JP7783962B2 - Anomaly detection device, anomaly detection method, and program - Google Patents

Description

The present invention relates to an anomaly detection device, an anomaly detection method, and a program.

Fattening cattle are known to be prone to developing abnormalities such as difficulty standing in the later stages of fattening. If this condition is left untreated, the cattle may die from suffocation, etc., so fattening cattle farmers patrol their barns to check whether the cattle are experiencing any abnormalities such as difficulty standing.

It is also known that breeding cows are prone to developing abnormalities such as difficulty standing before and after giving birth, and breeding farmers similarly patrol their barns to check whether their cows are experiencing any abnormalities such as difficulty standing.

In addition, cows may have difficulty standing up due to conditions such as bloat. Bloat is a disease that can occur in ruminant livestock such as cows, and can occur, for example, when they overeat fermented feed such as green grass. When bloat develops, the cow's abdomen becomes distended, making it difficult for the cow to stand up.

Methods for estimating the above-described difficulty in standing up have been known for some time (for example, Patent Document 1).

International Publication No. 2019/039118

However, conventional systems have not always been highly accurate in detecting abnormalities in livestock, such as difficulty standing up, resulting in false positives. For example, if a cow shakes its head or scratches its body while lying down, it may be mistakenly detected as having difficulty standing up.

One embodiment of the present invention has been made in light of the above points, and aims to detect abnormalities in livestock with high accuracy.

To achieve the above objective, one embodiment of an information processing device is an information processing device that determines whether an abnormality has occurred in livestock, and includes: a posture estimation unit that estimates the posture of the livestock based on measurement data measured by a sensor attached to the livestock; a score calculation unit that calculates a score representing the degree of abnormality of the livestock based on the posture and the measurement data; and a determination unit that determines whether an abnormality has occurred in the livestock based on the score.

It is possible to detect abnormalities in livestock with high accuracy.

1 is a diagram illustrating an example of the overall configuration of an anomaly detection system according to an embodiment of the present invention. FIG. 4 is a diagram showing an example of measurement data stored in a measurement data storage unit. FIG. 2 is a diagram illustrating an example of the functional configuration of a posture estimation processing unit according to the present embodiment. FIG. 10 is a diagram for explaining an example of the relationship between each attitude and each coordinate axis. FIG. 2 is a diagram illustrating an example of a functional configuration of an abnormality detection processing unit according to the present embodiment. 10 is a flowchart illustrating an example of a tag angle calculation process according to the present embodiment. 10 is a flowchart illustrating an example of a posture estimation process according to the present embodiment. 10 is a flowchart illustrating an example of an abnormality detection process according to the present embodiment.

One embodiment of the present invention will now be described. In this embodiment, an abnormality detection system 1 will be described that can accurately detect abnormalities such as difficulty standing up in cattle (for example, difficulty standing up that occurs in the later stages of fattening, difficulty standing up that occurs before and after childbirth, difficulty standing up due to bloat, etc.). However, cattle are only one example of livestock, and livestock are not limited to cattle. This embodiment can also be applied to various animals (for example, horses, sheep, pigs, etc.) that, like cattle, may experience abnormalities such as difficulty standing up.

<Overall configuration of anomaly detection system 1>
The overall configuration of an anomaly detection system 1 according to this embodiment is shown in Fig. 1. As shown in Fig. 1, the anomaly detection system 1 according to this embodiment includes an anomaly detection device 10 that detects abnormalities in cows, one or more tags 20 attached to each cow, and a reference air pressure sensor 30 that measures reference air pressure. It is preferable that the tags 20 be attached and fixed around the neck of the cow by a collar, belt, or the like.

The tag 20 is a device attached to the cow. One tag 20 is attached to each cow. The tag 20 includes an acceleration sensor that measures the acceleration (acceleration on three axes: x, y, and z) of the cow wearing the tag 20, and a barometric pressure sensor that measures air pressure. The coordinate system consisting of the x, y, and z axes is fixed to the tag 20. The tag 20 is attached to the cow when it is standing with its neck straight and not bent, so that the positive x-axis is the direction opposite to the direction of gravity, the positive y-axis is the direction the cow is moving, and the positive z-axis is to the right of the cow's moving direction. However, the directions of the axes (x, y, and z) relative to the cow are not limited to these, and the tag 20 can be attached to the cow so that each axis is oriented in any direction while maintaining the relative positional relationship between the axes in the coordinate system.

The tag 20 transmits measurement data including an acceleration sensor value measured by the acceleration sensor and an atmospheric pressure sensor value measured by the atmospheric pressure sensor at predetermined time intervals Δ ₁ (for example, Δ ₁ = 2 [s]) to the anomaly detection device 10. The measurement data transmitted to the anomaly detection device 10 is accumulated (stored) in a measurement data storage unit 300 described later.

The reference air pressure sensor 30 is installed at a predetermined position in the cowshed (for example, on the ground in the cowshed) and measures a reference air pressure. The reference air pressure sensor 30 transmits reference air pressure data including a reference air pressure sensor value indicating the measured air pressure to the anomaly detection device 10 every predetermined time _Δ2 (for example, _Δ2 = 2 [s]). The reference air pressure data transmitted to the anomaly detection device 10 is accumulated (stored) in the reference air pressure data storage unit 400, which will be described later. For simplicity, in the following, it is assumed that _Δ2 = _Δ1 and the sampling times of the reference air pressure sensor value and the air pressure sensor value are synchronized. Note that if _Δ2 ≠ _Δ1 , or if _Δ2 = _Δ1 but the sampling times of the reference air pressure sensor value and the air pressure sensor value are not synchronized, appropriate resampling or the like may be performed.

The abnormality detection device 10 is one or more computers that detect a cow's difficulty standing up as an abnormality. The abnormality detection device 10 has a posture estimation processing unit 100, an abnormality detection processing unit 200, a measurement data storage unit 300, a reference air pressure data storage unit 400, a tag angle data storage unit 500, and a posture data storage unit 600.

The posture estimation processing unit 100 estimates the posture of the cow using the measurement data stored in the measurement data storage unit 300 and the reference atmospheric pressure data stored in the reference atmospheric pressure data storage unit 400. Here, the inventors of the present application have observed actual cows and found that cow postures can be broadly classified into "standing" and "lying," and that lying postures can be further classified into "lying" and "prone." They have also found that the difference between lying and prone postures is particularly evident in the angle of neck rotation around the axis of the face forward. That is, when standing or lying, the neck is often not bent but straight, whereas when lying, the neck is not straight and is often rotated to some extent around the axis of the face forward. Based on this knowledge, the posture estimation processing unit 100 of this embodiment estimates whether the cow's posture is prone or reclining from the angle of neck rotation when the cow is lying.

Standing refers to the position in which the cow is standing. On the other hand, prone and lying both refer to the position in which the cow lies down, but generally, prone is a position in which the knees are bent and the chest is raised (a so-called sitting position), while lying is a position in which the cow is lying on its side with its limbs stretched out.

The abnormality detection processing unit 200 detects abnormalities in the cow (difficulty standing up) using the posture estimated by the posture estimation processing unit 100 and the measurement data stored in the measurement data storage unit 300. Here, the inventors of the present application have observed actual cows and discovered that even when cows are in the same sleeping position, lying on their backs means that there is almost no chance that they will have difficulty standing up, whereas lying on their sides means that they may have difficulty standing up. Based on this knowledge, the abnormality detection processing unit 200 increases a predetermined score only when the cow is lying down, and detects that the cow is having difficulty standing up when this score exceeds a certain threshold.

The posture estimation processing unit 100 and the anomaly detection processing unit 200 are realized, for example, by one or more programs installed in the anomaly detection device 10 being executed by a processor such as a CPU (Central Processing Unit). However, all or part of the functions of the posture estimation processing unit 100 and the anomaly detection processing unit 200 may be realized by a cloud server or the like connected to the anomaly detection device 10 via a communications network.

The measurement data storage unit 300 stores the measurement data received from each tag 20. The measurement data storage unit 300 stores the measurement data for each time period _Δ1 for each tag 20. The measurement data received from the tag 20 is accumulated (stored) in the measurement data storage unit 300 by, for example, the posture estimation processing unit 100 or the anomaly detection processing unit 200.

The reference atmospheric pressure data storage unit 400 stores the reference atmospheric pressure data received from the reference atmospheric pressure sensor 30. The reference atmospheric pressure data for each time period Δ ₂ is stored in the reference atmospheric pressure data storage unit 400. The reference atmospheric pressure data received from the reference atmospheric pressure sensor 30 is accumulated (stored) in the reference atmospheric pressure data storage unit 400 by, for example, the posture estimation processing unit 100 or the anomaly detection processing unit 200.

The tag angle data storage unit 500 stores tag angle data used to estimate whether the cow's posture is prone or recumbent. Details of the tag angle data will be provided later.

The posture data storage unit 600 stores posture data indicating the posture estimated by the posture estimation processing unit 100.

The measurement data storage unit 300, reference atmospheric pressure data storage unit 400, tag angle data storage unit 500, and attitude data storage unit 600 can be realized using storage devices such as HDDs (Hard Disk Drives) and SSDs (Solid State Drives). However, all or part of the measurement data storage unit 300, reference atmospheric pressure data storage unit 400, tag angle data storage unit 500, and attitude data storage unit 600 may also be realized by a database server or the like connected to the anomaly detection device 10 via a communications network.

<Measurement Data Stored in Measurement Data Storage Unit 300>
An example of measurement data stored in the measurement data storage unit 300 according to this embodiment is shown in FIG. 2. As shown in FIG. 2, the measurement data storage unit 300 stores one or more pieces of measurement data for each tag ID, which is information for identifying a tag. Since one tag 20 is attached to one cow, the tag ID may be information for identifying the cow (for example, individual identification information for the cow). In the following, the tag ID and information for identifying the cow will be considered to be the same, and it will be assumed that the tag ID can be used to identify the cow.

Each measurement data includes a date and time, an acceleration sensor value, and an air pressure sensor value. The date and time is, for example, the date and time when the tag 20 transmitted the measurement data. Note that the date and time may also be the date and time when the anomaly detection device 10 received the measurement data.

The acceleration sensor value is an acceleration value measured by an acceleration sensor included in the tag 20. The acceleration sensor value includes an x-component indicating the acceleration component in the x-axis direction, a y-component indicating the acceleration component in the y-axis direction, and a z-component indicating the acceleration component in the z-axis direction. For example, measurement data at date and time "t ₁ " includes the x-component of the acceleration sensor value "x ₁ ," the y-component of the acceleration sensor value "y ₁ ," and the z-component of the acceleration sensor value "z ₁ ." Similarly, measurement data at date and time "t ₂ " includes the x-component of the acceleration sensor value "x ₂ ," the y-component of the acceleration sensor value "y ₂ ," and the z-component of the acceleration sensor value "z ₂ ." Hereinafter, the x-component of the acceleration sensor value will be referred to as the "x-component acceleration sensor value," the y-component of the acceleration sensor value as the "y-component acceleration sensor value," and the z-component of the acceleration sensor value as the "z-component acceleration sensor value."

The air pressure sensor value is the value of air pressure measured by the air pressure sensor included in the tag 20. For example, the measurement data at date and time "t ₁ " includes the air pressure sensor value "p ₁ ". Similarly, for example, the measurement data at date and time "t ₂ " includes the air pressure sensor value "p ₂ ".

In this way, the measurement data stored in the measurement data storage unit 300 includes the date and time, acceleration sensor value, and atmospheric pressure sensor value for each tag ID, where Δ ₁ = t _i+1 - t _i (i is an integer equal to or greater than 1).

<Functional configuration of posture estimation processing unit 100>
The functional configuration of the posture estimation processing unit 100 according to this embodiment is shown in Fig. 3. As shown in Fig. 3, the posture estimation processing unit 100 according to this embodiment includes a tag angle calculation unit 101, a basic angle calculation unit 102, and a posture estimation unit 103.

The tag angle calculation unit 101 calculates the tag angle of each cow every predetermined time _Δ3 (for example, _Δ3 = 1 [m]). At this time, the tag angle calculation unit 101 extracts only the gravitational acceleration component from each of the x-component acceleration sensor values and z-component acceleration sensor values included in the current measurement data of the cow, and calculates the tag angle from the gravitational acceleration component in the x-axis direction and the gravitational acceleration component in the z-axis direction. The tag angle calculation unit 101 then stores the tag angle data including the date and time, tag ID, and tag angle in the tag angle data storage unit 500. Hereinafter, the gravitational acceleration component in the x-axis direction will also be referred to as the "x-component gravitational acceleration value," and the gravitational acceleration component in the z-axis direction will also be referred to as the "z-component gravitational acceleration value."

Here, the tag angle is the angle that represents how much the z-axis has rotated with the direction of gravity as the reference (i.e., 0) and the y-axis as the axis of rotation. Ideally, as shown in Figure 4, when the cow is standing or lying down, the positive direction of the x-axis is opposite to the direction of gravity, and the positive direction of the z-axis is to the right of the cow's front. Therefore, in this case, if the tag angle is θ, the x-component gravitational acceleration value is _gx , and the z-component gravitational acceleration value is _gz , then θ = tan ^-1 ( _gx / _gz ) = -π/2. On the other hand, when the cow is lying down, the positive direction of the x-axis is to the right (or left) of the cow's front, and the positive direction of the z-axis is the direction of gravity (or the opposite direction to the direction of gravity). Therefore, in this case, θ = tan ^-1 ( _gx / _gz ) = nπ (where n = -1, 0, or 1).

However, Figure 4 above shows an ideal situation in which each tag 20 is attached to the same position on each cow's neck and the collar, belt, etc. used to attach the tag 20 do not shift (or the influence of these can be ignored). In reality, the tag 20 may be attached to slightly different positions depending on the individual cow, and the collar, belt, etc. may shift depending on the cow's movements. Therefore, in this embodiment, an angle called the basic angle is introduced, and the basic angle of each cow is calculated every predetermined time _Δ4 (for example, _Δ4 = 10 [m]), and whether the cow is lying down or prone is estimated from the difference between the tag angle and the basic angle. Note that _Δ4 ≧ _Δ3 must be satisfied, and it is particularly preferable that _Δ4 = _N1 × _Δ3 , where _N1 is a natural number.

The basic angle calculation unit 102 calculates the basic angle of each cow every predetermined time _Δ4 . At this time, the basic angle calculation unit 102 calculates the basic angle of the cow using past tag angle data of the cow from the tag angle data stored in the tag angle data storage unit 500 and past posture data of the cow from the posture data stored in the posture data storage unit 600. Specifically, the basic angle calculation unit 102 calculates the average value of tag angles that satisfy predetermined conditions from tag angles of the cow over the past _T1 hours (for example, _T1 = 12 h or _T1 = 24 h), as the basic angle of the cow. Here, the tag angle that satisfies the predetermined condition is, for example, the tag angle when the cow is standing, but an additional condition may be that the amount of fluctuation in the tag angle is small.

The posture estimation unit 103 estimates whether the cow's posture is standing or lying down by barometric pressure analysis every predetermined time _Δ4 . Here, barometric pressure analysis refers to a method of estimating whether the cow's posture is standing or lying down from the difference between the barometric pressure sensor value included in the cow's measurement data and the reference barometric pressure sensor value included in the reference barometric pressure data at the same time on the same day. For details of barometric pressure analysis, see, for example, Japanese Patent Application Laid-Open Nos. 2019-97475, 2019-103442, 2020-198828, and 2020-198829.

Furthermore, if the posture estimation unit 103 estimates that the cow's posture is lying down, it further estimates whether the cow is lying down or prone using the cow's current tag angle and base angle. The posture estimation unit 103 then stores posture data including the date and time, tag ID, and estimated posture in the posture data storage unit 600.

<Functional configuration of the abnormality detection processing unit 200>
The functional configuration of the anomaly detection processing unit 200 according to this embodiment is shown in Fig. 5. As shown in Fig. 5, the anomaly detection processing unit 200 according to this embodiment includes a score calculation unit 201 and an anomaly detection unit 202.

The score calculation unit 201 calculates a predetermined score at a predetermined time Δ ₅ (for example, Δ ₅ = Δ ₄ ) using the measurement data of the cow from among the measurement data stored in the measurement data storage unit 300, in accordance with the posture represented by the latest posture data of the cow from among the posture data stored in the posture data storage unit 600. It is preferable that Δ ₅ ≧ Δ ₄ , and more preferably that Δ 5 can be expressed as Δ ₅ = N ₂ × Δ ₄ for a natural number N ₂ , with Δ ₅ = Δ ₄ being most suitable.

The abnormality detection unit 202 detects an abnormality in the cow (i.e., difficulty standing up) using the score calculated by the score calculation unit 201. At this time, the abnormality detection unit 202 detects that an abnormality has occurred in the cow, for example, if the score exceeds a predetermined threshold.

<Tag angle calculation process>
The tag angle calculation process according to this embodiment will be described with reference to Fig. 6. Note that the following steps S101 to S103 are repeatedly executed for each cow at predetermined intervals of _Δ3 . Below, a case where a tag angle for a certain cow is calculated will be described.

First, the tag angle calculation unit 101 obtains the x-component gravitational acceleration value and the z-component gravitational acceleration value from the x-component acceleration sensor value and the z-component acceleration sensor value contained in the latest measurement data for the cow (i.e., the latest measurement data containing the tag ID of the tag 20 attached to the cow) from the measurement data stored in the measurement data storage unit 300 (step S101).

Here, the acceleration sensor value for each component is a composite value of the acceleration sensor value due to the cow's movement and the acceleration sensor value due to gravitational acceleration. Furthermore, when the acceleration sensor value due to the cow's movement and the acceleration sensor value due to gravitational acceleration are converted into a frequency spectrum, the acceleration sensor value due to the cow's movement appears in the high-frequency components, while the acceleration sensor value due to gravitational acceleration appears in the low-frequency components. For this reason, the tag angle calculation unit 101 can obtain (extract) the x-component gravitational acceleration value from the x-component acceleration sensor value and the z-component gravitational acceleration value from the z-component acceleration sensor value, for example, by using a low-pass filter that attenuates acceleration sensor values with frequencies above a predetermined frequency.

Next, the tag angle calculation unit 101 calculates the tag angle using the x-component gravitational acceleration value and z-component gravitational acceleration value acquired in step S101 (step S102). That is, if the tag angle is θ, the x-component gravitational acceleration value is g _x , and the z-component gravitational acceleration value is g _z , the tag angle calculation unit 101 calculates the tag angle θ by θ = tan ^-1 (g _x /g _z ).

Then, the tag angle calculation unit 101 creates tag angle data including the current date and time, the tag ID of the tag 20 attached to the cow, and the tag angle θ calculated in step S102 above, and stores this tag angle data in the tag angle data storage unit 500 (step S103). As a result, the tag angle data for each cow is stored in the tag angle data storage unit 500 every predetermined time _Δ3 .

<Posture estimation processing>
The posture estimation process according to this embodiment will be described with reference to Fig. 7. Note that the following steps S201 to S211 are repeatedly executed for each cow at predetermined intervals of Δ _4. The following describes the case where the posture of a certain cow is estimated.

First, the basic angle calculation unit 102 calculates the basic angle of the cow using the cow's past tag angle data from the tag angle data stored in the tag angle data storage unit 500 and the cow's past posture data from the posture data stored in the posture data storage unit 600 (step S201).

Specifically, the basic angle calculation unit 102 extracts tag angles that satisfy predetermined conditions from the tag angles included in the tag angle data for the past T ₁ hour of the cow, and calculates the average value of the extracted tag angles as the basic angle of the cow. Here, as described above, the tag angle that satisfies the predetermined condition may be, for example, the tag angle when the cow is standing, or in addition, it may be a tag angle with little fluctuation over each time width of _Δ4 (that is, for example, a tag angle whose fluctuation falls within a certain predetermined range). Below, the basic angle of the cow will be represented by φ.

Next, the posture estimation unit 103 estimates whether the cow's posture is standing or lying down through air pressure analysis (step S202). That is, the posture estimation unit 103 estimates whether the cow's posture is standing or lying down based on the following (1-1) to (1-3). Note that, for simplicity, it is assumed below that there is no data loss and that noise removal is not required. For more details, please refer to, for example, Japanese Patent Application Laid-Open No. 2019-97475, Japanese Patent Application Laid-Open No. 2019-103442, Japanese Patent Application Laid-Open No. 2020-198828, and Japanese Patent Application Laid-Open No. 2020-198829.

(1-1) Obtain the barometric pressure sensor values included in the measurement data for the cow from Δ ₄ years ago to the present from among the measurement data stored in the measurement data storage unit 300, and the reference barometric pressure sensor values included in the reference barometric pressure data from Δ ₄ years ago to the present from among the reference barometric pressure data stored in the reference barometric pressure data storage unit 400. Hereinafter, let j = 1, ..., J be an index representing the date and time from Δ ₄ years ago to the present, and let the set of barometric pressure sensor values obtained here be {P _j ; j = 1, ..., J} and the set of reference barometric pressure sensor values be {Q _j ; j = 1, ..., J}.

(1-2) Calculate a differential barometric pressure sensor value by subtracting the reference barometric pressure sensor value from the barometric pressure sensor value at the same date and time. That is, calculate a set of differential barometric pressure sensor values {R _j := P _j −Q _j ; j=1, ..., J}.

(1-3) An index value such as the average value of the differential air pressure sensor values R _j (j = 1, ..., J) is calculated, and if this index value is higher than a predetermined threshold th ₁ (< 0), the cow's posture is estimated to be lying down, otherwise the cow's posture is estimated to be standing.

Next, the posture estimation unit 103 determines whether the posture estimated in step S202 above is a lying down posture (step S203).

If the cow is not determined to be in a lying position in step S203 above, the posture estimation unit 103 estimates that the cow's posture is standing (step S204).

On the other hand, if the cow is determined to be in a lying position in step S203, the posture estimation unit 103 compares the tag angle θ contained in the latest tag angle data for the cow among the tag angle data stored in the tag angle data storage unit 500 with the basic angle φ calculated in step S201 (step S205). That is, the posture estimation unit 103 calculates Δθ:=|θ-φ|.

Next, the posture estimation unit 103 determines whether the tag angle θ is sufficiently close to the base angle φ as a result of the comparison in step S205 (step S206). That is, the posture estimation unit 103 determines whether Δθ is smaller than a predetermined threshold value ε.

If it is determined in step S206 that the tag angle θ is sufficiently close to the base angle φ, the attitude estimation unit 103 calculates the degree of fluctuation in the tag angle over a certain period _T2 in the past (for example, _T2 = _Δ4 ) (step S207). Various index values can be used to represent the degree of fluctuation in the tag angle, and any index value can be calculated, but for example, variance or standard deviation can be used.

For example, if _Δ3 = 1 [m], _Δ4 = 10 [m], and _T2 = _Δ4 , then 10 tag angles _θ1 (= θ), _θ2 , ..., _θ10 are obtained as the tag angles for a certain period of time in the past _T2 . Therefore, in this case, the variance and standard deviation of these 10 tag angles _θ1 , _θ2 , ..., _θ10 can be calculated as the degree of fluctuation in the tag angles.

Next, the attitude estimation unit 103 determines whether the degree of fluctuation calculated in step S207 is large (step S208). That is, the attitude estimation unit 103 determines whether the degree of fluctuation of the tag angle is larger than a predetermined threshold value _th2 .

If it is not determined in step S208 above that the tag angle is significantly unstable, the posture estimation unit 103 estimates that the cow's posture is lying down (step S209).

On the other hand, if it is determined in step S206 above that the tag angle θ is not sufficiently close to the basic angle φ, or if it is determined in step S208 above that the tag angle is significantly unstable, the posture estimation unit 103 estimates that the cow's posture is lying down (step S210). In this way, even if the tag angle θ is sufficiently close to the basic angle φ, if the tag angle is significantly unstable, the cow's posture is estimated to be lying down. This is because even if the cow has difficulty standing up, it may raise its head several times, and in this case the tag angle θ may be sufficiently close to the basic angle φ.

Then, the posture estimation unit 103 stores posture data including the current date and time, the tag ID, and the posture estimated in step S204, step S209, or step S210 in the posture data storage unit 600 (step S211). As a result, the posture data of each cow is stored in the posture data storage unit 600 every predetermined time Δ ₄ .

Note that steps S207 and S208 do not necessarily have to be performed. For example, if it is determined in step S206 that the tag angle θ is sufficiently close to the basic angle φ, the cow's posture may be estimated to be lying down. However, performing steps S207 and S208 can further improve the accuracy of detecting difficulty in standing up.

<Abnormality detection processing>
The abnormality detection process according to this embodiment will be described with reference to Fig. 8. Note that the following steps S301 to S306 are repeatedly executed for each cow at predetermined intervals of Δ _5. The following describes the case where an abnormality in a certain cow is detected.

First, the score calculation unit 201 determines whether the posture represented by the latest posture data of the cow among the posture data stored in the posture data storage unit 600 (i.e., the current posture of the cow) is standing, lying down, or recumbent (step S301).

If it is determined in step S301 above that the cow's current posture is lying down, the score calculation unit 201 does nothing. If it is determined in step S301 above that the cow's current posture is standing up, the score calculation unit 201 resets the score S to 0 (step S302).

On the other hand, if the cow's current posture is determined to be recumbent in step S301 above, or following step S302, the score calculation unit 201 calculates the score s (step S303). That is, the score calculation unit 201 calculates the score s according to the following (2-1) to (2-4). Here, the score s is an index value for detecting abnormalities in the cow, and is calculated from the acceleration sensor values contained in the cow's measurement data from Δ ₅ years ago to the present. Note that, for simplicity, it is assumed below that there is no data loss and that noise removal is not required. For details on how to calculate the score s, please refer to, for example, Japanese Patent Application Laid-Open No. 2019-97475, Japanese Patent Application Laid-Open No. 2020-198828, Japanese Patent Application Laid-Open No. 2020-198829, etc.

(2-1) Of the measurement _data stored in the measurement data storage unit 300, acquire the acceleration sensor values (x-component acceleration sensor value, y-component acceleration sensor value, and z-component acceleration sensor value) included in the measurement data for the cow from Δ 5 years ago to the present. Hereinafter, the index representing the date and time from Δ ₅ years ago to the present will be k = 1, ..., K, and the set of acceleration sensor values acquired here will be {( _Xk , _Yk , _Zk ); k = 1, ..., K}. Note that _Xk represents the x-component acceleration sensor value, _Yk represents the y-component acceleration sensor value, and _Zk represents the z-component acceleration sensor value.

(2-2) Calculate the L2 norm of each acceleration sensor value ( _Xk , _Yk , _Zk ). That is, calculate _Lk = (| _Xk | ² + | _Yk | ² + | _Zk | ² ) ^1/2 for each k = 1, ..., K.

(2-3) Calculate the standard deviation of the L2 norm in units of a predetermined time Δ ₆ (where Δ ₆ < Δ ₅ ). Hereinafter, the calculated standard deviation is referred to as σ _d , and a set of standard deviations {σ _d ; d = 1, ..., D} is assumed to be obtained. Note that Δ ₆ can preferably be expressed as Δ ₅ = N ₃ × Δ ₆ where N ₃ is a natural number.

For example, when Δ ₁ = 2 [s], Δ ₅ = 10 [m], and Δ ₆ = 1 [m], 300 L2 norms L _k (k = 1, ..., 300) are obtained in the above (2-2), and 30 L2 norms L _k are included per minute. Therefore, one standard deviation σ _d is calculated from the 30 L2 norms L _k , and as a result, a set of 10 standard deviations {σ _d ; d = 1, ..., 10} is obtained.

(2-4) Then, using the set of standard deviations {σ _d ; d = 1, ..., D}, the number of times that the standard deviation exceeded a predetermined upper limit and then fell below a predetermined lower limit, and the number of times that the standard deviation fell below the predetermined lower limit and then exceeded the predetermined upper limit, from the past _Δ 5 years to the present, are counted, and the sum of these is calculated as the score s. Note that these upper and lower limits are preset parameters and are determined, for example, based on empirical rules, etc.

For example, let us say that the upper limit value is sup and the lower limit value is inf, and the standard deviations are arranged in chronological order as σ ₁ , ..., σ _D. In this case, for example, if sup < σ _d at a certain d, and then σ _d' < inf at a certain d'(> d), this is counted as one occurrence. Similarly, for example, if σ _d < inf at a certain d, and then sup < σ _d' at a certain d'(> d), this is counted as one occurrence.

Note that the score s is the sum of the number of times the standard deviation of the L2 norm of the acceleration sensor values over Δ ₆ exceeded the upper limit and then fell below the lower limit, and the number of times it fell below the lower limit and then exceeded the upper limit, so the value is high when the cow is continuously moving strongly. This is because, in general, when cows have difficulty standing up, they tend to struggle and therefore continuously move strongly (in other words, they tend to continuously move violently while lying down). In this sense, the score s can also be said to be an index value that represents the degree of the cow's violent behavior over the past Δ ₅ .

Next, the score calculation unit 201 adds the score s calculated in step S303 above to the score S (step S304).

Next, the anomaly detection unit 202 determines whether the score S exceeds a predetermined threshold value _th3 (step S305). The threshold value _th3 is a preset parameter, and its value can be set to any value as appropriate. For example, it can be set to about 15 to 35, and it is particularly preferable to set it to about 25.

If it is determined in step S305 above that the score S exceeds a predetermined threshold _th3 , the abnormality detection unit 202 detects that an abnormality (i.e., difficulty standing up) has occurred in the cow (step S306). The abnormality detection unit 202 may notify, for example, a terminal (e.g., a smartphone used by the farmer) connected to the abnormality detection device 10 via a communication network of the abnormality detection result. In this case, the abnormality detection result includes information such as the date and time of the abnormality occurrence, information identifying the cow in which the abnormality occurred (e.g., a tag ID), and information indicating the type of abnormality (e.g., a code indicating difficulty standing up). This enables the terminal to, for example, issue an alert indicating that an abnormality has occurred in the cow.

If it is not determined in step S305 that the score S exceeds the predetermined threshold value _th3 , the anomaly detection unit 202 does nothing.

<Effects of this embodiment>
As described above, the anomaly detection system 1 according to this embodiment classifies the posture of a cow into standing and lying down, and further classifies the lying down posture into prone and recumbent, and increases the score S, which captures the cow's continuous strong movements (i.e., the cow's continuous violent behavior), only when the cow is lying down. On the other hand, the score S is reset when the cow is standing up, and the value of the score S is maintained when the cow is recumbent. This prevents the score S from being increased for normal movements that are made when the cow is lying down (for example, shaking its head or scratching its body), thereby preventing false detection and making it possible to achieve highly accurate anomaly detection.

<Modification>
In the above embodiment, difficulty in standing up is detected based on the knowledge that when a cow has difficulty standing up, it generally makes strong, continuous movements (i.e., the cow thrashes about in a lying position). However, if an abnormality is detected but the cow is left in a state of difficulty in standing up, the cow may become lethargic and the abnormality may not be detected thereafter. In addition, in very rare cases, there may be cows that continue to have difficulty standing up while in a lethargic state.

Therefore, in order to be able to detect the above-mentioned difficulty in standing up in a lethargic state as an abnormality, the abnormality detection unit 202 may detect an abnormality, for example, when the duration that the cow remains lying down exceeds a predetermined threshold value th _4. The duration that the cow remains lying down may be calculated from the date and time and the posture included in the posture data of the cow stored in the posture data storage unit 600.

The present invention is not limited to the specifically disclosed embodiments above, and various modifications, alterations, and combinations with known technologies are possible without departing from the scope of the claims.

REFERENCE SIGNS LIST 1 Anomaly detection system 10 Anomaly detection device 100 Attitude estimation processing unit 101 Tag angle calculation unit 102 Basic angle calculation unit 103 Attitude estimation unit 200 Anomaly detection processing unit 201 Score calculation unit 202 Anomaly detection unit 300 Measurement data storage unit 400 Reference atmospheric pressure data storage unit 500 Tag angle data storage unit 600 Attitude data storage unit

Claims (5)

An abnormality detection device for detecting abnormalities in livestock,
a posture estimation unit that estimates the posture of the livestock based on measurement data including a triaxial acceleration sensor value and a barometric pressure sensor value measured by a triaxial acceleration sensor and a barometric pressure sensor included in equipment attached to the livestock, and reference barometric pressure data that indicates a barometric pressure value that is a reference for the barometric pressure sensor values ;
a score calculation unit that calculates a score for detecting an abnormality in the livestock based on the posture and the triaxial acceleration sensor value;
an abnormality detection unit that detects abnormalities in the livestock based on the score;
and
the device is fixedly attached to the neck of the livestock in such a manner that, when the livestock is standing and its neck is not bent, the direction opposite to the weight direction corresponds to the positive direction of the x-axis of the triaxial acceleration sensor, the moving direction of the livestock corresponds to the positive direction of the y-axis of the triaxial acceleration sensor, and the rightward direction with respect to the moving direction of the livestock corresponds to the positive direction of the z-axis of the triaxial acceleration sensor;
The posture estimation unit
estimating whether the posture of the livestock is standing or lying down based on the air pressure sensor value and the air pressure value represented by the reference air pressure data ;
If the posture of the livestock is estimated to be the lying posture, estimate whether the posture of the livestock is lying down or prone based on an average value of angles that satisfy a predetermined condition among angles of the device during a predetermined period in the past and the angle of the device ;
The angle of the device is an angle that represents how much the z-axis of the triaxial acceleration sensor has rotated with the y-axis of the triaxial acceleration sensor as the axis of rotation, relative to the direction of gravity . The score calculation unit
Calculating the score depending on whether the posture is standing or lying down ;
The anomaly detection device according to claim 1 , wherein the score is not calculated when the posture is the prone position . The score calculation unit
The anomaly detection device according to claim 2 , wherein the score is calculated as a value representing the degree of violence of the livestock while lying down. An abnormality detection device that detects abnormalities in livestock
a posture estimation step of estimating the posture of the livestock based on measurement data including a triaxial acceleration sensor value and a barometric pressure sensor value measured by a triaxial acceleration sensor and a barometric pressure sensor included in equipment attached to the livestock, and reference barometric pressure data representing a barometric pressure value that is a reference for the barometric pressure sensor values;
a score calculation step of calculating a score for detecting an abnormality in the livestock based on the posture and the triaxial acceleration sensor value;
an anomaly detection procedure for detecting an anomaly in the livestock based on the score;
Run
the device is fixedly attached to the neck of the livestock in such a manner that, when the livestock is standing and its neck is not bent, the direction opposite to the weight direction corresponds to the positive direction of the x-axis of the triaxial acceleration sensor, the moving direction of the livestock corresponds to the positive direction of the y-axis of the triaxial acceleration sensor, and the rightward direction with respect to the moving direction of the livestock corresponds to the positive direction of the z-axis of the triaxial acceleration sensor;
The pose estimation procedure includes:
estimating whether the posture of the livestock is standing or lying down based on the air pressure sensor value and the air pressure value represented by the reference air pressure data ;
If the posture of the livestock is estimated to be the lying posture, estimate whether the posture of the livestock is lying down or prone based on an average value of angles that satisfy a predetermined condition among angles of the device during a predetermined period in the past and the angle of the device ;
The anomaly detection method , wherein the angle of the device is an angle representing how much the z-axis of the triaxial acceleration sensor has rotated with the y-axis of the triaxial acceleration sensor as the axis of rotation, relative to the direction of gravity . Anomaly detection devices that detect abnormalities in livestock
a posture estimation step of estimating the posture of the livestock based on measurement data including a triaxial acceleration sensor value and a barometric pressure sensor value measured by a triaxial acceleration sensor and a barometric pressure sensor included in equipment attached to the livestock, and reference barometric pressure data representing a barometric pressure value that is a reference for the barometric pressure sensor values;
a score calculation step of calculating a score for detecting an abnormality in the livestock based on the posture and the triaxial acceleration sensor value;
an anomaly detection procedure for detecting an anomaly in the livestock based on the score;
Execute
the device is fixedly attached to the neck of the livestock in such a manner that, when the livestock is standing and its neck is not bent, the direction opposite to the weight direction corresponds to the positive direction of the x-axis of the triaxial acceleration sensor, the moving direction of the livestock corresponds to the positive direction of the y-axis of the triaxial acceleration sensor, and the rightward direction with respect to the moving direction of the livestock corresponds to the positive direction of the z-axis of the triaxial acceleration sensor;
The pose estimation procedure includes:
estimating whether the posture of the livestock is standing or lying down based on the air pressure sensor value and the air pressure value represented by the reference air pressure data ;
If the posture of the livestock is estimated to be the lying posture, estimate whether the posture of the livestock is lying down or prone based on an average value of angles that satisfy a predetermined condition among angles of the device during a predetermined period in the past and the angle of the device ;
The angle of the device is an angle that indicates how much the z-axis of the three-axis acceleration sensor has rotated with the y-axis of the three-axis acceleration sensor as the axis of rotation, relative to the direction of gravity .