Dead chicken detection method
By using a sliding device moving along the track in a cage chicken farming facility to capture stationary and dynamic images, combined with machine learning and infrared technology, the rapid and accurate detection of chicken deaths is achieved, and the problem of labor burden and egg distinction is solved, and the problem of differentiation of labor is improved. The management efficiency of the chicken farming environment.
Patent Information
- Application Number
- CN202280094077.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2022-04-05
- Publication Date
- 2024-11-12
AI Technical Summary
In large-scale cage chicken farming facilities, chicken deaths are difficult to detect quickly and accurately, resulting in a large labor burden and may affect the sanitary environment of live chickens. It is difficult to distinguish between old eggs and new eggs, which is easy to confuse.
A sliding device moving along the track is provided on the cage laminated body, and a camera is equipped to perform still image shooting and dynamic image shooting. Combined with the first and second life and death judgment devices, a machine learning model and infrared ray are used to compare the still image and dynamic image. Camera, accurately detect dead chickens.
It reduces the labor burden of operators, detects dead chickens quickly and accurately, avoids misjudgment of live chickens and confusion between old and new eggs, and improves the hygiene of the cage environment and the accuracy of egg recycling.
Smart Images

Figure CN118946264A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to a dead chicken detection method for detecting when chickens die in a chicken breeding facility where chickens are raised in cages. Background Art
[0002] In large-scale chicken farming facilities, a large number of chickens are raised using cage stacks, which are stacked with multiple layers of cages, each layer of which has multiple cages arranged in parallel in the horizontal direction. In such chicken farming facilities, feeding, watering, egg collection, and feces removal are mechanized, so that a large number of chickens can be raised with a small number of operators.
[0003] Therefore, even if there is chicken death in the cage, the operator is also difficult to find this situation. If the operator is to regularly patrol and check the situation of the chicken, then because the number of cages is very large, the operator's labor burden is large and also takes time.
[0004] If the dead chickens are left alone, the sanitary environment in the cages will deteriorate, and there is a possibility that it will also have an adverse effect on the live chickens. In addition, in chicken farming facilities that produce eggs, conveyor belts or trays for picking up eggs are generally set up along the cages, and the floor of the cages is tilted so that the eggs fall onto the conveyor belt by their own weight for collection. Therefore, the bodies of the dead chickens will become obstacles and hinder the eggs from falling. If the eggs are not collected until a long time later, the old eggs will be mixed with the new eggs. Since new eggs and old eggs cannot be distinguished from each other in appearance, such a situation must be avoided.
[0005] In response to this problem, the following suggestion has been made: multiple surveillance cameras are installed in the facility to film the cages and reflect the images on the surveillance television in the management room (see Patent Document 1). However, in this case, although the labor of operators patrolling the facility can be saved, the operators must always monitor the surveillance television, which still imposes a heavy labor burden.
[0006] Therefore, when chickens die in cages, a technology is desired that can reduce the workload of operators and detect the situation quickly and accurately.
[0007] Prior Art Literature
[0008] Patent Literature
[0009] Patent Document 1: Japanese Patent Publication No. 07-085696 Summary of the invention
[0010] Problems to be solved by the invention
[0011] Therefore, in view of the above-mentioned actual situation, the present invention aims to provide a dead chicken detection method that can reduce the workload of operators and detect the situation quickly and accurately when a chicken dies in a cage.
[0012] Solutions to Solve Problems
[0013] In order to solve the above-mentioned problems, the dead chicken detection method involved in the present invention is:
[0014] "A facility for raising chickens using a cage stack, wherein the cage stack comprises multiple layers of cages arranged side by side in a horizontal direction, wherein:
[0015] A track extending in the same direction as the cage array is provided on the cage array stack.
[0016] On the sliding device that reciprocates along the track, multiple cameras are provided corresponding to the height of each cage row.
[0017] The camera takes at least one still image of each of the cages, and the first determination device performs a first life or death determination on whether the still image contains dead chickens.
[0018] In this method, at least one still image is taken for each cage to determine whether the cage to be photographed contains dead chickens. The camera that takes the still image is equipped on a slide device that slides along a track provided on the cage row stack. If the camera is equipped on a robot device that moves along a passage in a facility, complex control is required to move the robot device along a predetermined path. In contrast, in this method, since the camera is equipped on a slide device that slides along a track provided on the cage row stack, control of moving the camera for photographing is simple.
[0019] In addition, since the slide device reciprocates along the track, a still image can be captured both during forward and reverse movement. That is, still images of the same cage can be acquired at a certain time interval. Therefore, still images taken at different times can be compared, and if chickens (unmoving chickens) are detected in the same position in the two images, there is a high possibility that the chickens are dead. Therefore, dead chickens can be detected more accurately than when still images taken at different times are not compared.
[0020] Furthermore, when a robot device that moves along a passage in a facility is equipped with a camera, a portable battery is usually used as a power source. Therefore, the distance that can be moved at one time is limited, and charging also takes time, so it is difficult for the robot device to take pictures frequently. In contrast, in the present invention, since a sliding device that moves along a structure fixed to a facility such as a cage column stack is equipped with a camera, power for movement and shooting can be supplied in a wired manner. Therefore, high-frequency shooting can be performed, and dead chickens can be detected more accurately by comparing multiple still images taken at different times.
[0021] In addition to the above-mentioned structure, the dead chicken detection method of the present invention can also be constituted as follows:
[0022] "As a result of the first life and death judgment, when it is determined that the still image contains a dead chicken, a second life and death judgment is performed using a moving image captured of a cage as the same shooting object to determine whether the moving image contains a dead chicken."
[0023] When determining whether there are dead chickens in a cage based only on a still image of the cage, there is a possibility that a sitting live chicken may be mistakenly determined as a dead chicken. In contrast, in this method, when it is determined that a dead chicken is included by the first life and death determination based on the still image, a second life and death determination is performed using a moving image taken for the same object. Therefore, even if the chicken looks like a dead chicken in the still image, the live chicken confirmed to be moving in the moving image can be distinguished from the dead chicken, so the life and death of the chicken in the cage can be accurately determined.
[0024] Here, as methods of "when it is determined as a result of the first life and death determination that the still image contains a dead chicken, performing a second life and death determination using a dynamic image shot of the cage as the same shooting object", there can be listed: a method in which a determination is made as a result of the first life and death determination that the still image contains a dead chicken, and a dynamic image is shot for the second life and death determination; a method in which a dynamic image is always shot, and when it is determined as a result of the first life and death determination that the still image contains a dead chicken, a second life and death determination is performed using dynamic images before and after that time point.
[0025] In the former case, a moving image is captured only when a dead chicken is determined as a result of the first life or death determination in the still image, and thus the moving image can be captured by the same camera as the camera mounted on the slide device and capturing the still image. Therefore, dead chickens can be detected with a simple device.
[0026] In the latter case, although the moving images are always captured, it is not necessary to monitor the moving images all the time, nor is it necessary to analyze all the moving images. Only when it is determined as a result of the first life and death determination that a dead chicken is included in the still image, the moving images before and after are used. Therefore, the moving images can be used efficiently.
[0027] In any case, at least the first judgment device performs the first life or death judgment. Therefore, in the process of judging the life or death of the chicken, many mechanized (automated) parts are used, so the workload of the operator can be reduced and the judgment can be performed quickly.
[0028] The dead chicken detection method involved in the present invention can be constructed as follows in the above-mentioned structure:
[0029] "The second life or death judgment is performed based on the result obtained by inputting the dynamic image into a learned model, which is generated by machine learning using teaching dynamic images containing only live chickens and teaching dynamic images containing dead chickens as learning data."
[0030] This is the structure that specifically specifies the second method of determining life and death.
[0031] The dead chicken detection method involved in the present invention can be constructed as follows in the above-mentioned structure:
[0032] "The above-mentioned first life or death judgment is any one of the following judgments: a judgment based on the result of image analysis of the above-mentioned still image regarding characteristic points specific to a dead chicken or a live chicken; a judgment based on the result obtained by inputting the above-mentioned still image into a learned model, wherein the learned model is generated by machine learning using teaching still images containing only live chickens and teaching still images containing dead chickens as learning data; a judgment based on the result of visualizing the temperature of the photographed object by an infrared camera; and a combination of two or more of these judgments."
[0033] This is the structure that specifically specifies the first method of determining life and death.
[0034] In addition to the above-mentioned structure, the dead chicken detection method of the present invention can also be constituted as follows:
[0035] "The cage array stack is provided with a feeding trough in each cage array.
[0036] The sliding device includes a plurality of hoppers for supplying bait to the feeding troughs respectively.
[0037] In this method, the slide device serves as both a device for moving the camera and a device for moving the hopper for supplying food to the trough. Therefore, it has the advantage that it contributes to both the automation of food supply and the automation of dead chicken detection as a single device.
[0038] Effects of the Invention
[0039] As described above, according to the present invention, when a chicken dies in a cage, it is possible to provide a dead chicken detection method that can reduce the workload of workers and detect the situation quickly and accurately. BRIEF DESCRIPTION OF THE DRAWINGS
[0040] Figure 1 This is a perspective view showing a cage array stacked body equipped with a sliding device with a portion omitted.
[0041] Figure 2 yes Figure 1 Side view of a cage array stack.
[0042] Figure 3 yes Figure 1 A three-dimensional view of the sliding device.
[0043] Figure 4 This is a block diagram illustrating the schematic configuration of a dead chicken detection system.
[0044] Figure 5 (a) is a schematic diagram of a cage with only live chickens. Figure 5 (b) is a schematic diagram of a cage with dead chickens.
[0045] Figure 6 This is a flowchart illustrating the processing in the dead chicken detection method according to the first embodiment.
[0046] Figure 7 This is a perspective view showing a cage array stacked body having other sliding devices with a portion omitted.
[0047] Figure 8 yes Figure 7 A three-dimensional view of the sliding device.
[0048] Fig. 9 (a) and Fig. 9 (b) are block diagrams showing other configurations of the dead chicken detection system. DETAILED DESCRIPTION
[0049] Hereinafter, a dead chicken detection method and a dead chicken detection system using the detection method as a specific embodiment of the present invention will be described with reference to the accompanying drawings. Figure 1 to Figure 6 The first embodiment will be described.
[0050] The chicken raising facility to which this embodiment is applied raises chickens using a cage array stack 90L having a plurality of cage arrays 90R stacked in a horizontal direction. A long feeding trough 85 for feeding is arranged horizontally along each cage array 90R.
[0051] The cage array stack 90L includes: a plurality of planar structures formed by a plurality of wires or rods arranged at intervals; and a plurality of longitudinal members 81 and a plurality of transverse members 82 forming a framework supporting the planar structures. The planar structure is a structure in which the wires or rods are crossed in a lattice, or a structure in which the wires or rods are adjacent to each other in parallel and at intervals in only one direction. In the cage array 90R, the side where the trough 85 is arranged is called the "front", and the opposite side is called the "rear". The front part 91 covering each cage 90 from the front, the back part 92 covering from the rear, the floor surface part 94 covering from the bottom, and the side surface part 93 dividing the adjacent cages 90 from each other are respectively composed of planar structures. In addition, the cage array 90R can also be configured to also include a top surface part covering the cage 90 from the top.
[0052] The trough 85 is provided at a position slightly higher than the floor surface 94 so that the chicken sticking its head out of the cage 90 can peck at the bait. Here, in the case of raising egg-laying chickens, an egg tray 86 for receiving eggs is arranged below the trough 85. The floor surface 94 is inclined so as to descend toward the egg tray 86. This is to allow the eggs laid by the chickens in the cage 90 to roll forward on the floor surface 94 by their own weight and be received by the egg tray 86 formed in front of the floor surface 94.
[0053] In most chicken farming facilities, as shown in the figure, two cage rows 90R are arranged adjacent to each other with their back sides 92 facing each other. In such a case, one side surface 93 is provided so as to cover two cages 90 belonging to the two adjacent cage rows 90R.
[0054] In addition, Figure 1 In the figure, the back portion, a portion of the front portion 91, the floor surface, a portion of the trough 85, and the egg tray are omitted. Figure 1 In the example, four layers of cage arrays 90R are stacked, but the number of layers is not limited to this.
[0055] The dead chicken detection system using the dead chicken detection method of the first embodiment is configured by adding the following devices to the cage array stack 90L configured as above: a slide device 10 having an imaging device 30; a slide drive mechanism 40; a control device 50; a first determination device 51; and a second determination device 52.
[0056] The slide device 10 is provided at each cage stack 90L and reciprocates in the horizontal direction in front of the cage stack 90L. The slide device 10 is photographed by the photographing device 30 during the reciprocating movement of the cage 90. Figure 1 In the figure, only one of the sliding devices 10 provided in the two cage column stacked bodies 90L is shown.
[0057] Each slide device 10 includes a slider 11 that moves along a rail 41, a support column 12 that is supported by the slider 11 and extends downward, and an imaging device 30. The slider 11 is formed with a long groove 15 that engages with the rail 41.
[0058] The number of the imaging devices 30 is the same as the number of the cage arrays 90R, which is four in this case. Each imaging device 30 is composed of a camera 31 and an illuminator 32. The camera 31 and the illuminator 32 are mounted on a bracket 39 extending horizontally from the support 12 corresponding to the height of each cage array 90R, so that the optical axis of the camera 31 faces the cage 90. The camera 31 can capture both still images and moving images. Two illuminators 32 are provided for one camera 31, located on both sides thereof, and illuminate the cage 90 when the camera 31 captures images.
[0059] The control device 50 is mounted on the support column 12. The control device 50 is composed of a computer, which mainly includes a storage device including a main storage device and an auxiliary storage device, a processor, and a communication device for communicating with the first determination device 51 and the second determination device 52. The storage device stores a program for causing the computer to function as a movement control mechanism for controlling the movement of the slide device 10 and a program for causing the computer to function as a shooting control mechanism for controlling the operation of the camera 31 and the illuminator 32.
[0060] The slide drive mechanism 40 includes, in addition to the rail 41 , a driving pulley 42 , a driven pulley 43 , a wire 44 wound around the driving pulley 42 and the driven pulley 43 , and a motor 45 that drives the driving pulley 42 to rotate.
[0061] In the cage array stack 90L, the rail 41 is provided along the upper end edge of the front end of the uppermost cage array 90R over the entire length of the cage array 90R. A driving pulley 42 is mounted on one end of the rail 41, and a driven pulley 43 is mounted on the other end, and the rail of the wire 44 wound around these pulleys is located just above the rail 41. Furthermore, by fixing a portion of the wire 44 to the slider 11, the wire 44 forms a closed loop via the slider 11.
[0062] The output shaft of the motor 45 is connected to the rotating shaft of the driving pulley 42. Thus, if the motor 45 is operated to rotate the driving pulley 42 in one direction, the wire 44 is moved in one direction while the driven pulley 43 is rotated, and the slider 11 is pulled by the wire 44, so that the sliding device 10 moves in one direction. In addition, if the motor 45 is operated to rotate the driving pulley 42 in the opposite direction, the wire 44 is moved in the opposite direction while the driven pulley 43 is rotated, and the slider 11 is pulled by the wire 44, so that the sliding device 10 moves in the opposite direction. That is, the sliding device 10 reciprocates along the cage column stack 90L through the forward and reverse rotation of the motor 45. The motor 45, the driving pulley 42, and the driven pulley 43 are supported on the frame of the cage column stack 90L. In addition, a square cylindrical guide pipe 46 is arranged along the trough 85 of the lowest cage row 90R, and a guide roller 47 installed on the support 12 abuts against the outer surface of the guide pipe 46 while rotating, thereby stabilizing the posture of the sliding device 10 and making the sliding movement smoother.
[0063] The movement control mechanism controls the slide drive mechanism 40 to move the slide device 10. When the imaging device 30 reaches a predetermined imaging point, the slide device 10 stops and sends a signal to the imaging control mechanism. The imaging control mechanism receives the signal and sends a signal to the imaging device 30 to irradiate light from the illuminator 32 and to make the camera 31 capture a still image. The imaging point is set so that at least one still image is captured for each cage 90. The captured still image is sent to the first determination device 51 in a state where cage specific information of the cage 90 of the specific imaging object is attached.
[0064] Here, cage specific information can be generated by comparing the arrangement order of the cages 90 of which the still images are captured as the slide device 10 moves and the number of still images captured until the capturing. The slide device 10 includes a plurality of imaging devices 30, and the plurality of cages 90 are captured simultaneously. The positional relationship with the cage 90 to be captured becomes clear according to the height at which each imaging device 30 is supported on the support column 12 of the slide device 10. Therefore, by attaching the identification code of each of the plurality of imaging devices 30 to the image captured by the imaging device 30, the plurality of cages 90 to be captured simultaneously can be distinguished, and cage specific information can be generated.
[0065] Alternatively, for each cage 90 , an identification code is displayed within a range captured by the imaging device 30 , and if an image can be captured including the identification code, cage-specific information can be obtained by image analysis of the identification code.
[0066] The first determination device 51 is a device that determines the life or death of the chicken in the cage 90 based on the still image captured by the imaging device 30. The first determination device 51 is composed of a computer, which mainly includes a storage device including a main storage device and an auxiliary storage device, a processor, and a communication device that communicates with the control device 50. The storage device stores a program that enables the computer to function as a first determination unit.
[0067] The control device 50 moves together with the slide device 10, whereas the first determination device 51 is separate from the slide device 10 and is installed in a management room or the like in the facility. The control device 50 and the first determination device 51 are connected via a communication network such as a LAN in the facility or the Internet.
[0068] The first determination mechanism includes: a still image acquisition mechanism, which acquires a still image captured by the camera 30; a first life or death determination mechanism, which determines the life or death of the chicken based on the still image; a moving signal sending mechanism, which, when it is determined that the still image does not contain dead chickens, sends a signal to the moving control mechanism by wired or wireless means to move the sliding device 10 to the next shooting point; a dynamic image sending mechanism, which, when it is determined that the still image contains dead chickens, causes the camera 30 to capture a dynamic image and sends the dynamic image with cage-specific information to the second determination mechanism; and a second moving signal sending mechanism, which, at the time when the shooting of the dynamic image is completed, sends a signal to the moving control mechanism by wired or wireless means to move the sliding device 10 to the next shooting point.
[0069] Methods for determining whether a still image contains a dead chicken by the first life or death determination unit include: (1-1) a method of performing image analysis on a still image (utilization of feature points); (1-2) a method of performing image analysis on a still image (comparison of multiple still images); (2) a method of using a learned model generated by machine learning of still images; (3) a method of simultaneously photographing the same photographing area by an infrared camera 31 and a visible light camera 31; and (4) a combination of these methods.
[0070] (1-1) Method of image analysis of still images (using feature points)
[0071] By extracting characteristic points of dead chickens or live chickens from a still image through image analysis, it is possible to determine whether a still image contains a dead chicken. Figure 5As shown in (a), since the live chicken 71 is standing, the feet of the live chicken 71 can be seen. In this figure, the feet of the live chicken 71 can be seen in the space between the feeding trough 85 and the egg tray 86 at the bottom of the cage 90. If the edge is extracted by analyzing the still image, the longitudinal contour line of the feet of the live chicken 71 can be detected.
[0072] On the other hand, Figure 5 As shown in (b), the dead chicken 72 is seen as a body, and the vertical contour of the feet cannot be detected even by image analysis. Therefore, by extracting the vertical contour of the feet as feature points from the still image, obtaining the feature amount and comparing it with a predetermined threshold, it is possible to determine whether the dead chicken 72 is included in the still image.
[0073] In addition, in the image of only the live chicken 71, the body of the chicken is almost not shown in the space below the feeding trough 85, whereas in the image of the dead chicken 72, the body of the chicken is shown in the space below the feeding trough 85. The body and feet of the chicken have different colors. Therefore, by using the color of the body of the chicken or the color of the feet as a feature point and using the ratio of the color in the area below the feeding trough 85 in the still image as a feature amount, life and death judgment can be performed on the still image.
[0074] (1-2) Method for analyzing still images (comparison of multiple still images)
[0075] The slide device 10 reciprocates along the track 41, so that the camera 30 can capture still images both during forward and reverse movement. For still images captured during forward movement and during reverse movement with the same cage as the shooting object, the difference in pixel values is calculated for each identical coordinate. In the two images, the difference in pixel values of the stationary parts is zero, and the moving parts have differences in pixel values. In addition to dead chickens, the stationary parts include the frame of the cage 90, the wires or rods that constitute the planar structure, and the feeding trough 85. Since the outlines of the frame, the planar structure, and the feeding trough 85 are linear, they can be distinguished from dead chickens whose outlines are nonlinear through image analysis. On the other hand, the part with a difference in pixel value can be determined as a live chicken.
[0076] The slide drive mechanism 40 including the motor 45 is supported on the cage array stack 90L as a structural body fixed to the facility, so that power can be supplied by wire. In addition, the photographing device 30 is a structure of the moving slide device 10, and the control device 50 is a structure that moves together with the slide device 10, and the slide device 10 is installed on the cage array stack 90L, so power for photographing or controlling can be supplied by wire. Therefore, unlike the case where a battery is used as a power source, there is no limit on the frequency of moving the slide device 10, and the slide device 10 can be reciprocated at a high frequency, and the same cage can be taken as a shooting object and photographed multiple times. Therefore, by performing the same comparison as above between multiple still images of the same cage as the shooting object (the difference in pixel value is calculated according to each same coordinate), dead chickens can be detected more accurately.
[0077] (2) Method using a learned model generated by machine learning of still images
[0078] A learned model is generated by machine learning using a plurality of still images containing only live chickens (hereinafter referred to as "live chicken teaching still images") and a plurality of still images containing dead chickens (hereinafter referred to as "dead chicken-containing teaching still images") as learning data. The generated learned model is stored in the storage device of the first determination device 51. By inputting the still image actually shot by the shooting device 30 into the learned model, the probability of the image containing only live chickens or the probability of the image containing dead chickens is output. In the case of machine learning through a neural network, the teaching images (live chicken teaching still images, dead chicken-containing teaching still images) to which information related to the life and death of chickens is added are used as learning data to generate the learned model, and then the teaching images without information related to life and death are input into the learned model, and the weighting coefficients between neurons are adjusted so as to output a probability suitable for life and death determination.
[0079] Preferably, the teaching images are photographed under the same conditions as the photographing conditions when the photographing device 30 actually photographs as the sliding device 10 moves. Preferably, the plurality of teaching still images of live chickens include various still images showing chickens facing in different directions and chickens at different positions in the depth direction or left-right direction of the cage 90. Preferably, the plurality of teaching still images of dead chickens include various still images showing chickens lying down with only the body visible but not the feet, chickens lying down with the body and horizontal feet visible, chickens with part of the body missing, and chickens with discolored feathers. Preferably, all the teaching images include images with eggs 79 on the egg tray 86, images with different numbers of eggs 79, and images without eggs 79.
[0080] In the case where the teaching image is an image of the inside of the cage 90, but the still image actually captured by the camera 30 (hereinafter sometimes referred to as the "input still image") shows the outside of the cage 90, or the teaching image is an image of the lower part of the cage 90 focusing on the chicken's feet, but the entire cage 90 is shown in the input still image, etc., the shooting areas of both parties are different, and the shooting areas of the input still image and the teaching image are made consistent and the sizes are set to be the same, and then the input still image is input into the learned model. The processing of making the shooting areas consistent can be performed, for example, by extracting the frame of the cage 90 or the feeding trough 85 through image analysis and using them as a reference.
[0081] By inputting a still image as input into the learned model and comparing the output value (probability) with a predetermined threshold, it is possible to determine whether the still image contains a dead chicken or not.
[0082] (3) Method of simultaneously photographing the same photographing area using the infrared camera 31 and the visible light camera 31
[0083] In this method, as the camera 31, a visible light camera and an infrared camera are used in pairs to simultaneously shoot the same shooting area. The infrared camera detects infrared rays radiated from an object, converts them into temperature, and images different temperatures as different colors, thereby visualizing the temperature of the object being photographed. The still image acquisition mechanism acquires the visible light image and the infrared image taken simultaneously in groups. The first life and death determination mechanism extracts the part where the chicken appears by image analysis of the visible light image, and specifies the part by coordinates. The extraction of the part where the chicken appears can be performed based on the color unique to the chicken, such as the color of the feathers.
[0084] The first life and death determination unit detects the color of a pixel in the infrared image at the same coordinates as the coordinates detected in the visible light image as the part where the chicken appears. Since the body temperature of a dead chicken is lower than that of a live chicken, if there are pixels of a color indicating a low temperature within a predetermined range in the part where the chicken should appear in the infrared image, it can be determined that the chicken is dead. Therefore, by setting the threshold to a temperature lower than the general temperature of the body temperature of a live chicken, it is possible to perform life and death determination of whether a dead chicken is included in a still image.
[0085] However, in the judgment based on still images, it is sometimes impossible to accurately judge whether a chicken is dead or alive. This is because, for example, when the chicken's feet are used as feature points in the judgment (1-1) of image analysis of a still image, or when the training image focusing on whether the chicken's feet are included in the image is used as learning data for machine learning in the judgment (2) of using a learned model generated by machine learning of still images, there is a possibility that a sitting (squatting) chicken will be judged as dead.
[0086] In addition, since the body temperature does not drop quickly after death, if the threshold is set low in the judgment (3) using an infrared camera, only chickens that have died a long time ago can be detected. If the threshold is set high in order to detect chickens that have died recently, there is a possibility that live chickens may be judged as dead because the body temperature of chickens also varies from one chicken to another.
[0087] Therefore, in this embodiment, when the first life and death determination unit determines that a dead chicken is included in the still image, a signal is sent from the first determination device 51 to the shooting control unit of the control device 50. The signal includes cage specific information of the still image determined to contain a dead chicken. The shooting control unit specifies the cage 90 of the shooting object based on the cage specific information, and controls the shooting device 30 to shoot a dynamic image for a predetermined time (for example, 5 seconds to 10 seconds). The dynamic image is sent to the second determination device 52 by the dynamic image sending unit in a state where the cage specific information is added.
[0088] Furthermore, when it is determined that the still images of only some of the cages 90 simultaneously photographed by the plurality of imaging devices 30 contain dead chickens, it is sufficient to photograph only the cage 90 with a moving image.
[0089] The second determination device 52 is a device that determines the life or death of the chicken in the cage 90 photographed based on the dynamic image. The device is composed of a computer, which mainly includes a storage device composed of a main storage device and an auxiliary storage device, a processor, a communication device for communicating with the control device 50, an alarm device, an input device such as a keyboard, and an output device including a display. The storage device stores a program that enables the computer to function as a second determination mechanism. The second determination device 52 is separated from the sliding device 10 and is set in a management room or the like in the facility.
[0090] The second determination unit includes: a moving image acquisition unit that acquires the moving image captured by the camera 30 and stores it in a storage device as a database associated with the cage specific information; a second life and death determination unit that determines the life and death of the chicken based on the moving image; and a warning unit that activates an alarm device when it is determined that the moving image contains a dead chicken, and displays the moving image containing the dead chicken and the cage specific information on the display. The alarm device can be a device that emits a warning sound, or a device that turns on or flashes a warning light.
[0091] The methods for determining whether a dynamic image contains a dead chicken by the second life and death determination mechanism include: (A) a method using a learned model generated by machine learning of dynamic images; and (B) a method of extracting an image at a certain point in time and an image after a specified period of time from the images constituting the dynamic image and comparing them.
[0092] (A) Method using a learned model generated by machine learning of dynamic images
[0093] A learned model for dynamic images is generated by machine learning using a plurality of dynamic images containing only live chickens (hereinafter referred to as "live chicken teaching dynamic images") and a plurality of dynamic images containing dead chickens (hereinafter referred to as "dead chicken-containing teaching dynamic images") as learning data. The generated learned model is stored in the storage device of the second determination device 52.
[0094] By inputting the dynamic image captured by the camera 30 into the learned model, the output is the probability of the dynamic image containing only live chickens or the probability of the dynamic image containing dead chickens. In machine learning, for example, a convolutional neural network with excellent image recognition and a regression neural network that is an extension of the neural network and can process time series data can be used in combination.
[0095] When machine learning is performed using these neural networks, teaching images to which information related to the life and death of chickens is attached (teaching dynamic images of live chickens, teaching dynamic images containing dead chickens) are used as learning data to generate a learned model. Then, teaching images without information related to life and death are input into the learned model, and the weighting coefficients between neurons are adjusted to output a probability suitable for life and death judgment.
[0096] Preferably, the teaching images are images captured by the camera 30 under the same conditions as those used when capturing dynamic images in the facility. Preferably, the plurality of teaching dynamic images of live chickens are composed of various dynamic images showing chickens facing in different directions and chickens at different positions in the depth direction or left-right direction of the cage 90. Preferably, the plurality of teaching dynamic images of dead chickens are composed of various dynamic images showing chickens lying down with only the body visible but not the feet, chickens lying down with the body and horizontal feet visible, chickens with part of the body missing, and chickens with discolored feathers. Preferably, the teaching images all include images of eggs 79 on the egg tray 86, images of different numbers of eggs 79, and images of no eggs 79.
[0097] In the second determination device 52 , the moving image captured by the imaging device 30 is input into the learned model, and the output value (probability) is compared with a predetermined threshold value, thereby making a life or death determination as to whether the moving image contains a dead chicken.
[0098] (B) A method of extracting an image at a certain point in time from the images constituting a dynamic image and comparing it with an image after a predetermined period of time has passed.
[0099] A still image at a certain time point T and a still image at a time point (T+△T) after a specified time have passed are extracted from the continuous images constituting the dynamic image, and a still image at a time point (T+2△T) after the same time has passed is extracted. Multiple still images are extracted at a time interval of △T, and the difference in pixel values for the same coordinates in the previous and next images is calculated along the time sequence. In a dynamic image, the difference in pixel values of the stationary parts is zero, and the difference in pixel values occurs in the moving parts. In addition to the dead chickens, the stationary parts include the frame of the cage 90, the wires or rods constituting the planar structure, and the feeding trough 85. Since the frame, the planar structure, and the feeding trough 85 have linear contours, they can be distinguished from the dead chickens whose contours are nonlinear through image analysis.
[0100] On the other hand, the part with a difference in pixel value can be determined as a live chicken. Here, when a dead chicken moves due to external force, such as when a live chicken kicks a dead chicken, a difference in pixel value also occurs, which can be a problem. However, since the time interval △T is short, in the case of a live chicken, the pixels with a difference in pixel value move little by little to the surrounding pixels (coordinates) with natural movement. In contrast, a dead chicken that moves due to external force will move greatly in an instant, so the pixels with a difference in pixel value move to the pixels far away. Therefore, by setting thresholds for the change in pixel value and the movement amount of the pixel (coordinate) where the pixel value changes, it is possible to determine whether a dead chicken is included in a dynamic image.
[0101] In the above description, the components of the dead chicken detection system and the role and function of each component are described. Figure 6 The following describes the processing flow of the dead chicken detection method of the first embodiment used in the system.
[0102] When the process starts, the slide 10 is moved to the initial shooting point by the control of the movement control mechanism (step P1), and the still image is shot by the shooting device 30 (step P2). The movement of the slide 10 and the shooting of the still image are preferably performed during a period when the chicken has a large amount of activity, such as a feeding period. The shot still image is sent to the first judgment device 51, and the first life and death judgment mechanism performs life and death judgment by the above-mentioned method (step P3).
[0103] As a result, if it is determined that the still image contains only live chickens and does not contain dead chickens (No in step P4), the movement control mechanism confirms whether the still image has been taken at all the shooting points based on the movement signal from the first determination device 51 (step P9). If there are still shooting points that have not been shot (No in step P9), the sliding device 10 is controlled to move toward the next shooting point (step P10), and then returns to the above-mentioned step P2 to shoot the still image.
[0104] On the other hand, when the first life and death determination mechanism determines that the still image contains a dead chicken (Yes in step P4), the camera 30 captures a moving image at the same shooting point as the still image (step P5). The captured moving image is sent to the second determination device 52, and the second life and death determination mechanism performs life and death determination according to the above-mentioned method (step P6).
[0105] As a result, when it is determined that the moving image contains dead chickens (Yes in step P7), the alarm device generates a warning sound and turns on or flashes a warning light. Then, cage-specific information that specifies the shooting point where the moving image containing the dead chickens was shot, in other words, information that specifies the cage 90 where the dead chickens are present, is displayed on the display of the second determination device 52 (step P8). After receiving the warning, the operator can go to the specific cage 90 and quickly remove the dead chickens from the cage 90.
[0106] As a result of the determination by the second life and death determination unit, if it is determined that the dynamic image contains only live chickens and does not contain dead chickens (No in step P7), the process proceeds to the above-mentioned step P9. In step P9, if it is confirmed that still images have been taken at all shooting points (Yes in step P9), the process ends.
[0107] As described above, according to the first embodiment, in addition to the first life and death determination of determining the life and death of the chickens in the cage 90 based on the still image, when it is determined by the first life and death determination that dead chickens are included, a moving image is captured at the same shooting point at a predetermined time, and a second life and death determination of determining the life and death of the chickens in the cage 90 based on the moving image is performed. Therefore, when a chicken in the cage 90 dies, this can be accurately detected.
[0108] Furthermore, the still image is captured under the control of the control device 50, the first life and death determination and the dynamic image capture are performed under the control of the first determination device 51 and the control device 50, and the second life and death determination and the generation of an alarm when a dead chicken is detected and the display of cage-specific information are performed under the control of the second determination device 52. Therefore, the only work that the operator needs to perform is to remove the dead chicken from the cage 90, which can greatly reduce the operator's workload.
[0109] Next, the dead chicken detection method of the second embodiment is described. In the first embodiment, as a result of the first life and death judgment, when it is determined that a dead chicken is included in the still image, a dynamic image is captured using the same camera 30 for the cage 90 as the same shooting object. In contrast, in the second embodiment, a dynamic image is always captured by a device different from the camera 30 for capturing still images. Therefore, the dead chicken detection system using the dead chicken detection method of the second embodiment is provided with a second camera (not shown) for always capturing dynamic images in addition to the sliding device 10 having the camera 30, the first judging device 51, and the second judging device 52.
[0110] A plurality of second camera devices are installed in various places in the facility, and they always capture dynamic images of all cages 90. The second camera devices and the cages 90 to be captured may correspond one-to-one, or one second camera device may capture dynamic images of a plurality of cages 90. In either case, dynamic image target cage information of the cage 90 to be captured is assigned to each second camera device. In addition, the second camera device is equipped with a device for counting time, and stores the captured dynamic images in association with the capturing time.
[0111] The slide device 10 is different from the first embodiment in the function of the camera 31. The camera 31 used in the second embodiment only captures still images and has a device for counting time.
[0112] The hardware structure of the first determination device 51 is the same as that of the first embodiment, but the functional structure is slightly different. The first determination mechanism of the second embodiment includes: a still image acquisition mechanism that acquires a still image captured by the shooting device 30; a first life and death determination mechanism that determines the life and death of the chicken based on the still image; a dead chicken detection signal sending mechanism that sends a dead chicken detection signal to the second determination device 52 when it is determined that the still image contains a dead chicken; and a moving signal sending mechanism that sends a signal to the moving control mechanism in a wired or wireless manner at the time when the shooting of the still image ends, so that the sliding device 10 moves to the next shooting point. The still image acquisition mechanism and the first life and death determination mechanism are the same as those of the first embodiment.
[0113] The hardware structure of the second determination device 52 is the same as that of the first embodiment, but a database that associates the identification information of the second camera device with the cage information of the dynamic image object is stored in the storage device. In addition, the second determination device 52 is slightly different from the first embodiment in terms of functional structure. The second determination mechanism of the second embodiment includes: a dynamic image acquisition mechanism, which acquires a dynamic image from the second camera device when receiving a dead chicken detection signal sent from the first determination device 51; a second life and death determination mechanism, which determines the life and death of the chicken based on the dynamic image; and a warning mechanism, which activates the alarm device when it is determined that the dynamic image contains a dead chicken, and displays the dynamic image containing the dead chicken and the cage-specific information on the display. The second life and death determination mechanism and the warning mechanism are the same as those of the first embodiment.
[0114] The processing flow of the dead chicken detection method of the second embodiment is described. The still image is captured as the slide device 10 moves, and the first life and death determination unit determines whether the still image contains a dead chicken, which is the same as the first embodiment.
[0115] When the first life and death determination unit determines that the still image contains a dead chicken, a dead chicken detection signal is sent to the second determination device 52. At this time, the dead chicken detection signal includes cage identification information that identifies the cage 90 in which the still image was captured and time information indicating the time when the still image was captured.
[0116] In the second determination device 52, upon receiving the dead chicken detection signal, the dynamic image acquisition mechanism acquires a dynamic image from the second camera. At this time, the dynamic image acquisition mechanism refers to the database that associates the identification information of the second camera with the dynamic image target cage information, and acquires the dynamic image from the second camera associated with the dynamic image target cage information corresponding to the cage specific information included in the dead chicken detection signal. That is, the dynamic image is acquired from the second camera that sets the cage 90, which is the shooting target of the still image determined to contain the dead chicken, as the shooting target of the dynamic image. In addition, the dynamic image acquisition mechanism refers to the time information included in the dead chicken detection signal, and acquires a dynamic image of a predetermined time before and after the time when the still image determined to contain the dead chicken is shot, for example, a dynamic image from 3 seconds before to 3 seconds after the time when the still image is shot.
[0117] The second life and death determination unit performs life and death determination on the acquired moving image. This life and death determination is the same as that described in the first embodiment. When it is determined that the moving image contains dead chickens, the alarm device issues a warning, and information specifying the cage 90 that is the object of the moving image containing dead chickens is displayed on the display of the second determination device 52, which is also the same as the first embodiment. The operator who receives the warning can go to the specific cage 90 and quickly remove the dead chickens from the cage 90.
[0118] As described above, in the second embodiment, although the dynamic image is always captured, it is not necessary to monitor the dynamic image all the time and analyze all the dynamic images. Only when it is determined that a dead chicken is included in the still image, the dynamic images before and after the still image are used to perform life and death determination. Therefore, the dynamic image can be efficiently used for life and death determination. In addition, by combining the life and death determination based on the dynamic image with the life and death determination based on the still image, the possibility of misjudging a live chicken as a dead chicken is reduced, and the dead chicken can be accurately detected, as in the first embodiment.
[0119] As mentioned above, although the preferred embodiment of the present invention is described, the present invention is not limited to the above-mentioned embodiment, and various improvements and design changes can be made without departing from the scope of the concept of the present invention.
[0120] For example, in the above-mentioned embodiment, the case where the first determination device 51 and the second determination device 52 are composed of different computers is exemplified (see Figure 4 ). This is not limited to this. As a hardware configuration, a single computer 61 may have the functions of the first determination device 51 and the second determination device 52 (see Fig. 9 (a)). In addition, in the above description, the control device 50 and the first determination device 51 are exemplified as being constituted by different computers. This is not limited to this, and as a hardware configuration, a single computer 62 may be installed on the support 12 of the slide device 10 and have the functions of the control device 50 and the first determination device 51 (see Fig. 9 (b)).
[0121] In addition, in the above description, the following occasion is exemplified: as a result of the first life and death judgment, when it is determined that dead chickens are included, a dynamic image is obtained by the second judgment device 52. It is not limited to this, and it can also be configured that the captured dynamic image is sent to the terminal used by the operator via a communication network together with an alarm signal and information about the specific cage 90. As the terminal used by the operator, in addition to a fixed personal computer, a mobile terminal such as a notebook personal computer, a tablet terminal, and a smart phone can also be used. In this way, the operator displays the dynamic image on the display of the terminal and can confirm the presence of dead chickens with his own eyes.
[0122] Furthermore, in the above description, an example is given in which a database associating a dynamic image with information on a specific cage 90 is stored in the storage device of the second determination device 52. This is not limited to this, and a database associating cage specific information with a still image may be stored when a still image is captured. The database may be stored in the storage device of the first determination device 51, or may be stored in the storage device of the second determination device 52 after being sent from the first determination device 51 to the second determination device 52.
[0123] In the above description, a slide device 10 is illustrated in which a support 12 is suspended from the slider 11 and the long groove 15 is engaged with the rail 41. Figure 7 as well as Figure 8 As shown, a slide device 20 is provided with a hopper 26 for supplying bait to the feeding trough 85.
[0124] The slider 21 of the sliding device 20 is provided with wheels 23 that engage with and move with the track 41 and a connector 24 that is connected to the wire 44 on both sides of the direction in which the track 41 extends. On the slider 21, a pair of pillars 22 extending downward are supported at intervals. A hopper 26 is supported between the pair of pillars 22. The number of hoppers 26 is equal to the number of layers of the cage array 90R in the cage array stack 90L, that is, the number of food troughs 85, and is supported on the pillars 22 corresponding to the height of the food trough 85. A plurality of hoppers 26 are connected by a pipe 27 extending in the up-down direction. If food is introduced into the opening 28 of the uppermost hopper 26 by a food supply device (not shown), the food is supplied to the food trough 85 via each hopper 26.
[0125] A plurality of imaging devices 30 are mounted on one of the pair of pillars 22. The bracket 39 extends horizontally from the pillar 22 in accordance with the height of each cage row 90R, and the camera 31 and the lighting device 32 are mounted thereon, which is the same as described above. In addition, the control device 50 is also mounted on one of the pair of pillars 22. The slide drive mechanism 40 has the same structure as described above.
[0126] The slide device 20 thus configured can also reciprocate along the cage array stack 90L similarly to the above-described slide device 10 , and can capture still images and moving images (first embodiment) or still images (second embodiment) by the imaging device 30 .
[0127] Since the slide device 20 also serves as a device for moving the hopper 26 for supplying food to the feeding trough 85, it has the advantage of contributing to both the automation of food supply and the automation of dead chicken detection as a single device.
[0128] In addition, it has been described above that still images are preferably taken during the feeding period when the chickens are most active. Since the hopper 26 operates during the feeding period, there is an advantage that the images taken by the image capturing device 30 of the slide device 20 are necessarily taken during the feeding period.
[0129] In addition, Figure 8 In the figure, the back portion, a portion of the front portion 91, the floor surface, a portion of the trough 85, and the egg tray are omitted. Figure 8 In the figure, only one of the sliding devices 20 provided on the two cage column stacked bodies 90L is shown.
[0130] In the above description, an embodiment is illustrated in which a second life and death judgment is performed based on a dynamic image when a dead chicken is determined to be included by the first life and death judgment. However, by making the sliding devices 10 and 20 reciprocate at a high frequency, when a large number of still images are taken of the same cage as the shooting object, the life and death of the chicken can be determined with a relatively high accuracy. In such a case, it is possible to set the system structure without the second judgment device 52 and set it as a dead chicken detection method that determines the life and death of the chicken only based on the first life and death judgment. The first judgment device 51 in this case is configured to have an alarm device that warns when it is determined that a dead chicken is included by the first life and death judgment, or a display that displays information related to the dead chicken. In addition, it is sufficient for the camera 31 to only take a still image.
Claims
1. A method for detecting dead chickens, characterized in that: The facility uses a cage stack to raise chickens. The cage stack is composed of multiple layers of cages arranged in parallel in a horizontal direction. In the above facility, A track extending in the same direction as the cage array is provided on the cage array stack. On the sliding device that reciprocates along the track, multiple cameras are provided corresponding to the height of each cage row. The camera takes at least one still image of each of the cages, and the first determination device performs a first life or death determination on whether the still image contains dead chickens.
2. The dead chicken detection method according to claim 1, characterized in that: If it is determined as a result of the first life and death determination that the still image contains a dead chicken, a second life and death determination is performed using a moving image captured of the cage as the same object to determine whether the moving image contains a dead chicken.
3. The dead chicken detection method according to claim 2, characterized in that: The second life or death determination is performed based on a result obtained by inputting the dynamic image into a learned model generated by machine learning using a teaching dynamic image containing only live chickens and a teaching dynamic image containing dead chickens as learning data.
4. The dead chicken detection method according to claim 1, characterized in that: The first life or death judgment is any one of the following judgments: a judgment based on the result of image analysis of the still image regarding characteristic points specific to a dead chicken or a live chicken; a judgment based on the result obtained by inputting the still image into a learned model, the learned model being generated by machine learning using teaching still images containing only live chickens and teaching still images containing dead chickens as learning data; a judgment based on the result of visualizing the temperature of the photographed object by an infrared camera; and a combination of two or more of these judgments.
5. The dead chicken detection method according to claim 1, characterized in that: The cage array stack is provided with a feeding trough in each cage array. The slide device includes a plurality of hoppers for supplying bait to the food trough, respectively.
Citation Information
Patent Citations
Fixed type poultry life information warning device
CN108460370A
Method for ventilating chicken coop
CN108882693A
Livestock and poultry survival detection method, device, equipment and computer program product
CN110287902A
Chicken flock state inspection monitoring system and method
CN111780811A
Carcass detection in henhouse and device therefor
JP1999276011A
