Ear mark sensor and respiratory organ disease determination system

Abstract

To provide an ear mark sensor and a respiratory organ disease determination system, capable of determining affection of a livestock individual more accurately and earlier.SOLUTION: An ear mark sensor 1 comprises a piezoelectric element 10, a housing 2, a penetration bar 3, a presser plate 4, and a spring unit 6. The housing 2 houses an amplifier for amplifying a signal detected by the piezoelectric element 10, a wireless unit for wirelessly outputting the signal amplified by the amplifier, and a battery for supplying power to the piezoelectric element 10 and the amplifier and is arranged on an external surface so that the piezoelectric element 10 is in close contact with an ear Y of a living bod. The presser plate 4 is arranged so as to sandwich the ear Y between it and the piezoelectric element 10. The penetration bar 3 extends from the housing 2 so as to penetrate the piezoelectric element 10, the ear Y, and the presser plate 4. Both ends of the spring unit 6 in a compressed state are inserted between a tip of the penetration bar 3 and the presser plate 4 to energize the presser plate 4 toward the ear Y.SELECTED DRAWING: Figure 1

Description

Technical Field

[0001] The present invention relates to an ear tag sensor and a respiratory disease determination system.

Background Art

[0002] Generally, livestock are raised in groups in a livestock house. However, if the detection of diseases such as respiratory diseases is delayed, there is a risk of disease exacerbation and spread. As a method to prevent this, a method for determining disease in groups using a microphone and a camera has been proposed (see, for example, Non-Patent Document 1). In addition, a method for determining disease by attaching an ear tag sensor to each livestock individual and recording body conduction sound that is resistant to noise has been disclosed (see, for example, Patent Documents 1 and 2).

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Patent Document 2

Non-Patent Documents

[0004]

Non-Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0005] Recently, the number of livestock raised in groups has been increasing steadily, making it difficult to determine the disease status of each individual. There is a need to develop a system that can more accurately and earlier determine the disease status of individual livestock.

[0006] The present invention has been made under the above circumstances, and an object thereof is to provide an ear tag sensor and a respiratory disease determination system that can more accurately and earlier determine the disease status of individual livestock among livestock raised in groups.

Means for Solving the Problems

[0007] To achieve the above object, an ear tag sensor according to a first aspect of the present invention includes a piezoelectric element, an amplifier that amplifies a signal detected by the piezoelectric element, a wireless unit that wirelessly outputs the signal amplified by the amplifier, a battery that supplies power to the piezoelectric element and the amplifier, and a housing that houses them and is disposed on an outer surface so that the piezoelectric element is in close contact with the ear of a living body; a holding plate disposed so as to sandwich the ear between the holding plate and the piezoelectric element; a through rod extending through the piezoelectric element, the ear, and the holding plate from the housing; a spring portion that is inserted between the tip of the through rod and the holding plate in a compressed state and biases the holding plate toward the ear; and is provided with.

[0008] The spring portion has a through hole through which the through rod passes and has a spherical shape curved along the direction in which the through rod extends, and is composed of a cylindrical elastic member that can expand and contract in the direction in which the through rod extends. It may also be.

[0009] It may also be provided with a cylindrical protective member that covers the side surface of the spring portion. It may also be.

[0010] Each time the wireless unit receives a detection command at regular intervals, The battery supplies power to the piezoelectric element and the amplifier, while the piezoelectric element is being supplied with power from the battery, it detects the body conduction sound of the living body, and the wireless unit wirelessly outputs the signal detected by the piezoelectric element. This may also be the case.

[0011] The respiratory disease determination system according to the second aspect of the present invention includes a data collection unit that collects waveform data of body conduction sound with biometric identification information added from a sensor attached to a living body for each living body, a preprocessing unit that performs preprocessing on the waveform data of the body conduction sound collected by the data collection unit, and a determination unit that determines, for each living body, the presence or absence of a respiratory disease by machine learning based on the waveform data of the body conduction sound preprocessed by the preprocessing unit. It is provided with.

[0012] The preprocessing unit outputs, as waveform data of the body conduction sound on which preprocessing has been performed, two-dimensional waveform data obtained by performing time-frequency analysis on the waveform data of the body conduction sound collected by the data collection unit. This may also be the case.

[0013] The determination unit may have a deep learning device that determines the presence or absence of a respiratory disease based on the waveform data preprocessed by the preprocessing unit. This may also be the case.

[0014] The deep learning device may include an LSTM (Long Short-Term Memory) network. This may also be the case.

[0015] The determination unit may include a feature quantity calculation unit that calculates the feature quantity of the waveform data preprocessed by the preprocessing unit. A learning unit having a machine learning device that determines the presence or absence of a respiratory disease based on the feature amount calculated by the feature amount calculation unit. It may be provided with as well.

Advantages of the Invention

[0016] According to the present invention, it is possible to more accurately and earlier determine the disease state of livestock individuals.

Brief Description of the Drawings

[0017]

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Figure 7

Figure 8

Figure 9

Figure 10

Figure 11

Figure 12

Embodiments for Carrying Out the Invention

[0018] Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings. In each drawing, the same or equivalent parts are denoted by the same reference numerals. In the following embodiments, expressions such as "having", "including", or "containing" also include the meaning of "consisting of" or "composed of".

[0019] Embodiment 1 First, Embodiment 1 of the present invention will be described. In this embodiment, a case will be described in which it is determined whether or not a pig P (see FIG. 3), which is a livestock animal group-housed in a livestock barn, is suffering from a respiratory disease.

[0020] [Ear tag sensor] It is necessary to determine whether or not each individual is suffering from a respiratory disease. For this reason, a sensor is attached to each pig P (see FIG. 3). This sensor is an ear tag sensor 1 (see FIG. 3) attached to the ear Y of the pig P.

[0021] The configuration of the ear tag sensor 1 will be described. As shown in FIG. 1(A), the ear tag sensor 1 includes a housing 2, a through rod 3, a holding plate 4, a tip portion 5, a spring portion 6, a piezoelectric element 10, and a sheet 11.

[0022] The housing 2 is a housing that houses a signal circuit inside. The housing 2 is a member having rigidity and is made of, for example, resin. An ear tag number (not shown), which is identification information for identifying the individual of the pig P, is displayed on the outer surface of the housing 2. The through rod 3 is a rigid rod-shaped member that extends in a direction orthogonal to the outer surface of the housing 2 from one end fixed to the housing 2. The housing 2 and the through rod 3 are integrated.

[0023] The holding plate 4, the tip portion 5, the spring portion 6, the piezoelectric element 10, and the sheet 11 are each provided with a through hole through which the through rod 3 penetrates at the center. The piezoelectric element 10 and the sheet 11 are stacked on the housing 2 in the order of the piezoelectric element 10 and the sheet 11 with the through rod 3 inserted into their respective through holes. In other words, the through rod 3 extends from the housing 2 through the piezoelectric element 10 and the sheet 11.

[0024] The piezoelectric element 10 is connected to a signal circuit inside the housing 2. The piezoelectric element 10 deforms in response to a force transmitted from the outside and outputs a voltage signal corresponding to the deformation to the signal circuit inside the housing 2. The sheet 11 is inserted to insulate the piezoelectric element 10 from a living body (the pig P in FIG. 3) as a conductor, and is made of a material that causes no problem even when it touches the living body.

[0025] The ear portion Y of the pig P (see FIG. 3) to which the ear tag sensor 1 is attached is provided with a through hole through which the through rod 3 penetrates. The through rod 3 is inserted into the through hole of the ear portion Y in a state where it is inserted into the through holes of the piezoelectric element 10 and the sheet 11. As a result, as shown in FIG. 1(B), on the housing 2, the piezoelectric element 10, the sheet 11, and the ear portion Y are stacked in this order.

[0026] As shown in FIGS. 1(A) and 1(B), the through rod 3 is inserted into the through holes of the holding plate 4, the spring portion 6, and the tip portion 5. As a result, the ear portion Y of the pig P is sandwiched between the set of the housing 2, the piezoelectric element 10, and the sheet 11 and the set of the holding plate 4, the spring portion 6, and the tip portion 5.

[0027] The holding plate 4 is a plate-like member made of a rigid material (for example, resin). The holding plate 4 is arranged so as to sandwich the ear portion Y between it and the piezoelectric element 10. The tip portion 5 is made of a rigid material (for example, resin). The tip portion 5 is fixed to the tip of the through rod 3 by a bolt portion 5a. The spring portion 6 is a spring arranged to expand and contract in the direction in which the through rod 3 extends. The spring portion 6 is inserted between the tip of the through rod 3 and the holding plate 4 in a compressed state, and biases the holding plate 4 toward the ear portion Y.

[0028] Due to the elastic force of the spring portion 6, the ear portion Y is sandwiched between the piezoelectric element 10 and the pressing plate 4, and the ear portion Y and the piezoelectric element 10 can be brought into close contact with each other. Therefore, the bioconductive sound transmitted from the ear portion Y by the piezoelectric element 10 can be detected with high sensitivity.

[0029] [Structure of the spring portion] The spring portion 6 is composed of a cylindrical elastic member. This cylindrical member has a through-hole through which the through-bar 3 passes and has a shape curved in a spherical shape along the direction in which the through-bar 3 extends, and is stretchable and contractible in the direction in which the through-bar 3 extends. The spring portion 6 is disposed between the tip of the through-bar 3 and the pressing plate 4 in a state of being contracted in the direction in which the through-bar 3 extends. By this spring portion 6, the pressing plate 4 is pressed against the ear portion Y of the pig P (see FIG. 3). Since the spring portion 6 is a cylindrical elastic member having the direction in which the through-bar 3 extends as the axial direction, it is less likely to deform in directions other than the direction in which the through-bar 3 extends compared to a coil spring.

[0030] The ear tag sensor 1 includes a skirt 7. The skirt 7 is made of a material having rigidity. The skirt 7 is a cylindrical protective member that covers the outer periphery of the side surface of the spring portion 6. The skirt 7 can suppress the entry of foreign matter (such as the feed of the pig P) around the spring portion 6 and damage to the spring portion 6 caused by being bitten by other pigs in group breeding. In the present embodiment, the skirt 7 is integrated with the pressing plate 4.

[0031] As shown in FIG. 2(A), the housing 2 houses an amplifier 20, a wireless unit 21, and a battery 22. The amplifier 20 amplifies the signal detected by the piezoelectric element 10 and also converts the analog signal output from the piezoelectric element 10 into a digital signal at a predetermined sampling frequency. The wireless unit 21 is a communication device capable of wireless communication with an external device, and wirelessly outputs the digital signal amplified by the amplifier 20. The piezoelectric element 10 and the amplifier 20 constitute a detection circuit for detecting a signal corresponding to the bioconductive sound of the pig P. The battery 22 supplies power to the signal circuit composed of the piezoelectric element 10 and the amplifier 20.

[0032] As shown in Fig. 2(B), the wireless unit 21 wirelessly receives a detection command from an external device at regular intervals T1 as described below. Each time the wireless unit 21 receives a wireless signal containing the detection command at regular intervals T1, the battery 22 supplies power to the signal circuit composed of the piezoelectric element 10 and the amplifier 20 for a predetermined time T2 (T2 < T1). The predetermined time T2 needs to be wide enough to obtain the signal waveform of the body conduction sound of the pig P. For example, T1 can be set to 60 minutes and T2 can be set to 10 minutes.

[0033] The piezoelectric element 10 detects the body conduction sound transmitted from the ear part Y while power is being supplied from the battery 22 (time T2). The wireless unit 21 wirelessly outputs the signal waveform data of the body conduction sound detected by the piezoelectric element 10 and amplified by the amplifier 20. In this way, the power consumption by the battery 22 can be made intermittent, so that the system can operate with low power consumption and the measurable time can be extended.

[0034] Note that the signal waveform data of the body conduction sound wirelessly output from the wireless unit 21 is attached with an ear tag, that is, an ear tag number for identifying the individual pig P. Thereby, it becomes possible to identify the individual of the pig P corresponding to the wireless signal.

[0035] The weight of the ear tag sensor 1 is preferably less than 20 g. For this reason, it is desirable to use the lightest possible material for parts such as the housing 2. This is to prevent ear lacerations.

[0036] [Respiratory disease determination system] As shown in Fig. 3, the ear tag sensor 1 is attached for each pig P. The respiratory disease determination system 100 according to the present embodiment determines the presence or absence of respiratory disease for each individual pig P based on the body conduction sound of the pig P from the ear tag sensors 1 attached to each of the plurality of pigs P.

[0037] The respiratory disease determination system 100 includes, in addition to the ear tag sensor 1 attached to each pig P, a router 30 and an information processing device 40. In FIG. 3, a plurality of routers 30 are provided, but the router 30 may be one.

[0038] The router 30 is wirelessly connected to the ear tag sensor 1 and is connected to the information processing device 40 via a wired communication network. The router 30 functions as a transmission unit that wirelessly outputs a detection command sent from the information processing device 40 to the ear tag sensor 1, and receives waveform data of the body conduction sound to which the ear tag number of the individual wirelessly output from the ear tag sensor 1 is attached, and functions as a reception unit that transmits it to the information processing device 40. By using a wired communication network between the router 30 and the information processing device 40, it is possible to reduce packet loss of the transmitted data.

[0039] The information processing device 40 collects waveform data of body conduction sound from the ear tag sensors 1 attached to each of a plurality of pigs P via the router 30. Specifically, the information processing device 40 transmits a detection command to the ear tag sensor 1 via the router 30 at regular intervals T1 (see FIG. 2(B)). The ear tag sensor 1 that has received the detection command detects the body conduction sound signal for a predetermined time T2, and wirelessly outputs the detected waveform data of the body conduction sound to the router 30 after adding the ear tag number. The information processing device 40 acquires the waveform data of the body conduction sound via the router 30. The information processing device 40 determines the presence or absence of respiratory disease for each individual based on the acquired waveform data of the body conduction sound.

[0040] [Information processing device] As shown in FIG. 3, the information processing device 40 includes a command unit 41, a data collection unit 42, a preprocessing unit 43, and a determination unit 44.

[0041] [Command unit] The command unit 41 transmits a detection command to the ear tag sensor 1 attached to each pig P via the router 30 at regular intervals T1 (see FIG. 2(B)).

[0042] [Data collection unit] The data collection unit 42 collects the waveform data of the body-conducted sound received from the ear tag sensor 1 via the router 30. The data collection unit 42 stores the collected waveform data of the body-conducted sound in a memory (not shown). The memory may be possessed by the data collection unit 42 or may be constructed in a cloud computer. As described above, the ear tag number is added to the waveform data of the bio-conducted sound. As shown in FIG. 4, the data collection unit 42 stores the signal waveform data of the body-conducted sound for each ear tag number (in FIG. 4, ear tag numbers A, B, ···), that is, for each individual, in a grouped manner.

[0043] [Preprocessing Unit] The preprocessing unit 43 performs preprocessing on the waveform data of the body-conducted sound. The preprocessing includes, for example, a normalization process of normalizing the waveform data of the body-conducted sound based on its maximum amplitude. The waveform data of the body-conducted sound has a large amplitude change when the state of the living body, for example, when the pig P walks, lies down, or eats. This amplitude change greatly affects the determination of the presence or absence of respiratory diseases. Therefore, the information processing device 40 reduces the influence of the amplitude change of the waveform data on the determination result by normalizing the waveform data.

[0044] Furthermore, the preprocessing unit 43 may remove outliers from the waveform data or remove data missing portions.

[0045] Furthermore, in the present embodiment, the preprocessing unit 43 outputs the two-dimensional waveform data obtained by performing time-frequency analysis, such as spectrum analysis, cepstrum analysis, delta cepstrum analysis, or real signal wavelet transform, on the waveform data of the body-conducted sound collected by the data collection unit 42 as the waveform data of the body-conducted sound on which the preprocessing has been performed.

[0046] In spectral analysis, two-dimensional waveform data showing the frequency spectrum obtained by cutting out the waveform data of the in-vivo conduction sound with a window function W and performing a short-time Fourier transform is obtained. This two-dimensional waveform data is called a spectrogram. As shown in Fig. 5(A), in the spectrogram, the horizontal axis represents time (number of frames), and the vertical axis represents frequency. The two-dimensional waveform data generated by spectral analysis is generated as waveform data F1 to FM (M is a natural number of 2 or more) showing the time change of the spectral intensity at each sample frequency. By spectral analysis, the spectral waveforms in the frequency bands of heartbeat and respiration with different frequency bands can be separated.

[0047] In cepstrum analysis, a cepstrum, which is two-dimensional waveform data obtained by further performing a Fourier transform on the frequency series signal obtained by cutting out the time waveform data of the in-vivo conduction sound with a window function W, is obtained. As shown in Fig. 5(B), in the cepstrum, the horizontal axis represents time (number of frames), and the vertical axis represents the number of dimensions. The two-dimensional waveform data generated by cepstrum analysis becomes waveform data F1 to FM (M is a natural number of 2 or more) showing the time change of the cepstrum intensity at each number of dimensions. It is the same as spectral analysis that the spectral waveforms in the frequency bands of heartbeat and respiration with different frequency bands can be separated by cepstrum analysis.

[0048] In delta cepstrum analysis, a delta cepstrum, which is two-dimensional waveform data obtained by calculating the difference between the previous and next two frames from the cepstrum, is obtained. The two-dimensional waveform data generated by delta cepstrum analysis becomes waveform data F1 to FM (M is a natural number of 2 or more) showing the time change of the intensity of the delta cepstrum at each number of dimensions. It is the same as cepstrum analysis that the spectral waveforms in the frequency bands of heartbeat and respiration with different frequency bands can be separated by delta cepstrum analysis.

[0049] The real signal wavelet transform analyzes self-similarity by compressing and stretching a wavelet function having characteristics similar to the waveform data of the intracorporeal conduction sound. By performing the real signal wavelet transform, two-dimensional waveform data showing a frequency spectrum can be obtained, similar to the Fourier transform. The two-dimensional waveform data generated by the real signal wavelet transform becomes waveform data F1 to FM (M is a natural number of 2 or more) showing the temporal change of the spectral intensity at each of a plurality of sample frequencies. By this real signal wavelet transform, the spectral waveforms of the heartbeat and respiration frequency bands with different frequency bands can be separated.

[0050] [Determination unit] Based on the waveform data F1 to FM of the intracorporeal conduction sound preprocessed by the preprocessing unit 43, the determination unit 44 determines, for each individual, the presence or absence of a respiratory disease by machine learning. More specifically, as shown in FIG. 6(A), the determination unit 44 has a deep learning unit 50 as a deep learning device that determines the presence or absence of a respiratory disease based on the waveform data F1 to FM that is the signal preprocessed by the preprocessing unit 43.

[0051] In the present embodiment, the deep learning unit 50 has a recurrent neural network including a Long Short-Term Memory (LSTM) network. The LSTM network, that is, the LSTM, includes a memory cell that stores data input in the past and three gates. The three gates are a forget gate that adjusts how much the value of the memory cell is retained at the next time, an input gate that adjusts the value added to the memory cell, and an output gate that adjusts how much the value of the memory cell affects the next layer. With the memory cell and the three gates, the LSTM enables processing considering time-series data over a long period.

[0052] The deep learning unit 50 can have a configuration shown in, for example, FIG. 6(B). For example, as shown in FIG. 6(B), the deep learning unit 50 includes an input layer (number of units M), a first LSTM (number of units 150) as an intermediate layer, DropOut (probability 0.2), a second LSTM (number of units 100), DropOut (probability 0.2), a fully connected layer (number of units 2), and an output layer. Waveform data F1 to FM are input to each unit of the input layer, and the output layer outputs a determination result of the presence or absence of a respiratory disease. DropOut is a mechanism for preventing overfitting of the deep learning unit 50.

[0053] [Hardware Configuration of Information Processing Apparatus 40] The information processing apparatus 40 shown in FIG. 3 is realized, for example, by a computer having a hardware configuration shown in FIG. 7 executing a software program. Specifically, the information processing apparatus 40 includes a CPU (Central Processing Unit) 61 that controls the entire apparatus, a main memory 62 that operates as a working area of the CPU 61, an external memory 63 that stores an operation program of the CPU 61, a man-machine interface 64, a communication interface 65, and an internal bus 68 that connects these components.

[0054] As will be described later, the CPU 61 realizes various functions of the information processing apparatus 40 by executing a program 69 stored in the main memory 62. The CPU 61 has a timer for timing, and executes processing according to the time measured by the timer.

[0055] The main memory 62 is composed of a RAM (Random Access Memory) or the like. The program 69 executed by the CPU 61 is loaded from the external memory 63 into the main memory 62. Further, the main memory 62 is also used as a working area (temporary data storage area) of the CPU 61. The functions of the instruction unit 41, data collection unit 42, preprocessing unit 43, and determination unit 44 shown in FIG. 3 are realized by the execution of the program of the CPU 61.

[0056] The external memory 63 is composed of non-volatile memories such as flash memories and hard disks. A program 69 for causing the CPU 61 to execute is stored in the external memory 63 in advance.

[0057] The man-machine interface 64 includes devices such as a keyboard and a mouse, and an interface device that connects these devices to the internal bus 68. Further, the man-machine interface 64 includes a display device such as a CRT (Cathode Ray Tube) and a liquid crystal monitor. A touch panel can be adopted as the man-machine interface 64. The determination result of the presence or absence of a disease or the like is displayed by the man-machine interface 64.

[0058] The communication interface 65 is an interface for data communication. Data communication with an external device becomes possible via the communication interface 65. Data communication with the router 30 or the cloud computer is performed via this communication interface 65.

[0059] The functions of the information processing apparatus 40 can be implemented in a computer system including one or more computers including one or more processors and one or more storage devices including a non-temporary storage medium. The plurality of computers communicate with each other via a connected communication network and implement the functions of the information processing apparatus 40. For example, a part of the plurality of functions of the information processing apparatus 40 may be implemented in one computer and another part may be implemented in another computer.

[0060] Next, the operation of the information processing apparatus 40 in the respiratory disease determination system 100 according to the present embodiment will be described.

[0061] [Command Output Processing] First, the command output process for outputting a detection command for body-conducted sound will be described. As shown in FIG. 8, the command unit 41 (see FIG. 3) of the information processing apparatus 40 waits until the transmission timing (step S1; No). As shown in FIG. 2(B), in the respiratory disease determination system, the transmission timing occurs at regular intervals T1. When the transmission timing arrives (step S1; Yes), the command unit 41 outputs a detection command to the router 30 (step S2). The detection command is sent to the ear tag sensor 1 attached to each individual via the router 30. When the ear tag sensor 1 receives the detection command, it detects the body-conducted sound. After executing step S2, the command unit 41 returns to step S1. Thereafter, the command output process is repeated every time a certain interval T1 elapses.

[0062] Note that the command unit 41 may shift the timing and transmit the detection command between the ear tag sensors 1. In this case, the ear tag number of the target ear tag sensor 1 is added to the detection command. The ear tag sensor 1 may be configured to detect the body-conducted sound when it receives the detection command and the ear tag number added to the detection command matches its own.

[0063] [Data collection process] Next, the data collection process for collecting the waveform data of the body-conducted sound will be described. As shown in FIG. 9, the data collection unit 42 of the information processing apparatus 40 waits until it receives the waveform data (step S11; No). When it receives the waveform data (step S11; Yes), the data collection unit 42 extracts the ear tag number added to the waveform data (step S12), and as shown in FIG. 4, stores the received body-conducted sound waveform data in the area corresponding to the extracted ear tag number (step S13). After executing step S13, the data collection unit 42 returns to step S11. Thereafter, the data collection process is repeated every time new waveform data is received.

[0064] In addition, when the detection commands are transmitted with a time shift between the ear tag sensors 1, the timing of receiving the waveform data of the body-conducted sound also shifts, so that it becomes possible to smoothly collect data. For a plurality of waveform data with close reception timings, those data may be accumulated in a temporary memory and processed in order.

[0065] [Data collection process] Next, the disease presence / absence determination process for determining the presence or absence of a respiratory disease in pig P will be described. As shown in FIG. 10, the information processing apparatus 40 waits until new waveform data is acquired by the data collection unit 42 (step S21; No).

[0066] When new waveform data is acquired (step S21; Yes), the preprocessing unit 43 of the information processing apparatus 40 performs preprocessing on the acquired new waveform data (step S22). By this preprocessing, conversion of the waveform data into two-dimensional waveform data (for example, spectrogram), normalization based on the maximum amplitude of the waveform data, etc. are performed as necessary. As a result, waveform data F1 to FM of the body-conducted sound are generated as shown in FIG. 5(A) or FIG. 5(B).

[0067] Subsequently, the determination unit 44 of the information processing apparatus 40 determines the presence or absence of a respiratory disease of the living body based on the waveform data on which the preprocessing has been performed (step S23). Here, the determination of the presence or absence of the disease in the deep learning unit 50 is performed based on the waveform data F1 to FM of the body-conducted sound. After executing step S23, the information processing apparatus 40 returns to step S21. Thereafter, each time new waveform data is received, the disease presence / absence determination process is repeated. Note that the data collection process in FIG. 9 and the disease presence / absence determination process in FIG. 10 may be performed at once.

[0068] [Evaluation result] Actually, an experiment was conducted to evaluate the determination accuracy using the respiratory disease determination system 100 according to the first embodiment. In this experiment, 5-week-old (at the time of inoculation) SPF (Specific Pathogen Free) pigs were inoculated with Pasteurella multocida, which is the pathogen of Pasteurella pneumonia, and deep learning was performed by the deep learning unit 50 to classify into two classes before and after the disease based on the waveform data of the in-vivo conduction sound obtained before and after the infection of three Pasteurella pneumonia-infected pigs. As the waveform data of the collected in-vivo conduction sound, about 100 to 400 pieces of waveform data for 5 seconds when the pig P was in a quiet state were cut out for each individual before and after the disease, and these were used. Those waveform data were divided into training data, verification data, and test data in the deep learning unit 50 and used for model creation and its evaluation.

[0069] Fig. 11(A) shows the classification accuracy rates for the training data, verification data, and test data at the end of the deep learning. As shown in Fig. 11(A), in the training data, the ratio of the number of data (463 + 206) for which the predicted class and the true class matched to the total number of data (463 + 25 + 11 + 206), that is, the correct answer rate, was 94.89%. Similarly, the correct answer rate when the determination was made with the verification data was 93.03% ((130 + 57) / 201). Also, the correct answer rate when the determination was made with the test data was 93.20% ((65 + 31) / 103). Thus, all showed high correct answer rates.

[0070] Next, using the determination model of the deep learning unit 50 created by deep learning using the waveform data of the in-vivo conduction sound of three individuals, an attempt was made to classify into two classes before and after the disease for the data of other individuals. As the prepared waveform data of the in-vivo conduction sound, those collected before inoculation, 3 days after inoculation, and 6 days after inoculation of one 5-week-old Pasteurella pneumonia-infected pig were used. Fig. 11(B) shows the respective classification results. As shown in Fig. 11(B), high correct answer rates were obtained both before inoculation, 3 days after inoculation, and 6 days after inoculation.

[0071] Embodiment 2 Next, Embodiment 2 of the present invention will be described. The respiratory disease determination system 100 according to this embodiment is the same as the respiratory disease determination system 100 according to the above Embodiment 1 in that it determines whether or not pigs P (see FIG. 3) group-housed in a livestock barn are suffering from a respiratory disease.

[0072] The respiratory disease determination system 100 according to this embodiment is different from the respiratory disease determination system 100 according to the above Embodiment 1 in the functional configuration of the determination unit 44. As shown in FIG. 12, the respiratory disease determination system 100 according to this embodiment includes a feature amount calculation unit 51 and a machine learning unit 52.

[0073] The feature amount calculation unit 51 calculates the feature amounts of the waveform data that has been preprocessed by the preprocessing unit 43. Various things can be used as the feature amounts. For example, the feature amounts may include statistical amounts such as the average of the signal level, the root mean square, the standard deviation, the shape factor, the kurtosis, and the strain. Further, the feature amounts may include impulse metrics such as the wave height factor, the impulse factor, and the clearance factor. Further, the feature amounts may include signal processing metrics such as the S / N ratio, the total harmonic distortion, and the SINAD (SIgnal-to-Noise And Distortion ratio).

[0074] The feature amounts calculated by the feature amount calculation unit 51 can be narrowed down to the feature amounts for which a significant difference is found between normal and diseased. Therefore, when examining the significance difference by t-test for normal (before inoculation) and diseased (after inoculation) specimens, a significant difference was obtained for the root mean square, SINAD, standard deviation, and total harmonic distortion. Therefore, it is possible to narrow down the feature amounts calculated by the feature amount calculation unit 51 to these four feature amounts.

[0075] The machine learning unit 52 determines the presence or absence of a respiratory disease based on the feature amounts calculated by the feature amount calculation unit 51. As the classification method in the machine learning unit 52, for example, a tree, linear discrimination, logistic regression, Bayes, support vector machine, k-nearest neighbor method, ensemble, etc. can be adopted, but there is no particular limitation on the classification method.

[0076] As described in detail above, according to the ear tag sensor 1 according to the present embodiment, the piezoelectric element 10 for detecting the body-conducted sound can be brought into close contact with the ear part Y of the pig P by the elastic force of the spring part 6. Therefore, it is possible to more accurately and earlier determine the disease of the pig P.

[0077] Further, according to the ear tag sensor 1 according to the present embodiment, the spring part 6 has a spherical curved shape along the direction in which the through rod 3 extends, and is composed of a cylindrical elastic member that can expand and contract in the direction in which the through rod 3 extends. For this reason, since the spring part 6 has a structure that is difficult to deform in a direction other than its expansion and contraction direction, the piezoelectric element 10 for detecting the body-conducted sound can be accurately brought into close contact with the ear part Y of the pig P by the spring part 6.

[0078] Further, according to the ear tag sensor 1 according to the present embodiment, the cylindrical protective member that covers the outer periphery of the side surface of the spring part 6 can prevent foreign matter from mixing into the periphery of the spring part 6 and reducing the biasing force of the spring part 6, or prevent damage to the spring part caused by being bitten by other pigs in group breeding.

[0079] Further, according to the ear tag sensor 1 according to the present embodiment, when a detection command is received from the outside, power is supplied from the battery 22 to detect the body-conducted sound. As a result, the battery 22 can be operated intermittently, so that it is possible to monitor the body-conducted sound for a long time. As shown in FIG. 2(B), for example, if a certain interval T1 is 60 minutes and the measurement time T2 is 10 minutes, it is possible to measure for a time more than six times longer than normal.

[0080] Further, according to the respiratory disease determination system 100 according to the present embodiment, the presence or absence of a respiratory disease in the pig P is determined by machine learning on the waveform data of the body-conducted sound. Therefore, it is possible to more accurately and earlier determine the disease of the pig P.

[0081] Further, according to the respiratory disease determination system 100 according to the present embodiment, machine learning is performed based on two-dimensional waveform data obtained by performing time-frequency analysis, such as spectrum analysis, cepstrum analysis, delta cepstrum analysis, or real signal wavelet transform, on the waveform data of the body-conducted sound, that is, the waveform decomposed for each specific component. Thereby, for example, machine learning can be performed in a state where the spectrum waveforms of respiration and heartbeat with different frequency bands are separated, so that the disease state of the livestock individual can be determined more accurately and earlier.

[0082] In the above embodiment, two-dimensional signal waveform data such as a spectrogram converted from the signal waveform data on the time axis is input to the machine learning device to determine the presence or absence of a respiratory disease. However, it is not limited to this. The signal waveform data (time series data) on the time axis may be input to the machine learning device to determine the presence or absence of a respiratory disease. Even in this case, in the preprocessing unit 43, normalization processing for normalizing the waveform data of the body-conducted sound based on its maximum amplitude is performed.

[0083] The machine learning performed by the determination unit 44 may be deep learning or other machine learning. It has been clarified that high determination accuracy can be obtained when a recurrent neural network including an LSTM network is used as the deep learning device.

[0084] Further, when performing other machine learning other than deep learning in the determination unit 44, it is necessary to obtain a feature amount from the waveform data of the body-conducted sound and use the feature amount as an input to the machine learning device. By selecting a feature amount having a significant difference in the presence or absence of the disease as the feature amount, it is possible to increase the determination accuracy of the presence or absence of a respiratory disease.

[0085] In the above embodiment, the determination target for the presence or absence of a respiratory disease is the pig P. However, it is not limited to this. Other animals such as cows, which are livestock raised in groups, may be the determination target.

[0086] In addition, the hardware configuration and software configuration of the information processing apparatus 40 are merely examples and can be arbitrarily changed and modified.

[0087] The central part that performs the processing of the information processing apparatus 40, which is composed of a CPU 61, a main memory 62, an external memory 63, a man-machine interface 64, a communication interface 65, an internal bus 68, etc., can be realized using a normal computer system without relying on a dedicated system as described above. For example, a computer program for executing the above operations can be stored and distributed in a computer-readable recording medium (flexible disk, CD-ROM, DVD-ROM, etc.), and the information processing apparatus 40 that executes the above processing may be configured by installing the computer program on a computer. Further, the computer program may be stored in a storage device of a server device on a communication network such as the Internet, and the information processing apparatus 40 may be configured by a normal computer system downloading the program.

[0088] When the functions of the computer are realized by sharing the OS (operating system) and the application program, or by the cooperation between the OS and the application program, only the application program part may be stored in a recording medium or a storage device.

[0089] It is also possible to superimpose a computer program on a carrier wave and distribute it via a communication network. For example, a computer program may be posted on a bulletin board (BBS, Bulletin Board System) on a communication network, and the computer program may be distributed via the network. Then, the computer program may be started and configured to be executable under the control of the OS in the same manner as other application programs to execute the above processing.

[0090] The present invention can be implemented in various embodiments and variations without departing from the broad spirit and scope of the present invention. Also, the above-described embodiments are for explaining the present invention and do not limit the scope of the present invention. That is, the scope of the present invention is indicated by the claims rather than the embodiments. And various modifications made within the scope of the claims and within the scope of the meaning of the invention equivalent thereto are considered to be within the scope of the present invention.

Industrial Applicability

[0091] The present invention is used to determine the morbidity of a living body.

Explanation of Signs

[0092] 1 Ear tag sensor, 2 Housing, 3 Through rod, 4 Suppressing plate, 5 Tip portion, 5a Bolt portion, 6 Spring portion, 7 Skirt, 10 Piezoelectric element, 11 Sheet, 20 Amplifier, 21 Wireless portion, 22 Battery, 30 Router, 40 Information processing device, 41 Command portion, 42 Data collection portion, 43 Preprocessing portion, 44 Judgment portion, 50 Deep learning portion, 51 Feature amount calculation portion, 52 Machine learning portion, 61 CPU, 62 Main memory, 63 External memory, 64 Man-machine interface (I / F), 65 Communication interface (I / F), 68 Internal bus, 69 Program, 100 Respiratory disease judgment system, P Pig, Y Ear

Claims

1. A piezoelectric element, an amplifier that amplifies a signal detected by the piezoelectric element, a wireless unit that wirelessly outputs the signal amplified by the amplifier, a battery that supplies power to the piezoelectric element and the amplifier, and a housing that houses these components and is disposed on an outer surface so that the piezoelectric element is in close contact with the ear of a living body; a pressing plate disposed so as to sandwich the ear between the pressing plate and the piezoelectric element; a through rod extending through the piezoelectric element, the ear, and the pressing plate from the housing; a spring portion that is inserted between the tip of the through rod and the pressing plate in a compressed state and biases the pressing plate toward the ear; An ear tag sensor comprising the above components.

2. The spring portion has a through hole through which the through rod passes and has a spherical curved shape along the direction in which the through rod extends, and is composed of a cylindrical elastic member that can expand and contract in the direction in which the through rod extends. The ear tag sensor according to Claim 1.

3. A cylindrical protective member that covers the side surface of the spring portion is provided. The ear tag sensor according to Claim 1 or 2.

4. Each time the wireless unit receives a detection command at regular intervals, the battery supplies power to the piezoelectric element and the amplifier, the piezoelectric element detects the in-vivo conduction sound of the living body while power is being supplied from the battery, the wireless unit wirelessly outputs the signal detected by the piezoelectric element. The ear tag sensor according to Claim 1.

5. A data collection unit that collects waveform data of in-vivo conduction sound with biological identification information added thereto from a sensor attached to a living body for each living body, A preprocessing unit that performs preprocessing on the waveform data of the in-vivo conduction sound collected by the data collection unit. A determination unit that determines, for each living body, whether or not a respiratory disease is present by machine learning based on the waveform data of the in-vivo conduction sound that has been pre-processed by the pre-processing unit. A respiratory disease determination system comprising the above.

6. The pre-processing unit Outputs two-dimensional waveform data obtained by performing time-frequency analysis on the waveform data of the in-vivo conduction sound collected by the data collection unit as the waveform data of the in-vivo conduction sound on which pre-processing has been performed. The respiratory disease determination system according to claim 5.

7. The determination unit Has a deep learning device that determines whether or not a respiratory disease is present based on the waveform data pre-processed by the pre-processing unit. The respiratory disease determination system according to claim 5 or 6.

8. The deep learning device Includes an LSTM (Long Short-Term Memory) network. The respiratory disease determination system according to claim 7.

9. The determination unit A feature quantity calculation unit that calculates the feature quantity of the waveform data on which pre-processing has been performed by the pre-processing unit, A learning unit having a machine learning device that determines whether or not a respiratory disease is present based on the feature quantity calculated by the feature quantity calculation unit, Comprising The respiratory disease determination system according to claim 5 or 6.

Images