A method for estimating the components of feed, a method for determining the proportions of multiple types of feed, and a computer program

Abstract

To provide a technology that can estimate feed components with a non-destructive method in a short period of time, or to provide a technology to appropriately blend feed according to feed components.SOLUTION: A method for estimating feed components comprises: (a) imaging each feed of the same type without pulverizing the feed for one or more types of feed with a spectral camera that can receive light with multiple pixels arranged in a two-dimensional array, comprising a spectral filter and an image sensor, and thereby obtaining a spectral reflectance from the light reception results at each of the multiple pixels, to obtain spectral reflectance information obtained according to the spectral reflectance for each of the one or more types of feed; (b) estimating the amount of the plurality of components for each of the one or more types of feed using a component estimation model that inputs the spectral reflectance information obtained from the light reception results of at least some of the multiple pixels and outputs the amount of the plurality of components; and (c) outputting the component amounts of the plurality of components.SELECTED DRAWING: Figure 5

Description

Technical Field

[0001] The present disclosure relates to a method for estimating components of feed, a method for determining the blending amounts of multiple types of feed, and a computer program.

Background Art

[0002] Patent Document 1 discloses a technique for automatically estimating the weight of an animal individual from animal individual image data by a computer and guiding the animal individual to one of two feeding areas according to the comparison result between the estimated weight and the reference weight. Different nutritious feeds are supplied to the two feeding areas.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] However, in the prior art, variations in feed components and feed blending methods are not considered. In reality, since there are variations in feed components, it is desirable to accurately measure or estimate feed components. Generally, there are methods using chemical drugs and methods using spectral information for measuring feed components. The former requires equipment, chemicals, and other specialized knowledge, so it is not a method that can be implemented by anyone who formulates feed. On the other hand, in the component analysis using the latter spectral information, since spectral information can only be obtained at one point of the feed, it was necessary to eliminate the unevenness in the component amounts for each part such as leaves and stems. In order to eliminate this unevenness, it was necessary to perform a pulverization process before measurement, and there was a problem that the result could not be obtained immediately at the feeding timing.

[0005] For these reasons, conventional component analysis methods all involve destructive processes requiring specialized knowledge and skills, and cannot be performed on-site in real time. Furthermore, there is a risk of feed deterioration during component analysis, which hinders the proper design of feed formulations.

[0006] Therefore, there is a need for a technology that allows anyone to quickly and non-destructively determine the components of feed at the feeding site. Furthermore, there is a need for a technology that allows for the appropriate formulation of feed based on its components. [Means for solving the problem]

[0007] According to the first form of this disclosure, Computers A method for estimating the components of feed is provided. This method includes (a) a step of obtaining spectral reflectance information for each of the one or more types of feed by using a spectral camera capable of receiving light with a plurality of pixels arranged in a two-dimensional array including a spectral filter and an image sensor, imaging each type of feed without crushing the feed, and obtaining spectral reflectance from the light reception results at each of the plurality of pixels; (b) a step of estimating the amount of the plurality of components for each of the one or more types of feed by using a component estimation model that takes the spectral reflectance information obtained from the light reception results at at least some of the plurality of pixels as input and outputs the amount of the plurality of components; and (c) a step of outputting the amount of the plurality of components.

[0008] According to the second form of this disclosure, ComputersA method is provided for determining the proportions of multiple types of feed for multiple captive animals. This method includes (a) using a spectral camera capable of receiving light with multiple pixels arranged in a two-dimensional array including a spectral filter and an image sensor, imaging each type of feed without crushing it, obtaining spectral reflectance from the light reception results at each of the multiple pixels, and obtaining spectral reflectance information for each of the multiple types of feed according to the spectral reflectance; and (b) using a component estimation model that takes the spectral reflectance information obtained from the light reception results at least some of the multiple pixels as input and outputs the component amounts of multiple components, to estimate the component amounts of the multiple components for each of the multiple types of feed. The process includes: (c) acquiring individual rearing information relating to the rearing status of each of the plurality of reared individuals; (d) determining the required amount of the plurality of components for each of the plurality of reared individuals using a required component estimation model that takes the individual rearing information as input and outputs the required amount of the plurality of components; and (e) determining the blending amount of the plurality of types of feed for each of the plurality of reared individuals from the amount of the plurality of components for each of the plurality of types of feed estimated in step (b) and the required amount of the plurality of components for each of the plurality of reared individuals determined in step (d).

[0009] A third embodiment of the present disclosure provides a computer program that causes a processor to perform a process for estimating the components of a feed. The computer program causes the processor to perform the following: (a) a process of obtaining spectral reflectance information for each of the one or more types of feed by using a spectral camera capable of receiving light with a plurality of pixels arranged in a two-dimensional array including a spectral filter and an image sensor, imaging each type of feed without crushing the feed, thereby obtaining spectral reflectance from the light-receiving results at each of the plurality of pixels; (b) a process of estimating the component amounts of the plurality of components for each of the one or more types of feed using a component estimation model that takes the spectral reflectance information obtained from the light-receiving results at at least some of the plurality of pixels as input and outputs the component amounts of the plurality of components; and (c) a process of outputting the component amounts of the plurality of components.

[0010] A fourth embodiment of this disclosure provides a computer program that causes a processor to perform a process to determine the proportions of multiple types of feed for multiple reared individuals. This computer program includes (a) a spectral camera capable of receiving light with multiple pixels arranged in a two-dimensional array including a spectral filter and an image sensor, which images each type of feed without crushing the feed, thereby obtaining spectral reflectance from the light-receiving results at each of the multiple pixels, and for each of the multiple types of feed, a process to obtain spectral reflectance information according to the spectral reflectance; and (b) a component estimation model which takes the spectral reflectance information obtained from the light-receiving results at at least some of the multiple pixels as input and outputs the component amounts of multiple components, which estimates the component amounts of the multiple components for each of the multiple types of feed. The processor is made to execute the following: (c) a process to acquire individual rearing information regarding the rearing status of each of the multiple rearing individuals; (d) a process to determine the required amount of the multiple components for each of the multiple rearing individuals using a required component estimation model that takes the individual rearing information as input and outputs the required amount of the multiple components; and (e) a process to determine the amount of the multiple types of feed for each of the multiple rearing individuals based on the amount of the multiple components for each of the multiple types of feed estimated in process (b) and the required amount of the multiple components for each of the multiple rearing individuals determined in process (d). [Brief explanation of the drawing]

[0011] [Figure 1] An explanatory diagram showing the configuration of the feeding system in the embodiment. [Figure 2] A block diagram showing the configuration of an information processing device. [Figure 3] A block diagram showing the internal configuration of the machine learning processing unit, the feed component quantity estimation unit, and the feed formulation processing unit. [Figure 4] A flowchart illustrating the processing steps for machine learning. [Figure 5]A flowchart illustrating the processing steps for estimating component amounts and determining blending amounts. [Modes for carrying out the invention]

[0012] Figure 1 is an explanatory diagram showing the configuration of a feeding system in an embodiment. This feeding system comprises a livestock barn 100, an automatic feeding device 200, an information processing device 300, and a spectroscopic camera 400.

[0013] The livestock shed 100 has multiple livestock pens 110. Each livestock pen 110 houses one livestock individual BA. A livestock individual BA is, for example, a cow or a pig. Each livestock pen 110 is equipped with a feeding trough 111 for compound feed.

[0014] The automatic feeding device 200 supplies compound feed, appropriately formulated according to the rearing condition of each individual animal BA, to the feeding trough 111 individually. The automatic feeding device 200 has multiple feed tanks 210 and a mixing and dispensing device 220. Each of the multiple feed tanks 210 contains a different feed FS1 to FSn, where the subscript n is an integer of 2 or more. The mixing and dispensing device 220 prepares compound feed by mixing feeds FS1 to FSn according to the rearing condition of each individual animal BA, according to the feed formulation information provided by the information processing device 300, and supplies it to the feeding trough 111.

[0015] The spectroscopic camera 400 is for measuring the spectral intensity distribution of feed FSj, where the subscript j is an integer from 1 to n. The spectroscopic camera 400 includes a spectroscopic filter 401 and an image sensor 402. As shown in an enlarged view at the bottom of Figure 1, in this embodiment, the image sensor 402 is a two-dimensional element in which a plurality of two-dimensionally arranged photodetectors function as a plurality of pixels Px, and these pixels Px are capable of receiving light. However, the spectroscopic camera 400 may also be configured to be capable of receiving light in a plurality of two-dimensionally arranged pixels Px by scanning a one-dimensional element in which the photodetectors are linearly arranged. The spectroscopic filter 401 is a tunable transmission filter with a narrow half-width of the transmitted wavelength and the ability to change the transmitted wavelength. Furthermore, it is preferable that the spectroscopic filter 401 allows only light in the near-infrared region to pass through. The near-infrared region typically refers to a wavelength range of about 700 to about 2500 nm. In this embodiment, the spectral filter 401 is an optical filter capable of changing the transmission wavelength in the near-infrared region, specifically in the wavelength range of 680 to 950 nm. Such a spectral filter 401 can be realized, for example, using a Fabry-Perot type spectral filter.

[0016] When imaging with the spectrophotometer 400, the feed FSj is placed in a container 410 such as a petri dish and illuminated obliquely from the left and right by illumination sources 420. For example, a halogen light source is used as the illumination source 420. The spectral reflectance of the feed FSj is the value obtained by dividing the spectral intensity I(λ) obtained by measuring the feed FSj by the reference spectral intensity I0(λ) obtained by measuring a standard white plate. Since the spectral intensities I(λ) and I0(λ) are measured for each of the multiple pixels Px arranged in a two-dimensional array, the spectral reflectance is calculated using the spectral intensities I(λ) and I0(λ) obtained at the same pixel position (u,v). Note that it is not necessary to use the spectral intensities obtained from all pixels Px of the spectrophotometer 400 as the spectral intensity I(λ) used to determine the components of the feed FSj; the spectral intensities obtained from only some of the pixels Px may be used.

[0017] By the way, as a spectroscopic measurement device, a device composed of a spectroscopic filter and a photoelectric conversion element such as a photodiode is known. However, in such a spectroscopic measurement device, since the area where spectroscopic measurement can be performed with high precision is limited, it is difficult to perform spectroscopic measurement on a large measurement sample with a side length of about several tens of cm in a short time. On the other hand, the spectroscopic camera 400 of the present embodiment includes a spectroscopic filter 401 and an imaging element 402, and is configured to be able to receive light by a plurality of pixels Px arranged in a two-dimensional array. Therefore, it has the advantage that a large measurement sample can be spectroscopically measured in a short time. Further, by using the spectroscopic camera 400 of the present embodiment, a feed image having intensity information for each wavelength can be obtained, so that the influence of unevenness of the feed can be reduced, and there is an advantage that it is not necessary to perform pretreatment such as pulverization or drying of the feed before spectroscopic measurement.

[0018] The information processing device 300 executes a process of estimating the component amounts of a plurality of components contained in the feed FSj by using the spectroscopic reflectance of each individual feed FSj. The information processing device 300 further executes a process of determining the blending amount of the feed suitable for the breeding individual BA according to the component amount of the feed FSj and the individual breeding information of each individual breeding individual BA.

[0019] FIG. 2 is a block diagram showing the configuration of the information processing device 300. The information processing device 300 includes a processor 310, a memory 320, an interface circuit 330, an input device 340 and a display device 350 connected to the interface circuit 330. The spectroscopic camera 400 is also connected to the interface circuit 330. The processor 310 not only has a function of executing the processes described in detail below, but also has a function of displaying the data obtained by the processes and the data generated in the process of the processes on the display device 350.

[0020] The processor 310 has the functions of a machine learning processing unit 510 that performs learning of a machine learning model, a feed component amount estimation unit 520 that estimates the amounts of components of a plurality of components included in the feed FSj, and a feed formulation processing unit 530 that determines the formulation amount of feed suitable for the breeding individual BA. These functions are respectively realized by the processor 310 executing a computer program stored in the memory 320. However, a part of these functions may be realized by a hardware circuit. The processor of the present disclosure is a term that also includes such a hardware circuit. Also, one or more processors that execute various processes may be processors included in one or more remote computers connected via a network.

[0021] The memory 320 stores a component estimation model 610, a required component estimation model 620, a breeding individual database 630, and a feed database 640. The component estimation model 610 is a machine learning model that takes the spectral reflectance information of the feed FSj as an input and outputs the amounts of components of a plurality of components included in the feed FSj. The required component estimation model 620 is a machine learning model that takes the individual breeding information regarding the breeding state of the breeding individual BA as an input and outputs the required amounts of components of a plurality of components. The breeding individual database 630 is a database in which the individual ID and breeding history are registered for each of a plurality of breeding individuals BA. The feed database 640 is a database in which the amounts of components of a plurality of components are registered for a plurality of types of feeds FSj.

[0022] FIG. 3 is a block diagram showing the internal configurations of the machine learning processing unit 510, the feed component amount estimation unit 520, and the feed formulation processing unit 530.

[0023] The machine learning processing unit 510 includes a preprocessing unit 511, a feed component quantity input unit 512, and a learning execution unit 513. The machine learning processing unit 510 performs machine learning of the component estimation model 610 using multiple training feed LFSi. Here, the subscript i is an integer from 1 to M, and M is an integer of 2 or greater. As described above, the component estimation model 610 is a machine learning model that takes spectral reflectance information of feed as input and outputs the amounts of multiple components contained in the feed. The component estimation model 610 can be constructed, for example, as a model that utilizes linear regression such as ridge regression or decision trees such as random forests. However, the component estimation model 610 may also be constructed using other types of models such as neural networks or support vector machines.

[0024] The preprocessing unit 511 creates spectral reflectance information by performing preprocessing on the spectral reflectance of the learning feed LFSi obtained by the spectral camera 400. For preprocessing, for example, any of the following calculations (1) to (3) can be used.

number

[0025] The spectral reflectance R(λ) is the result of dividing the spectral intensity I(λ) by the reference spectral intensity I0(λ), and is calculated for each pixel Px. To eliminate the influence of measurement noise, m spectral reflectance information may be generated by obtaining m spectral reflectances from the light reception results of Np pixels Px. Here, Np is the total number of pixels Px arranged in a two-dimensional array, and m is an integer between 2 and Np. One method for obtaining m spectral reflectances is to divide the pixel region composed of Np pixels into m sub-regions, average the spectral intensity I and the reference spectral intensity I0 within each sub-region, and then divide the averaged spectral intensity by the averaged reference spectral intensity to obtain m spectral reflectances. If each sub-region is a rectangular region containing M pixels, averaging means taking the average of the M pixels. Here, M is an integer between 2 and Np. Alternatively, the spectral reflectance may be calculated for each Np pixel, and then the spectral reflectance may be averaged within each sub-region. Another method for determining m spectral reflectances is to calculate the error between the average spectral reflectance (calculated by averaging Np spectral reflectances for each wavelength) and each individual spectral reflectance, and then select the spectral reflectances whose error value is less than or equal to a predetermined threshold. For example, the root mean square error can be used as the error. This allows for the exclusion of spectral reflectances with low reliability. Alternatively, spectral reflectance information may be generated using one representative spectral reflectance that represents the selected m spectral reflectances. The representative spectral reflectance can be the average or linear sum of the m spectral reflectances. Furthermore, as another method to remove the influence of measurement noise, M spectral reflectances may be calculated from Np pixels, and m spectral reflectances may be selected from the M spectral reflectances. Here, M is an integer between 2 and Np, and m is an integer between 2 and M. By performing the above-mentioned calculations on the m spectral reflectances obtained by any of these methods, the spectral reflectance information for m can be obtained. This process corresponds to obtaining m spectral reflectance information using the light reception results of m pixels selected from among multiple pixels.

[0026] In equations (1) to (3) above, the base of the logarithm is set to 10, but the base of the logarithm may be a value other than 10. As can be seen from these examples, it is preferable that the spectral reflectance information be calculated from the logarithm of the spectral reflectance R(λ). According to the inventor's experiments in this disclosure, it was found that using the second derivative U(λ) of the logarithm of the spectral reflectance R(λ) as spectral reflectance information is preferable in that the estimated values ​​of the feed components by the component estimation model 610 become more accurate. The reason for taking the logarithm of the spectral reflectance R(λ) is that in the absorption phenomenon by the target component, the logarithm of the spectral reflectance R(λ) is proportional to the concentration of the target component. Furthermore, the second derivative is performed because different target components absorb different wavelengths more strongly, and this appears as a peak in the spectral reflectance distribution. In other words, by taking the second derivative of the logarithm of the spectral reflectance R(λ), it becomes possible to extract the peak of the spectral reflectance distribution with high sensitivity.

[0027] The feed component quantity input unit 512 accepts input of the component quantities of multiple components for each of the multiple learning feed LFSi. For each learning feed LFSi, the component quantities of multiple components are accurately measured in advance using a chemical component analyzer. The multiple components are those that are of high importance as feed for the reared individual BA, and for example, multiple components including water, crude protein, crude fat, neutral detergent fiber, and acidic detergent fiber are used. It is particularly preferable that the multiple components include water. As described later, anyone can estimate the components of the feed non-destructively in a short time, eliminating concerns about feed deterioration that occurred during conventional manual component analysis, and allowing for accurate and rapid estimation of multiple components, including water, at the feeding site.

[0028] The learning execution unit 513 uses spectral reflectance information for multiple training feed LFSi and the amounts of multiple components as training data to perform machine learning on the component estimation model 610.

[0029] The feed component quantity estimation unit 520 includes a preprocessing unit 521, an estimation execution unit 522, and a component estimation model 610. The component estimation model 610 is stored in the memory 320 after machine learning by the machine learning processing unit 510 is completed. The preprocessing unit 521 performs the same preprocessing as the preprocessing unit 511 of the machine learning processing unit 510. The estimation execution unit 522 performs the process of estimating the component quantities of multiple components by inputting the spectral reflectance information of the feed FSj for each of the multiple types of feed FSj into the machine learning-trained component estimation model 610. These feed FSj are the feeds actually contained in the multiple feed tanks 210 of the automatic feeding device 200.

[0030] The feed formulation processing unit 530 includes a livestock information acquisition unit 531, a required component amount determination unit 532, a feed formulation amount determination unit 533, a required component estimation model 620, a livestock individual database 630, and a feed database 640.

[0031] The rearing information acquisition unit 531 acquires individual rearing information for each of the multiple rearing individuals BA from the rearing individual database 630, regarding the rearing status of each individual BA. For example, two or more pieces of information can be used as individual rearing information, such as individual ID, age in months, weight, feeding history, and BCS (Body Condition Score). Alternatively, individual rearing information may be acquired using measurement results from various sensors, such as a weighing scale or stereo camera, without using the rearing individual database 630.

[0032] The required component quantity determination unit 532 uses the required component estimation model 620 to determine the required component quantities for each of the multiple livestock individuals BA. The required component estimation model 620 is a model that takes individual livestock information as input and outputs the required component quantities for multiple components. In addition to the individual livestock information of livestock individuals BA, the required component estimation model 620 may also be configured to take value indicators related to the value of livestock individuals BA as input. As these value indicators, milk components for dairy cows or meat quality for beef cattle can be used. The required component estimation model 620 may be configured as a machine learning model such as a neural network, or it may be configured as a function or a lookup table. It is assumed that the training of the required component estimation model 620 has been performed in advance.

[0033] The feed formulation determination unit 533 determines the formulation amounts of multiple types of feed FSj for each individual animal BA based on the component amounts of multiple components for each feed FSj estimated by the feed component amount estimation unit 520 and the required component amounts for the animal BA determined by the required component amount determination unit 532. The feed formulation information indicating the formulation amounts of multiple types of feed FSj is supplied to the automatic feeding device 200.

[0034] Figure 4 is a flowchart showing the processing steps for machine learning. This machine learning process is performed by the machine learning processing unit 510.

[0035] In step S110, the feed component amount input unit 512 acquires the component amounts of multiple components for multiple training feed LFSi. Here, the component amounts are, for example, the component amounts per unit mass. As mentioned above, the component amounts of multiple components contained in the training feed LFSi have been accurately measured in advance. In step S120, the machine learning processing unit 510 acquires the spectral reflectance R(λ) for each of the multiple training feed LFSi using the spectroscopic camera 400.

[0036] In step S130, the preprocessing unit 511 generates spectral reflectance information by performing preprocessing on the spectral reflectance R(λ). As mentioned above, the spectral reflectance R(λ) is the result of dividing the spectral intensity I(λ) by the reference spectral intensity I0(λ), and is calculated for each pixel Px. Note that spectral reflectance and spectral reflectance information may be generated using only a portion of multiple pixels Px. For example, in order to remove the influence of measurement noise, m spectral reflectances may be obtained from the light reception results of Np pixels Px, and m spectral reflectance information may be generated. Furthermore, one spectral reflectance information may be generated using one representative spectral reflectance that represents the m spectral reflectances. As mentioned above, it is preferable that the spectral reflectance information be calculated from the logarithm of the spectral reflectance R(λ), and it is particularly preferable that it be calculated by taking the second derivative of the logarithm of the spectral reflectance.

[0037] In step S140, the learning execution unit 513 performs machine learning on the component estimation model 610 using learning data that includes the component amounts and spectral reflectance information of the learning feed LFSi. Once learning is complete, the trained component estimation model 610 is stored in the memory 320.

[0038] Thus, the machine learning processing of this embodiment can successfully perform machine learning on the component estimation model 610 that estimates the components of feed. In particular, since information calculated from the logarithm of the spectral reflectance R(λ) is used as spectral reflectance information, it is possible to improve the estimation accuracy of the component estimation model 610.

[0039] Figure 5 is a flowchart showing the processing procedures for estimating component amounts and determining blending amounts. These processes are performed by the feed component amount estimation unit 520 and the feed blending processing unit 530.

[0040] In step S210, the feed component amount estimation unit 520 uses the spectroscopic camera 400 to acquire the spectral reflectance R(λ) for several types of feed FSj. The spectroscopic measurement of feed FSj is performed without grinding the feed FSj. It is also preferable that the spectroscopic measurement of feed FSj is performed without drying the feed FSj. In this embodiment, since the spectroscopic measurement of feed FSj is performed without grinding or drying the feed FSj, the feed FSj that was the subject of the spectroscopic measurement can be used as feed for the reared individuals BA as is.

[0041] In step S220, the preprocessing unit 521 generates spectral reflectance information by performing preprocessing on the spectral reflectance R(λ). This preprocessing is the same as the preprocessing in step S130.

[0042] In step S230, the estimation execution unit 522 uses the trained component estimation model 610 to estimate the amounts of multiple components for each feed FSj. That is, the estimation execution unit 522 inputs spectral reflectance information of each feed FSj into the component estimation model 610 and obtains the amounts of multiple components output from the component estimation model 610. The estimated component amounts here are, for example, the component amounts per unit mass.

[0043] Furthermore, when using m spectral reflectance information calculated from m light reception results as spectral reflectance information for one type of feed FSj, it is preferable to determine the representative value of m estimated component amounts, estimated from the m spectral reflectance information, as the estimated component amount for each component. Here, m is an integer between 2 and Np, where Np is the total number of pixels Px. As the representative value of the m estimated component amounts, the average, maximum, or minimum value of the m estimated component amounts can be used. However, using the average value of the m estimated component amounts as the representative value can improve estimation accuracy.

[0044] In step S240, the estimation execution unit 522 outputs estimated values ​​for the amounts of multiple components for each feed FSj. In this embodiment, the estimation execution unit 522 registers the estimated values ​​for the amounts of multiple components in the feed database 640. In step S240, other indicators such as the total digestible nutrients calculated from the amounts of multiple components may also be output. Alternatively, the estimated values ​​for the amounts of multiple components may be output to an output device such as a display device 350 or a printer.

[0045] In step S240, the estimation execution unit 522 may further output the distribution of component amounts in an image region containing Np pixels Px. The lower part of Figure 1 shows an example of this output, which is a component amount distribution CDi displayed on the display device 350. The component amount distribution CDi is an image that shows the component amount of the i-th component within an image region containing Np pixels Px. In the example in Figure 1, the component amount is represented by pixel density, but the component amount may also be represented numerically. It is not necessary to create such a component amount distribution CDi for image regions containing all Np pixels Px; it may be created for image regions containing at least some of the Np pixels Px. Furthermore, it is preferable to create a component amount distribution CDi for at least one of several components. By outputting such a component amount distribution CDi, the operator can know the distribution of component amounts in the feed FSj.

[0046] The processes described in steps S210 to S240 above correspond to the process of estimating the components of the feed. In the example described above, the component amounts were estimated for each of several types of feed FSj, but the method disclosed herein is also applicable when estimating the component amounts for one or more types of feed. However, in the processes from step S250 onward, the estimated component amounts for several types of feed FSj are used to determine the feed mixture for each individual animal BA.

[0047] In step S250, the rearing information acquisition unit 531 acquires individual rearing information for one rearing individual BA. Individual rearing information can be obtained by referring to the rearing individual database 630 using the individual ID of the rearing individual BA. Individual rearing information includes two or more pieces of information such as individual ID, age in months, weight, feeding history, and BCS (Body Condition Score). Alternatively, some or all of the individual rearing information may be obtained from each rearing individual BA using various sensors. The following processes in steps S260 to S280 are performed for the one rearing individual BA selected in step S250.

[0048] In step S260, the required component amount determination unit 532 uses the required component estimation model 620 to determine the required component amounts for multiple components for the livestock individual BA. That is, the required component amount determination unit 532 inputs the livestock information of the livestock individual BA into the required component estimation model 620 and obtains the required component amounts for multiple components output from the required component estimation model 620. These required component amounts are, for example, the weight of the components. If the required component estimation model 620 is configured to accept a value index related to the livestock value of the livestock individual BA in addition to the livestock information of the livestock individual BA, the required component amount determination unit 532 may also accept the input of the value index. Alternatively, the value index may be set in advance before starting the process shown in Figure 5.

[0049] In step S270, the feed formulation determination unit 533 determines the formulation amounts of multiple types of feed FSj for the individual animal BA. That is, the feed formulation determination unit 533 determines the formulation amounts of multiple types of feed FSj so as to achieve the required amounts of multiple components determined in step S260. In this case, the values ​​of multiple components contained in each feed FSj are those registered in the feed database 640 in step S230. The formulation of multiple types of feed FSj is performed such that the evaluation value E, given by, for example, the following formula, is minimized.

number

[0050] By determining the amount Mj of multiple types of feed FSj so as to minimize the evaluation value E given by equation (4) above, the required amount Ci of multiple components i can be achieved.

[0051] Furthermore, as shown in equation (5) below, it is also possible to set a necessary condition that the error in the amount of each component i is less than or equal to the allowable component amount error σi.

number

[0052] Instead of the condition in (5) above, or in addition to the condition in equation (5) above, it is also possible to set a necessary condition that each component i contains an amount of component equal to or greater than the required component amount Ci, as shown in equation (6) below.

number

[0053] In step S280, the feed compounding unit 530 supplies feed compounding information indicating the compounding amounts Mj of multiple types of feed FSj determined in step S270 to the automatic feeding device 200, and the mixing and dispensing device 220 performs the compounding and dispensing of feed FSj accordingly.

[0054] Alternatively, the feed mixture FSj may be prepared and fed manually without using the automatic feeding device 200. In this case, the information processing device 300 may output feed mixture information, indicating the amount Mj of feed mixture FSj, using an output device such as a display device 350 or a printer. Alternatively, the information processing device 300 may transmit the feed mixture information to another information processing device.

[0055] In step S290, it is determined whether the processing in steps S250 to S280 has been completed for all reared individuals BA. If the processing for all reared individuals BA has not been completed, the process returns to step S250, the next reared individual BA is selected, and the processing in steps S250 to S280 is executed.

[0056] The component quantity estimation process described above allows anyone to estimate the components of feed FSj non-destructively in a short amount of time. Furthermore, it eliminates concerns about feed deterioration that occurred during conventional component analysis, enabling more accurate feed design. In addition, the formulation amount determination process described above allows for the appropriate determination of the formulation amounts of multiple types of feed FSj according to the rearing conditions of the reared individual BA.

[0057] Other forms: This disclosure is not limited to the embodiments described above, and can be implemented in various forms without departing from its spirit. For example, this disclosure can also be implemented in the following forms (aspects). The technical features in the embodiments described above that correspond to the technical features in each of the forms described below can be replaced or combined as appropriate in order to solve some or all of the problems of this disclosure, or to achieve some or all of the effects of this disclosure. Furthermore, if such technical features are not described as essential in this specification, they can be deleted as appropriate.

[0058] (1) According to a first embodiment of the present disclosure, a method for estimating the components of feed is provided. This method includes (a) a step of obtaining spectral reflectance information for each of the one or more types of feed, by using a spectral camera capable of receiving light with a plurality of pixels arranged in a two-dimensional array including a spectral filter and an image sensor, and imaging each type of feed without crushing the feed, thereby obtaining spectral reflectance from the light reception results at each of the plurality of pixels; (b) a step of estimating the component amounts of the plurality of components for each of the one or more types of feed, using a component estimation model that takes the spectral reflectance information obtained from the light reception results at at least some of the plurality of pixels as input and outputs the component amounts of the plurality of components; and (c) a step of outputting the component amounts of the plurality of components. This method allows anyone to estimate the components of feed non-destructively in a short amount of time. Furthermore, it eliminates concerns about feed deterioration that occurred during conventional component analysis, enabling more accurate feed design.

[0059] (2) A second embodiment of the present disclosure provides a method for determining the proportions of multiple types of feed for multiple reared individuals. This method includes (a) a step of obtaining spectral reflectance information for each of the multiple types of feed by using a spectral camera capable of receiving light with multiple pixels arranged in a two-dimensional array including a spectral filter and an image sensor, imaging each type of feed without crushing the feed, and obtaining spectral reflectance from the light reception results at each of the multiple pixels, and (b) a step of estimating the amount of the multiple components for each of the multiple types of feed by using a component estimation model that takes the spectral reflectance information obtained from the light reception results at at least some of the multiple pixels as input and outputs the amount of the multiple components. The process includes: (c) acquiring individual rearing information relating to the rearing status of each of the plurality of reared individuals; (d) determining the required amount of the plurality of components for each of the plurality of reared individuals using a required component estimation model that takes the individual rearing information as input and outputs the required amount of the plurality of components; and (e) determining the blending amount of the plurality of types of feed for each of the plurality of reared individuals from the amount of the plurality of components for each of the plurality of types of feed estimated in step (b) and the required amount of the plurality of components for each of the plurality of reared individuals determined in step (d). This method allows for the appropriate determination of the proportions of multiple types of feed depending on the rearing conditions of the animals.

[0060] (3) In the above method, step (a) may include a step of determining m spectral reflectance information using the light reception results of m pixels selected from the plurality of pixels when m is an integer of 2 or more, and step (b) may include a step of determining a representative value of m component quantity estimates obtained from the component estimation model according to each of the m spectral reflectance information as the component quantity of the plurality of components. This method allows for accurate estimation of component quantities using m pieces of spectral reflectance information.

[0061] (4) In the above method, the spectral reflectance information may be information calculated from the logarithm of the spectral reflectance. This method allows for accurate estimation of component quantities using a component estimation model.

[0062] (5) In the above method, the spectral reflectance information may be information calculated by taking the second derivative of the logarithm of the spectral reflectance. This method allows for more accurate estimation of component quantities using a component estimation model.

[0063] (6) In the above method, the spectral reflectance information may include information in the near-infrared region. This method allows for accurate estimation of component quantities using a component estimation model.

[0064] (7) In the above method, step (c) may include a step of outputting the distribution of the amount of the component in an image region that includes at least some of the pixels among the plurality of pixels, for at least some of the components of the plurality of components. This method allows us to determine the distribution of component amounts in animal feed.

[0065] (8) In the above method, step (d) may be to input the value index relating to the value of each feed individual as livestock, along with the individual rearing information, into the required component estimation model to determine the required amount of each of the multiple components for each rearing individual. This method allows for the appropriate determination of the proportions of multiple types of feed based on the rearing conditions and value indicators of the captive animals.

[0066] (9) According to a third embodiment of the present disclosure, a computer program is provided that causes a processor to perform a process for estimating the components of a feed. The computer program causes the processor to perform the following: (a) a process of obtaining spectral reflectance information for each of the one or more types of feed, by using a spectral camera capable of receiving light with a plurality of pixels arranged in a two-dimensional array including a spectral filter and an image sensor, and imaging each type of feed without crushing the feed, thereby obtaining spectral reflectance from the light-receiving results at each of the plurality of pixels; (b) a process of estimating the component amounts of the plurality of components for each of the one or more types of feed, using a component estimation model that takes the spectral reflectance information obtained from the light-receiving results at at least some of the plurality of pixels as input and outputs the component amounts of the plurality of components; and (c) a process of outputting the component amounts of the plurality of components.

[0067] (10) According to a fourth embodiment of the present disclosure, a computer program is provided which causes a processor to perform a process of determining the proportions of multiple types of feed for multiple reared individuals. This computer program includes (a) a spectral camera capable of receiving light with multiple pixels arranged in a two-dimensional array including a spectral filter and an image sensor, which images each type of feed without crushing the feed, thereby obtaining spectral reflectance from the light-receiving results at each of the multiple pixels, and for each of the multiple types of feed, a process of obtaining spectral reflectance information obtained according to the spectral reflectance, and (b) a component estimation model which takes the spectral reflectance information obtained from the light-receiving results at at least some of the multiple pixels as input and outputs the component amounts of multiple components, which estimates the component amounts of the multiple components for each of the multiple types of feed. The processor is made to execute the following: (c) a process to acquire individual rearing information regarding the rearing status of each of the multiple rearing individuals; (d) a process to determine the required amount of the multiple components for each of the multiple rearing individuals using a required component estimation model that takes the individual rearing information as input and outputs the required amount of the multiple components; and (e) a process to determine the amount of the multiple types of feed for each of the multiple rearing individuals based on the amount of the multiple components for each of the multiple types of feed estimated in process (b) and the required amount of the multiple components for each of the multiple rearing individuals determined in process (d).

[0068] This disclosure can also be implemented in various forms other than those described above. For example, it can be implemented in the form of an apparatus for carrying out the above-described method, or a non-transitory storage medium on which a computer program is recorded. [Explanation of symbols]

[0069] 100…Livestock shed, 110…Livestock pen, 111…Feeding trough, 200…Automatic feeding device, 210…Feed tank, 220…Mixing and dispensing device, 300…Information processing device, 310…Processor, 320…Memory, 330…Interface circuit, 340…Input device, 350…Display device, 400…Spectroscopic camera, 401…Spectroscopic filter, 402…Image sensor, 410…Containment container, 420…Illumination light source, 510… Machine learning processing unit, 511…Preprocessing unit, 512…Feed component quantity input unit, 513…Learning execution unit, 520…Feed component quantity estimation unit, 521…Preprocessing unit, 522…Estimation execution unit, 530…Feed compounding processing unit, 531…Breeding information acquisition unit, 532…Required component quantity determination unit, 533…Feed compounding quantity determination unit, 610…Component estimation model, 620…Required component estimation model, 630…Breeding individual database, 640…Feed database

Claims

1. A method for a computer to estimate the components of feed, (a) Using a spectral camera capable of receiving light with a plurality of pixels arranged in a two-dimensional array including a spectral filter and an image sensor, for one or more types of feed, imaging the feed without crushing the feed, thereby obtaining spectral reflectance from the light-receiving results at each of the plurality of pixels, and for each of the one or more types of feed, obtaining spectral reflectance information according to the spectral reflectance, (b) A step of estimating the amount of the multiple components for each of the one or more types of feed, using a component estimation model that takes the spectral reflectance information obtained from the light reception results of at least some of the multiple pixels as input and outputs the amount of the multiple components as output, (c) A step of outputting the amounts of the plurality of components, Methods that include...

2. A method for a computer to determine the proportions of multiple types of feed for multiple captive animals, (a) Using a spectral camera capable of receiving light with a plurality of pixels arranged in a two-dimensional array including a spectral filter and an image sensor, imaging each of the plurality of types of feed without crushing the feed, thereby obtaining spectral reflectance from the light reception results at each of the plurality of pixels, and obtaining spectral reflectance information for each of the plurality of types of feed according to the spectral reflectance, (b) A step of estimating the amount of the multiple components for each of the multiple types of feed using a component estimation model that takes the spectral reflectance information obtained from the light reception results of at least some of the multiple pixels as input and outputs the amount of the multiple components, (c) A step of acquiring individual rearing information regarding the rearing status of each of the multiple reared individuals, (d) A step of determining the required amount of the multiple components for each of the multiple reared individuals using a required component estimation model that takes the individual rearing information as input and outputs the required amount of the multiple components, (e) A step of determining the amount of each of the multiple types of feed for each of the multiple types of feed estimated in step (b), and the required amount of each of the multiple types of feed for each of the multiple individuals determined in step (d), Methods that include...

3. A method according to claim 1 or 2, Step (a) includes a step of determining m spectral reflectance information using the light reception results of m pixels selected from the plurality of pixels, when m is an integer of 2 or more. Step (b) includes determining a representative value of the m component quantity estimates obtained from the component estimation model according to each of the m spectral reflectance information, as the component quantity of the plurality of components. method.

4. A method according to claim 1 or 2, The spectral reflectance information is calculated from the logarithm of the spectral reflectance, in a method.

5. The method according to claim 4, The spectral reflectance information is calculated by taking the second derivative of the logarithm of the spectral reflectance, in this method.

6. A method according to claim 1 or 2, The spectral reflectance information includes information in the near-infrared region, in a method.

7. The method according to claim 1, The method further includes step (c) of outputting the distribution of the amount of the component in an image region that includes at least some of the pixels among the plurality of pixels, for at least some of the components of the plurality of components.

8. The method according to claim 2, The step (d) is a method for determining the required amounts of the plurality of components for each individual being raised by inputting the value index relating to the value of each individual being raised as livestock, along with the individual rearing information, into a required component estimation model.

9. A computer program that causes a processor to perform a process to estimate the components of animal feed, (a) Using a spectral camera capable of receiving light with a plurality of pixels arranged in a two-dimensional array including a spectral filter and an image sensor, for one or more types of feed, imaging is performed for each type of feed without crushing the feed, thereby obtaining spectral reflectance from the light-receiving results at each of the plurality of pixels, and for each of the one or more types of feed, spectral reflectance information obtained according to the spectral reflectance is obtained, (b) A process to estimate the amount of the multiple components for each of the one or more types of feed, using a component estimation model that takes the spectral reflectance information obtained from the light reception results of at least some of the multiple pixels as input and outputs the amount of the multiple components as output, (c) A process to output the amounts of the plurality of components, A computer program that causes the aforementioned processor to execute.

10. A computer program that causes a processor to perform a process to determine the proportions of multiple types of feed for multiple captive animals, (a) Using a spectral camera capable of receiving light with a plurality of pixels arranged in a two-dimensional array including a spectral filter and an image sensor, imaging each of the plurality of types of feed without crushing the feed, thereby obtaining spectral reflectance from the light reception results at each of the plurality of pixels, and processing to obtain spectral reflectance information for each of the plurality of types of feed according to the spectral reflectance, (b) A process to estimate the amount of the multiple components for each of the multiple types of feed, using a component estimation model that takes the spectral reflectance information obtained from the light reception results of at least some of the multiple pixels as input and outputs the amount of the multiple components, (c) A process to obtain individual rearing information regarding the rearing status of each of the multiple rearing individuals, (d) A process to determine the required amount of the multiple components for each of the multiple individuals being raised, using a required component estimation model that takes the individual rearing information as input and outputs the required amount of the multiple components, (e) A process to determine the amount of each of the multiple types of feed for each of the multiple types of feed estimated in process (b) and the required amount of each of the multiple types of feed for each of the multiple individuals determined in process (d), A computer program that causes the aforementioned processor to execute.

Images