A method and system for estimating the market value of a pig based on visual processing

Abstract

The application discloses a pork pig market value estimation method and system based on visual processing, and relates to the technical field of intelligent pig breeding management. The method comprises the following steps: acquiring pig passing video stream of a pork pig market channel; performing pig target identification and cross-frame trajectory association processing on the pig passing video stream to determine pig number information; determining body size feature information according to pig trajectory information meeting preset stable conditions, and combining current weight estimation parameters to determine single body weight estimation information; determining visual weight estimation results according to the single body weight estimation information; acquiring value evaluation reference information including pig price information, pricing rule information and cost reference information to generate market value estimation information; determining weight estimation deviation information according to actual measurement weight feedback information and the visual weight estimation results, and updating the current weight estimation parameters. The application can improve the real-time performance and accuracy of the batch number and weight estimation of the pork pig, timely support the market value judgment of the pork pig with the market data, and improve the long-term weight estimation stability.

Description

Technical Field

[0001] This application relates to the field of intelligent management technology for pig farming, and in particular to a method and system for estimating the market value of pork based on visual processing. Background Technology

[0002] As pig farming becomes increasingly large-scale and intensive, farms are placing greater emphasis on controlling the number, weight, and batch management data of pigs during slaughter, transportation, and sales. The weight and batch quantity of pigs before market entry not only affect loading scheduling and transaction settlement but also influence farms' judgments on slaughter timing, sales prices, and profitability. Currently, farms typically rely on manual observation, manual recording, or centralized weighing to obtain relevant data during the slaughtering process. While these methods may meet basic management needs in small-batch slaughtering scenarios, they are susceptible to influences from personnel experience, on-site order, and operational rhythm in scenarios involving continuous loading and batch slaughtering, making it difficult to reflect the overall situation of each batch in a timely and accurate manner.

[0003] Currently, some livestock management systems or intelligent monitoring devices are being used to assist in pig management. For example, image acquisition devices capture images of pigs in pigsties or passageways, and combined with image recognition, target detection, or data statistics methods, the number, body size, or growth status of pigs can be analyzed. This type of technology reduces the workload of manual statistics and judgment to some extent, but its application scenarios and output results still have limitations. On the one hand, the slaughtering site typically involves rapid pig movement, mutual obstruction, changes in posture, changes in lighting, and limited passageway space, which can easily affect the stability of the collected data and the reliability of the analysis results. On the other hand, existing systems often focus on the collection or identification of single data points, such as quantity statistics, body condition observation, or weight estimation, making it difficult to generate comprehensive data results that can directly support sales decisions during the slaughtering process. Furthermore, some systems are highly dependent on network conditions, manual verification, or subsequent data processing, and still lack sufficient real-time processing capabilities and long-term stable operation.

[0004] Therefore, in the management of hog slaughter, the following issues urgently need to be addressed: the acquisition of on-site data is easily affected by complex operating environments; the real-time and accuracy of the assessment of slaughter batch weight and quantity are insufficient; the relevant data is difficult to support the judgment of market value in a timely manner; and the long-term reliability of the system is insufficient. Summary of the Invention

[0005] This application provides a method and system for estimating the market value of hogs based on visual processing, aiming to solve the problems in existing technologies for hog slaughter management, such as the susceptibility of on-site data acquisition to complex operating environments, insufficient real-time performance and accuracy of slaughter batch weight and quantity assessment, difficulty in timely support of market value judgment by relevant data, and insufficient long-term operational reliability of the system.

[0006] Firstly, a method for estimating the market value of pork based on visual processing, applied to an edge computing terminal, the method comprising: Acquire video streams of pigs passing through an image acquisition device set up in the pig slaughtering channel; The pig target recognition and cross-frame trajectory association processing are performed on the pig passage video stream to obtain pig trajectory information, and the number of pigs to be slaughtered is determined based on the pig trajectory information; Based on the pig trajectory information that meets the preset stability conditions, determine the body size feature information corresponding to each target pig, and based on the body size feature information and the current weight estimation parameters, determine the individual weight estimation information corresponding to each target pig. Based on the individual weight estimation information of each target pig, determine the visual weight estimation result corresponding to the current batch of pigs to be slaughtered; Obtain the value assessment reference information corresponding to the current batch of pigs being slaughtered, the value assessment reference information including pig price information, pricing rule information and cost reference information; Based on the information on the number of pigs slaughtered, the visual weight estimation results, and the value assessment reference information, the estimated market value information corresponding to the current slaughter batch is generated. Obtain the measured weight feedback information corresponding to the current batch of animals slaughtered, and determine the weight estimation deviation information based on the measured weight feedback information and the visual weight estimation result; The current weight estimation parameters are updated based on the weight estimation deviation information, so as to estimate the weight of subsequent batches of animals to be slaughtered based on the updated weight estimation parameters.

[0007] Optionally, in the above scheme, pig target recognition and cross-frame trajectory association processing are performed on the pig passage video stream to obtain pig trajectory information, including: Video frames are extracted from the video stream of the pigs passing through, resulting in multiple video frames to be identified; Pig target detection is performed on each of the video frames to be identified to obtain pig detection information in each of the video frames to be identified; wherein, the pig detection information includes pig detection region and detection confidence level; Based on the pig detection regions and pig target features in adjacent video frames, determine the motion association information of the pig target; Based on the motion association information, the trajectory of pig targets across frames is associated to obtain the trajectory information of each pig target.

[0008] Optionally, in the above scheme, determining the number of pigs to be slaughtered based on the pig trajectory information includes: A preset virtual detection line is determined in the video frame corresponding to the pig slaughtering channel to obtain counting reference information; Based on the pig trajectory information, identify the trajectory information to be matched where the trajectory is interrupted; The trajectory fusion result is determined based on the spatiotemporal matching relationship between the trajectory information to be matched and the historical trajectory information; Based on the trajectory fusion results and the counting reference information, the passage status information corresponding to each pig target is determined; Based on the passage status information, a valid passage trajectory that has completely passed through the preset virtual detection line is determined from the pig trajectory information; The number of pigs to be slaughtered is determined based on the number of valid passage trajectories.

[0009] In the above scheme, optionally, based on the pig trajectory information that meets preset stability conditions, the body size feature information corresponding to each target pig is determined, including: Based on the pig trajectory information, determine the trajectory continuity information and trajectory integrity information corresponding to each pig target; Based on the trajectory continuity information and the trajectory completeness information, a stable pig trajectory that meets the preset stability condition is determined from the pig trajectory information; Based on the stable pig trajectory, the corresponding target measurement image is determined from the pig passage video stream; Contour extraction is performed on the pig region in the target measurement image to obtain pig contour information; Based on the pig outline information, determine the body size feature information corresponding to the target pig.

[0010] Optionally, in the above scheme, based on the body size feature information and the current weight estimation parameters, the individual weight estimation information corresponding to each of the target pigs is determined, including: Obtain the calibration parameters corresponding to the image acquisition device to obtain image scale conversion information; The pig's outline information is scaled according to the image scale conversion information to obtain the actual body size data; Based on the actual body size data, determine the chest circumference and body length characteristics of the target pig; Based on the chest circumference feature information, the body length feature information, and the current weight estimation parameters, the individual weight estimation information corresponding to the target pig is determined.

[0011] Optionally, in the above scheme, based on the information on the number of pigs slaughtered, the visual weight estimation results, and the value assessment reference information, the estimated market value information corresponding to the current slaughter batch is generated, including: Based on the visual weight estimation results, determine the total batch weight and average batch weight information corresponding to the current batch being slaughtered; Based on the average weight of the batch and the pricing rule information, determine the pricing matching information corresponding to the current batch being sold; Based on the pricing matching information, hog price information, and total batch weight information, the expected revenue information corresponding to the current slaughter batch is determined; Based on the cost reference information and the total weight information of the batch, the estimated cost information corresponding to the current batch of animals slaughtered is determined; Based on the projected revenue information and the projected cost information, the estimated listing value information corresponding to the current batch of pigs slaughtered is generated.

[0012] Optionally, in the above scheme, the measured weight feedback information corresponding to the current batch of animals being slaughtered is obtained, and the weight estimation deviation information is determined based on the measured weight feedback information and the visual weight estimation result, including: Obtain the measured total weight information corresponding to the current batch of animals being slaughtered, fed back by the weighing equipment; Based on the measured total weight information and the number of pigs slaughtered, determine the measured average weight information corresponding to the current slaughter batch; Based on the visual weight estimation results, determine the visual average weight information corresponding to the current batch of animals being slaughtered; The weight estimation deviation information is determined based on the measured average weight information and the visual average weight information.

[0013] Optionally, in the above scheme, updating the current weight estimation parameters based on the weight estimation deviation information includes: Based on the weight estimation deviation information, a calibration factor is determined to characterize the degree of deviation between the visual weight estimation result and the measured weight feedback information; The current weight estimation parameters are corrected according to the calibration factor to obtain the updated weight estimation parameters; The updated weight estimation parameters are associated with and stored with the batch identification information corresponding to the current slaughter batch to obtain calibration record information; If subsequent batches of pigs are identified, the updated weight estimation parameters are used to estimate the weight of the pigs in those subsequent batches.

[0014] Optionally, in the above scheme, before performing pig target identification on the pig passage video stream, the method further includes: Image quality analysis is performed on the video stream of the pigs passing through to obtain channel environmental state information; wherein, the channel environmental state information includes at least one of brightness state information and motion blur state information; Based on the brightness status information, the exposure adjustment information corresponding to the image acquisition device is determined; Based on the exposure adjustment information, the image acquisition device is controlled to adjust the image acquisition parameters to obtain a video frame with adjusted brightness. Based on the motion blur state information, candidate measurement frames for body size measurement are determined from the brightness-adjusted video frames; Pig targets are identified based on the brightness-adjusted video frames, and body size features are determined based on the candidate measurement frames.

[0015] Secondly, a device for estimating the market value of pork based on visual processing, the system comprising: Image acquisition device, used to acquire video streams of pigs passing through the pig slaughtering channel; An edge computing terminal, connected to the image acquisition device, includes: The trajectory counting module is used to perform pig target recognition and cross-frame trajectory association processing on the pig passage video stream to obtain pig trajectory information, and determine the number of pigs to be slaughtered based on the pig trajectory information; The visual weight estimation module is used to determine the body size feature information of each target pig based on the pig trajectory information that meets the preset stability conditions, determine the individual weight estimation information of each target pig based on the body size feature information and the current weight estimation parameters, and determine the visual weight estimation result corresponding to the current slaughter batch based on the individual weight estimation information of each target pig. The value estimation module is used to obtain value assessment reference information corresponding to the current slaughter batch, and generate market value estimation information corresponding to the current slaughter batch based on the slaughter pig quantity information, the visual weight estimation result and the value assessment reference information; wherein, the value assessment reference information includes live pig price information, pricing rule information and cost reference information; The parameter calibration module is used to obtain the measured weight feedback information corresponding to the current batch of slaughter, determine the weight estimation deviation information based on the measured weight feedback information and the visual weight estimation result, and update the current weight estimation parameter based on the weight estimation deviation information, so as to estimate the weight of subsequent batches of slaughter based on the updated weight estimation parameter.

[0016] Compared with the prior art, this application has at least the following beneficial effects: This application, based on further analysis and research of existing technical problems, recognizes that existing technologies in hog slaughter management suffer from several issues: first, data acquisition is easily affected by complex operating environments; second, the real-time performance and accuracy of batch weight and quantity assessment are insufficient; third, relevant data cannot promptly support market value judgments; and fourth, the long-term reliability of the system is inadequate. This application addresses these issues by acquiring video streams of pigs passing through the slaughter channel and performing pig target identification and cross-frame trajectory association processing on these streams. This allows pigs during the slaughter process to be continuously represented by their trajectories, reducing counting errors caused by occlusion, pauses, or repeated entry and exit during manual observation and single-frame recognition. Furthermore, based on the pig trajectory information meeting preset stability conditions, the body size characteristics of each target pig are determined, and combined with current weight estimation parameters, individual weight estimation information is obtained. Finally, the visual weight estimation result for the current slaughter batch is formed from the individual weight estimation information. This system enables the simultaneous acquisition of pig quantity and weight information at the slaughter site, improving the real-time nature and accuracy of slaughter batch weight assessment. Furthermore, by acquiring value assessment reference information including hog price information, pricing rule information, and cost reference information, and combining this with slaughter pig quantity information and visual weight estimation results to generate market value prediction information, the system allows the quantity and weight data collected on-site to be directly converted into value data for market judgment, thus addressing the problem of slaughter data failing to support sales decisions in a timely manner. Simultaneously, by acquiring the measured weight feedback information of the current slaughter batch, and determining the weight estimation deviation information based on the measured weight feedback information and visual weight estimation results, the system updates the current weight estimation parameters based on the weight estimation deviation information. This allows subsequent slaughter batches to be weighted based on the updated weight estimation parameters, thereby suppressing the accumulation of weight estimation deviations during long-term operation and improving the stability and reliability of the system for continuous use. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating a vision-based method for estimating the market value of pork, provided in one embodiment of this application. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0019] like Figure 1As shown in the illustration, this application provides a method for estimating the market value of pork based on visual processing. This method can be applied to edge computing terminals. Edge computing terminals can be deployed at the pork slaughtering site, for example, in a dustproof box near the loading channel or loading platform. An image acquisition device can be installed above the pork slaughtering channel to capture video streams of pigs passing through. As a specific example, the image acquisition device can be a global shutter industrial camera, which can be installed approximately 2.5m above the entrance to the loading platform, with a resolution of 1920×1080, a frame rate of 30fps, and IP67 protection capability to adapt to the dust, moisture, and vibration conditions of the farm environment. The edge computing terminal can be an embedded computing device with image inference capabilities, such as an NVIDIA Jetson Orin NX, or other edge devices with neural network inference, video decoding, and data processing capabilities. The edge computing terminal can be connected to and powered by the image acquisition device via PoE, or connected to a local server, cloud platform, or data interface via 4G, WiFi, or wired network.

[0020] After the pig slaughtering operation begins, the image acquisition device continuously captures video footage of the pigs moving along the slaughtering channel. The edge computing terminal then acquires the video stream of pigs moving along the channel. This video stream can consist of consecutive video frames, each containing one or more pigs moving along the loading channel. The edge computing terminal can decode, extract frames, normalize their dimensions, and perform image preprocessing on the video stream to obtain a sequence of video frames suitable for subsequent recognition and processing.

[0021] The edge computing terminal performs pig target recognition and cross-frame trajectory association processing on the video stream of pigs passing through, obtaining pig trajectory information. Pig target recognition can be achieved through a pre-trained target detection model. This model can be a lightweight target detection model; in specific implementations, an improved YOLOv8 detection model can be used, or other target detection models capable of outputting pig detection regions and detection confidence scores can be employed. Cross-frame trajectory association processing can be implemented through a multi-target tracking model. Specifically, the DeepSORT multi-target tracking algorithm can be used, or other multi-target tracking algorithms combining detection regions, target appearance features, and motion state prediction can be employed. Through pig target recognition and cross-frame trajectory association, corresponding trajectory identifiers can be assigned to each continuously moving pig within the channel, thus obtaining pig trajectory information. Pig trajectory information can include trajectory identifiers, trajectory start and end frames, detection region sequence, center point position sequence, movement direction, trajectory duration, and trajectory integrity status. The edge computing terminal determines the number of pigs to be slaughtered based on the pig trajectory information. For example, it only counts the trajectory of pigs that have completely passed through the preset counting position into the number of pigs to be slaughtered, thereby avoiding the repeated counting of the same pig when it stops, retreats, or re-enters the screen in the channel.

[0022] After obtaining the pig trajectory information, the edge computing terminal filters trajectories that meet preset stability conditions. These preset stability conditions may include a trajectory duration reaching a preset frame threshold, continuity of the trajectory detection area meeting preset requirements, no severe occlusion of the pig target, and the pig's posture being suitable for body size measurement. As an example, a pig trajectory appearing continuously for at least 5 frames can be identified as meeting the preset stability conditions. Based on the pig trajectory information meeting the preset stability conditions, the edge computing terminal determines the body size feature information corresponding to each target pig. Body size feature information may include one or more of the following: chest circumference feature information, body length feature information, back contour feature information, width feature information, and length feature information. In a specific implementation, a target measurement image can be selected from the video frames corresponding to the stable trajectory. Contour extraction, ellipse fitting, or geometric fitting processing can be performed on the pig region in the target measurement image to obtain the chest circumference and body length feature information of the target pig.

[0023] The edge computing terminal determines the individual weight estimate for each target pig based on body size characteristics and current weight estimation parameters. The current weight estimation parameters can be calibration constants in the weight estimation formula or model parameters in the visual weight estimation model. As a specific example, the individual weight estimate W can be calculated based on chest circumference (HG), body length (BL), and the current weight estimation parameter K, expressed as: W = (HG² × BL) / K. The initial value of the current weight estimation parameter K can be determined based on historical samples, farm-specific measurements, or empirical data; for example, the initial K could be 12000. Based on the individual weight estimate for each target pig, the edge computing terminal determines the visual weight estimate for the current slaughter batch. The visual weight estimate result can include total batch weight, average batch weight, individual weight estimate, estimated quantity, and estimation time.

[0024] Edge computing terminals acquire value assessment reference information corresponding to the current batch of pigs being sold. This reference information includes pig price information, pricing rules, and cost reference information. Pig price information can be obtained from local cache, farm management systems, remote API interfaces, or cloud platforms. Pricing rules can include segmented pricing rules based on weight ranges, such as a premium price for 110kg to 125kg. Cost reference information can include unit weight breeding costs, batch fixed costs, transportation costs, or other cost data related to market value assessment. For example, cost reference information could include a preset breeding cost of 14.5 yuan / kg, and pig price information could include a price of 18.2 yuan / kg corresponding to a specific weight range.

[0025] The edge computing terminal generates estimated market value information for the current batch of pigs slaughtered based on the number of pigs slaughtered, visual weight estimation results, and value assessment reference information. This estimated market value information may include the number of pigs slaughtered, total batch weight, average batch weight, applicable price, projected revenue, projected cost, projected gross profit, profit margin, and profit / loss indications. The edge computing terminal can output the estimated market value information to a web interface, LED display screen, or mobile terminal page. It can also support users scanning a QR code to export PDF or Excel files, thus creating a valuation file for the current batch of pigs slaughtered.

[0026] After the current batch of animals is weighed, the edge computing terminal obtains the actual weight feedback information corresponding to that batch. This feedback information can come from a weighbridge, weighing system, farm management system, or manual input interface. Based on the actual weight feedback and visual weight estimation results, the edge computing terminal determines the weight estimation deviation information. This deviation can include the difference between the actual total weight and the visually estimated total weight, the difference between the actual average weight and the visual average weight, relative error, or a calibration factor. The edge computing terminal updates the current weight estimation parameters based on the deviation information, using the updated parameters to estimate the weight of subsequent batches. For example, in a batch where the visually estimated average weight is 122.0 kg and the weighbridge-measured average weight is 118.7 kg, the calibration factor α can be determined as 118.7 / 122.0 ≈ 0.973. The current weight estimation parameter K is then corrected based on this calibration factor, for example, by updating K to K / α², so that the updated weight estimation parameters are used for the next batch. In addition to formula parameter correction, the visual weight estimation model can also be updated through lightweight model fine-tuning, such as online fine-tuning of some model parameters, to reduce weight estimation deviation in long-term operation.

[0027] Through the above implementation method, the edge computing terminal can complete video acquisition, target recognition, trajectory counting, body size estimation, batch value prediction, and actual measurement feedback calibration at the pig slaughter site. This allows slaughter data to be generated in real time and directly used for market value judgment. Simultaneously, updating the current weight estimation parameters through actual weight feedback information can suppress long-term accuracy drift caused by camera installation status, differences in pig body size, changes in ambient lighting, and site variations. Therefore, this implementation method can improve the real-time performance and accuracy of pig slaughter quantity and weight assessment, enhance the timeliness of market value judgment, and improve the long-term reliability of the system.

[0028] In some embodiments, performing pig target recognition and cross-frame trajectory association processing on the pig passage video stream to obtain pig trajectory information may include the following process: The edge computing terminal first extracts video frames from the pig passage video stream to obtain multiple video frames to be identified. Video frame extraction can be performed at a fixed frame rate, or adaptive frame extraction can be performed based on the computing resources of the edge computing terminal, the pig passage speed, or the degree of image change. For example, when computing resources are sufficient, the 30fps video stream can be processed frame by frame, while when the image change in the channel is small, frame extraction can be performed at intervals to reduce the computing load.

[0029] The edge computing terminal performs pig target detection on each video frame to be identified, obtaining pig detection information for each frame. This information includes the pig detection region and detection confidence score, and may also include target category, target center point, detection box size, and target appearance features. Pig target detection can employ a lightweight detection model deployed on the edge computing terminal. This model can be trained based on sample images of pigs leaving the pigs' enclosure, enabling it to adapt to conditions such as dense pig populations, occlusion, rapid movement, and changing lighting in the loading lane.

[0030] The edge computing terminal determines the motion association information of pig targets based on the pig detection regions and pig target features in adjacent video frames. Pig target features can include the detection region location, detection region size, target appearance features, center point movement direction, velocity prediction information, and historical trajectory status. Motion association information represents the correspondence between pig targets in the current video frame and pig targets in the previous or historical video frames. The edge computing terminal can comprehensively determine motion association information by considering factors such as target center distance, detection region overlap, consistency of movement direction, and similarity of appearance features.

[0031] The edge computing terminal performs trajectory association on pig targets across frames based on motion correlation information, obtaining pig trajectory information corresponding to each pig target. As a specific implementation, the DeepSORT multi-target tracking algorithm can be used to associate pig detection information in each video frame, combining Kalman filter prediction and feature matching to generate a unique trajectory identifier for each pig target. Pig trajectory information can be established when a pig enters the frame and terminates when a pig leaves the frame or crosses a preset virtual detection line.

[0032] Through the above implementation method, this solution transforms the frame-by-frame pig detection results into a continuous trajectory across frames, which can avoid the problems of repeated statistics, missed statistics and target confusion caused by relying solely on single-frame detection for counting and weight estimation, thereby providing a stable data foundation for subsequent determination of the number of pigs to be slaughtered and extraction of body size features.

[0033] In some embodiments, determining the number of pigs slaughtered based on pig trajectory information may include the following process: An edge computing terminal determines a preset virtual detection line in the video frame corresponding to the pig slaughtering channel to obtain counting reference information. The preset virtual detection line may be set in the exit area of ​​the loading channel, the entrance area of ​​the loading platform, or other locations that indicate the completion of pig slaughtering. The counting reference information may include the position, direction, effective detection area, and passage direction of the preset virtual detection line.

[0034] The edge computing terminal identifies trajectory information to be matched due to interruptions in the pig's trajectory based on the pig's trajectory information. In the pig loading channel, multiple pigs may pass closely together and block each other, causing a pig's trajectory marker to be temporarily lost. The edge computing terminal can identify trajectory interruptions that may be caused by occlusion based on the trajectory end position, trajectory end time, target movement direction, and the starting position of the subsequent newly created trajectory, and obtain the trajectory information to be matched.

[0035] The edge computing terminal determines the trajectory fusion result based on the spatiotemporal matching relationship between the trajectory information to be matched and historical trajectory information. The spatiotemporal matching relationship can include trajectory interruption time, distance between positions before and after trajectory interruption, consistency of movement direction, and similarity of target appearance. As an example, if the trajectory of a pig is lost due to occlusion for no more than 2 seconds, and the newly appearing trajectory meets the matching conditions with the historical trajectory in terms of spatial location and movement direction, then the newly appearing trajectory and the historical trajectory can be merged into the trajectory of the same pig, resulting in a trajectory fusion result. Trajectory fusion avoids duplicate counting caused by assigning multiple trajectory identifiers to the same pig due to short-term occlusion.

[0036] The edge computing terminal determines the passage status information corresponding to each pig target based on the trajectory fusion results and counting benchmark information. Passage status information can include states such as not passing, passing, fully passed, and reverse retreat. Based on the passage status information, the edge computing terminal identifies valid passage trajectories that have fully passed the preset virtual detection line from the pig trajectory information, and determines the number of pigs to be slaughtered based on the number of valid passage trajectories. Trajectories that have not fully passed the preset virtual detection line, trajectories that repeatedly enter and exit but have not formed a valid slaughter passage, or trajectories belonging to the same pig after fusion with historical trajectories, are not counted repeatedly in the number of pigs to be slaughtered.

[0037] Through the above implementation methods, this solution can achieve trajectory fusion and deduplication counting under conditions of dense pig passage, short-term obstruction, and repeated entry and exit from the passage, thereby reducing missed and duplicate counts and improving the accuracy of information on the number of pigs slaughtered.

[0038] In some embodiments, determining the body size feature information corresponding to each target pig based on pig trajectory information that meets preset stability conditions may include the following process: The edge computing terminal determines the trajectory persistence information and trajectory integrity information corresponding to each target pig based on the pig trajectory information. The trajectory persistence information may include the trajectory persistence frame number, trajectory duration, and number of consecutive detections; the trajectory integrity information may include the target occlusion degree, detection box integrity, trajectory interruption status, and whether the target is within a preset measurement area.

[0039] The edge computing terminal determines stable pig trajectories that meet preset stability conditions from the pig trajectory information based on trajectory continuity and trajectory integrity information. Preset stability conditions may include a trajectory duration of no less than a preset number of frames, no severe truncation in the target detection area, the pig target being located in the middle of the measurement area of ​​the channel, and the target pose being suitable for acquiring body size data. As an example, a trajectory with continuous detection of no less than 5 frames and a complete detection area can be determined as a stable pig trajectory.

[0040] The edge computing terminal determines the corresponding target measurement image from the pig's passage video stream based on the stable pig trajectory. The target measurement image can be a single frame from the stable pig trajectory, or a frame with higher clarity, less occlusion, and more stable posture among multiple candidate frames. In cases where high-speed movement causes motion blur, the edge computing terminal can prioritize frames with lower movement speed or relatively stationary positions as the target measurement image.

[0041] The edge computing terminal extracts the contour of the pig region in the target measurement image to obtain the pig's contour information. Contour extraction may include target region segmentation, edge detection, background removal, contour smoothing, and outlier removal. Based on the pig's contour information, the edge computing terminal determines the corresponding body size features of the target pig. As a specific implementation, ellipse fitting can be performed on the pig's contour information, and body length and chest circumference features can be estimated based on the changes in the major axis, minor axis, and contour width of the fitted ellipse.

[0042] Through the above implementation method, this solution does not directly estimate the weight of all detected pig images. Instead, it first selects stable pig trajectories and determines the target measurement image from them, and then performs contour extraction and body size feature determination. This can reduce the impact of occlusion, abnormal posture and motion blur on body size measurement and improve the reliability of subsequent individual weight estimation information.

[0043] In some embodiments, determining the individual weight estimate of each target pig based on body size feature information and current weight estimate parameters may include the following process: An edge computing terminal acquires the calibration parameters corresponding to the image acquisition device to obtain image scale conversion information. The calibration parameters may include the installation height, installation angle, focal length, distortion parameters, field of view, channel ground calibration points, and pixel scale conversion relationships of the image acquisition device. The image scale conversion information is used to convert the pixel dimensions in the target measurement image into actual spatial dimensions.

[0044] The edge computing terminal performs scale transformation on the pig's outline information based on image scale transformation information to obtain actual body size data. This actual body size data may include the pig's actual length in the walking direction, its actual width perpendicular to the walking direction, the length of the principal axis of the outline, the width of the minor axis of the outline, and the projected size of the target area. Based on this actual body size data, the edge computing terminal determines the chest circumference and body length features of the target pig. The chest circumference feature can be determined based on the pig's outline width, minor axis, and a preset chest circumference mapping relationship; the body length feature can be determined based on the pig's outline length, major axis, and a preset body length mapping relationship.

[0045] The edge computing terminal determines the individual weight estimate of the target pig based on chest circumference, body length, and current weight estimation parameters. As a specific example, chest circumference can be denoted as HG, body length as BL, and the current weight estimation parameter as K. The individual weight estimate W can then be determined using the formula W = (HG² × BL) / K. The current weight estimation parameter K can be set and updated based on different farms, pig breeds, camera installation conditions, or historical calibration data.

[0046] Through the above implementation method, this solution uses the calibration parameters of the image acquisition device to convert the image contour into actual body size data, and determines the individual body weight estimation information based on chest circumference feature information, body length feature information and current weight estimation parameters. This makes the visual weight estimation process have a clear data conversion chain, which can reduce the scale uncertainty caused by direct weight estimation of monocular images and improve the interpretability and accuracy of the weight estimation results.

[0047] In some embodiments, generating estimated market value information for the current slaughter batch based on the number of pigs slaughtered, visual weight estimation results, and value assessment reference information may include the following process: The edge computing terminal determines the total batch weight information and average batch weight information for the current slaughter batch based on the visual weight estimation results. The total batch weight information can be obtained by summing the individual weight estimation information for each target pig, and the average batch weight information can be calculated from the total batch weight information and the number of pigs slaughtered.

[0048] The edge computing terminal determines the pricing matching information corresponding to the current batch based on the average batch weight information and pricing rules. Pricing rules can include price levels, base prices, premium prices, discounted prices, and deduction rules for different weight ranges. For example, the pricing rules can define 110kg to 125kg as the premium weight range. If the average batch weight falls within this range, the current batch is matched with the premium price; if the average batch weight is lower or higher than this range, the corresponding ordinary price or deducted price is matched.

[0049] The edge computing terminal determines the expected revenue for the current batch of pigs to be sold based on pricing matching information, pig price information, and total batch weight information. Pig price information can be the price of the day, a user-defined transaction price, a contract price, or a real-time price from a remote price interface. The edge computing terminal also determines the expected cost for the current batch of pigs to be sold based on cost reference information and total batch weight information. Cost reference information can include the unit weight breeding cost, such as 14.5 yuan / kg, or one or more of the following: feeding costs, medicine costs, transportation costs, and management costs.

[0050] The edge computing terminal generates estimated market value information for the current batch of pigs to be sold based on projected revenue and cost information. This estimated market value information can include projected revenue, projected costs, total gross profit, profit margin, average weight, total weight, quantity, price range, and profit / loss alerts. For example, the system can calculate total gross profit based on the batch's total weight, the current day's live pig price, and the cost per unit weight of pigs raised, and generate a profit margin based on the ratio of total gross profit to projected revenue. The estimated market value information can be updated in real time during the pig loading process and displayed via a web interface or LED screen, or exported as PDF or Excel format after the batch is completed.

[0051] Through the above implementation methods, this solution links the visual weight estimation results with pig price information, pricing rules information, and cost reference information. It not only outputs weight data but also generates operating results such as expected revenue, expected costs, gross profit, and profit margin, enabling the data collected at the slaughter site to support timely market value judgments and sales decisions.

[0052] In some embodiments, obtaining the measured weight feedback information corresponding to the current batch of livestock being slaughtered, and determining the weight estimation deviation information based on the measured weight feedback information and the visual weight estimation result, may include the following process: The edge computing terminal obtains the measured total weight information corresponding to the current batch of livestock being slaughtered, fed back by the weighing equipment. The weighing equipment can be a weighbridge, an electronic weighing platform, or an existing weighing system in the farm. The measured total weight information can be automatically transmitted to the edge computing terminal by the weighing equipment, or it can be entered by the operator after confirmation through an APP, a web interface, or a local input device.

[0053] The edge computing terminal determines the measured average weight of the current batch based on the measured total weight and the number of pigs slaughtered. It also determines the visual average weight of the current batch based on visual weight estimation. The visual average weight can be calculated from the total batch weight and the number of pigs slaughtered, or it can be obtained by averaging the individual weight estimations for each target pig.

[0054] The edge computing terminal determines the weight estimation deviation based on the measured average weight information and the visual average weight information. The weight estimation deviation can be the difference between the measured average weight information and the visual average weight information, or it can be the ratio or relative error between the two. For example, if the visual average weight is higher than the measured average weight, the weight estimation deviation information can indicate that the current weight estimation result is too high; if the visual average weight is lower than the measured average weight, the weight estimation deviation information can indicate that the current weight estimation result is too low.

[0055] Through the above implementation method, this solution can use the actual weight feedback information of the current batch of animals slaughtered to objectively evaluate the visual weight estimation results, and form weight estimation deviation information that can be used for subsequent parameter updates, thereby providing a data foundation for the long-term accuracy maintenance of the system.

[0056] In some embodiments, updating the current weight estimation parameters based on the weight estimation deviation information may include the following process: The edge computing terminal determines a calibration factor to characterize the degree of deviation between the visual weight estimation result and the measured weight feedback information based on the weight estimation deviation information. The calibration factor can be determined based on the ratio between the measured average weight information and the visual average weight information, or it can be determined based on the ratio between the measured total weight information and the visual estimated total weight.

[0057] The edge computing terminal corrects the current weight estimation parameters based on a calibration factor, resulting in updated weight estimation parameters. As a specific example, if the visually estimated average weight is 122.0 kg and the actual measured average weight is 118.7 kg, the calibration factor α can be calculated as α = 118.7 / 122.0 ≈ 0.973. The current weight estimation parameter K is then corrected based on this calibration factor, for example, updating K to K / α². The updated weight estimation parameters can be immediately used for visual weight estimation of the next batch. In addition to correcting formula parameters, the edge computing terminal can also perform lightweight model fine-tuning of the visual weight estimation model based on weight estimation deviation information, such as updating some model parameters or adaptation layer parameters, thereby achieving online calibration.

[0058] The edge computing terminal associates and stores the updated weight estimation parameters with the batch identification information corresponding to the current slaughter batch, obtaining calibration record information. Batch identification information may include the slaughter batch number, date, pig house number, passage number, camera number, breed information, and operator confirmation information. Calibration record information can be stored on the edge computing terminal or encrypted and uploaded to a local server or cloud platform for subsequent traceability, cross-farm model aggregation, or remote upgrades. When the edge computing terminal identifies a subsequent slaughter batch, it uses the updated weight estimation parameters to estimate the weight of the pigs in that batch.

[0059] Through the above implementation methods, this solution can automatically correct the current weight estimation parameters based on actual measurement feedback, realize online accuracy maintenance without stopping the machine or retraining, reduce weight estimation accuracy drift caused by environmental changes, equipment installation offset or differences in pig body size during long-term operation, and improve the stability of the system throughout its entire life cycle.

[0060] In some embodiments, before performing pig target recognition on the pig passage video stream, the edge computing terminal can also perform image quality analysis on the pig passage video stream to obtain channel environmental state information. The channel environmental state information includes at least one of brightness state information and motion blur state information. Brightness state information can be determined by statistically analyzing the average brightness, local brightness, contrast, or exposure distribution of video frames; motion blur state information can be determined by the inter-frame pixel change rate, target center point velocity, edge sharpness, or blur evaluation metrics.

[0061] The edge computing terminal determines the exposure adjustment information corresponding to the image acquisition device based on the brightness status information. For example, when the edge computing terminal detects that the ambient light is below a preset threshold, it can determine that the exposure time or gain parameter needs to be increased. As an example, the preset threshold can be 50 lux. When the ambient light in the brightness status information characterizing channel is below 50 lux, the edge computing terminal can generate exposure adjustment information and control the image acquisition device to adjust the image acquisition parameters according to the exposure adjustment information, obtaining a video frame with adjusted brightness. Through this processing, the output image can have sufficient contrast to support subsequent pig target recognition and contour extraction.

[0062] The edge computing terminal determines candidate measurement frames for body size measurement from the brightness-adjusted video frames based on motion blur status information. For example, the edge computing terminal can determine the pig's movement speed based on the pixel change rate between consecutive frames. When the pig's movement speed exceeds a preset speed threshold, it is determined that the current frame has a risk of motion blur. As an example, the preset speed threshold can be 1.5 m / s. When the motion blur status information indicates that the pig's movement speed is high, the edge computing terminal can perform sharpening filtering on the video frame and prioritize relatively static or high-resolution frames in the trajectory as candidate measurement frames.

[0063] The edge computing terminal performs pig target recognition based on brightness-adjusted video frames and determines body size features based on candidate measurement frames. In other words, the video frames used for target recognition can be brightness-adjusted to improve detection stability, and the images used for body size measurement can be selected from candidate measurement frames to reduce the impact of motion blur on chest circumference and body length feature information.

[0064] Through the above implementation methods, this solution can perform pre-emptive quality control for complex working conditions such as unstable lighting, rapid movement of pigs, and blurred images in the pig loading channel, improve the stability of target detection, trajectory association, and body size measurement, thereby improving the reliability of information on the number of pigs at slaughter and visual weight estimation results.

[0065] This embodiment also provides a vision-based system for estimating the market value of market-ready pigs. The system includes an image acquisition device and an edge computing terminal. The image acquisition device is used to acquire video streams of pigs moving through the pig loading channel. The image acquisition device can be installed above the loading channel or at other locations that can completely cover the pigs' passage area. The image acquisition device can be an industrial camera, preferably a global shutter industrial camera, to reduce image distortion caused by rapid movement. The image acquisition device can have a dustproof, waterproof, and vibration-resistant structure to adapt to the farm environment.

[0066] The edge computing terminal connects to the image acquisition device. The edge computing terminal can be housed in a stainless steel dustproof enclosure and connected to the image acquisition device via PoE. The edge computing terminal can connect to a local server or cloud platform via 4G, WiFi, or wired network to obtain hog price information, upload calibration records and valuation reports, or receive remote upgrade data. The edge computing terminal includes a trajectory counting module, a visual weight estimation module, a value prediction module, and a parameter calibration module.

[0067] The trajectory counting module is used to perform pig target recognition and cross-frame trajectory association processing on the pig passage video stream to obtain pig trajectory information, and determine the number of pigs to be slaughtered based on the pig trajectory information. The trajectory counting module can deploy target detection models and multi-target tracking models, such as an improved YOLOv8 detection model and a DeepSORT tracking model, to achieve pig detection, tracking, trajectory reconnection, and deduplication counting in densely occluded scenes.

[0068] The visual weight estimation module determines the body size characteristics of each target pig based on pig trajectory information that meets preset stability conditions. Based on these body size characteristics and current weight estimation parameters, it determines the individual weight estimation of each target pig and, based on this individual weight estimation, determines the visual weight estimation result for the current slaughter batch. The module performs processes such as contour extraction, ellipse fitting, scale conversion, and weight estimation formula calculation to determine the total batch weight and average batch weight.

[0069] The value estimation module retrieves value assessment reference information corresponding to the current batch of pigs being slaughtered. Based on the number of pigs slaughtered, visual weight estimation results, and value assessment reference information, it generates estimated market value information for the current batch. The value assessment reference information includes hog price information, pricing rules, and cost reference information. The value estimation module can retrieve daily hog price information from local cache or a remote API, can call user-preset weight-based pricing rules, and can calculate expected revenue, expected costs, total gross profit, and profit margin based on preset breeding costs. The value estimation module can also output the estimated market value information to a web interface, LED display screen, or mobile page, and supports exporting PDF or Excel files via QR code scanning.

[0070] The parameter calibration module acquires the measured weight feedback information corresponding to the current batch of animals being slaughtered. Based on the measured weight feedback information and the visual weight estimation result, it determines the weight estimation deviation information and updates the current weight estimation parameters according to the weight estimation deviation information. This updated weight estimation parameters are then used to estimate the weight of subsequent batches being slaughtered. The measured weight feedback information can come from a weighbridge or other weighing equipment. The parameter calibration module can generate calibration record information from the updated weight estimation parameters, weight estimation deviation information, and batch identification information. This calibration record information is then encrypted and uploaded to the cloud platform for cross-field model aggregation and OTA upgrades.

[0071] Through the above implementation methods, this system can achieve end-side video acquisition, edge-side visual reasoning, trajectory counting, visual weight estimation, market value prediction, and parameter self-calibration at the pig slaughter site. The system completes the core reasoning tasks on the edge computing terminal, reducing reliance on cloud networks and improving real-time performance. By combining the visual weight estimation results with pig price information, pricing rules, and cost reference information, it can output market value prediction information with operational decision-making value. The parameter calibration module uses measured weight feedback information to update the current weight estimation parameters, suppressing long-term accuracy drift and improving the system's long-term stability and data traceability.

[0072] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

Claims

1. A method for estimating the market value of pork based on visual processing, characterized in that, Applied to edge computing terminals, the method includes: Acquire video streams of pigs passing through an image acquisition device set up in the pig slaughtering channel; The pig target recognition and cross-frame trajectory association processing are performed on the pig passage video stream to obtain pig trajectory information, and the number of pigs to be slaughtered is determined based on the pig trajectory information; Based on the pig trajectory information that meets the preset stability conditions, determine the body size feature information corresponding to each target pig, and based on the body size feature information and the current weight estimation parameters, determine the individual weight estimation information corresponding to each target pig. Based on the individual weight estimation information of each target pig, determine the visual weight estimation result corresponding to the current batch of pigs to be slaughtered; Obtain the value assessment reference information corresponding to the current batch of pigs being slaughtered, the value assessment reference information including pig price information, pricing rule information and cost reference information; Based on the information on the number of pigs slaughtered, the visual weight estimation results, and the value assessment reference information, the estimated market value information corresponding to the current slaughter batch is generated. Obtain the measured weight feedback information corresponding to the current batch of animals slaughtered, and determine the weight estimation deviation information based on the measured weight feedback information and the visual weight estimation result; The current weight estimation parameters are updated based on the weight estimation deviation information, so as to estimate the weight of subsequent batches of animals to be slaughtered based on the updated weight estimation parameters.

2. The method for estimating the market value of pork based on visual processing according to claim 1, characterized in that, The pig target recognition and cross-frame trajectory association processing are performed on the pig passage video stream to obtain pig trajectory information, including: Video frames are extracted from the video stream of the pigs passing through, resulting in multiple video frames to be identified; Pig target detection is performed on each of the video frames to be identified to obtain pig detection information in each of the video frames to be identified; wherein, the pig detection information includes pig detection region and detection confidence level; Based on the pig detection regions and pig target features in adjacent video frames, determine the motion association information of the pig target; Based on the motion association information, the trajectory of pig targets across frames is associated to obtain the trajectory information of each pig target.

3. The method for estimating the market value of pork based on visual processing according to claim 2, characterized in that, The number of pigs to be slaughtered is determined based on the pig trajectory information, including: A preset virtual detection line is determined in the video frame corresponding to the pig slaughtering channel to obtain counting reference information; Based on the pig trajectory information, identify the trajectory information to be matched where the trajectory is interrupted; The trajectory fusion result is determined based on the spatiotemporal matching relationship between the trajectory information to be matched and the historical trajectory information; Based on the trajectory fusion results and the counting reference information, the passage status information corresponding to each pig target is determined; Based on the passage status information, a valid passage trajectory that has completely passed through the preset virtual detection line is determined from the pig trajectory information; The number of pigs to be slaughtered is determined based on the number of valid passage trajectories.

4. The method for estimating the market value of pork based on visual processing according to claim 2, characterized in that, Based on the pig trajectory information that meets the preset stability conditions, determine the body size feature information corresponding to each target pig, including: Based on the pig trajectory information, determine the trajectory continuity information and trajectory integrity information corresponding to each pig target; Based on the trajectory continuity information and the trajectory completeness information, a stable pig trajectory that meets the preset stability condition is determined from the pig trajectory information; Based on the stable pig trajectory, the corresponding target measurement image is determined from the pig passage video stream; Contour extraction is performed on the pig region in the target measurement image to obtain pig contour information; Based on the pig outline information, determine the body size feature information corresponding to the target pig.

5. The method for estimating the market value of pork based on visual processing according to claim 4, characterized in that, Based on the body size feature information and the current weight estimation parameters, determine the individual weight estimation information for each of the target pigs, including: Obtain the calibration parameters corresponding to the image acquisition device to obtain image scale conversion information; The pig's outline information is scaled according to the image scale conversion information to obtain the actual body size data; Based on the actual body size data, determine the chest circumference and body length characteristics of the target pig; Based on the chest circumference feature information, the body length feature information, and the current weight estimation parameters, the individual weight estimation information corresponding to the target pig is determined.

6. The method for estimating the market value of pork based on visual processing according to claim 1, characterized in that, Based on the information on the number of pigs slaughtered, the visual weight estimation results, and the value assessment reference information, the estimated market value information corresponding to the current slaughter batch is generated, including: Based on the visual weight estimation results, determine the total batch weight and average batch weight information corresponding to the current batch being slaughtered; Based on the average weight of the batch and the pricing rule information, determine the pricing matching information corresponding to the current batch being sold; Based on the pricing matching information, hog price information, and total batch weight information, the expected revenue information corresponding to the current slaughter batch is determined; Based on the cost reference information and the total weight information of the batch, the estimated cost information corresponding to the current batch of animals slaughtered is determined; Based on the projected revenue information and the projected cost information, the estimated listing value information corresponding to the current batch of pigs slaughtered is generated.

7. The method for estimating the market value of pork based on visual processing according to claim 1, characterized in that, Obtain the measured weight feedback information corresponding to the current batch of animals slaughtered, and determine the weight estimation deviation information based on the measured weight feedback information and the visual weight estimation result, including: Obtain the measured total weight information corresponding to the current batch of animals being slaughtered, fed back by the weighing equipment; Based on the measured total weight information and the number of pigs slaughtered, determine the measured average weight information corresponding to the current slaughter batch; Based on the visual weight estimation results, determine the visual average weight information corresponding to the current batch of animals being slaughtered; The weight estimation deviation information is determined based on the measured average weight information and the visual average weight information.

8. The method for estimating the market value of pork based on visual processing according to claim 7, characterized in that, Updating the current weight estimation parameters based on the weight estimation deviation information includes: Based on the weight estimation deviation information, a calibration factor is determined to characterize the degree of deviation between the visual weight estimation result and the measured weight feedback information; The current weight estimation parameters are corrected according to the calibration factor to obtain the updated weight estimation parameters; The updated weight estimation parameters are associated with and stored with the batch identification information corresponding to the current slaughter batch to obtain calibration record information; If subsequent batches of pigs are identified, the updated weight estimation parameters are used to estimate the weight of the pigs in those subsequent batches.

9. The method for estimating the market value of pork based on visual processing according to claim 2, characterized in that, Before performing pig target identification on the pig passage video stream, the method further includes: Image quality analysis is performed on the video stream of the pigs passing through to obtain channel environmental state information; wherein, the channel environmental state information includes at least one of brightness state information and motion blur state information; Based on the brightness status information, the exposure adjustment information corresponding to the image acquisition device is determined; Based on the exposure adjustment information, the image acquisition device is controlled to adjust the image acquisition parameters to obtain a video frame with adjusted brightness. Based on the motion blur state information, candidate measurement frames for body size measurement are determined from the brightness-adjusted video frames; Pig targets are identified based on the brightness-adjusted video frames, and body size features are determined based on the candidate measurement frames.

10. A system for predicting the market value of pork based on visual processing, characterized in that, The system includes: Image acquisition device, used to acquire video streams of pigs passing through the pig slaughtering channel; An edge computing terminal, connected to the image acquisition device, includes: The trajectory counting module is used to perform pig target recognition and cross-frame trajectory association processing on the pig passage video stream to obtain pig trajectory information, and determine the number of pigs to be slaughtered based on the pig trajectory information; The visual weight estimation module is used to determine the body size feature information of each target pig based on the pig trajectory information that meets the preset stability conditions, determine the individual weight estimation information of each target pig based on the body size feature information and the current weight estimation parameters, and determine the visual weight estimation result corresponding to the current slaughter batch based on the individual weight estimation information of each target pig. The value estimation module is used to obtain value assessment reference information corresponding to the current slaughter batch, and generate market value estimation information corresponding to the current slaughter batch based on the slaughter pig quantity information, the visual weight estimation result and the value assessment reference information; wherein, the value assessment reference information includes live pig price information, pricing rule information and cost reference information; The parameter calibration module is used to obtain the measured weight feedback information corresponding to the current batch of slaughter, determine the weight estimation deviation information based on the measured weight feedback information and the visual weight estimation result, and update the current weight estimation parameter based on the weight estimation deviation information, so as to estimate the weight of subsequent batches of slaughter based on the updated weight estimation parameter.

Images