Chicken Flock Health Status Analysis System Based on Data Analysis

Through the chicken health status analysis system based on data analysis, using image and sound signals, combined with activity trajectory, to evaluate the chicken health status, the problems of traditional time-consuming and labor-intensive evaluation and low accuracy of neural network models are solved, and more efficient health is achieved Status Assessment.

CN118779832BActive Publication Date: 2024-11-12大连云智信科技发展有限公司
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411259740.7
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-09-10
Publication Date
2024-11-12
Estimated Expiration
2044-09-10

AI Technical Summary

Technical Problem

It is time-consuming and labor-intensive to evaluate the health status of traditional breeding chickens, and neural network models are difficult to dynamically capture the stable posture of chickens, resulting in low accuracy and the normal-scale models are prone to overfitting or underfitting.

Method used

Design a chicken flock health status analysis system based on data analysis. By obtaining the chicken flock images, sound signals and activity trajectories, the health rate of each chicken's head-up posture, the health rate of head-down posture and the pecking order are determined, and the health status of the chicken is comprehensively analyzed. . The system includes a acquisition module, a gesture module, a pecking module and a health module, and uses feature point recognition, static image processing and pecking analysis to calculate the probability of health abnormalities.

Benefits of technology

It improves the accuracy of health analysis, simplifies the process of evaluating chicken health status, solves the problems of difficulty in labeling and low accuracy in the existing technology, and can more simply and efficiently evaluate the health status of chickens.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118779832B_ABST
    Figure CN118779832B_ABST
Patent Text Reader

Abstract

The present invention relates to the technical field of electronic digital data processing, and in particular to a chicken flock health status analysis system based on data analysis. The system obtains chicken flock images, sound signals and the activity trajectory of each farmed chicken in a chicken farm within a preset time interval, and determines the head-up posture health rate and head-down posture health rate of each farmed chicken according to the chicken flock images. The head-up posture health rate and head-down posture health rate corresponding to the farmed chickens are determined based on static chicken flock images, which is simpler and more efficient. The chicken flock's pecking order is determined according to the sound signals and the activity trajectory, reflecting the sociality of each farmed chicken. The health status of each farmed chicken is determined according to the head-up posture health rate, the head-down posture health rate and the pecking order. The sociality, the head-up posture health rate and the head-down posture health rate are considered to comprehensively determine the health status of the farmed chickens, which is conducive to improving the accuracy of health analysis.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention relates to the technical field of electronic digital data processing, and in particular to a chicken health status analysis system based on data analysis. Background Art

[0002] In the traditional chicken farming process, it is usually necessary to manually observe the appearance, posture, behavior, appetite and feather status of the chickens, and then conduct health status assessment, which is time-consuming and laborious, and easy to miss. Therefore, some existing methods use deep learning models to automatically identify sick chickens by learning the chicken's movement and behavior feature data.

[0003] However, due to the physiological structure of chickens, their movement patterns are not coherent, and they always present short, intermittent irregular movements. This discontinuous movement pattern makes it difficult for the neural network model to dynamically capture the stable posture of the chickens. At the same time, the posture changes of different chickens have different frequencies within the same sampling time. Dynamic capture can easily lead to large differences in sample information, which not only increases the difficulty of data labeling, but also requires the model to have extremely high generalization capabilities. Conventional neural network models may have problems of overfitting or underfitting, resulting in low accuracy. Summary of the invention

[0004] In order to solve the above technical problems, the purpose of the present invention is to provide a chicken health status analysis system based on data analysis, and the technical solutions adopted are as follows:

[0005] In a first aspect, an embodiment of the present application provides a chicken health status analysis system based on data analysis, comprising:

[0006] An acquisition module is used to acquire images of chickens in a chicken farm, sound signals, and the activity trajectory of each chicken in a preset time interval;

[0007] A posture module, used to determine the head-up posture health rate and head-down posture health rate of each farmed chicken based on the chicken flock image;

[0008] A pecking order module, used to determine the pecking order of the chickens according to the sound signal and the activity trajectory;

[0009] The health module is used to determine the health status of each farmed chicken according to the head-up posture health rate, the head-down posture health rate and the pecking order.

[0010] In one embodiment, the posture module includes a first posture unit, a second posture unit, a third posture unit, a fourth posture unit, and a fifth posture unit;

[0011] The first posture unit is used to identify the head area and the tail area of ​​each frame of the chicken image in the chicken image, and determine the head feature points of the head area and the tail feature points of the tail area;

[0012] The second posture unit is used to determine the shooting direction of the camera that shoots the chicken flock image, and to determine the orientation of each farmed chicken in each of the chicken flock images;

[0013] The third posture unit is used to determine a static valid image of each farmed chicken from each frame of the chicken group image according to the shooting direction and the orientation;

[0014] The fourth posture unit is used to cut out the corresponding image slice of the farmed chicken from each of the static valid images, and in the vertical direction, when the head feature point in the image slice is higher than the tail feature point, the image slice is determined to be a head-up image, and when the head feature point in the image slice is lower than the tail feature point, the image slice is determined to be a head-down image, and a head-up image set of each farmed chicken is determined based on a plurality of head-up images, and a head-down image set of each farmed chicken is determined based on a plurality of head-down images;

[0015] The fifth posture unit is used to determine the head-up posture health rate and the head-down posture health rate of each farmed chicken based on the head-up image set and the head-down image set.

[0016] In one embodiment, determining a static valid image of each farmed chicken from each frame of the chicken flock image according to the shooting direction and the orientation includes:

[0017] Determine the angle between each of the directions and the mapping direction of the shooting direction;

[0018] Determine a first absolute value of a difference between the direction angle and a first preset angle;

[0019] When the first absolute value is less than or equal to a first angle threshold, the image of the chicken flock corresponding to the orientation is used as a target image, and the target image is used as a static valid image of the farmed chickens corresponding to the orientation.

[0020] In one embodiment, determining the head-up posture health rate and the head-down posture health rate of each farmed chicken according to the head-up image set and the head-down image set includes:

[0021] Determine a first centroid point of each head-up image in each head-up image set, a first connecting line between each first centroid point and a head feature point of the corresponding head-up image, a first length of the first connecting line, and a first angle between the first connecting line and a vertical line in a vertical direction and upward;

[0022] Obtaining a head-up coefficient of each head-up image corresponding to each farmed chicken according to a Euclidean norm of a cosine value of the first length and the first angle;

[0023] The head-raising coefficients of each head-raising image corresponding to each farmed chicken are averaged to obtain the head-raising posture health rate of each farmed chicken;

[0024] The head-down posture health rate of each farmed chicken is determined based on the head-down image set.

[0025] In one embodiment, determining the head-down posture health rate of each farmed chicken according to the head-down image set includes:

[0026] Determine a second centroid point of each of the head-down images in each of the head-down image sets, a second connecting line between each of the second centroid points and the corresponding head feature point of the head-down image, a second length of the second connecting line, and a second angle between the second connecting line and a vertical line in a vertical direction and downward;

[0027] Determine a second absolute value of a difference between the second angle and a second preset angle;

[0028] Determine a third connecting line between the second centroid point of each of the head-down images and the corresponding tail feature point of the head-down image, and a third angle between the second connecting line and the third connecting line;

[0029] Calculate the difference between the third angle and the third preset angle to obtain a fourth angle, retain the angle value of the fourth angle that is greater than the second angle threshold, and reset the angle value of the fourth angle that is less than or equal to the second angle threshold to zero;

[0030] Obtaining a head-down coefficient of each head-down image corresponding to each farmed chicken according to the second length, the cosine value of the second absolute value, and the Euclidean norm of the cosine value of the fourth angle;

[0031] Determine the clustering factor of each farmed chicken according to the acquisition time of each of the head-down images in the head-down image set;

[0032] The head-down posture health rate of each farmed chicken is determined according to the clustering factor and the head-down coefficient of each head-down image corresponding to each farmed chicken.

[0033] In one embodiment, determining the clustering factor of each farmed chicken according to the acquisition time of each of the head-down images in the head-down image set includes:

[0034] Determine the acquisition time of each of the head-down images in the head-down image set, determine a target head-down image set from the head-down image set, determine a target head-down image from the target head-down image set, and determine adjacent acquisition times of head-down images closest to the acquisition time of the target head-down image from other head-down image sets;

[0035] Calculate the absolute value of the time difference between the acquisition time of the target head-down image and each of the adjacent acquisition times;

[0036] Determine the total absolute value of the time difference according to the absolute values ​​of the time differences and the number of farmed chickens;

[0037] Obtaining a clustering factor corresponding to the target head-down image according to a first ratio of the total absolute value of the time difference to the number of farmed chickens;

[0038] Return to the step of determining the target head-down image from the target head-down image set until the corresponding clustering factor of each target head-down image in the target head-down image set is determined, and return to the step of determining the target head-down image set from the head-down image set until the corresponding clustering factor of each head-down image in each head-down image set is determined;

[0039] The clustering factor of each farmed chicken is determined according to the clustering factors corresponding to each of the head-down images in each of the head-down image sets.

[0040] In one embodiment, determining the head-down posture health rate of each farmed chicken according to the clustering factor and the head-down coefficient of each head-down image corresponding to each farmed chicken includes:

[0041] Calculating respectively a second ratio of the head-down coefficient of each head-down image corresponding to each farmed chicken to the clustering factor, and determining a normalized result of the second ratio;

[0042] The head-down posture health rate of each farmed chicken is obtained by averaging multiple normalized results of the head-down image set of each farmed chicken.

[0043] In one embodiment, the pecking module includes a first pecking unit, a second pecking unit, a third pecking unit, a fourth pecking unit, and a fifth pecking unit;

[0044] A first pecking unit, used for determining a plurality of data points of the sound signal, a sound source position of each of the data points, and a sound amplitude and frequency value of each of the data points;

[0045] A second pecking unit is used to determine a search range with the sound source position of each data point as the center according to a preset radius, determine the number of other data points in each search range, and determine an average number of points according to the sum of the number of points and the number of the search range;

[0046] The third pecking unit is used to mark the data point as a maximum point when the sound amplitude and frequency value of the data point located at the center are both greater than the sound amplitude and frequency value of other data points, retain the target maximum point centered on the maximum point, the number of corresponding points of which is greater than or equal to the average number of points, and take each farmed chicken corresponding to the target maximum point as a suspected leader chicken;

[0047] A fourth pecking order unit is used to determine a target activity trajectory of each of the suspected leader chickens from the activity trajectory;

[0048] The fifth pecking order unit is used to determine the pecking order of the chickens according to the activity trajectory and the target activity trajectory.

[0049] In one embodiment, determining the pecking order of the chickens according to the activity trajectory and the target activity trajectory includes:

[0050] Calculating the mean square error between the activity trajectory of each farmed chicken and the target activity trajectory of each suspected leader chicken respectively;

[0051] Determine, based on each of the mean square errors, the trajectory similarity between the activity trajectory of each farmed chicken and the target activity trajectory of each of the suspected leading chickens;

[0052] The activity trajectory of each farmed chicken with the maximum trajectory similarity is taken as the followability of the farmed chicken;

[0053] The following properties are arranged in descending order to obtain the pecking order of the chickens.

[0054] In one embodiment, the health module includes a first health unit, a second health unit, a third health unit, a fourth health unit, and a fifth health unit;

[0055] A first health unit is used to determine a first ratio of the head-up image of each farmed chicken to all images of the chicken flock, and a second ratio of the head-down image of each farmed chicken to all images of the chicken flock;

[0056] A second health unit is used to perform weighted calculation according to the head-up posture health rate, the first proportion, the head-down posture health rate and the second proportion to obtain a health posture coefficient of each farmed chicken, and sort the health posture coefficients of all farmed chickens from large to small to obtain a posture sequence corresponding to all farmed chickens;

[0057] A third health unit is used to respectively determine the sum and the absolute value of the difference between the first sequence number of each farmed chicken in the posture sequence and the second sequence number in the pecking sequence;

[0058] A fourth health unit, used for determining a calculation result according to a natural exponential function and an absolute value of the difference respectively;

[0059] The fifth health unit is used to determine a third ratio of the average value of the sum value to the calculation result, and normalize the third ratio to obtain the health abnormality probability of each farmed chicken, wherein the health abnormality probability represents the health status of the farmed chicken.

[0060] The present invention has the following beneficial effects:

[0061] By obtaining the images of the chicken flock, sound signals and the activity trajectory of each farmed chicken in a preset time interval, the head-up posture health rate and head-down posture health rate of each farmed chicken are determined based on the images of the chicken flock. The corresponding head-up posture health rate and head-down posture health rate of the farmed chicken are determined based on the static images of the chicken flock, which is simpler and more efficient; the pecking order of the chicken flock is determined based on the sound signals and activity trajectories to reflect the sociality of each farmed chicken; the health status of each farmed chicken is determined based on the head-up posture health rate, head-down posture health rate and pecking order, and the sociality, head-up posture health rate, and head-down posture health rate are considered to comprehensively determine the health status of the farmed chickens, which is conducive to improving the accuracy of health analysis and solving the problems of difficult labeling and low accuracy in the existing neural network model method. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0063] Figure 1 A structural block diagram of a chicken health status analysis system based on data analysis provided by an embodiment of the present invention;

[0064] Figure 2 A schematic diagram of the shooting direction and the orientation of farmed chickens according to an embodiment of the present invention;

[0065] Figure 3 A schematic diagram of a first centroid point, a first connecting line and a first angle according to an embodiment of the present invention;

[0066] Figure 4 A schematic diagram of a second connecting line, a second angle, a third connecting line, and a third angle according to an embodiment of the present invention;

[0067] Figure 5 A schematic diagram of the relationship between the acquisition time of a target head-down image and other head-down images in a set of head-down images according to an embodiment of the present invention;

[0068] Figure 6 This is a schematic diagram of visualizing data point projection in a two-dimensional plane according to an embodiment of the present invention. DETAILED DESCRIPTION

[0069] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the following is a detailed description of the specific implementation, structure, characteristics and effects of a chicken health status analysis system based on data analysis proposed by the present invention in combination with the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" does not necessarily refer to the same embodiment. In addition, specific features, structures or characteristics in one or more embodiments may be combined in any suitable form.

[0070] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0071] The specific scheme of a chicken health status analysis system based on data analysis provided by the present invention is described in detail below in conjunction with the accompanying drawings.

[0072] See also Figure 1 , which shows a schematic diagram of a chicken health status analysis system based on data analysis provided by an embodiment of the present invention. The chicken health status analysis system based on data analysis may at least include an acquisition module, a posture module, a pecking order module and a health module. Specifically:

[0073] An acquisition module is used to acquire images of chickens in a chicken farm, sound signals, and activity tracks of each chicken in a preset time interval;

[0074] The posture module is used to determine the head-up posture health rate and head-down posture health rate of each farmed chicken based on the chicken group image;

[0075] The pecking order module is used to determine the pecking order of the chickens based on the sound signals and activity trajectories;

[0076] The health module is used to determine the health status of each farmed chicken based on the head-up posture health rate, head-down posture health rate and pecking order.

[0077] The technical solution of the embodiment of the present application obtains the images of the chicken flock, sound signals and the activity trajectory of each farmed chicken in a chicken farm within a preset time interval, and determines the head-up posture health rate and head-down posture health rate of each farmed chicken based on the images of the chicken flock, and determines the corresponding head-up posture health rate and head-down posture health rate of the farmed chicken based on the static chicken flock images, which is simpler and more efficient; determines the pecking order of the chicken flock based on the sound signals and activity trajectories to reflect the sociality of each farmed chicken; determines the health status of each farmed chicken based on the head-up posture health rate, head-down posture health rate and pecking order, and considers the sociality, head-up posture health rate, and head-down posture health rate to comprehensively determine the health status of the farmed chicken, which is conducive to improving the accuracy of health analysis and solving the problems of difficult labeling and low accuracy in the existing neural network model method.

[0078] In one embodiment, the acquisition module may include a camera, a sensor unit, and a "chicken tag". The sensor unit includes several different types of sensors installed at different locations inside the farm to obtain data such as temperature, humidity, and ammonia concentration. A sound sensor consisting of a circular microphone array composed of several microphones is installed on the central roof of the chicken coop to obtain sound signals for monitoring whether the breeding environment in the chicken coop of the farm is abnormal. There are multiple "chicken tags", and each farmed chicken is installed with a "chicken tag". The "chicken tag" has a built-in positioning tracker and a gyroscope, and the positioning tracker can obtain real-time position data. The camera has a built-in gyroscope to determine the direction of the camera. The number of cameras can also be multiple to cover different areas inside the farm. The cameras are all installed in a corner of the chicken coop and are relatively high, so as to facilitate the acquisition of images of the chickens from a wider perspective.

[0079] It should be noted that the preset time interval can be based on actual settings, such as 1 hour, 4 hours, etc., without specific limitations. In the embodiment of the present application, 4 hours is taken as an example, and the interval monitoring process does not require real-time performance; the data acquired by the acquisition module can be stored in the terminal or server, such as in the monitoring center of the server, for the monitoring center to analyze the data, such as using the trained YOLO target recognition model to identify each farmed chicken in each frame of the chicken flock image contained in the chicken flock image, and then combining the position data of the positioning tracker to generate the activity trajectory of each farmed chicken, so as to achieve tracking of the farmed chicken; and the sound signal can be analyzed to identify the source of the call according to the sound source location. Among them, in addition to real-time collection and overall planning of data, the monitoring center can also pre-process (clean, denoise) the data sent by the sensor and the monitoring acquisition module, and transmit the processed data to the monitoring center for subsequent analysis and evaluation.

[0080] It should be noted that the chicken's movements are often not continuous, showing a "stop-and-go" feeling. This is because the chicken's visual system is very sharp, but unlike humans, it does not have continuous visual processing capabilities. The chicken's eyes cannot process visual information at the same time when moving quickly. At the same time, in addition to the characteristics of the visual system, chickens also have the characteristics of foraging behavior, neural control methods, balance and coordination requirements, and predation and defense strategies, which make chickens pause frequently during the action, thus forming the "stop-and-go" action pattern we observed. It is precisely because of this action pattern of chickens that it is difficult for the neural network model to capture the stable posture of the chickens. For different chickens, a number of equal-length video frame data are captured. In the same time, some chickens may be stationary, and some chickens may change their postures several times. The information contained in these samples is quite different, requiring the network model to have extremely high generalization capabilities to process. Conventional posture recognition models are very prone to overfitting or underfitting problems. Therefore, this application provides a simpler and more efficient method. Specifically, in view of the frequent pauses of chickens, it is necessary to first analyze and capture the posture of the chickens frame by frame: In order to intuitively identify the health status of farmed chickens, it can generally be judged from the posture of the chickens. Under normal and healthy conditions, the heads of farmed chickens are usually upright, showing an alert and active state. Healthy chickens will maintain a natural upright posture of the head and neck, with bright eyes and quick response to external stimuli. This upright head posture helps chickens foraging, vigilance and social activities; while they may lower their heads when foraging or resting normally.

[0081] In one embodiment, the posture module includes a first posture unit, a second posture unit, a third posture unit, a fourth posture unit and a fifth posture unit.

[0082] The first posture unit is used to identify the head area and the tail area of ​​each frame of the chicken image in the chicken image, and determine the head feature points of the head area and the tail feature points of the tail area.

[0083] Optionally, the flock image includes several frames of flock images, and the first posture unit identifies all farmed chickens in each frame of the flock image through a pre-trained YOLO target recognition model, and identifies the head region and tail region of each farmed chicken, thereby obtaining the head region and tail region of each frame of the flock image. At the same time, the centroid of the head region of each farmed chicken is used as a head feature point, denoted as G, and the centroid of the tail region is used as a tail feature point, denoted as B.

[0084] The second posture unit is used to determine the shooting direction of the camera that shoots the chicken flock image, and to determine the orientation of each farmed chicken in each chicken flock image.

[0085] Optionally, the shooting direction of the camera for shooting the image of the chicken flock is determined based on the gyroscope inside the camera, such as Figure 2 As shown, taking the orientation of a camera and a farmed chicken as an example, the relationship between the shooting direction, the orientation of the farmed chicken (referred to as chicken), the mapping direction of the shooting direction, and the vertical direction is shown in FIG. Figure 2 As shown in the figure, the orientation of the farmed chicken is the projection direction of the line connecting the head feature point and the tail feature point on the ground. It should be noted that, for example, there are cameras A1 and A2. In the image of the flock of chickens captured by camera A1, there is a flock of chickens image B, and there is a farmed chicken C in the flock of chickens image B. At this time, the shooting direction of the flock of chickens image B refers to the shooting direction of camera A1, and the orientation of the farmed chicken C refers to its orientation in the flock of chickens image B at this time.

[0086] The third posture unit is used to determine the static effective image of each farmed chicken from each frame of the chicken group image according to the shooting direction and orientation.

[0087] First, determine the angle between each orientation and the mapping direction of the shooting direction (i.e. the projection direction of the camera's shooting direction on the ground). .

[0088] Secondly, determine the direction angle A first absolute value of the difference from the first preset angle.

[0089] Then, when the first absolute value is less than or equal to the first angle threshold, the image of the chicken flock corresponding to the orientation is used as the target image, and the target image is used as the static valid image of the farmed chickens corresponding to the orientation.

[0090] It should be noted that the first angle threshold is based on actual settings. In this application, 30 For example; the first preset angle is not limited to 90 , because if the direction angle The closer to 90 (i.e. close to the side view), the clearer the chicken’s posture information will be. Therefore, determine the direction angle With 90 The first absolute value of the difference: , when the first absolute value Less than or equal to 30 , the angle between the directions The image of the flock of chickens corresponding to the orientation is used as the target image, and the target image is used as the static effective image of the farmed chickens corresponding to the orientation, so as to determine the static effective image of each farmed chicken. The static effective image is used to ensure that the head and tail of each farmed chicken can be clearly identified.

[0091] The fourth posture unit is used to cut out the corresponding image slice of the farmed chicken from each static valid image. In the vertical direction, when the head feature point in the image slice is higher than the tail feature point, the image slice is determined to be a head-up image; when the head feature point in the image slice is lower than the tail feature point, the image slice is determined to be a head-down image; a head-up image set of each farmed chicken is determined based on several head-up images, and a head-down image set of each farmed chicken is determined based on several head-down images.

[0092] Optionally, after all static valid images are determined, corresponding image slices of farmed chickens are cut out from each static valid image. Figure 2 As shown in the figure, in the vertical direction, when the head feature point in the image slice is higher than the tail feature point, the image slice is determined to be a head-up image; and when the head feature point in the image slice is lower than the tail feature point, the image slice is determined to be a head-down image. Finally, a set of head-up images of each farmed chicken is constructed using several head-up images, and a set of head-down images of each farmed chicken is constructed using several head-down images, that is, each farmed chicken has a corresponding head-up image set and head-down image set.

[0093] The fifth posture unit is used to determine the head-up posture health rate and the head-down posture health rate of each farmed chicken based on the head-up image set and the head-down image set.

[0094] It should be noted that due to the influence of physiological structure and survival habits, the basic posture of chickens is very monotonous and can be divided into two types: raising the head and lowering the head. The action of raising the head can often easily and directly reflect the mental state of the chicken, while the factors that lead to the action of lowering the head are more complicated, so further analysis is needed.

[0095] First, if Figure 3 As shown, taking a farmed chicken as an example, the first centroid point O of each head-up image in each head-up image set, the first connecting line between each first centroid point O and the head feature point G of the corresponding head-up image are determined. , First Connection The first length and the first connection The vertical line pointing upward (i.e. Figure 3 The first angle of 90° in the direction of .

[0096] Secondly, according to the first length and the first angle The cosine value of The Euclidean norm of the head-up image is obtained to obtain the head-up coefficient of each head-up image corresponding to each farmed chicken. Among them, the first length represents the elongated length of the chicken head, and the first angle The smaller the cosine value, the larger the chicken's head is. Therefore, the larger the Euclidean norm of the two, the better the chicken's head-raising posture.

[0097] Then, the head-up coefficients of each head-up image corresponding to each farmed chicken are averaged to obtain the head-up posture health rate of each farmed chicken. For example, there are 10 head-up coefficients of each head-up image corresponding to a farmed chicken. At this time, the 10 head-up coefficients are summed and divided by 10 to obtain the average. The average value obtained at this time is the head-up posture health rate of each farmed chicken, which is recorded as , that is, The head-raising posture health rate of farmed chickens.

[0098] Finally, the head-down posture health rate of each farmed chicken is determined based on the head-down image collection.

[0099] 1. If Figure 4 As shown, taking a farmed chicken as an example, the second centroid point of each head-down image in each head-down image set, and the second connecting line between each second centroid point and the head feature point G of the corresponding head-down image are determined. , Second connection The second length and the second connection The vertical line pointing downward (i.e. Figure 4 The second angle of 270° .

[0100] 2. Determine a second absolute value of the difference between the second included angle and the second preset angle.

[0101] Optionally, the second preset angle is Figure 4 Take 315° in as an example, and then determine the second angle The second absolute value of the difference from the second preset angle is recorded as , = It should be noted that when a healthy chicken lowers its head to peck, its neck will be stretched as long and far as possible. Therefore, 315° is used as the second preset angle. When the chicken lowers its head, it will be 45 degrees downward (i.e. Figure 4 315° in the figure is the theoretically achievable maximum ground contact distance. The closer to this angle, the The smaller the number, the more energetic the chicken is, and the closer it is to stretching its neck to peck. The bigger the number, the more likely the chicken is lowering its head for non-pecking purposes, which may indicate health problems.

[0102] 3. Determine the third line between the second centroid point of each head-down image and the tail feature point B of the corresponding head-down image And the second connection Connect with the third The third angle between .

[0103] 4. Calculate the difference between the third angle and the third preset angle to obtain a fourth angle, retain the angle value of the fourth angle that is greater than the second angle threshold, and return the angle value of the fourth angle that is less than or equal to the second angle threshold to zero.

[0104] Optionally, the third preset angle is 180° and the second angle threshold is 0°, for example, the third angle It is the angle between the head feature point, the center of mass, and the tail feature point of the farmed chicken. When a healthy farmed chicken lowers its head to peck normally, its tail will remain stable. Therefore, the line connecting the head, body, and tail is close to a straight line. By calculating the third angle The difference between the third preset angle and the fourth angle is obtained , = When the fourth angle When it is greater than 0°, it means that the head of the chicken is too droopy or the tail is too droopy. The angle value is retained, and the fourth angle less than or equal to 0° All angle values ​​are reset to zero.

[0105] 5. According to the second length and the second absolute value The cosine value of and the fourth angle The cosine value of The Euclidean norm of each chicken is used to obtain the head-down coefficient of each head-down image corresponding to each farmed chicken. , that is, The head-down coefficient of the head-down image.

[0106] 6. Determine the clustering factor of each farmed chicken based on the acquisition time of each head-down image in the head-down image set.

[0107] It should be noted that, unlike the action of raising its head, the chicken's head-lowering action is generally purposeful. For gregarious and social chickens, if the chickens all lower their heads at the same time at a similar time, it is most likely normal. Otherwise, it means that a certain chicken may be abnormal. This is because the chickens' daily routines are almost the same. Therefore, when the chickens all lower their heads, it means that they are eating or resting together. If individual chickens always lower their heads when they are not eating or resting, it is more likely to be abnormal. The analysis is based on this principle.

[0108] First, determine the acquisition time of each head-down image in the head-down image set, determine the target head-down image set from the head-down image set, determine the target head-down image from the target head-down image set, and determine the adjacent acquisition time of the head-down image closest to the acquisition time of the target head-down image from other head-down image sets, for example Figure 5 As shown, taking four sets of head-down images as an example, i.e., four farmed chickens, the target head-down image is 61, and in other head-down image sets, the closest head-down images can be determined as 62, 63, and 64 respectively, and the acquisition times corresponding to 62, 63, and 64 are the respective adjacent acquisition times.

[0109] Secondly, calculate the absolute value of the time difference between the acquisition time of the target head-down image and each adjacent acquisition time. The head-down image, whose acquisition time is recorded as , the rest The adjacent acquisition time of the head-down image closest to the acquisition time of the target head-down image in the set of head-down images of chickens is , calculate the absolute value of the time difference between the acquisition time of the target head-down image and each adjacent acquisition time: .

[0110] Then, according to the absolute value of the time difference And the number of farmed chickens N, determine the absolute value of the total time difference According to the first ratio of the absolute value of the total time difference and the number of farmed chickens, we get the first Clustering factors of head-down images , .

[0111] Furthermore, return to the step of determining the target head-down image from the target head-down image set until the corresponding clustering factor of each target head-down image in the target head-down image set is determined, and return to the step of determining the target head-down image set from the head-down image set until the corresponding clustering factor of each head-down image in each head-down image set is determined.

[0112] Finally, the clustering factor of each farmed chicken is determined according to the clustering factors corresponding to each head-down image in each head-down image set. That is, the clustering factor of each farmed chicken includes the clustering factors corresponding to each head-down image in its head-down image set. Among them, the cluster factor The smaller the value, the more likely it is that the chickens were eating or resting in unison at the time the head-down image was acquired. Therefore, at this acquisition time, the head-down posture of the farmed chickens may be a normal activity.

[0113] 7. Determine the head-down posture health rate of each farmed chicken based on the clustering factor and the head-down coefficient of each head-up image corresponding to each farmed chicken.

[0114] First, calculate the head-down coefficient of each head-down image corresponding to each farmed chicken The corresponding clustering factors The second ratio , and determine the normalized result of the second ratio , the larger the numerator and the smaller the denominator, the more The more normal the head-down posture of the corresponding farmed chicken in the lowered head portrait is.

[0115] Then, the average processing is performed based on the multiple normalized results of the head-down image set of each farmed chicken, that is, the multiple normalized results of the head-down image set of each farmed chicken are summed up, and then divided by the number of head-down images, and finally the head-down posture health rate of each farmed chicken is obtained. , that is, the health rate of the head-down posture of the oth chicken .

[0116] It should be noted that chickens are gregarious animals. They exhibit complex social behaviors in both natural and artificial environments. There is usually a clear hierarchy in the flock, also known as the "pecking order". This hierarchy is established through interactions between individuals, usually reflected in pecking, roosting and mating. Chickens in a higher position will have priority in enjoying food and resources, so the "pecking order" of the flock is extracted, because only the healthier the chickens are, the more motivated and active they are, and the higher the "pecking order".

[0117] In one embodiment, the pecking module includes a first pecking unit, a second pecking unit, a third pecking unit, a fourth pecking unit and a fifth pecking unit.

[0118] The first pecking unit is used to determine a number of data points of the sound signal, the sound source position of each data point, and the sound amplitude and frequency value of each data point.

[0119] Optionally, in the field of chicken farming, in order to facilitate the management of the flock, the breeder generally only introduces one flock of chickens into a chicken pen. However, in addition to the leader chicken, there are generally multiple secondary leader chickens in the flock. There may even be phenomena such as the leader chicken being replaced and multiple small flocks being formed inside. Therefore, it is impossible to directly locate and identify the leader chicken. Therefore, in this application, the acquired sound signal is used to identify each chicken call through a pre-trained chicken call sound recognition model, and the sound source is identified by sound source positioning, which can correspond to a number of call data, that is, a number of data points of the sound signal, each data point includes its sound source position, sound amplitude and frequency value, such as Figure 6 As shown, the data points are projected onto a two-dimensional plane for visualization, resulting in a scatter plot of each data point.

[0120] It should be noted that since chickens high in the pecking order usually make louder and higher-frequency calls, such calls can convey their dominance and warn other chickens not to invade their territory or resources; the calls of chickens low in the pecking order may be lower and gentler, with a lower frequency and volume, to avoid provoking chickens high in the pecking order, thereby reducing conflicts and attacks; data points with low amplitude and frequency may correspond to chickens low in the pecking order, or may be chickens with health problems and are not credible, while data points with high amplitude and high frequency must correspond to chickens high in the pecking order.

[0121] The second pecking unit is used to determine the search range based on a preset radius and with the sound source position of each data point as the center, determine the number of other data points in each search range, and determine the average number of points based on the total number of points and the number of search ranges.

[0122] Optionally, taking the preset radius of 1m as an example, according to the preset radius, the search range is determined with the sound source position of each data point as the center, the number of other data points in each search range is determined, and the average number of points is determined according to the sum of the number of points and the total number of data points, which is recorded as For example, there are 3 search ranges, and the number of points of other data points in the search ranges are 10, 8, and 6 respectively. At this time, the total is 24, and there are 3 search ranges, so the average number of points is 8.

[0123] The third pecking unit is used to mark the data point as a maximum point when the sound amplitude and frequency value of the data point at the center are correspondingly greater than the sound amplitude and frequency value of other data points, retain the target maximum point centered on the maximum point and whose corresponding number of points is greater than or equal to the average number of points, and regard each farmed chicken corresponding to the target maximum point as a suspected leading chicken.

[0124] Optionally, when the sound amplitude and frequency value of the data point located at the center are both greater than the sound amplitude and frequency value of other data points, the data point is marked as a maximum point. This is because each flock of chickens follows the leading chicken and the flock of chickens is in a state of clustering and surrounding the leading chicken. The leading chicken is surrounded by chickens that are behind it in the pecking order. Then, in order to avoid the influence of isolated data points on the analysis results, the target maximum point with the maximum point as the center and the number of corresponding points greater than or equal to the average number of points is retained. Assuming that there are Q target maximum points remaining, each farmed chicken corresponding to the Q target maximum points is regarded as a suspected leading chicken, so that the target maximum point can locate the sound source of each farmed chicken, and Q suspected leading chickens corresponding to the Q target maximum points can be obtained.

[0125] The fourth pecking unit is used to determine the target activity trajectory of each suspected leader chicken from the activity trajectory.

[0126] Optionally, based on the Q suspected leader chickens, a target activity trajectory of each suspected leader chicken is determined from previously acquired activity trajectories.

[0127] The fifth pecking order unit is used to determine the pecking order of the chickens according to the activity trajectory and the target activity trajectory.

[0128] 1. Calculate the mean square error between the activity trajectory of each farmed chicken and the target activity trajectory of each suspected leader chicken: , is the activity trajectory of any farmed chicken, It is the target activity trajectory of any suspected leader.

[0129] 2. Based on each mean square error, determine the similarity between the activity trajectory of each farmed chicken and the target activity trajectory of each suspected leader chicken , e represents a natural constant, and the specific formula is:

[0130]

[0131] 3. Through the above calculation, each farmed chicken can obtain Q trajectory similarities with Q suspected leader chickens, and then the maximum trajectory similarity is used as the followability of the chicken, that is, the activity trajectory of the maximum trajectory similarity of each farmed chicken is used as the followability of the farmed chicken.

[0132] 4. Then, arrange the followers in order from large to small to get the pecking order of the chicken flock.

[0133] In one embodiment, the health module includes a first health unit, a second health unit, a third health unit, a fourth health unit and a fifth health unit; wherein:

[0134] The first health unit is used to determine a first ratio of images of each farmed chicken raising its head to all images of the chicken flock, and a second ratio of images of each farmed chicken lowering its head to all images of the chicken flock.

[0135] Optionally, determine the first proportion of each farmed chicken's head-up image to all chicken images , and the image of each farmed chicken with its head down accounts for the second largest proportion of all chicken images .

[0136] The second health unit is used to perform weighted calculation based on the head-up posture health rate, the first proportion, the head-down posture health rate and the second proportion to obtain the health posture coefficient of each farmed chicken, and sort the health posture coefficients of all farmed chickens from large to small to obtain the posture sequence corresponding to all farmed chickens.

[0137] Optionally, a weighted calculation is performed based on the head-up posture health rate, the first proportion, the head-down posture health rate, and the second proportion to obtain a health posture coefficient for each farmed chicken. The specific calculation formula is:

[0138]

[0139] in, For the The health posture coefficient of only farmed chickens, For the The head-raising posture health rate of farmed chickens, For the The head-down posture health rate of farmed chickens. It should be noted that the health posture coefficient can describe the static posture characteristics of each farmed chicken. The larger the value, the healthier the static posture of the chicken; then, the health posture coefficients of all farmed chickens are sorted from large to small to obtain the posture sequence corresponding to all farmed chickens.

[0140] In the embodiment of the present application, the posture sequence represents the posture health of the farmed chickens, and the "pecking order" represents the social characteristics of the farmed chickens. The lower the posture health of the farmed chickens and the lower the pecking order, the higher the possibility of pathology. In addition, for each farmed chicken, the greater the difference in its sequence number between the two sequences, the less the chicken's posture conforms to its social characteristics. There are two possibilities when the difference between the two sequences is too large: the first is that the posture seems healthy but is rejected by the flock; the other is that the chicken is active in the flock but the posture is abnormally sluggish. Both represent that the farmed chickens may have health problems. The present application performs subsequent calculations based on this principle.

[0141] The third health unit is used to respectively determine the sum and absolute value of the difference between the first sequence number in the posture sequence and the second sequence number in the pecking sequence of each farmed chicken.

[0142] Optionally, determine the first sequence number of each farmed chicken in the posture sequence and the second number in the pecking order The sum of and the absolute value of the difference , Representative Only raise chickens.

[0143] The fourth health unit is used to determine the calculation result according to the natural exponential function and the absolute value of the difference.

[0144] Optionally, for the natural exponential function and the absolute value of the difference , determine the calculation results . e represents a natural constant.

[0145] Fifth health unit, used to determine the average of the sum values With the calculation results The third ratio , and normalize the third ratio , get the health abnormality probability of each farmed chicken ,Right now , representing The abnormal health probability of farmed chickens represents the health status of farmed chickens.

[0146] It should be noted that The larger the value, the Farming only chickens is further back in both sequences; The larger the value, the The more the chicken is bred, the less it conforms to its social characteristics; The larger the numerator, the smaller the denominator, which means The more abnormal it is to only raise chickens.

[0147] Optionally, a health threshold can be set in advance, for example, the health threshold is 0.7. Chickens with a health abnormality probability greater than or equal to the health threshold are determined to be potentially abnormal farmed chickens and need to be captured, diagnosed, and isolated for observation to prevent possible disease risks in a timely manner.

[0148] In the embodiment of the present application, in order to avoid the problem of sample information differences caused by the action characteristics of chickens, the health status of farmed chickens is not analyzed with continuous posture change data, but the head feature points and tail feature points are analyzed based on the static chicken group image of each chicken to judge its health status, especially for the samples of head-down posture, the purpose of chicken pecking and cluster characteristics are combined to obtain the head-down posture health rate. In addition, judging the health status of farmed chickens based on the head-up and head-down actions is simple and efficient compared to the complex deep learning recognition method; in the embodiment of the present application, it is believed that due to the action characteristics of chickens, a large amount of status information is contained in the static frame of chickens, and there is no need to spend a lot of resources to learn invalid behaviors and action patterns. According to the head-up posture health rate and the head-down posture health rate in the static frame, combined with multimodal data analysis, whether the sociality of the chicken group is consistent with the posture health characteristics, the health status of farmed chickens can be comprehensively evaluated.

[0149] It should be noted that the sequence of the above embodiments of the present invention is only for description and does not represent the advantages and disadvantages of the embodiments. The processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0150] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referenced to each other, and each embodiment focuses on the differences from other embodiments.

Claims

1. Chicken health status analysis system based on data analysis, It is characterized in that include: An acquisition module is used to acquire images of chickens in a chicken farm, sound signals, and activity tracks of each chicken in a preset time interval; A posture module, used to determine the head-up posture health rate and head-down posture health rate of each farmed chicken based on the chicken flock image; A pecking order module, used to determine the pecking order of the chickens according to the sound signal and the activity trajectory; A health module, used to determine the health status of each farmed chicken according to the head-up posture health rate, the head-down posture health rate and the pecking order; The posture module includes a first posture unit, a second posture unit, a third posture unit, a fourth posture unit and a fifth posture unit; The first posture unit is used to identify the head area and the tail area of ​​each frame of the chicken image in the chicken image, and determine the head feature points of the head area and the tail feature points of the tail area; The second posture unit is used to determine the shooting direction of the camera that shoots the chicken flock image, and to determine the orientation of each farmed chicken in each of the chicken flock images; The third posture unit is used to determine a static valid image of each farmed chicken from each frame of the chicken group image according to the shooting direction and the orientation; The fourth posture unit is used to cut out the corresponding image slice of the farmed chicken from each of the static valid images, and in the vertical direction, when the head feature point in the image slice is higher than the tail feature point, the image slice is determined to be a head-up image, and when the head feature point in the image slice is lower than the tail feature point, the image slice is determined to be a head-down image, and a head-up image set of each farmed chicken is determined based on a plurality of head-up images, and a head-down image set of each farmed chicken is determined based on a plurality of head-down images; The fifth posture unit is used to determine the head-up posture health rate and the head-down posture health rate of each farmed chicken according to the head-up image set and the head-down image set; The pecking module includes a first pecking unit, a second pecking unit, a third pecking unit, a fourth pecking unit and a fifth pecking unit; A first pecking unit, used for determining a plurality of data points of the sound signal, a sound source position of each of the data points, and a sound amplitude and frequency value of each of the data points; A second pecking unit is used to determine a search range with the sound source position of each data point as the center according to a preset radius, determine the number of other data points in each search range, and determine an average number of points according to the sum of the number of points and the number of the search range; The third pecking unit is used to mark the data point as a maximum point when the sound amplitude and frequency value of the data point located at the center are both greater than the sound amplitude and frequency value of other data points, retain the target maximum point centered on the maximum point, the number of corresponding points of which is greater than or equal to the average number of points, and take each farmed chicken corresponding to the target maximum point as a suspected leader chicken; A fourth pecking order unit is used to determine a target activity trajectory of each of the suspected leader chickens from the activity trajectory; a fifth pecking order unit, configured to determine the pecking order of the chickens according to the activity trajectory and the target activity trajectory; Determining the pecking order of the chickens according to the activity trajectory and the target activity trajectory includes: Calculating the mean square error between the activity trajectory of each farmed chicken and the target activity trajectory of each suspected leader chicken respectively; Determine, based on each of the mean square errors, the trajectory similarity between the activity trajectory of each farmed chicken and the target activity trajectory of each of the suspected leading chickens; The activity trajectory of each farmed chicken with the maximum trajectory similarity is taken as the followability of the farmed chicken; The following properties are arranged in descending order to obtain the pecking order of the chickens.

2. According to the chicken health status analysis system based on data analysis according to claim 1, Features: Determining a static valid image of each farmed chicken from each frame of the chicken group image according to the shooting direction and the orientation includes: Determine the angle between each of the directions and the mapping direction of the shooting direction; Determine a first absolute value of a difference between the direction angle and a first preset angle; When the first absolute value is less than or equal to a first angle threshold, the image of the chicken flock corresponding to the orientation is used as a target image, and the target image is used as a static valid image of the farmed chickens corresponding to the orientation.

3. According to the chicken health status analysis system based on data analysis in claim 1, Features: Determining the head-up posture health rate and the head-down posture health rate of each farmed chicken according to the head-up image set and the head-down image set includes: Determine a first centroid point of each head-up image in each head-up image set, a first connecting line between each first centroid point and a head feature point of the corresponding head-up image, a first length of the first connecting line, and a first angle between the first connecting line and a vertical line in a vertical direction and upward; Obtaining a head-up coefficient of each head-up image corresponding to each farmed chicken according to a Euclidean norm of a cosine value of the first length and the first angle; The head-raising coefficients of each head-raising image corresponding to each farmed chicken are averaged to obtain the head-raising posture health rate of each farmed chicken; The head-down posture health rate of each farmed chicken is determined based on the head-down image set.

4. The chicken health status analysis system based on data analysis according to claim 3, Features: Determining the head-down posture health rate of each farmed chicken according to the head-down image set includes: Determine a second centroid point of each of the head-down images in each of the head-down image sets, a second connecting line between each of the second centroid points and the corresponding head feature point of the head-down image, a second length of the second connecting line, and a second angle between the second connecting line and a vertical line in a vertical direction and downward; Determine a second absolute value of a difference between the second angle and a second preset angle; Determine a third connecting line between the second centroid point of each of the head-down images and the corresponding tail feature point of the head-down image, and a third angle between the second connecting line and the third connecting line; Calculate the difference between the third angle and the third preset angle to obtain a fourth angle, retain the angle value of the fourth angle that is greater than the second angle threshold, and reset the angle value of the fourth angle that is less than or equal to the second angle threshold to zero; Obtaining a head-down coefficient of each head-down image corresponding to each farmed chicken according to the second length, the cosine value of the second absolute value, and the Euclidean norm of the cosine value of the fourth angle; Determine the clustering factor of each farmed chicken according to the acquisition time of each of the head-down images in the head-down image set; The head-down posture health rate of each farmed chicken is determined according to the clustering factor and the head-down coefficient of each head-down image corresponding to each farmed chicken.

5. The chicken health status analysis system based on data analysis according to claim 4, Features: Determining the clustering factor of each farmed chicken according to the acquisition time of each of the head-down images in the head-down image set includes: Determine the acquisition time of each of the head-down images in the head-down image set, determine a target head-down image set from the head-down image set, determine a target head-down image from the target head-down image set, and determine adjacent acquisition times of head-down images closest to the acquisition time of the target head-down image from other head-down image sets; Calculate the absolute value of the time difference between the acquisition time of the target head-down image and each of the adjacent acquisition times; Determine the total absolute value of the time difference according to the absolute values ​​of the time differences and the number of farmed chickens; Obtaining a clustering factor corresponding to the target head-down image according to a first ratio of the total absolute value of the time difference to the number of farmed chickens; Return to the step of determining the target head-down image from the target head-down image set until the corresponding clustering factor of each target head-down image in the target head-down image set is determined, and return to the step of determining the target head-down image set from the head-down image set until the corresponding clustering factor of each head-down image in each head-down image set is determined; The clustering factor of each farmed chicken is determined according to the clustering factors corresponding to each of the head-down images in each of the head-down image sets.

6. The chicken health status analysis system based on data analysis according to claim 4, Features: Determining the head-down posture health rate of each farmed chicken according to the clustering factor and the head-down coefficient of each head-down image corresponding to each farmed chicken includes: Calculating respectively a second ratio of the head-down coefficient of each head-down image corresponding to each farmed chicken to the clustering factor, and determining a normalized result of the second ratio; The head-down posture health rate of each farmed chicken is obtained by averaging multiple normalized results of the head-down image set of each farmed chicken.

7. The chicken health status analysis system based on data analysis according to claim 1, Features: The health module includes a first health unit, a second health unit, a third health unit, a fourth health unit and a fifth health unit; A first health unit is used to determine a first ratio of the head-up image of each farmed chicken to all images of the chicken flock, and a second ratio of the head-down image of each farmed chicken to all images of the chicken flock; A second health unit is used to perform weighted calculation according to the head-up posture health rate, the first proportion, the head-down posture health rate and the second proportion to obtain a health posture coefficient of each farmed chicken, and sort the health posture coefficients of all farmed chickens from large to small to obtain a posture sequence corresponding to all farmed chickens; A third health unit is used to respectively determine the sum and the absolute value of the difference between the first sequence number of each farmed chicken in the posture sequence and the second sequence number in the pecking sequence; A fourth health unit, used for determining a calculation result according to a natural exponential function and an absolute value of the difference respectively; The fifth health unit is used to determine a third ratio of the average value of the sum value to the calculation result, and normalize the third ratio to obtain the health abnormality probability of each farmed chicken, wherein the health abnormality probability represents the health status of the farmed chicken.