Method for estimating body condition score of animal, method for estimating weight of animal and computer program

Abstract

To provide technology that can estimate a body condition score or weight of an animal using a proper frame image.SOLUTION: A method includes the steps of: (a) acquiring a plurality of frame images by imaging an animal; (b) selecting one frame image from the plurality of frame images as a processing target image; and (c) obtaining an estimated BCS that is an estimated value of a body condition score of the animal from the processing target image using a BCS estimation model that has finished machine learning.SELECTED DRAWING: Figure 3

Description

[Technical field]

[0001] The present disclosure relates to a method for estimating the body condition score of an animal, a method for estimating the weight of an animal, and a computer program. [Background technology]

[0002] Patent Document 1 discloses a technology for determining the body condition score of a cow. The body condition score is a numerical representation of the degree of fat accumulation, and is basically expressed as a value ranging from 1 to 5. In this disclosure, the body condition score is also referred to as "BCS." The technology in Patent Document 1 detects the cow's contour line from an input image of the cow, determines an approximation curve for the contour line, extracts characteristic parameters from the relationship between the contour line and the approximation curve, and determines the BCS according to the characteristic parameters. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] JP 2019-187277 A Summary of the Invention [Problem to be solved by the invention]

[0004] However, the conventional technology had a problem that it could not calculate the BCS well when the input image was inappropriate. This problem was not limited to calculating the BCS of cows, but was also common when calculating the BCS of other livestock such as pigs and pets such as dogs and cats. In addition, a technology that can estimate the weight of animals in the same way as BCS is desired. [Means for solving the problem]

[0005] According to a first aspect of the present disclosure, there is provided a method for estimating a body condition score of an animal, the method including: (a) acquiring a plurality of frame images by photographing the animal, (b) selecting one of the plurality of frame images as a processing target image, and (c) determining an estimated BCS, which is an estimate of the body condition score of the animal, from the processing target image using a machine-learned BCS estimation model.

[0006] According to a second aspect of the present disclosure, there is provided a method for estimating a body weight of an animal, the method including: (a) acquiring a plurality of frame images by photographing the animal, (b) selecting one of the plurality of frame images as a processing target image, and (c) obtaining an estimated body weight, which is an estimate of the body weight of the animal, from the processing target image using a machine-learned body weight estimation model.

[0007] According to a third aspect of the present disclosure, there is provided a computer program for causing a processor to execute a process for estimating a body condition score of an animal. The computer program causes the processor to execute the following processes: (a) acquiring a plurality of frame images by photographing the animal, (b) selecting one of the plurality of frame images as a processing target image, and (c) determining an estimated BCS, which is an estimate of the body condition score of the animal, from the processing target image using a machine-learned BCS estimation model.

[0008] According to a fourth aspect of the present disclosure, there is provided a computer program for causing a processor to execute a process for estimating a weight of an animal. The computer program causes the processor to execute the following processes: (a) acquiring a plurality of frame images by photographing the animal, (b) selecting one of the plurality of frame images as a processing target image, and (c) determining an estimated weight, which is an estimate of the weight of the animal, from the processing target image using a machine-learned weight estimation model. [Brief description of the drawings]

[0009] [Figure 1] FIG. 1 is a block diagram showing the configuration of a BCS estimation system according to a first embodiment. [Diagram 2] 4 is a flowchart showing the processing procedure of a BCS estimation process in the first embodiment. [Diagram 3] FIG. 4 is an explanatory diagram showing the contents of a BCS estimation process in the first embodiment. [Figure 4] FIG. 11 is an explanatory diagram showing the content of a weight estimation process in the second embodiment. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0010] A. First embodiment 1 is an explanatory diagram showing the configuration of a BCS estimation system in the first embodiment. The BCS estimation system includes an information processing device 300 and a camera 410.

[0011] The camera 410 is for taking an image of the animal AA that is the target of the BCS estimation process. In this embodiment, the animal AA is a cow, but other animals such as pigs, dogs, and cats may be the processing target. The camera 410 is preferably installed so as to take an image of the animal AA from above. As the camera 410, a video camera that takes a moving image may be used, or a still image camera that takes a still image may be used. The camera 410 is preferably one that takes a color image. However, a depth camera that takes a depth image instead of a color image may be used. In this embodiment, the camera 410 as a video camera is used to take a moving image of the animal AA.

[0012] The information processing device 300 executes a process of estimating the BCS (body condition score) of the animal AA using an image of the animal AA captured by the camera 410. The information processing device 300 has a processor 310, a memory 320, an interface circuit 330, and an input device 340 and a display device 350 connected to the interface circuit 330. The camera 410 is also connected to the interface circuit 330. The processor 310 not only has a function of executing the processes described in detail below, but also has a function of displaying on the display device 350 data obtained by the processes and data generated in the course of the processes.

[0013] The processor 310 has the functions of a frame image acquisition unit 510, a feature recognition unit 520, an individual ID determination unit 530, and a BCS estimation unit 540. The frame image acquisition unit 510 acquires a plurality of frame images from an image captured by the camera 410. The feature recognition unit 520 recognizes features related to the shape of the animal AA from the frame images. The individual ID determination unit 530 determines the individual ID of the animal AA. The BCS estimation unit 540 estimates the BCS of the animal AA using the frame images. The functions of each of these units are respectively realized by the processor 310 executing a computer program stored in the memory 320. However, some of these functions may be realized by a hardware circuit. The processor in the present disclosure is a term that also includes such a hardware circuit. In addition, one or more processors that execute various processes may be processors included in one or more remote computers connected via a network.

[0014] The memory 320 stores a feature recognition model 610, an individual ID estimation model 620, a BCS estimation model 630, and a BCS history database 640. The feature recognition model 610 is a machine learning model that receives a frame image of an animal AA as an input and outputs features related to the shape of the animal AA. The individual ID estimation model 620 is a machine learning model that receives a frame image of an animal AA as an input and outputs an individual ID of the animal AA. The BCS estimation model 630 is a machine learning model that receives a frame image of an animal AA as an input and outputs an estimated BCS of the animal AA. The recognition model 610, the individual ID estimation model 620, and the BCS estimation model 630 have been trained. The BCS history database 640 is a database in which the individual ID and the history of the BCS are registered for each of a plurality of animals AAs.

[0015] FIG. 2 is a flowchart showing the procedure of the BCS estimation process in the first embodiment, and FIG. 3 is an explanatory diagram showing the process contents.

[0016] In step S110, the frame image acquisition unit 510 acquires a plurality of frame images FGj by photographing the animal AA using the camera 410. The camera 410 is preferably installed so as to photograph the animal AA from directly above. In this way, as shown in FIG. 3, a frame image FGj showing the state in which the animal AA is observed from above can be obtained. For example, if the camera 410 is installed above the passage through which the animal AA moves, a video of the animal AA passing by can be captured. For example, if the time it takes for the animal AA to pass through the angle of view of the camera 410 is 3 seconds, 180 frame images FGj can be acquired from a video with a frame rate of 60 fps. Here, j is an ordinal number for distinguishing frame images. Note that a plurality of frame images FGj may be acquired using a still image camera or a depth camera instead of a video camera. As can be understood from this explanation, in this disclosure, the term "frame image" is used in a broad sense to include not only still images extracted from a video, but also still images captured by a still image camera and depth images captured by a depth camera. Also, a combination of a still image and a depth image may be used as one frame image.

[0017] The frame image FGj shown in Fig. 3 includes the rear part of the animal AA including the tail and the lumbar region. Since the BCS has a strong correlation with the shape of the rear part of the animal AA, it is preferable that the frame image FGj includes the rear part of the animal AA including the tail and the lumbar region.

[0018] In step S120, the frame image acquisition section 510 selects one frame image from among the plurality of frame images FGj as an image to be processed. In the following, the same reference symbol "FGj" as for the frame image will be used for the selected image to be processed.

[0019] In step S130, the feature recognition unit 520 recognizes features related to the shape of the animal AA using the processing target image FGj. In this embodiment, the feature recognition unit 520 recognizes a plurality of feature points CPn present on the outline CTR of the animal AA as "features". Specifically, as shown in FIG. 3, the feature recognition unit 520 acquires a plurality of feature points CPn by inputting the processing target image FGj to the feature recognition model 610. The feature recognition model 610 is a machine learning model that receives the processing target image FGj as an input and outputs a plurality of feature points CPn. The feature recognition model 610 may be configured to output pixel coordinates (u, v) of the feature points CPn. The feature recognition model 610 may be configured using, for example, ResNet, which is a deep learning model. The feature recognition model 610 may be configured using a machine learning model other than ResNet. In addition, the recognition process of the feature points CPn may be performed using a method other than the deep learning model.

[0020] The feature recognition unit 520 may determine that the feature point CPn output from the feature recognition model 610 has been recognized only when the reliability of the output of the feature recognition model 610 is equal to or greater than a predetermined reliability threshold. In this way, the BCS can be determined using a processing target image having a highly reliable feature point CPn. The reliability of the output of the feature recognition model 610 means the value of the activation function of the final layer. The reliability threshold is preferably set in the range of 0.85 to 0.95, for example.

[0021] The subscript n added to the symbols of the multiple feature points CPn is an integer from 1 to N that distinguishes the feature points, where N is an integer of 2 or more. In the example of Fig. 3, five feature points CP1 to CP5 that exist on the outline CTR are recognized. These feature points CP1 to CP5 are points that indicate the position of the base of the tail, the positions of the left and right ischial bones, and the positions of the left and right hip angles. Note that the feature recognition model 610 may be configured to recognize feature points that indicate other positions.

[0022] In step S140, the feature recognition unit 520 determines whether the recognized feature satisfies a preset allowable condition. In this embodiment, the following two allowable conditions C1 and C2 can be used. <Acceptable condition C1> A preset number N of feature points CP1 to CPN are all recognized. <Acceptable condition C2> The characteristic points CP1 to CPN are located in the center of the processing target image FGj.

[0023] The above-mentioned permissible condition C1 is a condition as to whether or not all of the five feature points CP1 to CPN can be recognized when N is set to 5. If one or more of the five feature points CP1 to CPN cannot be recognized, the process returns to step S120, another frame image is selected as a new image to be processed, and the processes from step S130 onwards are executed again.

[0024] The above-mentioned permissible condition C2 is, for example, a condition as to whether or not the minimum value of the distance calculated from the center position of the processing target image FGj to each feature point CPn is equal to or less than a distance threshold. The distance threshold is preferably set to a value of 20% to 30% of the length of the short side of the processing target image FGj, for example. If the feature points CP1 to CPN are not in the center of the processing target image FGj, the reliability of the feature point CPn is low, so the process returns to step S120, another frame image is selected as a new processing target image, and the processes from step S130 onwards are executed again.

[0025] One of the above-mentioned permissible conditions C1 and C2 may be omitted. Also, the judgment in step S140 may be omitted. In this case, step S130 can also be omitted. However, it is preferable to perform the judgment in step S140 using at least the permissible condition C1, since this can improve the estimation accuracy of the BCS.

[0026] The above-described processing in steps S130 and S140 is a type of image determination processing in which features related to the shape of the animal AA are recognized from the processing target image FGj, and whether or not the features satisfy a preset allowable condition is used to determine whether or not the processing target image FGj is an appropriate image. By performing such image determination processing, it is possible to select a processing target image FGj suitable for estimating the BCS.

[0027] In this embodiment, a plurality of feature points CPn existing on the outline CTR of the animal AA are recognized as features related to the shape of the animal AA, but features other than the feature points CPn may be recognized. For example, the bounding box of the animal AA may be recognized as a feature from the processing target image FGj, and depending on whether the bounding box satisfies the tolerance condition, it may be determined whether the processing target image FGj is appropriate or not. The process of recognizing the bounding box of the animal AA from the processing target image FGj can also be executed using a machine learning model. In addition, when the bounding box is used, the following tolerance conditions C1a and C2a can be used instead of the above-mentioned tolerance conditions C1 and C2. <Acceptable condition C1a> The area of ​​the bounding box is greater than or equal to a predefined area threshold. <Acceptable Condition C2a> The center of the bounding box is in the center of the processing target image FGj. It should be noted that one of the permissible conditions C1a and C2a may be omitted.

[0028] Instead of recognizing the bounding box, semantic segmentation may be performed on the processing target image FGj to recognize the image area of ​​the animal AA as a feature. In this case, the following allowable conditions C1b and C2b can be used instead of the above-mentioned allowable conditions C1a and C2a. <Acceptable Condition C1b> The area of ​​the image region of animal AA is greater than or equal to a pre-set area threshold. <Acceptable Condition C2b> The center of the image area of ​​the animal AA is located in the center of the processing target image FGj. It should be noted that one of the permissible conditions C1b and C2b may be omitted.

[0029] In step S150, the individual ID determination unit 530 determines the individual ID of the animal AA using the individual ID estimation model 620. The individual ID estimation model 620 is a machine learning model that receives the processing target image FGj as an input and outputs an individual ID. The individual ID estimation model 620 can be configured using, for example, a Siamese net, which is a deep learning model. The individual ID estimation model 620 may be configured using a machine learning model other than the Siamese net.

[0030] Instead of using the individual ID estimation model 620 to determine the individual ID, an ear tag or an RFID tag (Radio Frequency IDentification tag) of the animal AA may be used to determine the individual ID. In this case, a specific part such as the face of the animal AA may be photographed with a camera other than the camera photographing the body of the animal AA, and the individual ID may be determined using a frame image of the specific part. However, it is quite difficult to photograph the ear tag so that the individual ID can be determined from the image of the ear tag. In addition, when reading the RFID tag, a process of installing the RFID tag on the animal AA is required, and a reading unit for reading the RFID tag is also required. In contrast, if the individual ID is estimated from the processing target image FGj using the individual ID estimation model 620 as in this embodiment, there is an advantage that both the identification of the individual of the animal AA and the estimation of the BCS can be performed using the processing target image FGj. In particular, when the body of the animal AA has a unique pattern that differs from individual to individual, the individual ID can be accurately estimated from the processing target image FGj. When an individual ID can be estimated from the processing target image FGj, the processing target image FGj contains the characteristics of the animal AA in an appropriate state, and therefore a highly reliable estimated BCS can be obtained by using the processing target image FGj.

[0031] In step S160, the individual ID determination unit 530 judges whether the individual ID has been determined. Whether the individual ID has been determined is determined, for example, according to whether the reliability of the output of the individual ID estimation model 620 satisfies a predetermined allowable condition. As the allowable condition, a condition that the reliability of the estimated individual ID is equal to or higher than a reliability threshold can be used. The reliability of the output of the individual ID estimation model 620 corresponds to the inverse of the distance between the registered individual ID and the estimated individual ID, or a value obtained by multiplying the inverse by minus 1. For example, in a Siamese network, the values ​​of the final layer are used as coordinates in a certain space, and the coordinates of the preregistered individual ID and the coordinates of the input data that are closest are output as the estimated individual ID. Therefore, a value that is correlated with the distance in the output space of the final layer can be used as the reliability. If the individual ID cannot be determined, the process returns to step S120, another frame image is selected as a new image to be processed, and the process from step S130 onwards is executed again. If the individual ID can be determined, the process proceeds to step S170.

[0032] In step S170, the BCS estimation unit 540 obtains an estimated BCS value Sc of the animal AA using the BCS estimation model 630. The BCS estimation model 630 is a machine learning model that receives the processing target image FGj as an input and outputs the estimated BCS value Sc. The BCS estimation model 630 can be configured using, for example, ResNet, which is a deep learning model. The BCS estimation model 630 may be configured using a machine learning model other than the ResNet network. Also, a process of obtaining the estimated BCS value Sc may be performed using a method other than the deep learning model.

[0033] The BCS of a cow is expressed as one of 17 candidate values, ranging from 1 to 5 in increments of 0.25. The BCS estimation model 630 may be configured as a classification model that indicates one of these 17 candidate values, or may be configured as a regression model with a single BCS output that can take any numeric value.

[0034] If the reliability of the BCS output of the BCS estimation model 630 is less than a predetermined reliability threshold, the process may return to step S120, select another frame image as a new image to be processed, and execute the processes from step S130 onward again. The reliability of the BCS output means the value of the activation function of the final layer.

[0035] In step S180, BCS estimation unit 540 uses the individual ID to refer to BCS history database 640 to determine the allowable range of the estimated BCS value Sc. The allowable range of the estimated BCS value Sc is set, for example, as follows: Sp-δ1≦Sc≦Sp+δ2 … (1) Here, Sp is a past BCS value determined from the BCS history of animal AA, and δ1 and δ2 are allowable deviations. The most recent BCS value can be used as the past BCS value Sp. Alternatively, the average value of the most recent BCS values ​​or an estimated value estimated from the most recent BCS values ​​may be used as the past BCS value Sp. In this embodiment, the most recent BCS value is used as the past BCS value Sp.

[0036] The negative allowable deviation δ1 and the positive allowable deviation δ2 may be set to equal values ​​or different values. For example, the negative allowable deviation δ1 may be set to zero, and the positive allowable deviation δ2 may be set to a positive value. These allowable deviations δ1 and δ2 are set as a range in which the BCS may change between the previous and current BCS estimation processes in FIG. 2. For example, when the BCS estimation process is performed periodically once a week, the negative allowable deviation δ1 may be set to 0.25, and the positive allowable deviation δ2 may be set to 0.5. When the BCS estimation process is performed irregularly, the values ​​of the allowable deviations δ1 and δ2, i.e., the allowable range of the estimated BCS value Sc, may be adjusted according to the length of the elapsed time between the previous BCS estimation process and the current BCS estimation process.

[0037] In step S190, the BCS estimation unit 540 determines whether the estimated BCS value Sc is within the allowable range. In this embodiment, whether the estimated BCS value Sc is within the allowable range is determined depending on whether the above formula (1) is satisfied. If the estimated BCS value Sc is not within the allowable range, the reliability of the estimated BCS value Sc is low, so the process returns to step S120, another frame image is selected as a new image to be processed, and the processes from step S130 onwards are executed again. On the other hand, if the estimated BCS value Sc is within the allowable range, the process proceeds to step S200.

[0038] In step S200, the BCS estimation unit 540 uses the estimated BCS value Sc to determine the final BCS of the animal AA. For example, the estimated BCS value Sc may be determined as the final BCS as it is. Alternatively, the processes of steps S120 to S190 may be performed for a plurality of frame images FGj, and after a predetermined number M of appropriate estimated BCS values ​​Sc are obtained, a representative value of these M estimated BCS values ​​Sc may be determined as the final BCS. Here, M is an integer of 2 or more. As the representative value of the M estimated BCS values ​​Sc, for example, the estimated BCS value Sc with the smallest difference from the past BCS value Sp or the average value of the M estimated BCS values ​​Sc may be used. The final BCS determined in this manner is output to an output device such as the display device 350 or a printer, and is also registered in the BCS history database 640.

[0039] As mentioned above, the BCS of a cow is usually expressed as one of 17 candidate values ​​that range from 1 to 5 in increments of 0.25. Therefore, it is preferable to adjust the final BCS to one of these 17 candidate values. In this case, the candidate value that is closest to the estimated BCS value Sc or the candidate value that is closest to the representative value of the multiple estimated BCS values ​​Sc is determined as the final BCS.

[0040] The order of steps S120 to S190 can be changed as appropriate. If the result of steps S120 to S190 shows that a suitable frame image FGj does not exist and the final BCS cannot be obtained, a message indicating that the BCS cannot be estimated may be output. Alternatively, the final BCS may be obtained by taking measures such as reducing the required number N of feature points CPn or relaxing various conditions. Some or all of steps S130 to S160 and S180 to S200 may be omitted.

[0041] According to the above-described first embodiment, the body condition score of the animal AA can be estimated using an appropriate frame image from among the multiple frame images FGj.

[0042] B. Second embodiment Fig. 4 is an explanatory diagram showing the contents of the animal weight estimation process in the second embodiment. The main differences from the BCS estimation process shown in Fig. 3 are as follows. <Difference 1> In the second embodiment, in addition to the camera 410 that photographs the animal AA from above, a camera 420 that photographs the animal AA from the side is provided, and a pair of images obtained by the two cameras 410, 420 are used as a "frame image." However, one of the cameras 410, 420 may be omitted. <Difference 2> In the second embodiment, the frame images FGj captured by the cameras 410 and 420 each include an image that includes almost the entire body of the animal AA. <Difference 3> In the second embodiment, the feature recognition unit 520 recognizes feature points CP11 to CP15 that exist on the outline CTR of the entire body of the animal AA as features related to the shape of the animal AA. <Difference 4> In the second embodiment, the BCS estimation unit 540 of the first embodiment is replaced with a body weight estimation unit 550, the BCS estimation model 630 is replaced with a body weight estimation model 650, and the BCS history database 640 is replaced with a body weight history database 660.

[0043] Although the procedure for estimating the weight of the animal AA in the second embodiment is not shown, it is possible to use a procedure in which "BCS" is read as "weight" in the BCS estimation procedure shown in Fig. 2. Also, the various forms and modified examples described in the first embodiment are applicable to the second embodiment in almost the same way.

[0044] According to the second embodiment, it is possible to estimate the weight of the animal AA by using an appropriate frame image from among the multiple frame images FGj.

[0045] Other forms: The present disclosure is not limited to the above-mentioned embodiment, and can be realized in various forms without departing from the spirit of the present disclosure. For example, the present disclosure can be realized in the following aspects. The technical features in the above-mentioned embodiments corresponding to the technical features in each aspect described below can be appropriately replaced or combined in order to solve some or all of the problems of the present disclosure, or to achieve some or all of the effects of the present disclosure. Furthermore, if the technical feature is not described as essential in this specification, it can be appropriately deleted.

[0046] (1) According to a first aspect of the present disclosure, there is provided a method for estimating a body condition score of an animal, the method including: (a) acquiring a plurality of frame images by photographing the animal, (b) selecting one of the plurality of frame images as a processing target image, and (c) determining an estimated BCS, which is an estimate of the body condition score of the animal, from the processing target image using a machine-learned BCS estimation model. According to this method, the body condition score of an animal can be estimated using an appropriate frame image from among a plurality of frame images.

[0047] (2) In the above method, step (b) may include a step of executing a process for recognizing features related to the shape of the animal from the image to be processed, and a step of selecting another frame image as the image to be processed if the features do not satisfy a preset tolerance condition. According to this method, an appropriate frame image can be selected as the image to be processed.

[0048] (3) In the above method, step (b) may include a step of determining an individual ID that identifies an individual animal from the image to be processed, and a step of selecting another frame image as the image to be processed if the reliability of the individual ID does not satisfy an acceptable condition. According to this method, an appropriate frame image can be selected as the image to be processed.

[0049] (4) In the above method, step (b) may include the steps of determining an individual ID that identifies the individual animal, using the individual ID to refer to the BCS history of the animal to determine an acceptable range for the estimated BCS, and selecting another frame image as the image to be processed if the estimated BCS is not within the acceptable range. According to this method, an appropriate frame image can be selected as the image to be processed.

[0050] (5) The method may further include the steps of: (d) determining an individual ID that identifies the individual animal; (e) using the individual ID to reference the BCS history of the animal to determine an acceptable range for the estimated BCS; and (f) if the estimated BCS is within the acceptable range, determining a body condition score of the animal using the estimated BCS. This method allows for an appropriate range of body condition scores to be estimated from BCS history.

[0051] (6) The above method may further include the step of: (g) if the estimated BCS is not within the allowable range, returning to step (b), selecting another frame image as the image to be processed, and executing steps (c) to (f) again. According to this method, an appropriate frame image can be selected as the image to be processed.

[0052] (7) In the above method, the step (d) may include a step of estimating the individual ID from the processing target image using a machine-learned individual estimation model. This method allows both identification of individual animals and estimation of the BCS using the processed image.

[0053] (8) According to a second aspect of the present disclosure, there is provided a method for estimating a body weight of an animal, the method including: (a) acquiring a plurality of frame images by photographing the animal, (b) selecting one of the plurality of frame images as a processing target image, and (c) determining an estimated body weight, which is an estimate of the body weight of the animal, from the processing target image using a machine-learned body weight estimation model.

[0054] (9) According to a third aspect of the present disclosure, there is provided a computer program for causing a processor to execute a process for estimating a body condition score of an animal. The computer program causes the processor to execute the following processes: (a) acquiring a plurality of frame images by photographing the animal, (b) selecting one of the plurality of frame images as a processing target image, and (c) determining an estimated BCS, which is an estimate of the body condition score of the animal, from the processing target image using a machine-learned BCS estimation model.

[0055] (10) According to a fourth aspect of the present disclosure, there is provided a computer program for causing a processor to execute a process for estimating a weight of an animal. The computer program causes the processor to execute the following processes: (a) acquiring a plurality of frame images by photographing the animal, (b) selecting one of the plurality of frame images as a processing target image, and (c) calculating an estimated weight, which is an estimate of the weight of the animal, from the processing target image using a machine-learned weight estimation model.

[0056] The present disclosure may be realized in various forms other than those described above, such as a device for implementing a BCS estimation process, a device for implementing a weight estimation process, or a non-transitory storage medium on which a computer program is recorded. [Explanation of symbols]

[0057] 300...information processing device, 310...processor, 320...memory, 330...interface circuit, 340...input device, 350...display device, 410...camera, 420...camera, 510...frame image acquisition unit, 520...feature recognition unit, 530...individual ID determination unit, 540...BCS estimation unit, 550...weight estimation unit, 610...feature recognition model, 620...individual ID estimation model, 630...BCS estimation model, 640...BCS history database, 650...weight estimation model, 660...weight history database

Claims

1. A method for estimating the body condition score of an animal using a computer, comprising: (a) acquiring a plurality of frame images by photographing the animal; (b) selecting one frame image from the plurality of frame images as a processing target image; (c) calculating an estimated BCS, which is an estimate of the body condition score of the animal, from the image to be processed using a machine-learned BCS estimation model; A method comprising:

2. 10. The method of claim 1, The step (b) a step of executing a process of recognizing features related to the shape of the animal from the processing target image; a step of selecting another frame image as the processing target image when the feature does not satisfy a preset acceptable condition; A method comprising:

3. 10. The method of claim 1, The step (b) determining an individual ID that identifies an individual animal from the image to be processed; a step of selecting another frame image as the processing target image when the reliability of the individual ID does not satisfy an allowable condition; A method comprising:

4. 10. The method of claim 1, The step (b) determining an individual ID that identifies the individual animal; determining an acceptable range for the estimated BCS by referencing the animal's BCS history using the individual ID; selecting another frame image as the processing target image when the estimated BCS is not within the allowable range; A method comprising:

5. 10. The method of claim 1 further comprising: (d) determining an individual ID that identifies the individual animal; (e) using the individual ID to reference the animal's BCS history to determine an acceptable range for the estimated BCS; (f) if the estimated BCS is within the acceptable range, determining a body condition score for the animal using the estimated BCS; A method comprising:

6. 6. The method of claim 5, further comprising: (g) if the estimated BCS is not within the allowable range, returning to the step (b), selecting another frame image as the processing target image, and executing the steps (c) to (f) again; A method comprising:

7. 6. The method of claim 5, The method, wherein the step (d) includes a step of estimating the individual ID from the processing target image using a machine-learned individual estimation model.

8. A method for estimating the weight of an animal by a computer, comprising: (a) acquiring a plurality of frame images by photographing the animal; (b) selecting one frame image from the plurality of frame images as a processing target image; (c) calculating an estimated weight, which is an estimate of the weight of the animal, from the image to be processed using a machine-learned weight estimation model; A method comprising:

9. A computer program that causes a processor to execute a process for estimating a body condition score of an animal, comprising: (a) acquiring a plurality of frame images by photographing the animal; (b) selecting one frame image from the plurality of frame images as an image to be processed; (c) using a machine-learned BCS estimation model to obtain an estimated BCS, which is an estimated value of the body condition score of the animal, from the processing target image; A computer program that causes the processor to execute the above.

10. A computer program that causes a processor to execute a process for estimating the weight of an animal, (a) acquiring a plurality of frame images by photographing the animal; (b) selecting one frame image from the plurality of frame images as an image to be processed; (c) calculating an estimated weight, which is an estimate of the weight of the animal, from the image to be processed using a machine-learned weight estimation model; A computer program that causes the processor to execute the above.

Images