Intelligent supervision system for pig weighing

Abstract

This invention is an intelligent monitoring system specifically designed for large-scale pig farm transactions. The system analyzes the video stream from the image acquisition device in real time through a data processing unit, using AI models to analyze the images during weighing to obtain the number of pigs and the estimated weight of each pig. Simultaneously, the system can count the total number of pigs entering and leaving the weighbridge in real time, effectively preventing duplicate weighing. A remote monitoring device connects to the weighbridge to obtain the weighing weight of N pigs. Finally, the collected multi-dimensional data is uploaded to the data storage unit, and users can view relevant information through the system's interactive interface. This system effectively improves the accuracy and efficiency of weighing, reduces manual intervention and human error, and provides real-time data analysis support to optimize the transaction process. Furthermore, the system's automation functions help improve pig farm management efficiency, providing efficient and reliable intelligent support for pig farming enterprises during transactions.

Description

Technical Field

[0001] This invention relates to the field of intelligent systems, specifically to trading scenarios in large-scale pig farms, aiming to optimize the pig weighing process through image analysis and data processing technologies. The system combines computer vision, wireless communication, and data storage technologies to improve the efficiency of farm management and the accuracy of transactions. Background Technology

[0002] Traditional pig weighing relies primarily on manual operation and mechanical weighbridges. This method is time-consuming and susceptible to human error, leading to inaccurate data. In large-scale farms, manually recording and calculating the number and weight of pigs is cumbersome and prone to errors. While some existing systems incorporate electronic weighbridges and barcode technology to reduce manual recording, they still require human intervention and lack real-time analysis capabilities.

[0003] Some systems attempt to use image recognition technology to capture images of pigs via cameras and perform simple processing to estimate their number and weight. However, these systems suffer from shortcomings in recognition accuracy and real-time performance, making it difficult to accurately identify pigs in complex environments. Furthermore, their data storage and management mechanisms are inadequate, failing to provide comprehensive data support for farms.

[0004] This invention employs advanced YOLOv8 image recognition algorithms and deep learning models, improving recognition accuracy and efficiency. It can analyze video streams in real time to obtain the number of pigs and estimate their weight, and combine this with actual weight data obtained from remote monitoring equipment to provide accurate weighing results. The system also features data storage and interactive functions, allowing users to view and analyze data through an interface, generate visual reports, and optimize management processes. This system overcomes the shortcomings of existing technologies, providing reliable intelligent support for livestock enterprises. Summary of the Invention

[0005] To enable intelligent support for trading scenarios in large-scale pig farms, this invention provides an intelligent monitoring system for weighing pigs, employing the following technical solution: An intelligent monitoring system for weighing pigs, characterized in that it comprises: Image acquisition device (1) is used to acquire high-resolution image data of pigs in the weighing area; The data processing unit (2), acting as the system control unit, is responsible for receiving video data from the image acquisition device (1) and coordinating the interaction and collaboration between various devices and functional units. Simultaneously, it performs the following functions based on deep learning algorithms: Image recognition and number counting are performed using object detection algorithms; the location of pigs is detected and bounding boxes are generated. Image segmentation of the pig region was performed using a semantic segmentation algorithm to extract boundary contours; Pig weight is estimated based on pixel area and a pre-trained regression formula; A wireless transmission device (3) is used to enable bidirectional transmission of real-time data and instructions between the user and the data processing unit (2); The remote monitoring device (4) is used to access the weight data of the old weighbridge sensor in real time and transmit the data to the data processing unit (2) via wireless Wi-Fi. Data storage unit (5), used to store data during the weighing process of pigs, including: Real-time video streams and processed images; The results of image recognition and segmentation; Estimated weight, actual weight, and count data of pigs; The system interactive interface (6) is used for users to view stored data and generate visual analysis reports. It also serves as a bridge between the wireless transmission device (3) and the data processing unit (2) to realize information interaction and transmission.

[0006] Furthermore, the image acquisition device (1) is characterized in that: a high-resolution monocular camera is used and installed at a predetermined height directly above the midpoint of the pig weighbridge stall to obtain a complete top view of the weighbridge stall, obtain the original image (7), and eliminate the image distortion caused by the lens or installation angle through a calibration-based radial distortion correction algorithm to obtain a corrected image (8).

[0007] Furthermore, the data processing unit (2) is characterized in that it includes: Real-time analysis of the video stream (9) transmitted by the image acquisition device (1); Equipped with a pig identification model (10) and a pig segmentation model (11), data on the number of pigs (12), the estimated weight of each pig (13), and the total number of pigs entering and leaving the weighbridge (14) are obtained through image recognition and contour segmentation. It supports connection to a remote monitoring device (4) via Wi-Fi to obtain the weighbridge weight of N pigs in real time (15) and upload the analysis results to the data storage unit (5).

[0008] Furthermore, the pig identification model (10) and pig segmentation model (11) are characterized by comprising: Built based on convolutional neural networks and deep learning algorithms; The pig identification model (10) is preferably trained using the YOLOv8 algorithm. By analyzing the corrected image (8) of the pigs when they are weighed, the number of pigs (12) and the coordinates of their identification boxes (16) are obtained. The pig segmentation model (11) is based on the SAM model. It extracts the outline pixel region of a single pig by recognizing the bounding box coordinates (16), calculates the pixel area, and obtains the estimated weight of each pig (13) by combining the regression formula. The above models and algorithms can be used with other object detection or segmentation methods, but are not limited to the YOLOv8 algorithm and the SAM model.

[0009] Furthermore, the data processing unit (2) is characterized by its function of real-time analysis of the video stream (9) transmitted by the image acquisition device (1), comprising: Based on the pig identification model (10), analyze each frame of video stream image and mark the entry and exit lines at the entry and exit positions, and count the behavior of pigs passing through the entry or exit lines. When pigs cross the entry line, the number of pigs entering the pen increases by 1, and vice versa, it decreases by 1. When pigs cross the slaughter line, the number of pigs slaughtered increases by 1, and vice versa, it decreases by 1. After receiving the weighing instruction sent by the wireless transmission device (3), the data processing unit (2) uploads data such as the total number of pigs entering and leaving the weighbridge (14) and the number of pigs (12) to the data storage unit (5).

[0010] Furthermore, the remote monitoring device (4) is characterized in that it includes: A wireless serial port server (17) is used to achieve the connection, which solves the problem that old weighbridges do not have wireless function and it is difficult to connect the weighbridge directly to the data processing unit (2). A wireless serial server (17) is configured next to each weighbridge head, wherein the RS485 port is connected to the weighbridge head and data is accessed to the local network via Wi-Fi. The data processing unit (2) can realize the real-time transmission and monitoring of weighbridge data through the wireless connection, thereby obtaining the weighbridge weight of N pigs when the pigs are weighed (15).

[0011] Furthermore, the wireless transmission device (3) is characterized in that: it refers to mobile terminals such as mobile phones, computers, and tablets, which realize interactive operation with the data processing unit (2) through wireless networks (Wi-Fi or 4 / 5G), and send weighing instructions to the data processing unit (2) through the system interactive interface (6) for execution, and can obtain data such as a 10-second video stream (9), original image (7), corrected image (8), number of pigs (12), estimated weight of each pig (13), total number of pigs entering and leaving the weighbridge (14), and weighbridge weight of N pigs (15), so as to realize data support work.

[0012] Furthermore, the data storage unit (5) is characterized in that it employs a storage server, which is mainly used to store structured data, images, and videos generated during the use of the monitoring system.

[0013] Furthermore, the system interactive interface (6) is characterized in that: Staff members obtain the system interaction interface (6) through the wireless transmission device (3) and interact with the data processing unit (2) through commands; Display the data in the data storage unit (5) to achieve a visual interface; Users can generate charts and reports based on the data, helping them to understand and analyze the data more intuitively, so as to more comprehensively demonstrate the data characteristics of the intelligent supervision process.

[0014] Furthermore, the intelligent monitoring system for weighing pigs is characterized in that: the data processing unit (2) analyzes the video stream (9) of the image acquisition device (1) in real time; the staff sends instructions to the data processing unit (2) through the wireless transmission device (3) to extract the video stream (9) and obtain the original image (7) of the pig weighing, and obtains the corrected image (8) through distortion correction processing. The corrected image (8) is analyzed by AI model to obtain the number of pigs (12) and the estimated weight of each pig (13), and the total number of pigs entering and leaving the weighbridge is counted (14); the remote monitoring device (4) is connected to the weighbridge to obtain the weighbridge weight (15) of N pigs entering the weighbridge; the wireless transmission device (3) sends instructions to the data processing unit (2) to save all the acquired data to the data storage unit (5), and the user can view it through the system interactive interface (6).

[0015] The beneficial effects of this invention are as follows: First, by employing the advanced YOLOv8 image recognition algorithm and deep learning model, the system can analyze video streams in real time and automatically acquire data on the number of pigs (9) and the estimated weight of each pig (10). This reduces the need for manual operation and lowers the possibility of human error. By counting the total number of pigs entering and leaving the weighbridge (11), duplicate weighing is effectively prevented, achieving more accurate asset monitoring. Second, the system is compatible with older weighbridge equipment through remote monitoring equipment (4), enabling wireless data transmission and monitoring. This automated supervision and data processing significantly improves weighing efficiency and saves time and labor costs. Finally, the system's interactive interface (6) provides a visual display of the data, allowing users to generate charts and reports for a more intuitive understanding and analysis of the data.

[0016] After trials in actual pig farms, the invention has shown excellent results. These beneficial effects enable the invention to effectively improve efficiency and accuracy in large-scale farm applications, providing reliable intelligent support for trading scenarios in livestock enterprises. Attached Figure Description

[0017] Figure 1 This is a diagram of the main components.

[0018] Figure 2 This is a diagram showing the remote monitoring equipment connected to the weighbridge head.

[0019] Figure 3 The video stream (9) stored in the data storage unit (5), the original image (7) of the pig being weighed, and the corrected image (8) are also stored in the data storage unit (5).

[0020] Figure 4 It is the line for entering and exiting the weighbridge.

[0021] Figure 5 The number of pigs stored in the data storage unit (5) (12), the total number of pigs entering and leaving the weighbridge (14), the estimated weight of each pig (13), and the weighbridge weight of N pigs (15).

[0022] Figure 6 It is the system interface for users to operate the weighing process (6).

[0023] Figure 7 It is the dataset used for model training. Detailed Implementation

[0024] The present invention will be described in detail below with reference to the accompanying drawings and preferred embodiments. The purpose and effects of the present invention will become clearer. It should be understood that the specific embodiments described herein are merely for explaining the present invention and are not intended to limit the present invention.

[0025] Example 1: System Overall Architecture and Workflow: An intelligent monitoring system for weighing pigs, such as Figure 1 As shown, it includes the following components: image acquisition device (1), data processing unit (2), wireless transmission device (3), remote monitoring device (4), data storage unit (5), and system interaction interface (6).

[0026] The system achieves intelligent monitoring of the pig weighing process through the following steps: 1. Image acquisition equipment (1) Installation and calibration: The image acquisition device (1) uses a high-resolution monocular camera, installed at a predetermined height directly above the center of the pig weighbridge stall to ensure that a complete top-down view can be captured, such as... Figure 3 As shown, the camera captures the video stream (9) when the pigs are weighed and generates the original image (7). The original image (7) is processed by a calibration-based radial distortion correction algorithm to obtain the corrected image (8), providing high-quality image input for subsequent intelligent analysis.

[0027] The distortion correction method uses a checkerboard calibration. First, more than 50 images containing a checkerboard pattern are captured at different angles and distances to calculate the camera's intrinsic parameter matrix (mtx) and distortion coefficients (dist). mtx is used to convert 3D world coordinates to image coordinates, and dist is used to correct distortion in the image. An optimized intrinsic parameter matrix (newmtx) is obtained using mtx and dist, which helps remove image distortion and optimize the field of view. Then, the original image (7) is distorted using the image processing algorithm f and the parameters mtx, dist, and newmtx to obtain the corrected image (8), ensuring that the captured image is more realistic and accurate, thereby improving the accuracy and robustness of subsequent analysis.

[0028] Corrected image (8) = f(original image (7), mtx, dist, newmtx).

[0029] 2. Real-time video stream analysis: The data processing unit (2) receives the video stream (9) transmitted by the image acquisition device (1), performs target detection on each frame of the image through the pig recognition model (10), counts the number of pigs (12), and generates the recognition box coordinates (16). The recognition box coordinates (16) are further passed to the pig segmentation model (11), which extracts the outline pixel region of a single pig, calculates the pixel area, and estimates the weight of each pig (13) by combining the regression formula.

[0030] Meanwhile, the data processing unit (2) analyzes the behavior of pigs passing through the entry and exit lines based on the position markers of the entry and exit lines, and updates the total number of pigs entering and exiting the weighbridge (14).

[0031] By using image recognition and analysis technology, the weighing process of pigs can be automated and precisely managed, which effectively improves the efficiency of the system, avoids human error, and ensures the real-time and reliability of the weighing data.

[0032] 3. Linkage with weighbridge sensors: The remote monitoring device (4) uses a wireless serial port server (17) to connect with the old weighbridge sensor, such as... Figure 2 As shown. After the RS485 interface is connected to the weighbridge head, the weighbridge weight data is transmitted to the data processing unit (2) via Wi-Fi. The weighbridge sensor collects the total weight of N pigs entering the weighbridge in real time (15), solving the problem that old weighbridges do not have wireless function and it is difficult to connect the weighbridge directly to the data processing unit (2).

[0033] 4. Data storage and interaction: The data processing unit (2) uploads the analysis results, including the number of pigs (12), estimated weight (13), total number of pigs entering and leaving the weighbridge (14), and the weighbridge weight of N pigs (15), to the data storage unit (5). In addition, the video stream (9), original image (7), and corrected image (8) are also saved, such as... Figure 3 and Figure 5 As shown.

[0034] The system's interactive interface (6) provides data display and operation functions. Staff can view data and generate charts and reports (such as...) through the wireless transmission device (3). Figure 6 As shown in the figure, this enables comprehensive monitoring and analysis of the weighing process. This process improves data accessibility and transparency, supports farm managers in making faster and more accurate decisions, optimizes management processes, and ensures data storage and long-term traceability.

[0035] This embodiment describes the overall system architecture and workflow. By integrating an image acquisition device (1), a data processing unit (2), a wireless transmission device (3), and a remote monitoring device (4), the system achieves automated and precise management of the pig weighing process. The system uses a high-resolution camera for image acquisition and corrects distortion through image processing algorithms to ensure the quality of image input. Combining a pig recognition model (10) and a pig segmentation model (11), the system can count the number of pigs (12) in real time and estimate the weight of each pig (13), while simultaneously obtaining the total weight of N pigs (15) in conjunction with the weighbridge sensor. The data processing unit (2) stores the analysis results and displays them through the system's interactive interface (6), enabling managers to comprehensively monitor and analyze the weighing process, improve decision-making efficiency and data management capabilities, effectively solve the limitations of traditional weighbridge systems, and enhance the intelligent level of farm management.

[0036] Example 2: Pig identification, weight estimation, and entry / exit statistics 1. Pig identification construction and training: The pig identification model (10) is preferably trained using the YOLOv8 real-time target detection algorithm. More than 2,000 images of pigs being weighed are acquired through the image acquisition device (1), and the data is processed, labeled, trained and fine-tuned.

[0037] During data processing, image data under different lighting conditions are selected, and various noise types are simulated, such as image blurring, cluttered backgrounds, and target occlusion, to ensure the diversity of sample data, thereby enhancing the robustness of the model and improving its recognition accuracy and stability in complex environments.

[0038] During the data annotation process, we ensured that the location of the pigs and the bounding box of each image were accurately labeled. We combined manual annotation with automated annotation and used high-quality annotation tools to ensure that the model could obtain an accurate target detection framework during training, thereby improving detection accuracy.

[0039] During data training, a batch training approach is adopted, splitting the dataset into training and validation sets. Cross-validation is used to avoid overfitting, and data augmentation techniques such as random cropping, rotation, and scaling are employed. Figure 7 (As shown), further improve the model's generalization ability and robustness.

[0040] During data fine-tuning, based on the performance of the initially trained model, detailed parameter adjustments are made on specific tasks or datasets to optimize the model's recognition accuracy for pigs. Especially when encountering special weighing environments and changes in lighting, the model's hyperparameters are fine-tuned to better adapt to real-world application scenarios, improving detection stability and accuracy.

[0041] Finally, a pig identification model (10) was obtained, and its accuracy reached 99% after comparative analysis.

[0042] 2. Pig butchering and weight estimation: The pig segmentation model (11) preferably uses the SAM model, and the model structure includes three main modules: Image Encoder: Responsible for extracting features from images and outputting image embeddings, providing high-dimensional features for subsequent processing; Prompt Encoder: Receives the input encoded information (such as point information or box information) and combines it with imageembedding to provide further information for the mask decoder; Mask Decoder: Combines information from the image encoder and prompt encoder to generate the final mask output, thereby extracting the outline of the pig; By using the SAM model, the complexity and cost of model training can be effectively reduced, while meeting the high accuracy requirements for pig contour extraction, ensuring the accuracy and efficiency of segmentation. By calculating the pig contour in the pig weighing correction image (8), the pixel area can be obtained. Combined with the weighing weight of N pigs (15), a regression formula can be established, and the formula can be obtained: Estimated weight of each pig (13) = A×(pixel area)^2 + B×pixel area + C.

[0043] 3. Inbound and Outbound Statistics: like Figure 4As shown, the data processing unit (2) obtains the track_id and midpoint coordinates of the pigs by combining the marked coordinates of the entry and exit lines with the pig identification model (10), and determines whether each pig has passed through the entry and exit lines. Thus, it updates the total number of pigs entering and exiting the weighbridge in real time (14), and displays the relevant data on the system interactive interface (6) through the wireless transmission device (3) for staff to view and manage. This process effectively prevents pigs from being weighed repeatedly. The formula is: number of pigs (12) = number of pigs entering the weighbridge - number of pigs exiting the weighbridge.

[0044] This embodiment details the implementation process of pig identification, weight estimation, and entry / exit statistics. The system uses the YOLOv8 algorithm to build and train a pig identification model (10). Through diverse data processing and annotation, the model is ensured to have high accuracy and robustness in complex environments, effectively supporting manual pig counting. The pig segmentation model (11) adopts the SAM model. With efficient feature extraction and contour segmentation, it accurately calculates the pixel area of ​​the pigs and combines it with the weighbridge weight of N pigs (15). The estimated weight of each pig is obtained through a regression formula (13), supporting the supervision requirement of weighing uniformity. The entry / exit statistics update the number of pigs entering and exiting in real time by marking the entry / exit lines to avoid duplicate weighing.

[0045] In summary, the above embodiments enhance the intelligence level of data management and provide reliable intelligent support for pig trading scenarios in large-scale breeding enterprises.

Claims

1. An intelligent supervision system for pig weighing, characterized in that, include: Image acquisition device (1) is used to acquire high-resolution image data of pigs in the weighing area; The data processing unit (2), acting as the system control unit, is responsible for receiving video data from the image acquisition device (1) and coordinating the interaction and collaboration between various devices and functional units. Simultaneously, it performs the following functions based on deep learning algorithms: Image recognition and number counting are performed using object detection algorithms; the location of pigs is detected and bounding boxes are generated. Image segmentation of the pig region was performed using a semantic segmentation algorithm to extract boundary contours; Pig weight is estimated based on pixel area and a pre-trained regression formula; A wireless transmission device (3) is used to enable bidirectional transmission of real-time data and instructions between the user and the data processing unit (2); The remote monitoring device (4) is used to access the weight data of the old weighbridge sensor in real time and transmit the data to the data processing unit (2) via wireless Wi-Fi. Data storage unit (5), used to store data during the weighing process of pigs, including: Real-time video streams and processed images; The results of image recognition and segmentation; Estimated weight, actual weight, and count data of pigs; The system interactive interface (6) is used for users to view stored data and generate visual analysis reports. It also serves as a bridge between the wireless transmission device (3) and the data processing unit (2) to realize information interaction and transmission.

2. The image acquisition device (1) according to claim 1, characterized in that: A high-resolution monocular camera was installed at a predetermined height directly above the midpoint of the pig weighbridge to obtain a complete top view of the weighbridge and acquire the original image (7). The image distortion caused by the lens or installation angle was eliminated by a calibration-based radial distortion correction algorithm to acquire the corrected image (8).

3. The data processing unit (2) according to claim 1, characterized in that, include: Real-time analysis of the video stream (9) transmitted by the image acquisition device (1); Equipped with a pig identification model (10) and a pig segmentation model (11), data on the number of pigs (12), the estimated weight of each pig (13), and the total number of pigs entering and leaving the weighbridge (14) are obtained through image recognition and contour segmentation. It supports connection to a remote monitoring device (4) via Wi-Fi to obtain the weighbridge weight of N pigs in real time (15) and upload the analysis results to the data storage unit (5).

4. The pig identification model (10) and pig segmentation model (11) according to claim 3, characterized in that, include: Built based on convolutional neural networks and deep learning algorithms; The pig identification model (10) is preferably trained using the YOLOv8 algorithm. By analyzing the corrected image (8) of the pigs when they are weighed, the number of pigs (12) and the coordinates of their identification boxes (16) are obtained. The pig segmentation model (11) is based on the SAM model. It extracts the outline pixel region of a single pig by recognizing the bounding box coordinates (16), calculates the pixel area, and obtains the estimated weight of each pig (13) by combining the regression formula. The above models and algorithms can be used with other object detection or segmentation methods, but are not limited to the YOLOv8 algorithm and the SAM model.

5. The data processing unit (2) according to claim 3, which performs real-time analysis of the video stream (9) transmitted by the image acquisition device (1), is characterized in that, include: Based on the pig identification model (10), analyze each frame of video stream image and mark the entry and exit lines at the entry and exit positions, and count the behavior of pigs passing through the entry or exit lines. When pigs cross the entry line, the number of pigs entering the pen increases by 1, and vice versa, it decreases by 1. When pigs cross the slaughter line, the number of pigs slaughtered increases by 1, and vice versa, it decreases by 1. After receiving the weighing instruction sent by the wireless transmission device (3), the data processing unit (2) uploads data such as the total number of pigs entering and leaving the weighbridge (14) and the number of pigs (12) to the data storage unit (5).

6. The remote monitoring device (4) according to claim 1, characterized in that, include: A wireless serial port server (17) is used to achieve the connection, which solves the problem that old weighbridges do not have wireless function and it is difficult to connect the weighbridge directly to the data processing unit (2). A wireless serial server (17) is configured next to each weighbridge head, wherein the RS485 port is connected to the weighbridge head and data is accessed to the local network via Wi-Fi. The data processing unit (2) can realize the real-time transmission and monitoring of weighbridge data through the wireless connection, thereby obtaining the weighbridge weight of N pigs when the pigs are weighed (15).

7. The wireless transmission device (3) according to claim 1, characterized in that: This refers to the interaction between mobile terminals such as mobile phones, computers, and tablets and the data processing unit (2) via wireless networks (Wi-Fi or 4 / 5G). The system sends weighing instructions to the data processing unit (2) through the system interface (6) for execution. It can obtain data such as a 10-second video stream (9), the original image (7), the corrected image (8), the number of pigs (12), the estimated weight of each pig (13), the total number of pigs entering and leaving the weighbridge (14), and the weighbridge weight of N pigs (15), thus enabling data support work.

8. The data storage unit (5) according to claim 1, characterized in that: A storage server is used, primarily to store structured data, images, and videos generated during the use of the monitoring system.

9. The system interactive interface (6) according to claim 1, characterized in that: Staff members obtain the system interaction interface (6) through the wireless transmission device (3) and interact with the data processing unit (2) through commands; Display the data in the data storage unit (5) to achieve a visual interface; Users can generate charts and reports based on the data, helping them to understand and analyze the data more intuitively, so as to more comprehensively demonstrate the data characteristics of the intelligent supervision process.

10. The intelligent monitoring system for weighing pigs according to claim 1, characterized in that: The data processing unit (2) analyzes the video stream (9) of the image acquisition device (1) in real time; the staff sends instructions to the data processing unit (2) through the wireless transmission device (3) to extract the video stream (9) and obtain the original image (7) of the pigs being weighed, and obtains the corrected image (8) through distortion correction processing. The corrected image (8) is analyzed by AI model to obtain the number of pigs (12) and the estimated weight of each pig (13), and the total number of pigs entering and leaving the weighbridge is counted (14); the remote monitoring device (4) connects to the weighbridge and obtains the weighbridge weight (15) of N pigs entering the weighbridge; the wireless transmission device (3) sends instructions to the data processing unit (2) to save all the acquired data to the data storage unit (5), which can be viewed by the user through the system interactive interface (6).

Images