Disease monitoring and preventing system for goose breeding based on big data

Abstract

The invention discloses a disease monitoring and preventing system for goose breeding based on big data, and relates to the technical field of poultry breeding monitoring. The system comprises a sensing front end, a storage end and a big data processing end, the big data end comprises an individual health monitoring module, a group behavior recognition module and a comprehensive evaluation patrol module; individual health monitoring and group behavior recognition are carried out on the goose house, and comprehensive analysis is carried out by combining output results of the individual health monitoring and the group behavior recognition to plan a goose house inspection route, so that omnibearing perception and intelligent analysis of the environment, physiology and behaviors of the goose house are realized, weak physiological or behavior abnormalities can be captured in the early stage of a disease; and the operation and maintenance cost and the computing power occupation are obviously reduced through dynamic inspection route optimization and intermittent sampling, so that the epidemic disease prevention and control efficiency and the production safety of the goose farm are comprehensively improved.

Description

Technical Field

[0001] The present invention relates to the technical field of poultry breeding monitoring, and in particular to a disease monitoring and prevention system for goose breeding based on big data. Background Art

[0002] With the large-scale development of the breeding industry and the large demand for poultry in the market, goose breeding has gradually adopted modern breeding methods to meet market demand. The environmental parameters of the geese at each stage are monitored throughout the process and the deviation of environmental conditions is timely adjusted. For example, if the ammonia concentration is too high, the ventilation is turned on; if the temperature is too low, the heating lamp is turned on, etc., to provide the optimal conditions required for the geese at each stage (brooding period, rearing period, fattening period), thereby effectively improving the survival rate and growth rate; however, the judgment of diseases in the goose breeding process is mainly through the staff observing the appearance of the geese and manually measuring the body temperature during the patrol. This method requires a relatively large waste of manpower, and it is believed that the judgment standards are inconsistent and easy to miss. Summary of the Invention

[0003] Based on this, it is necessary to provide a disease monitoring and prevention system for goose farming based on big data to address the problems mentioned in the above background technology.

[0004] The purpose of the present invention can be achieved by the following technical solutions: a disease monitoring and prevention system for goose farming based on big data, including a sensing front end and a storage end, the sensing front end is used to communicate with various sensors in the goose house to collect individual physiological information and group behavior characteristic information of the goose house, and the big data processing end has a built-in group behavior recognition module, an individual health monitoring module, and a comprehensive assessment and patrol module; The group behavior recognition module performs group recognition based on high-definition group images of the goose house to determine the growth stage of the goose house and output it. It then recognizes the appearance of the goose house according to the growth stage to output an appearance score, and regularly updates the appearance score of the goose house. It recognizes and analyzes the group behavior of the goose house according to the growth stage and outputs a behavior score. If the behavior score is less than the behavior warning threshold, an interference warning is directly triggered. Otherwise, it performs posture estimation and recognition analysis to output a posture score. If the posture score is less than the posture warning threshold, an interference warning is directly triggered. Otherwise, the goose house number and its corresponding appearance score, behavior score, and posture score are sent to the comprehensive evaluation and patrol module. The individual health monitoring module performs physiological health analysis based on the individual physiological information of the geese sampled in the goose house, and outputs the health score of each sampled goose within a fixed period of time. If the health score is less than the health warning threshold, the sampled goose is recorded as a physiological abnormality warning goose, and the number of physiological abnormality warning geese in the goose house is counted. If there is one or more, an intervention warning is directly triggered; otherwise, the health scores of each sampled goose are averaged to obtain the physiological health score of the goose house, and the score is sent to the comprehensive assessment and patrol module; The comprehensive assessment inspection module compares the goose house numbers received from the group behavior recognition module and the individual health monitoring module, filters out the goose house numbers that exist at the same time, and classifies them into the goose houses to be inspected; then it plans the route based on the goose houses to be inspected, outputs the final inspection route, and sends it to the patrol personnel.

[0005] In some embodiments, the growth stage determines the output process: Extract high-definition group images and perform image preprocessing on them. Use a lightweight object detection network to detect the goose body in each frame of the preprocessed high-definition group images, and output the rectangular detection frame of each goose and its corresponding confidence. Only rectangular detection frames with a confidence level ≥ 0.5 are retained. Perform hand-eye calibration on the high-definition camera in advance to obtain the intrinsic parameter matrix K and distortion parameters. Use a floor calibration plate to calculate the relationship between the image coordinates and the ground truth distance, and obtain the linear mapping relationship between pixel width and true width. For each rectangular detection frame, take its width as the equivalent pixel length on the ground plane and convert it into The front width of the goose is used as the body width of the goose. The body widths of all geese detected in the current frame are summarized, and the average body width and standard deviation are calculated. The system is set with body width intervals corresponding to different stages by default, which can identify the growth stage of each frame image. The growth stage of several consecutive frames is counted, and the growth stage with the largest number of frames is selected as the current growth stage of the goose house. The corresponding breeding environment parameters are matched according to the current growth stage. The specific breeding environment parameters include breeding temperature, breeding humidity, ammonia concentration upper limit, and carbon dioxide concentration upper limit. The environmental conditions of the goose house are controlled based on this breeding environment.

[0006] In some embodiments, the process of identifying appearance is: Based on high-definition local close-up images, a lightweight classification network is used to analyze the plumage and color uniformity of feathers. Image superpixel segmentation is used to extract the local feather area of ​​the goose body, calculate the feather coverage rate, collect RGB pixel distribution of the feather area, and calculate the color difference standard deviation. The current growth stage of the goose house is extracted and compared with the default growth stages of this system to match the corresponding ideal feather coverage threshold and color difference standard threshold, which are recorded as and , It represents the minimum feather coverage at this growth stage under ideal conditions; It represents the maximum color difference of feathers at this growth stage under ideal conditions; according to the formula and The feather coverage rate and color difference standard deviation are normalized and then weightedly fused to obtain the appearance score. The appearance scores of all high-definition local close-up images in the goose house are averaged to obtain the appearance score of the goose house, which is recorded as F3. An update time is set, and high-definition local close-up images of all geese in the goose house are re-photographed according to the update time. The above-mentioned feather fullness and feather color uniformity are analyzed to obtain the appearance score.

[0007] In some embodiments, the process of behavior recognition analysis is: Extract several consecutive frames within a fixed time period and identify the feeding, drinking, walking, stationary, clustering, and abnormal fighting behaviors. Count the number of geese feeding, drinking, walking, stationary, clustering, and abnormal fighting behaviors, and divide the total number of geese by the total number of visible geese to obtain the feeding ratio, drinking ratio, walking ratio, stationary ratio, clustering ratio, and abnormal fighting ratio. The system defaults to an ideal behavior ratio interval for each growth stage. The current growth stage of the goose house is extracted and compared with the growth stages defaulted by the system to match the corresponding ideal behavior ratio interval. For each type of behavior ratio b, b∈{feeding ratio, drinking ratio, walking ratio, stationary ratio, clustering ratio, abnormal fighting ratio}, a single scoring function is defined , according to the measured proportion b and the ideal interval [ p min , p max ] , according to the formula S b = { 1 b ∈ [ p min , p max ] 1 − p min − b p min b ＜ p min 1 − b − p max 1 − p max b ＞ p max Get the score for each type of behavior And the various behavior scores are weighted and fused to obtain the behavior score F1.

[0008] In some embodiments, the process of gesture recognition analysis is: The posture estimation model is used to identify the key bone points of each goose in several consecutive frames within a fixed time length to determine the abnormal state of each goose in a single frame. The system defaults to normal standing posture, abnormal hunched posture and unstable gait posture. The number of normal standing posture, abnormal hunched posture and unstable gait posture in the consecutive frames is counted respectively, and the proportion of abnormal hunched posture, unstable gait posture and normal standing posture is calculated respectively, and recorded as and According to the formula The two posture proportions are normalized and then weighted fused to obtain the posture score F2; The maximum rickets ratio set by the system by default. This is the default maximum allowable limp percentage of this system.

[0009] In some embodiments, the process of physiological health analysis is: Extract the growth stage of the goose house and compare it with the default growth stage of this system to match the corresponding physiological standard parameters. The specific physiological standards include temperature range, heart rate range and activity range, and mark each physiological standard parameter as [ h min , h max ] ; Each index of the individual physiological information of each sampled goose in the goose house is marked as a, where a∈{temperature, heart rate, activity}, and the normalized score is defined as , and mark each physiological standard parameter as [ h min , h max ] ; Each index of the individual physiological information of each sampled goose in the goose house is marked as a, where a∈{temperature, heart rate, activity}, and the normalized score is defined as , according to the formula Calculate the score for each indicator , and then perform weighted fusion calculation to obtain the physiological score of each sampled goose per unit time; synchronously take a fixed time length, analyze the physiological health development trend of each sampled goose within the fixed time length to obtain the health score of each sampled goose within the fixed time length; compare and analyze the health score with the default health warning threshold of this system. If the health score is less than the health warning threshold, it means that the sampled goose has a serious physiological health risk, then the sampled goose will be recorded as a physiological abnormality warning goose, and the number of physiological abnormality warning geese in the goose house will be counted. If there is one or more, the intervention warning will be directly triggered and sent to the management personnel so that the management personnel can directly intervene in the goose house; otherwise, the health scores of each sampled goose will be averaged to obtain the physiological health score of the goose house.

[0010] In some embodiments, the process of analyzing the physiological health development trend of each sampled goose within a fixed period of time is as follows: The physiological scores of each sampled goose at each unit time within a fixed time were counted, and a two-dimensional rectangular coordinate system was established with time as the horizontal coordinate and the physiological score as the vertical coordinate. The physiological scores of the sampled geese at each unit time within a fixed time were plotted in the two-dimensional rectangular coordinate system, and line segments were used to obtain the physiological score line graph of the sampled geese in sequence; the mean of the physiological scores at each unit time in the physiological score line graph was calculated to obtain the horizontal score, and then the global trend of all sampling points in the physiological score line graph was fitted with the least squares straight line, and the global slope was recorded as g; an ideal maximum rising slope was set. and the maximum descending slope - , according to the formula The global slope is normalized, and then the normalized global slope and the horizontal score are weightedly fused to obtain the health score of the sampled geese within a fixed time period.

[0011] In some embodiments, the inspection route planning process is: The locations of the goose houses to be inspected are marked on the goose breeding map, and one goose house is selected as the starting point and the other goose house is selected as the ending point. The path algorithm is used to generate several inspection paths starting from the starting point, ending at the ending point, and covering all the goose houses to be inspected; for each path, one or two goose houses to be inspected at the middle position are selected as the interruption position, and the inspection route is divided into two parts by the interruption position, one part contains the starting point and the other part contains the ending point; the group representation scores corresponding to the goose houses to be inspected in the part containing the starting point are summed up to obtain the front-end representation score, and the group representation scores corresponding to the goose houses to be inspected in the part containing the ending point are summed up to obtain the back-end representation score, and the front-end representation score is divided by the back-end representation score to obtain the representation ratio, and only the inspection routes with the representation ratio ≥ 1 are retained, and then the shortest path is selected as the final inspection route and sent to the patrol personnel.

[0012] Compared with the prior art, the present invention has the following beneficial effects: 1. The group behavior recognition module uses a multi-view camera network and edge inference to perform high-precision detection and multi-target tracking of key goose behaviors in real time, and combines feather density and color difference analysis to derive an appearance score. De-noising, distortion correction, and ROI segmentation are performed on each image frame to ensure data quality. Then, using the YOLO series of lightweight models and multi-target tracking algorithms, each goose is behaviorally labeled, achieving quantitative statistics of the goose flock's behavioral patterns. Based on growth-stage ideal behavior intervals and feather cover detection using superpixel segmentation, the system can quickly determine collective feeding enthusiasm, stress clustering, and individual feather health. It can eliminate "missed / false positives" caused by single-dimensional thresholds and identify problems such as malnutrition and parasites through the fusion of appearance and behavior. This provides a multimodal, multi-layered, and refined perspective for group-level health assessments, greatly enhancing the accuracy and reliability of overall early warnings. 2. The individual health monitoring module uses the temperature, heart rate, and activity data collected by random inspection patches to map real-time indicators into normalized scores with a unified dimension using growth-stage reference intervals. It then generates a comprehensive health score (physiological score) that represents the broken line trend and average level within a fixed time window. This score can detect subtle physiological changes in individual geese at an early stage—such as a mild fever, abnormal heart rate, or a sudden drop in activity—and compare them with thresholds to provide second- or minute-level warnings. Once a goose's health score falls below the warning line, an intervention warning is issued, allowing managers to isolate the goose in this emergency. Compared with traditional "whole-flock temperature measurement" or "subjective observation," this mechanism significantly improves the sensitivity of early disease detection while avoiding the high cost of monitoring all geese. Furthermore, trend slope analysis can determine the direction of disease progression, providing managers with more timely diagnostic evidence, effectively reducing the risk of disease spread and saving medical and population losses. 3. The comprehensive assessment patrol module summarizes the appearance score, behavior score, posture score and physiological health score of each goose house into a single "group representation score", and uses this score to screen the goose houses to be inspected on the map and plan the optimal patrol route; first, all houses that have issued immediate intervention warnings are eliminated, and then the risk ranking is synchronized according to weighted fusion to ensure that high-risk houses are included in the patrol priority; multiple paths covering all houses to be inspected are generated between the starting point (current location of the personnel) and the end point (exit or dormitory), and a second screening is performed based on the "front and back risk ratio" and "path length", and finally an inspection route that is both the shortest and puts the higher-risk houses in the front section is output; on the one hand, the walking distance and time cost are minimized; on the other hand, the houses with the highest risk are placed in the most accessible location, ensuring that patrol personnel can conduct on-site inspections of key areas in the shortest time, and improving patrol efficiency and response speed, truly realizing "minimum input, maximum output" intelligent patrol.

[0013] In summary, the present invention conducts individual health monitoring and group behavior recognition on goose houses, and conducts comprehensive analysis based on the output of both to plan goose house inspection routes, thereby achieving all-round perception and intelligent analysis of the goose house environment, physiology, and behavior. It can not only capture subtle physiological or behavioral abnormalities in the early stages of the disease and prioritize the goose houses with the highest risks for on-site inspections, but also significantly reduce operation and maintenance costs and computing power usage through dynamic inspection route optimization and intermittent sampling, thereby comprehensively improving the disease prevention and control efficiency and production safety of goose farms. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the conventional technology, the following briefly introduces the drawings required for use in the embodiments or the conventional technology descriptions. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0015] Figure 1 This is a schematic diagram of the system module connection of the present invention. DETAILED DESCRIPTION

[0016] To make the above-mentioned objects, features, and advantages of the present invention more readily apparent, specific embodiments of the present invention are described in detail below with reference to the accompanying drawings. The following description sets forth numerous specific details to facilitate a full understanding of the present invention. However, the present invention can be implemented in many other ways than those described herein, and those skilled in the art may make similar modifications without departing from the scope of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0017] like Figure 1As shown in the figure, the disease monitoring and prevention system for goose farming based on big data includes: a perception front end, a storage end and a big data processing end, wherein the perception front end is used to communicate with various sensors in the goose house to collect individual physiological information and group behavior characteristic information of the goose house, and send it to the storage end for storage; the big data end includes an individual health monitoring module, a group behavior recognition module, and a comprehensive assessment and patrol module.

[0018] Goose breeding farms cover a relatively large area, and the entire goose breeding area is divided into several goose houses. Each goose house is equipped with various sensors to collect environmental information, individual physiological information and group behavior characteristic information of the goose house. It should be explained that environmental sensors include temperature and humidity sensors, ammonia / carbon dioxide sensors, light intensity sensors, etc.; physiological information is collected using collar patches, which have built-in temperature detection needles, three-axis accelerometers and heart rate monitors, etc.; group behavior is collected using multiple high-definition cameras in the goose house, etc.; due to the large number of geese in goose breeding, if all geese are equipped with physiological monitoring collar patches, the collection of this behavior The purchase and maintenance costs are very high. Although the comprehensive individual physiological parameters of geese can be monitored, this has no practical application value. However, in order to monitor the individual physiological information of each goose house, our technicians selected a fixed number of geese in each goose house (5% of each goose house) wearing collar patches as samples to collect individual physiological information, which can not only control costs, but also reflect the overall individual physiological status to a certain extent. The environmental information includes temperature, humidity, ammonia concentration, and carbon dioxide concentration; the individual physiological information includes temperature, heart rate, and activity level; the group behavior characteristic information includes high-definition group images and high-definition local close-up images.

[0019] The system has default settings for the cultivation parameters of each growth stage in goose farming, which include the brooding period, the rearing period, and the fattening period. The cultivation parameters include the cultivation temperature, the cultivation humidity, the upper limit of ammonia concentration, and the upper limit of carbon dioxide concentration; the feather coverage threshold, the color difference standard threshold, the ideal behavior ratio interval, and the physiological standard parameters of each growth stage.

[0020] The big data processing end includes a group behavior recognition module, an individual health monitoring module, and a comprehensive assessment and patrol module.

[0021] The group behavior recognition module uses high-definition group images to identify the growth stage of the goose house and evaluate the appearance of the geese by identifying group behavior characteristics. Specifically: Each goose house is installed with at least 2-4 high-definition cameras (1080p, night vision IR fill light), covering different angles in the house to ensure a panoramic view and be able to fully capture most of the goose flocks, feeding areas, drinking areas, and rest areas in the house; the fields of view of adjacent cameras overlap by about 20-30%, ensuring that the same goose can be recognized by any two cameras, which is conducive to multi-perspective fusion; the cameras are usually hung 2.5-3.5m from the roof of the house, with a bird's-eye view angle of about 45°-60° to avoid "blocking" and "direct light" problems.

[0022] Extract high-definition group images and perform image preprocessing on them: 1. Denoising and enhancement: For nighttime or low-light environments, use adaptive histogram equalization (CLAHE) or gamma correction to enhance details in low-light areas; perform local exposure compensation for overexposed or shadowed areas; 2. Distortion correction: If a wide-angle lens is used, first calibrate and correct the fisheye or barrel distortion to ensure that the detection frame truly corresponds to the size ratio of the goose body on the ground; 3. Segment ROI (Region of Interest): Use a static background model (Background Subtraction) to cut off the area outside the shed or where no one is active, leaving only the "goose activity area" to reduce the amount of subsequent calculations; Use a lightweight target detection network to detect the goose body in each frame of the preprocessed high-definition group image, and output the rectangular detection frame (bounding box) of each goose. box) and its corresponding confidence level, only the rectangular detection frame with a confidence level ≥ 0.5 is retained to eliminate occasional errors and improve the detection results; this frame has already covered the widest left and right areas of the goose on the pixel plane, pre-calibrated the HD camera by hand and eye, obtained the intrinsic parameter matrix K and distortion parameters, used the floor calibration board (known actual size) to calculate the relationship between the image coordinates and the ground truth distance, obtained the linear mapping relationship between pixel width and true width, for each rectangular detection frame, took the equivalent pixel length of its width on the ground plane, converted it into the front width of the goose as the body width of the goose; the body widths of all geese detected in the current frame were summarized, and the average body width and standard deviation were calculated; the system defaults to setting body width intervals corresponding to different stages, the average body width at the end of the brooding period (3-4 weeks old) is about 12-15cm; during the rearing period The average body width in the middle stage (8-10 weeks of age) is about 18-22 cm; the average body width in the late fattening stage (16-18 weeks of age) is about 25-30 cm; if the average width belongs to the brooding period and the standard deviation is less than 3 cm, it is determined to be in the brooding period; if the average width belongs to the growing period and the standard deviation is less than 4 cm, it is determined to be in the growing period; if the average width belongs to the fattening period and the standard deviation is less than 5 cm, it is determined to be in the fattening period; this can identify the growth stage of each frame image, count the growth stages of 10 consecutive frames, and select the growth stage with the largest number of frames as the current growth stage of the goose house. According to the current growth stage, the corresponding breeding environment parameters are matched. The specific breeding environment parameters include breeding temperature, breeding humidity, ammonia concentration upper limit, and carbon dioxide concentration upper limit. The environmental conditions of the goose house are controlled based on this breeding environment.

[0023] After determining the current growth stage of the goose house, the behavioral distribution and individual appearance characteristics of the goose group are further identified and judged to assess the health status of the goose house. For the rectangular detection frame of each detected goose, ID assignment and trajectory tracking are performed in the video sequence to obtain the real-time position of each goose. The speed and acceleration of each goose are calculated based on the unique vector of the same ID in consecutive frames, and used as the basic features for subsequent behavioral classification.

[0024] Based on high-definition local close-up images captured by a high-definition camera, a lightweight classification network is used to analyze the plumage and color uniformity of the feathers. The superpixel segmentation of the image (SLIC) is used to extract the local feather area of ​​the goose body, and the feather coverage rate (feather pixels / background pixels) is calculated and recorded as Y. The higher the feather coverage rate, the denser the feathers. The RGB pixel distribution of the feather area is collected, and the color difference standard deviation is calculated and recorded as C. The larger the color difference standard deviation, the presence of local fading, or even parasites or malnutrition. The current growth stage of the goose house is extracted and compared with the default growth stages of this system to match the corresponding ideal feather coverage threshold and color difference standard threshold, which are recorded as and , It represents the minimum feather coverage at this growth stage under ideal conditions; It represents the maximum color difference of feathers at this growth stage under ideal conditions; according to the formula and , normalize the feather coverage rate and the color difference standard deviation, and then perform weighted fusion to obtain the appearance score. Then, calculate the average of the appearance scores of all high-definition local close-up images in the goose house to obtain the appearance score of the goose house and record it as F3; set an update time, for example, the update time is 5 days, and re-take high-definition local close-up images of all geese in the goose house every 5 days and perform the above-mentioned feather fullness and feather color uniformity analysis to obtain the appearance score. Therefore, the appearance score of the vague goose house can be updated every 5 days.

[0025] This system has a pre-trained behavior recognition model by default, and the key behaviors that can be automatically identified include feeding, drinking, walking, stationary, clustering, and abnormal fighting. This behavior recognition model is based on a large amount of image and video data collected on-site at the goose house, which is professionally labeled for training, so it can accurately distinguish and detect the above-mentioned behaviors in real time. It identifies the number of geese feeding, drinking, walking, stationary, clustering, and abnormal fighting in several consecutive frames (taking 10 minutes of continuous frames), and divides them by the total number of visible geese to obtain the feeding ratio, drinking ratio, walking ratio, stationary ratio, clustering ratio, and abnormal fighting ratio.

[0026] This system has a default ideal behavior ratio range for each growth stage. For example, during the brooding period: the feeding range is 0.60-0.75, the drinking range is 0.10-1.15, the walking range is 0.15-0.25, the static range is 0.05-0.10, the clustering range is 0.00-0.10, and the abnormal fighting range is 0.00-0.02 (zero or very low, that is, no fighting under ideal conditions); during the brooding period, goslings need higher feeding and activity to promote rapid development. At the same time, excessive clustering or fighting should be avoided as much as possible during this stage; during the rearing period, the growth of geese tends to be stable and they begin to move closer to "adult geese". Appropriate Increase activity and reduce excessive rest, while paying attention to the frequency of gathering and fighting; the fattening period is mainly focused on weight gain and fat accumulation, the activity level can be moderately reduced, the feeding ratio is still important, but the rest time will be relatively increased, and long-term gathering or frequent fighting should be avoided; therefore, the ideal behavior ratios of each growth stage are different, the current growth stage of the goose house is extracted and compared with the default growth stages of this system to match the corresponding ideal behavior ratio interval, for each type of behavior ratio b, b∈{feeding ratio, drinking ratio, walking ratio, stationary ratio, gathering ratio, abnormal fighting ratio}, define a single scoring function , according to the measured proportion b and the ideal interval [ p min , p max ] , according to the formula S b = { 1 b ∈ [ p min , p max ] 1 − p min − b p min b ＜ p min 1 − b − p max 1 − p max b ＞ p max Get the score for each type of behavior The behavior scores of various types are weighted and fused to obtain the behavior score F1; the behavior score of the goose house is compared with the default behavior warning threshold of this system. If the behavior score is less than the behavior warning threshold, it means that there is a serious behavioral abnormality in the goose house at this time, and the intervention warning is directly triggered and sent to the goose house management personnel for intervention; otherwise, it means that there is no obvious abnormal behavior in the goose house, and posture estimation and recognition analysis is performed.

[0027] The pose estimation model is used to identify the key bone points (neck, waist, wing roots, leg joints, etc.) of each goose in several consecutive frames (10 minutes of continuous frames) to determine the abnormal state of each goose in a single frame. The system defaults to a normal standing posture: the neck maintains a certain angle (45°–60°), and the legs are separated and standing stably; the abnormal kyphotic posture is: the neck angle is less than 30°, and the head is drooped ≥45°; the unstable gait posture is: one leg is raised and stayed for more than 2s (limping) or the center of gravity fluctuates within n consecutive frames (movement disorder); the number of normal standing postures, abnormal kyphotic postures, and unstable gait postures in the continuous frames are counted respectively, and the proportion of abnormal kyphotic postures, unstable gait postures and normal standing postures are calculated respectively, and recorded as and According to the formula The two posture proportions are normalized and then weighted fused to obtain the posture score F2; The default maximum rickets ratio of this system (such as 0.1), is the maximum allowable lameness ratio (0.05) by default of this system. The formula clearly shows that the more normal the state of each goose in the goose house is, the greater the posture score of the goose house will be. The posture score of the goose house will be compared with the posture warning threshold preset by the system. If the posture score is less than the posture warning threshold, it means that there is a serious posture abnormality in the goose house at this time, which directly triggers the intervention warning and sends it to the goose house manager for intervention. Otherwise, the goose house number and its corresponding appearance score, behavior score and posture score will be sent to the comprehensive assessment patrol module.

[0028] The group behavior recognition module uses a multi-view camera network and edge reasoning to perform high-precision detection and multi-target tracking of the goose's key behaviors in real time, and combines feather density and color difference analysis to derive an appearance score. In each frame of the image, data quality is first ensured through denoising, distortion correction and ROI segmentation. Then, using the YOLO series of lightweight models and multi-target tracking algorithms, each goose is behaviorally labeled to achieve quantitative statistics of the goose flock's behavior patterns. Based on the ideal behavior interval of growth stages and feather cover detection using superpixel segmentation, the system can quickly determine collective feeding enthusiasm, the degree of stress clustering and the health of individual feathers. It can not only eliminate the "missed / false positives" caused by a single-dimensional threshold, but also identify problems such as malnutrition or parasites through the fusion of appearance and behavior, providing a multi-modal, multi-level and refined perspective for group-level health assessment, greatly enhancing the accuracy and reliability of the overall warning.

[0029] The individual health monitoring module performs physiological health analysis based on the physiological information of individual geese sampled in the goose house to determine the physiological health level of the goose house. Specifically: Extract the growth stage of the goose house and compare it with the default growth stage of this system to match the corresponding physiological standard parameters. The specific physiological standards include temperature range, heart rate range and activity range, and mark each physiological standard parameter as [ h min , h max ] ; Each index of the individual physiological information of each sampled goose in the goose house is marked as a, where a∈{temperature, heart rate, activity}, and the normalized score is defined as , according to the formula Calculate the score for each indicator , and then perform weighted fusion calculation to obtain the physiological score of each sampled goose per unit time; synchronously take a fixed time length (10 minutes), and count the physiological scores of each sampled goose per unit time within the fixed time length, use time as the horizontal coordinate and the physiological score as the vertical coordinate in a two-dimensional rectangular coordinate system, and plot the physiological scores of the sampled geese per unit time within the fixed time length in the two-dimensional rectangular coordinate system, and use line segments to obtain the physiological score line graph of the sampled geese in sequence; calculate the mean of the physiological scores per unit time in the physiological score line graph to obtain the horizontal score, and then perform least squares linear fitting on the global trend of all sampling points in the physiological score line graph, and obtain the global slope recorded as g; set an ideal maximum rising slope and the maximum descending slope - , according to the formula The global slope is normalized, and then the normalized global slope and the horizontal score are weighted and fused to calculate the health score of the sampled goose within a fixed time. The health score is compared with the health warning threshold of the system by default. If the health score is less than the health warning threshold, it means that the sampled goose has a serious physiological health risk. The sampled goose is recorded as a physiological abnormality warning goose, and the number of physiological abnormality warning geese in the goose house is counted. If there is one or more, an intervention warning is directly triggered and sent to the management personnel so that the management personnel can directly intervene in the goose house; otherwise, the health of each sampled goose is recorded as a physiological abnormality warning goose. The scores are averaged to obtain the physiological health score of the goose house, recorded as F4, and sent to the comprehensive assessment patrol module. It should be noted that the geese will not have much appearance and physiological changes in a short period of time. In practical applications, it is not necessary to conduct in-depth monitoring and analysis of the goose house all day long, which will cause data redundancy and occupy computing power. Therefore, this application adopts an intermittent method (full collection every 30 minutes, 1 hour or 1.5 hours, including behavior, appearance, individual physiology, etc.) to timely detect environmental mutations, behavioral abnormalities or health warnings. Most pre-disease signs will be captured at these monitoring points.

[0030] The individual health monitoring module uses the body temperature, heart rate and activity level collected by random inspection patches, and maps the real-time indicators into normalized scores with a unified dimension through growth stage reference intervals. It also generates a comprehensive health score - physiological score - that shows the broken line trend and average level within a fixed time window. It can capture subtle physiological changes in individual geese at an early stage - such as mild fever, abnormal heart rate or sudden decrease in activity - and compare with the threshold to achieve second-level or minute-level warnings. Once the health score of a goose falls below the warning line, an intervention warning will be issued, which can facilitate management personnel to isolate this emergency situation. Compared with traditional "whole-flock temperature measurement" or "subjective observation", this mechanism greatly improves the sensitivity of early disease detection and avoids the high-cost monitoring of all geese. At the same time, trend slope analysis can also determine the direction of disease progression, providing management personnel with more timely diagnostic basis, thereby effectively reducing the risk of disease spread and saving medical and population loss costs.

[0031] The comprehensive assessment and patrol module conducts a comprehensive analysis of the group characteristics of the goose house to form a patrol route, specifically: Compare the goose house numbers received from the group behavior recognition module and the individual health monitoring module, filter out the goose house numbers that exist at the same time, and classify them into the goose houses to be inspected; it should be noted that if they exist at the same time, it means that the goose house has not triggered the interference warning, and it needs to be inspected and planned; extract the appearance score, behavior score, posture score and physiological health score of the goose house to be inspected, and perform weighted fusion calculation to obtain the group representation score of the goose house, and sort the goose houses to be inspected in ascending order according to their corresponding group representation scores. It should be noted that the smaller the group representation score of the goose house, the greater the group risk it has; mark the location of the goose house to be inspected on the goose breeding map, and select a goose house as the starting point and a goose house as the end point. Usually, the goose house used as the starting point is the one closest to the patrol personnel in the goose farm to reduce invalid walking distance; select the one closest to the farm exit (or back to the dormitory / equipment) The nearest one to the room (the location of the room) is used as the end point so that the inspection can be quickly evacuated or archived after the inspection is completed; a path algorithm is used to generate several inspection paths starting from the starting point, ending at the end point, and covering all goose houses to be inspected; for each path, one or two goose houses to be inspected at the middle position are selected as the interruption position, and the inspection route is divided into two parts by the interruption position, one part includes the starting point and the other part includes the end point; the group representation scores corresponding to the goose houses to be inspected in the part including the starting point are summed up to obtain the front-end representation score, and the group representation scores corresponding to the goose houses to be inspected in the part including the end point are summed up to obtain the back-end representation score, and the front-end representation score is divided by the back-end representation score to obtain the representation ratio. If the representation ratio is less than 1, it will be discarded, and only the inspection routes with the representation ratio ≥ 1 will be retained. The shortest path is then selected as the final inspection route and sent to the patrol personnel to guide them to quickly complete the goose house inspection.

[0032] The comprehensive assessment patrol module summarizes the appearance score, behavior score, posture score and physiological health score of each goose house into a single "group representation score", and uses this score to screen the goose houses to be inspected on the map and plan the optimal patrol route; first, all houses that have issued immediate intervention warnings are eliminated, and then the risk ranking is synchronized according to weighted fusion to ensure that high-risk houses are included in the patrol first; multiple paths covering all houses to be inspected are generated between the starting point (current location of the personnel) and the end point (exit or dormitory), and a second screening is performed based on the "front and back risk ratio" and "path length", and finally an inspection route that is both the shortest and puts the higher-risk houses in the front section is output; on the one hand, the walking distance and time cost are minimized; on the other hand, the houses with the highest risk are placed in the most accessible location, ensuring that patrol personnel can conduct on-site inspections of key areas in the shortest time, and improving patrol efficiency and response speed, truly realizing intelligent patrol with "minimum input and maximum output".

[0033] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, 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.

[0034] The formula set above is an example of an empirical formula customized by technical experts in this field. Any formula that conforms to the relationship between the parameters of this application is acceptable. The specific size of the weight factor in the formula is reasonably set by technical personnel in this field according to actual use.

[0035] The above embodiments merely illustrate several implementations of the present invention, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the patent. It should be noted that a person skilled in the art would be able to make various modifications and improvements without departing from the spirit of the present invention, all of which fall within the scope of protection of the present invention. Therefore, the scope of protection of the patent for this invention shall be determined by the appended claims.

Claims

1. A disease monitoring and prevention system for goose farming based on big data, including a sensing front end and a storage end. The sensing front end is used to communicate with various sensors in the goose house to collect individual physiological information and group behavioral characteristics of the geese; it is characterized by: Also includes: The big data processing end has built-in group behavior recognition module, individual health monitoring module, and comprehensive assessment and patrol module; The group behavior recognition module uses high-definition group images of the goose house to identify the group and determine the growth stage of the goose house. It then identifies the appearance of the goose house based on the growth stage and outputs an appearance score. The appearance score of the goose house is updated regularly. The group behavior of the goose house is identified and analyzed based on the growth stage and outputs a behavior score. If the behavior score is less than the behavior warning threshold, an intervention warning is directly triggered. Otherwise, posture estimation and recognition analysis is performed to output a posture score. If the posture score is less than the posture warning threshold, an interference warning is directly triggered; otherwise, the goose house number and its corresponding appearance score, behavior score, and posture score are sent to the comprehensive evaluation and patrol module; The individual health monitoring module performs physiological health analysis based on the individual physiological information of the geese sampled in the goose house, and outputs the health score of each sampled goose within a fixed period of time. If the health score is less than the health warning threshold, the sampled goose is recorded as a physiological abnormality warning goose. The number of physiological abnormality warning geese in the goose house is counted. If there is one or more physiological abnormality warning geese, an intervention warning is directly triggered. Otherwise, the health scores of all sampled geese are averaged to obtain the physiological health score of the goose house, and the score is sent to the comprehensive assessment and inspection module; The comprehensive assessment inspection module compares the goose house numbers received from the group behavior recognition module and the individual health monitoring module, filters out the goose house numbers that exist at the same time, and classifies them into the goose houses to be inspected; Then, the route is planned based on the goose house to be inspected and the final inspection route is output.

2. The disease monitoring and prevention system for goose farming based on big data according to claim 1 is characterized in that: Growth stage judgment output process: High-definition group images are extracted and preprocessed. A lightweight object detection network is used to detect the goose body in each frame of the preprocessed high-definition group images. A rectangular detection box for each goose and its corresponding confidence score are output, and only rectangular detection boxes with a confidence score greater than or equal to 0.5 are retained. The high-definition camera is pre-calibrated by hand and eye to obtain the intrinsic parameter matrix and distortion parameters. The relationship between the image coordinates and the ground truth distance is calculated using a floor calibration plate, and a linear mapping relationship between pixel width and true width is obtained. For each rectangular detection box, the equivalent pixel length of its width on the ground plane is taken and converted into the frontal width of the goose as the body width of the goose. The body widths of all geese detected in the current frame are summarized, and the average body width and standard deviation are calculated. The system defaults to body width intervals corresponding to different stages, which can be used to identify the growth stage of each frame. The growth stages of several consecutive frames are counted, and the growth stage with the most frames is selected as the current growth stage of the goose house. The corresponding breeding environment parameters, including breeding temperature, breeding humidity, ammonia concentration upper limit, and carbon dioxide concentration upper limit, are matched according to the current growth stage. The environmental conditions of the goose house are controlled based on this breeding environment.

3. The disease monitoring and prevention system for goose farming based on big data according to claim 2 is characterized in that: The process of identifying by appearance is as follows: Based on high-definition local close-up images, a lightweight classification network is used to analyze the feather fullness and feather color uniformity. The local feather area of ​​the goose body is extracted using image superpixel segmentation, the feather coverage rate is calculated, the RGB pixel distribution of the feather area is collected, and the color difference standard deviation is calculated; the current growth stage of the goose house is extracted and compared with the default growth stages of this system to match the corresponding ideal feather coverage threshold and color difference standard threshold, and the feather coverage rate and color difference standard deviation of the current high-definition local close-up image are normalized based on them, and then the processed feather coverage rate and color difference standard deviation are weightedly fused to obtain the appearance score of the high-definition local close-up image, and then the appearance scores of all high-definition local close-up images in the goose house are averaged to obtain the appearance score of the goose house; an update time is set, and high-definition local close-up images of all geese in the goose house are re-shot according to the update time and the above-mentioned feather fullness and feather color uniformity analysis is performed to obtain the appearance score.

4. The disease monitoring and prevention system for goose farming based on big data according to claim 3 is characterized in that: The process of behavior recognition analysis is as follows: Extract several consecutive frames within a fixed time period and identify the feeding, drinking, walking, stationary, clustering, and abnormal fighting behaviors. Count the number of geese feeding, drinking, walking, stationary, clustering, and abnormal fighting behaviors, and divide the total number of geese by the total number of visible geese to obtain the feeding ratio, drinking ratio, walking ratio, stationary ratio, clustering ratio, and abnormal fighting ratio. This system has a default ideal behavior ratio interval for each growth stage. The current growth stage of the goose house is extracted and compared with the default growth stages of this system to match the corresponding ideal behavior ratio interval. For each type of behavior ratio b, b∈{feeding ratio, drinking ratio, walking ratio, stationary ratio, gathering ratio, abnormal fighting ratio}, a single scoring function is defined. The proportion of each type of behavior is normalized according to the ideal behavior ratio interval, and the scores of each type of behavior after processing are weighted and fused to obtain the behavior score.

5. The disease monitoring and prevention system for goose farming based on big data according to claim 4 is characterized in that: The process of posture recognition analysis is as follows: The posture estimation model is used to identify the key bone points of each goose in several consecutive frames within a fixed time length to determine the abnormal state of each goose in a single frame. This system defaults to normal standing posture, abnormal hunched posture and unstable gait posture. The number of normal standing posture, abnormal hunched posture and unstable gait posture in consecutive frames is counted respectively, and the posture proportion of abnormal hunched posture, unstable gait posture and normal standing posture is calculated respectively. The postures are then normalized and weighted fused to obtain the posture score.

6. The disease monitoring and prevention system for goose farming based on big data according to claim 5, characterized in that: The process of physiological health analysis is: Extract the growth stage of the goose house and compare it with the default growth stage of this system to match the corresponding physiological standard parameters, including temperature range, heart rate range and activity range, and mark each physiological standard parameter as ; Each index of the individual physiological information of each sampled goose in the goose house is marked as a, where a∈{temperature, heart rate, activity}, and the normalized score is defined as , normalize each indicator of individual physiological information according to physiological standard parameters to obtain the score of each indicator , and then perform weighted fusion calculation to obtain the physiological score of each sampled goose per unit time; synchronously take a fixed time length, analyze the physiological health development trend of each sampled goose within the fixed time length to obtain the health score of each sampled goose within the fixed time length; The health score is compared and analyzed with the default health warning threshold of this system. If the health score is less than the health warning threshold, it means that the sampled goose has a serious physiological health risk. The sampled goose is recorded as a physiological abnormality warning goose, and the number of physiological abnormality warning geese in the goose house is counted. If there is one or more, an intervention warning is directly triggered and sent to the management personnel so that the management personnel can directly intervene in the goose house; otherwise, the health scores of each sampled goose are averaged to obtain the physiological health score of the goose house.

7. The disease monitoring and prevention system for goose farming based on big data according to claim 6 is characterized in that: The process of analyzing the physiological health development trend of each sampled goose within a fixed period of time is as follows: The physiological scores of each sampled goose at each unit time within a fixed time period are counted, and a two-dimensional rectangular coordinate system is established with time as the horizontal coordinate and the physiological score as the vertical coordinate. The physiological scores of the sampled geese at each unit time within the fixed time period are plotted in the two-dimensional rectangular coordinate system, and line segments are used to obtain a line graph of the physiological scores of the sampled geese in sequence; The horizontal score was obtained by averaging the physiological scores of each unit time in the physiological score line graph. The global trend of all sampling points in the physiological score line graph was fitted with the least squares straight line, and the global slope was recorded as g. Set the ideal maximum slope and the maximum descending slope - , according to the formula The global slope is normalized, and then the normalized global slope and the horizontal score are weightedly fused to obtain the health score of the sampled geese within a fixed time period.

8. The disease monitoring and prevention system for goose farming based on big data according to claim 7 is characterized in that: The inspection route planning process is as follows: Mark the locations of the goose houses to be inspected on the goose breeding map, select one goose house as the starting point and another goose house as the end point, and use the path algorithm to generate several inspection paths starting from the starting point and ending at the end point, covering all the goose houses to be inspected; For each path, one or two goose houses to be inspected in the middle are selected as the interruption position, and the inspection route is divided into two parts by the interruption position, one part includes the starting point and the other part includes the end point; the group representation scores corresponding to the goose houses to be inspected in the part including the starting point are summed up to obtain the front-end representation score, and the group representation scores corresponding to the goose houses to be inspected in the part including the end point are summed up to obtain the back-end representation score, and the front-end representation score is divided by the back-end representation score to obtain the representation ratio, and only the inspection routes with representation ratios ≥ 1 are retained, and then the shortest path is selected as the final inspection route and sent to the patrol personnel.

9. The disease monitoring and prevention system for goose farming based on big data according to claim 1, characterized in that: The sensing front end is used to communicate with various sensors in the goose house to collect individual physiological information and group behavioral characteristic information of the goose house, and send it to the storage end for storage.

Images