lines marked on the image, to ensure the accuracy and visualization of the results.

[0193] Result display and export module: The result display and export module is responsible for intuitively displaying the analysis results in the user interface in the form of charts or text. Users can clearly see the specific values ​​of backfat thickness and eye muscle depth, and mark the measurement lines and key points on the processed images. This module also provides data export function, and users can choose to export the analysis results as TXT format files for subsequent recording, report writing and further data analysis.

[0194] Data Management Module: The data management module processes and maintains the storage and management of image files and analysis results to ensure data integrity and reliability. The system automatically stores the analyzed key indicators and processed image files in the local file system, following certain naming conventions to prevent file conflicts. The module also supports batch processing, which can automatically filter non-image format files to ensure the smooth progress of the entire analysis process. In addition, users can save the analysis results as TXT files through the export function to facilitate further data processing and analysis.

[0195] Exception handling and logging module: The exception handling and logging module uses a global exception capture mechanism to promptly capture and record runtime errors, ensuring that the system can provide timely feedback to users when encountering problems. For foreseeable exceptions, such as file read and write permission restrictions or unsupported image formats, the system will provide friendly error prompts to guide users on how to solve the problem. The logging function includes operation logs and system logs, which record the user's operation history and detailed information on system operation to facilitate problem diagnosis and user support. Users can use the log export function to save operation logs and system logs as TXT files to facilitate problem tracking and technical support from the support team.

[0196] Another object of the present invention is to provide a computer device, which includes a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the method and system for analyzing pig live B-ultrasound image data based on deep learning.

[0197] Another object of the present invention is to provide a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, the processor executes the steps of the method and system for analyzing pig live B-ultrasound image data based on deep learning.

[0198] Another object of the present invention is to provide an information data processing terminal, which is used to implement the deep learning-based pig living B-ultrasound image data analysis method and system.

[0199] Figure 3 It is the construction of a B-ultrasound image analysis dataset, in which 304 pig data were enhanced to form 3040 training sets and 304 validation sets, and 100 pigs were used as a test set.

[0200] Figure 4 are the annotations of the detection boxes, (A) is the interspersed detection box, (B) is the small detection box, (C) is the medium detection box, and (D) is the large detection box.

[0201] Figure 5 It is the marking and correction of key points. (A) is three-point marking, (B) is two-point marking, (C) is the initial marking map, and (D) is the corrected marking map, all on a straight line.

[0202] Figure 6 It is a data enhancement method, (A) is image data enhancement, and (B) is rotation data enhancement.

[0203] Figure 7 It is the flow chart of the model, which inputs the image into the deep neural network, obtains the structure of key point prediction, and finally draws it on the original image.

[0204] Figure 8 It is a comparison of the effects of two methods. (A) is a comparison of the effects of the three-point method, and (B) is a comparison of the effects of the two-point method.

[0205] Fig. 9 It is a method to convert pixel values ​​into real values, showing the process of converting pixel values ​​(540px) into actual length (150mm) through conversion factors.

[0206] Fig.10 It is a flow chart of the pig live B-ultrasound image analysis software, showing the complete process from image uploading, model reasoning, result display to log recording and exporting.

[0207] Fig.11 This is a real screen shot of the pig live B-ultrasound image analysis software, showing the actual operation screen of the software interface, image uploading and analysis, key point measurement and result display.

[0208] Fig.12 is a visualization effect diagram of the feature map, (A) is a visualization effect diagram of the interspersed detection box, (B) is a visualization effect diagram of the small detection box, (C) is a visualization effect diagram of the medium detection box, and (D) is a visualization effect diagram of the large detection box.

[0209] Fig.13 is the evaluation of the model, (A) is the F1-confidence curve, (B) is the precision-recall curve, (C) is the normalized confusion matrix, (D) is the precision-confidence curve, (E) is the recall-confidence curve, and (F) is the normalized confusion matrix.

[0210] Fig.14 It is a multiple measurement jump interval diagram, showing the changes in eye muscle depth and back fat thickness in multiple measurements.

[0211] Fig.15 : are correlation analysis diagrams, (A) is the correlation analysis of backfat thickness prediction, and (B) is the correlation analysis of eye muscle depth prediction.

[0212] Fig.16 These are deviation analysis diagrams. (A) is the back fat thickness deviation analysis, and (B) is the eye muscle depth deviation analysis.

[0213] Application Example: Automated backfat thickness and eye muscle depth measurement system in Wuhan Zhule Tianyuan Xinzhou pig farm

[0214] In order to demonstrate the creativity and technical value of the technical solution of the present invention, the application example of the deep learning-based pig live B-ultrasound image data analysis method and system in Wuhan Zhule Tianyuan Xinzhou Pig Farm will be introduced in detail below.

[0215] Example: Wuhan Zhule Tianyuan Xinzhou Pig Farm

[0216] System Overview

[0217] At Wuhan Zhule Tianyuan Xinzhou Pig Farm, in order to achieve efficient measurement and management of pig production and breeding, the pig live B-ultrasound image data analysis system based on deep learning of the present invention was applied to build an automated backfat thickness and eye muscle depth measurement system.

[0218] Specific steps

[0219] (1) Data Collection

[0220] B-ultrasound image acquisition: Several portable veterinary B-ultrasound imagers are installed in the pig farm to collect real-time B-ultrasound image data of the pigs' backs and eye muscles. These data come from different breeds of pigs, including Duroc, Yorkshire, Landrace and Berkshire.

[0221] Data upload: Through the image upload module, the pig farm staff uploads the collected B-ultrasound image data to the system for processing and analysis.

[0222] (2) Data processing and annotation

[0223] Labeling tool: Use the Labelme tool to label all images, including detection boxes and key point locations, to ensure data consistency and accuracy.

[0224] Data format conversion: The annotation results are saved in JSON format and converted into YOLO format to prepare for subsequent model training.

[0225] (3) Data enhancement and model training

[0226] Data enhancement: Through data enhancement techniques (such as color transformation, exposure adjustment, saturation adjustment, brightness change, and simulated image noise and blur), the training data is expanded and the generalization ability of the model is enhanced.

[0227] Model selection and training: Select the YOLOv8Pose model for training, set appropriate hyperparameters (such as epochs, batch size), and use the RTX 3090 GPU for training to ensure the efficiency and accuracy of the model.

[0228] (4) Real-time measurement and analysis

[0229] Automated analysis: In a production environment, the system automatically analyzes uploaded B-ultrasound images, uses deep learning models to detect key points, and calculates backfat thickness and eye muscle depth.

[0230] Result display: The measurement results are displayed in real time on the user interface and saved in the local database for subsequent analysis and management.

[0231] (5) Data export and report generation

[0232] Result export: The system provides data export function, and pig farm managers can export measurement results into TXT format files for recording and report writing.

[0233] Report generation: The system automatically generates a detailed measurement report, including the specific values ​​of backfat thickness and eye muscle depth and detailed description of the measurement process, providing a scientific basis for production management and breeding.

[0234] Implementation Effect

[0235] (1) Measurement accuracy: Through the deep learning model, the accuracy of the measurement results is significantly improved, reducing the errors and uncertainties of manual measurements and ensuring the reliability of each measurement.

[0236] (2) Measurement efficiency: The system can perform real-time measurements while pigs are moving, greatly improving measurement efficiency, meeting the needs of large-scale production, and avoiding production stagnation caused by measurement delays.

[0237] (3) Production cost: The automated measurement system reduces the reliance on expensive mechanical equipment and a large amount of manual operations, reduces production costs, and improves economic benefits.

[0238] (4) Scientific breeding support: The high-precision back fat thickness and eye muscle depth data provided by the system provide a scientific basis for pig production and breeding, optimize breeding strategies, and improve breeding efficiency and effectiveness.

[0239] In the application example of Wuhan Zhule Tianyuan Xinzhou Pig Farm, the deep learning-based pig live ultrasound image data analysis method and system of the present invention, through advanced deep learning technology and system integration, realizes the efficient and accurate measurement of the back fat thickness and eye muscle depth of live pigs, significantly improves the measurement accuracy and efficiency, reduces production costs, provides reliable data support for scientific breeding, and has broad application prospects and significant economic benefits.

[0240] The embodiments of the present invention have achieved some positive effects in the process of research and development or use, and have significant advantages over the prior art. The following content is described in combination with data, charts, etc. of the test process.

[0241] Significant improvement in recognition accuracy: In the experiments conducted, the system performed well in key point recognition. Fig.13 (A) shows the F1-confidence curve, indicating that the system can maintain high efficiency at different confidence levels. Fig.13 (D) shows the precision-confidence curve. Fig.13 (B) and Fig.13(E) shows the precision-recall curve and recall-confidence curve respectively. These results show that the recognition accuracy of this system on the test dataset is high and stable.

[0242] Accuracy performance analysis

[0243] Fitting analysis of backfat thickness and eye loin depth:

[0244] Fig.15 (A) shows the correlation analysis of backfat thickness prediction. The equation of the fitting line is y = 0.8960x + 2.0122, and the fitting degree R 2 is 0.8833, and the correlation coefficient r is 0.9399. Fig.15 (B) shows the correlation analysis of eye muscle depth prediction. The equation of the fitting line is y = 0.9915x + 0.9089, and the fitting degree R 2 The correlation coefficient is 0.9285 and r is 0.9636. These results illustrate the stability and accuracy of the system under different conditions.

[0245] Deviation analysis of backfat thickness and loin depth:

[0246] Fig.16 (A) shows the deviation analysis between the true value and the predicted value of backfat thickness, with a mean square error (MSE) of 0.9730 and a root mean square error (RMSE) of 0.9864. Fig.16 (B) shows the deviation analysis of the true value and predicted value of eye muscle depth, with MSE of 2.6513 and RMSE of 1.6283. These results show that the correlation between the predicted value and the true value using this system is high, the deviation is small, and the prediction effect is better than the conventional method.

[0247] Multiple measurement jump intervals:

[0248] Fig.14 The jump range of eye muscle depth and back fat thickness in multiple measurements was demonstrated, and the results showed that the measured values ​​were highly consistent, further verifying the stability and reliability of the system.

[0249] Actual application effect: Fig.11 The actual operation interface of the system was demonstrated, showing that the system can measure the eye muscle area of ​​pig carcasses in real time, mark key points and record data. These data not only help companies improve production efficiency, but also support scientific breeding decisions.

[0250] Feature map visualization effect: Fig.12 The feature map visualization effects of different detection boxes are shown, (A) is an interspersed detection box, (B) is a small detection box, (C) is a medium detection box, and (D) is a large detection box. These visualization effects help us understand the model's feature extraction capabilities in different areas and improve the model's interpretability.

[0251] Through the above tests and analysis, the in vivo B-ultrasound image intelligent measurement system of the present invention has significant innovation and practicality in technology, and also shows great potential and value in practical applications. The high precision, high efficiency and high reliability of the system make it have a wide range of application prospects in actual production scenarios, significantly improving production efficiency and product quality, and providing scientific breeding decision support and precise production management basis for enterprises.

[0252] The embodiment of the present invention provides a method for analyzing pig in vivo B-ultrasound image data based on deep learning, comprising the following steps:

[0253] S1: Unique dataset construction

[0254] Ultrasound imaging data from different breeds of pigs were collected at the pig farm of Zhule Tianyuan Ecological Agriculture Co., Ltd. using a portable veterinary ultrasound imager. After data sorting, ultrasound imaging data were obtained, and each data set was equipped with the true value information of the pig's breed, weight, age, back fat thickness and eye muscle depth;

[0255] S2: Detection box and key point annotation

[0256] All images were annotated by the Labelme tool, and all detection boxes and key points were annotated by the same worker. The annotation results were saved in json format and then converted to YOLO format. During the image annotation process, the conventional backfat thickness and eye muscle depth were annotated separately;

[0257] S3: Data Augmentation Strategy

[0258] Expand training data through data augmentation methods such as color transformation, exposure adjustment, saturation adjustment, brightness change, and simulated image noise and blur;

[0259] S4: Selection and training of deep learning models

[0260] S5: Develop B-ultrasound image analysis and calculation algorithms

[0261] In terms of image processing and key point detection, a deep learning model is used to analyze B-ultrasound images. By detecting and obtaining key point coordinate information, the pixel measurements of eye muscle depth and back fat thickness are calculated and converted into actual physical dimensions.

[0262] S6: Building a deep learning-based pig in vivo ultrasound imaging data analysis system

[0263] S7: Model performance and B-ultrasound image analysis evaluation

[0264] 1. Model level: Use "artificial intelligence model" or "deep learning model" as a superordinate concept to cover more types of models.

[0265] Use any AI model to perform image analysis and key point detection to calculate relevant physical parameters.

[0266] 2. Analysis content: Expanded to the analysis of all B-ultrasound images, not limited to eye muscle depth and back fat thickness, including other potential physiological parameters.

[0267] Analyze image data including but not limited to loin depth, backfat thickness, and other physiological parameters.

[0268] 3. Holistic approach: By providing new processing ideas, the rights coverage is expanded as much as possible, covering the entire process of data collection, processing, analysis and system construction.

[0269] It includes dataset construction, image annotation, data enhancement, model training and selection, image analysis and computing algorithm development, system construction and performance evaluation, which can comprehensively analyze various physiological parameters of living pigs.

[0270] It should be noted that the embodiments of the present invention can be implemented by hardware, software, or a combination of software and hardware. The hardware part can be implemented using dedicated logic; the software part can be stored in a memory and executed by an appropriate instruction execution system, such as a microprocessor or dedicated design hardware. It can be understood by a person of ordinary skill in the art that the above-mentioned devices and methods can be implemented using computer executable instructions and / or contained in a processor control code, such as a carrier medium such as a disk, CD or DVD-ROM, a programmable memory such as a read-only memory (firmware), or a data carrier such as an optical or electronic signal carrier. Such code is provided on the carrier medium. The device and its modules of the present invention can be implemented by hardware circuits such as very large-scale integrated circuits or gate arrays, semiconductors such as logic chips, transistors, etc., or programmable hardware devices such as field programmable gate arrays, programmable logic devices, etc., can also be implemented by software executed by various types of processors, and can also be implemented by a combination of the above-mentioned hardware circuits and software, such as firmware.

[0271] The above description is only a specific implementation mode of the present invention, but the protection scope of the present invention is not limited thereto. Any modifications, equivalent substitutions and improvements made by any technician familiar with the technical field within the technical scope disclosed by the present invention and within the spirit and principle of the present invention should be covered by the protection scope of the present invention.

Claims

1. A method for analyzing pig in vivo B-ultrasound image data based on deep learning, characterized in that: By building a unique data set and using artificial intelligence models to conduct a comprehensive analysis of live pig B-ultrasound images, including image annotation, data enhancement, model training, and real-time system construction; the steps include collecting B-ultrasound image data in pig farms, annotating detection frames and key points, applying data enhancement strategies, selecting and training deep learning models, developing algorithms for analysis and calculation, and finally building a deep learning-based analysis system and evaluating its performance. This method innovatively achieves accurate measurement and analysis of pig backfat thickness and eye muscle depth.

2. The method for analyzing pig in vivo B-ultrasound image data based on deep learning as claimed in claim 1, characterized in that: The following steps are involved: S1, unique dataset construction: Ultrasound imaging data from different breeds of pigs were collected at Zhule Tianyuan Xinzhou Pig Farm using a portable veterinary ultrasound imaging device. After data collation, ultrasound imaging data were obtained, and each data set was equipped with the true value information of the pig's breed, weight, age, back fat thickness and eye muscle depth. S2, detection box and key point annotation: All images were annotated using the Labelme tool, and all detection boxes and key points were annotated by the same worker. The annotation results were saved in json format and then converted into YOLO format. During the image annotation process, the conventional backfat thickness and eye muscle depth were annotated separately. S3, data enhancement strategy: The training data was expanded through data augmentation methods, including color transformation, exposure adjustment, saturation adjustment, brightness change, and simulated image noise and blur; S4, selection and training of deep learning models; S5, Develop B-ultrasound image analysis and calculation algorithms: In terms of image processing and key point detection, a deep learning model is used to analyze B-ultrasound images. By detecting and obtaining key point coordinate information, the pixel measurements of eye muscle depth and back fat thickness are calculated and converted into actual physical dimensions. S6, build a pig in vivo B-ultrasound image data analysis system based on deep learning; S7, Evaluation of model performance and B-ultrasound image analysis.

3. The method for analyzing pig in vivo B-ultrasound image data based on deep learning as claimed in claim 2, characterized in that: In S1, 100 data sets are randomly selected as test sets and do not participate in training and verification; after image data enhancement is performed on 304 data sets other than the test set, 304 images are extracted as verification sets, and the remaining 3040 images are used as training sets.

4. The method for analyzing pig in vivo B-ultrasound image data based on deep learning as claimed in claim 2, characterized in that: The mechanism based on YOLOv8-pose model learning in S2 requires first marking the detection box and then marking the key points in the image; S21, box-point relationship design: In order to determine the detection box design that optimizes the model performance, four box-point relationship experiments were designed, ranging from small detection boxes that are close to key points to large detection boxes that cover a wider image area, to evaluate the impact of different detection box configurations on key point recognition accuracy; S22, detection box annotation: All images are annotated by the Labelme tool. All detection boxes and key points are annotated by the same measurement personnel who measure the true value to ensure uniformity and reduce subjective errors. The annotation results are saved in json format and then converted to YOLO format by the label conversion algorithm; S23, key point marking: In B-ultrasound image analysis, the measurement definition of eye muscle depth and back fat thickness is based on the vertical distance between the third and fourth ribs of the pig; because the upper point of the eye muscle and the lower point of the back fat are the same point, three key points are marked: the lower point of the eye muscle, the upper point of the eye muscle and the upper point of the back fat, and the three should be arranged vertically in space; S24, key point annotation correction: The accurate annotation of key points is crucial to the accuracy of measurement; therefore, a key point annotation correction process was implemented, which includes adjusting the x-coordinates of other key points based on the x-coordinate of the subocular point; S3 is used to construct a pig B-ultrasound image dataset; S31, keypoint image data augmentation: changing the visual characteristics of an image without affecting the labels of keypoints to simulate changes in the real world, including color transformation, exposure adjustment, saturation adjustment, brightness change, and simulated image noise and blur; S32, key point rotation data enhancement: Considering rotation data enhancement, not only the image is rotated, but also the label of the rotated image is adapted accordingly.

5. The method for analyzing pig in vivo B-ultrasound image data based on deep learning as claimed in claim 2, characterized in that: In S4, the selected YOLOv8pose model is used for key point detection; S41, training parameter setting: YOLOv8 model is the core of this study. Ultralytics library is used to train YOLOv8 model. YOLOv8n is used as the main model. The training cycle is set to 500 epochs, which is medium. As a control, a fully trained model is made, that is, 100 epochs of early stopping patience is set. In these 100 epochs, the training is terminated without any prompt. S42, YOLOv8pose model training: the input image size has a resolution of 640*640, and RTX 3090GPU is used for training. Other parameters including learning rate, batch size, etc. all use Ultralytics' default weights for YOLOv8 training; during the training process, data enhancement is also used, including hue, saturation, brightness adjustment intensity, as well as translation, scaling, and mosaic data enhancement.

6. The method for analyzing pig in vivo B-ultrasound image data based on deep learning as claimed in claim 2, characterized in that: The S5 analysis steps for B-ultrasound images include using the OpenCV software library to convert the image from the original BGR color space to the RGB color space required by the YOLO model; using the Ultralytics library to support YOLO detection, performing deep learning analysis on the image, thereby identifying key points in the image and obtaining the coordinate information of these points; S51, secondary drawing and calculation: In order to ensure the accuracy of the calculation of eye muscle depth and back fat thickness, secondary drawing and calculation were implemented on the analysis results; first, the coordinates of the lower point of the eye muscle were determined as a reference, and then a vertical line was drawn upward from this point to intersect with the y-axis coordinate of the upper point of the eye muscle to determine a new upper point position of the eye muscle; S52, pixel value conversion to real value method: Since the model outputs the pixel position of the key point, and these pixel values ​​need to be converted to the actual physical size, a conversion factor is introduced to achieve this conversion; considering the resolution of B-ultrasound images and the needs of actual measurement, the height of 540 pixels of each image corresponds to the actual length of 150 mm as a benchmark to obtain the conversion factor; S53, Two-point algorithm for B-ultrasound analysis: A two-point algorithm was proposed based on the improvement and innovation of the three-point algorithm.

7. The method for analyzing pig in vivo B-ultrasound image data based on deep learning as claimed in claim 2, characterized in that: The S6 constructs a pig live B-ultrasound image data analysis system based on deep learning, which includes the following main modules: image uploading, model management, image analysis and result display, data export, exception handling and log recording, and model performance evaluation; specifically: S61, Image Upload Module: The image upload module supports users to select a single B-ultrasound image file by clicking the "Upload Image" button, or upload image files in batches by clicking the "Select Folder" button; this module is compatible with common image formats such as JPG and PNG, ensuring that users can upload images smoothly in various application scenarios; After the upload succeeds or fails, the system will provide feedback to the user on the upload status through an interface prompt or message window; S62, Model Management Module: The model management module allows users to load and replace different YOLO model files to meet different analysis needs; users can select and load new model files from the local disk through the "Change Model" button on the interface; the system supports dynamic loading technology, and the model can be replaced without restarting the software; through this module, users can choose the most suitable model according to the specific analysis goals; S63, Image Analysis Module: The image analysis module is the core of the system, responsible for processing images uploaded by users and performing deep learning model analysis; this module uses the YOLOv8Pose model to detect key points, automatically identify and calculate important biometric indicators such as backfat thickness and eye muscle depth; During the analysis, the system pre-processes the image, uses a deep learning model to detect key points and outputs coordinate information; The analysis results will be displayed in real time in the designated area of ​​the user interface, including key points and measurement lines marked on the image; S64, result display and export module: The result display and export module is responsible for intuitively displaying the analysis results in the user interface in the form of charts or texts; users can clearly see the specific values ​​of backfat thickness and eye muscle depth, and mark the measurement lines and key points on the processed images; this module also provides data export function, and users can choose to export the analysis results into TXT format files to facilitate subsequent records, report writing and further data analysis; S65, Data Management Module: The data management module processes and maintains the storage and management of image files and analysis results to ensure the integrity and reliability of the data; the system automatically stores the analyzed key indicators and processed image files in the local file system, following certain naming conventions to prevent file conflicts; S66, Exception handling and logging module: The exception handling and logging module uses a global exception capture mechanism to timely capture and record runtime errors, ensuring that the system can provide timely feedback to users when encountering problems; for foreseeable exceptions, such as file read and write permission restrictions or unsupported image formats, the system will provide friendly error prompts to guide users on how to solve the problem; the logging function includes operation logs and system logs, which record the user's operation history and detailed information on system operation; S7 introduces the innovative model evaluation index of multiple measurement jump intervals to evaluate the stability and repeatability of the results based on conventional model performance analysis such as precision, recall rate, F1 score, confusion matrix and normalized confusion matrix. In addition, the correlation analysis and deviation analysis of B-ultrasound analysis results and true values ​​are also included in the evaluation system, specifically: S71, multiple measurement jump interval: In order to evaluate the consistency and repeatability of the measurement, the multiple measurement jump interval is introduced; this indicator measures the maximum difference between the results when multiple measurements are made on the same image; ideally, this difference should be zero, indicating that the model gives completely consistent results in multiple measurements; S72, B-ultrasound image analysis and evaluation: To further verify the effectiveness of the model, correlation analysis and deviation analysis were performed between the predicted values ​​and the true values; the correlation coefficient between the predicted values ​​and the true values ​​was calculated.

8. A pig in vivo B-ultrasound image data analysis system based on deep learning that implements the B-ultrasound image data analysis method according to any one of claims 1 to 7, characterized in that: include: Image upload module: The image upload module supports users to select a single B-ultrasound image file by clicking the "Upload Image" button, or upload image files in batches by clicking the "Select Folder" button; The module is compatible with common image formats such as JPG and PNG, ensuring that users can upload images smoothly in various application scenarios; After the upload succeeds or fails, the system will provide feedback to the user on the upload status through an interface prompt or message window; Model management module: The model management module allows users to load and replace different YOLO model files to meet different analysis needs; through the "Change Model" button on the interface, select and load a new model file from the local disk; the system supports dynamic loading technology, and the model can be replaced without restarting the software, thereby improving the flexibility and convenience of operation; Image analysis module: The image analysis module is the core of the system, responsible for processing images uploaded by users and performing deep learning model analysis; Result display and export module: The result display and export module is responsible for intuitively displaying the analysis results in the user interface in the form of charts or text; Data management module: The data management module processes and maintains the storage and management of image files and analysis results; Exception handling and logging module: The exception handling and logging module uses a global exception capture mechanism to promptly capture and record runtime errors, ensuring that the system can provide timely feedback to users when encountering problems; for foreseeable exceptions.

9. A computer-readable storage medium storing a computer program, wherein when the computer program is executed by a processor, the processor executes the steps of the method for analyzing pig live B-ultrasound image data based on deep learning as described in any one of claims 1 to 7.

10. An information data processing terminal, characterized in that: The information data processing terminal is used to implement the pig living B-ultrasound image data analysis system based on deep learning as described in claim 8.