Specific living body section weighing method, specific living body section weighing device, and trained model

Abstract

To enable the weight of a specific section attached to a living body consisting of a plant or an animal and separable from the surrounding to be nondestructively measured.SOLUTION: Provided is a specific living body section weighing method for nondestructively measure the weight of a specific section Wa attached to a living body W and a separable from the surrounding. An arithmetic model is configured to, by irradiating a sample living body with infrared light in advance, perform arithmetic operation to calculate the weight of the specific section Wa of the living body W on the basis of a secondary differential spectrum in absorbance regarding the wavelength of the infrared light region of diffusion reflected light or diffusion transmitted light from the sample living body. Then, the living body W to be weighed is irradiated with near-infrared light, diffusion reflected light or diffusion transmitted light from the living body W to be weighed is received, the absorbance of received light is measured, and the weight of the specific section Wa of the living body W to be weighed is calculated from the absorbance and the arithmetic model.SELECTED DRAWING: Figure 6

Description

[Technical Field]

[0001] The present invention relates to a method for weighing a specific part of a living organism, which non-destructively measures the weight of a part of a living organism such as a plant or animal using near-infrared spectrum, a device for weighing a specific part of a living organism, and a trained model that can be used with this device. [Background technology]

[0002] A known method for nondestructively measuring biological elements using near-infrared spectra is disclosed in, for example, Japanese Patent No. 5973275 (Patent Document 1). This method receives near-infrared light that passes through biological tissue and reflects from specific sites, and noninvasively measures the amounts of various substances present between specific sites, such as glucose, cholesterol, triglycerides, uric acid, blood proteins, total albumin, total globulin, red blood cell concentration, and other body fluid components and water present locally within the biological tissue. The method uses a confocal optical system to simultaneously or chronologically acquire reflected light from two focal planes at different depths, distinguish between light absorption and light reflection at the regular and irregular focal planes, and cancel out the light reflection component, leaving only light absorption, to determine the amount of substance present in the region between the regular and irregular focal planes according to the Beer-Lambert law. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Patent No. 5973275 Summary of the Invention [Problem to be solved by the invention]

[0004] However, this conventional method for weighing a living organism measures the amount of various substances present between specific parts within the tissue of the living organism, but has the problem that it cannot be used to measure the weight of a specific part of the living organism, such as an organ that is attached to the living organism and can be separated from its surroundings. For example, in the case of a living organism with a shell, such as sea urchin, if you want to know the weight of the edible gonads as a specific part before cracking the shell and extracting them, the conventional technology can measure the weight of the substance, but cannot measure the gonads as a specific part. Furthermore, in the case of a living organism that is covered by a shell, such as sea urchin, there is a problem that near-infrared light does not penetrate sufficiently, making the amount of substance less accurate.

[0005] The present invention has been made in consideration of these problems, and aims to provide a method and device for weighing a specific part of a living organism, which enables the weight of a specific part attached to a living organism made of a plant or animal and separable from its surroundings to be measured non-destructively, and a trained model that can be used with this device. [Means for solving the problem]

[0006] In order to achieve the above object, the method for weighing a specific part of a living body of the present invention is a method for non-destructively measuring the weight of a specific part of a living body that is attached to the living body and can be separated from its surroundings, comprising: a calculation model is constructed in advance, which performs calculations to calculate the weight of a specific part of a living organism based on a second derivative spectrum of the absorbance of the received light, by irradiating a sample living organism with near-infrared light, receiving reflected light from or transmitted through the sample living organism, and measuring the weight of the specific part of the living organism; Near-infrared light is irradiated onto the living body to be weighed, diffuse reflected light or diffuse transmitted light from the living body to be weighed is received, the absorbance of the received light is measured, and the weight of a specific part of the living body to be weighed is calculated from the absorbance and the above-mentioned calculation model.

[0007] Here, a living organism may be a plant or an animal, and may be alive or dead. Examples include edible or ornamental plants such as vegetables and fruits, the human body, seafood, and livestock. The specific part refers to a specific tissue mass comprised of tissue that is attached to the living organism and can be separated from the surrounding tissue. The specific part may be entirely within the living organism, or partly exposed from the living organism. For example, in plants, examples include the flesh and seeds inside fruit. In animals, examples include various internal organs such as the liver and ovaries (eggs), as well as tumors such as cancer that do not constitute organs. The gonads of the sea urinary tract, described below, are internal organs that can be separated from the surrounding tissue. Furthermore, the specific part is not limited to the tissue of the target living organism itself, but may also be a tissue mass of another living organism attached to the living organism, such as a parasite.

[0008] This involves irradiating a living organism to be weighed with near-infrared light, receiving diffusely reflected or transmitted light from the organism, measuring the absorbance of the received light, and calculating the weight of a specific part of the organism from the absorbance and a computational model. This allows the weight of a specific part attached to the organism and separable from its surroundings to be measured non-destructively, allowing the weight of the specific part to be known and evaluated without removing it, and allowing the organism to be weighed alive, which is useful and causes little damage to the organism. For example, in the case of an organism whose specific part is attached to the organism and used for food, and which is distributed as a whole without removing the specific part, it is useful to know the weight of the required specific part in advance and evaluate its quality. In particular, when there are many such organisms, it is extremely useful for sorting the quality of the organisms. Furthermore, since the method causes little damage to the organism and allows weighing while alive, it can be applied, for example, in the case of seafood, to understanding the growth status and resource management in aquaculture farms and fishing grounds.

[0009] and, as necessary, the computational model is an analytical model constructed using statistical analysis or a trained model constructed using machine learning; The statistical analysis is selected from discriminant analysis, multiple regression analysis, PLS regression analysis, and support vector machine (SVM), The above machine learning is configured to be selected from tree analysis such as decision trees (CART), regression trees, random forests, and gradient boosting trees; neural networks such as perceptrons, convolutional neural networks (CNNs), recurrent neural networks (RNNs), residual networks (ResNets), and generative adversarial networks (GANs); deep learning (DNNs); and ensemble analysis consisting of a combination of these.

[0010] The computational model consists of an analytical model constructed using statistical analysis or a trained model constructed using machine learning, which can improve weighing accuracy. In particular, when deep learning (DNN) is used, weighing accuracy can be significantly improved. The wavelengths in the computational model can be selected and combined from statistically significant wavelengths in the near-infrared region ranging from 700 nm to 2500 nm. This allows for accurate weighing of specific parts of a living body that can be separated from the surrounding area.

[0011] Furthermore, if necessary, the total weight of the living body to be weighed is measured, and the weight ratio of the specific part calculated above to the total weight of the living body to be weighed is calculated. This allows the weight of the specific part relative to the total weight of the living body to be evaluated. Therefore, for example, this is useful in the case of a living body in which the specific part is attached to the living body and used for food, and the living body is distributed as a whole without removing the specific part.

[0012] Furthermore, if necessary, a plurality of weight ratio ranges are set by dividing the weight ratio of a specific portion of the living body to the total weight of the living body into predetermined ranges, and the living body to be weighed is sorted into the weight ratio ranges corresponding to the calculated weight ratios of the living body. Since the sorting is performed, when there are a large number of living bodies, the quality of the living bodies can be sorted more reliably, and sorting efficiency can be improved.

[0013] The living organism has a shell covering the specific portion, and the shell itself is formed to be transmissive to the irradiated near-infrared rays and / or is formed to have a transmissive portion through which the irradiated near-infrared rays can pass. The transmissive portion refers to a portion through which near-infrared rays pass more easily than other portions. The shell itself may be formed to be transmissive to near-infrared rays, and may have a transmissive portion through which near-infrared rays can pass. Examples of such living organisms that have edible parts include mollusks such as shellfish and snails, arthropods such as shrimp and crabs, sea urchins which are echinoderms, and sea squirts which are chordates.

[0014] In this case, it is effective that the shell is formed with a permeable portion through which the irradiated near-infrared rays can pass, and that the near-infrared rays for the irradiation are directed toward the permeable portion. If the shell has a portion through which near-infrared rays can easily pass compared to other portions, such as a partially thin portion, a hole, or an opening, this portion is a permeable portion and can be made into a permeable portion, and therefore, by directing near-infrared rays toward this portion, it becomes possible to reliably perform weighing.

[0015] Furthermore, if necessary, the living organism is a sea urchin having a shell with a mouth opening as the transmissible part with the mouth exposed at the center at the bottom and having gonads inside the shell, and the specific part to be measured is the gonad, and the near-infrared light for the irradiation is irradiated toward the mouth opening as the transmissible part.

[0016] Sea urchins are echinoderms, with over 100 species found in the waters off Japan alone. Representative examples include the pulcherrima, the Siberian sea urchin, the purple sea urchin, the northern purple sea urchin, and the red sea urchin. Traditionally, the gonads inside the shell have been eaten, but in recent years, they have increasingly been sold with the shell intact. Therefore, it is useful to measure the gonads inside the shell. In this case, the inventors discovered that while near-infrared light can be irradiated onto the general part of the shell, since the shell of the sea urchin is relatively thin and has many small holes, it is particularly effective to irradiate near-infrared light into the interior through the mouth opening, where the mouth is exposed in the center of the lower part of the shell, and then measure the absorbance by receiving the light from this mouth opening. This allows for accurate gonad measurement for a large number of sea urchins.

[0017] In particular, by training an AI model on many different types of gonads, rather than on individual types, it becomes possible to estimate the weight of the gonads with high accuracy. Similarly, by training many different types of sea urinary bladder simultaneously, it becomes possible to estimate the weight of the entire sea urinary bladder without building models for each individual type.

[0018] In order to achieve the above object, the device for weighing a specific part of a living body of the present invention is a device that realizes the method for weighing a specific part of a living body of the present invention, and is a device for weighing a specific part of a living body that is attached to a living body and can be separated from its surroundings in a non-destructive manner, a light source unit that irradiates a living body to be weighed with light in the near-infrared region; a light receiving unit that receives diffuse reflected light or diffuse transmitted light from the living body to be weighed; and a control unit that calculates the weight of a specific part of the living body to be weighed based on the absorbance of the light received by the light receiving unit, The control unit is configured to include a calculation model storage unit that stores a calculation model constructed based on the second derivative spectrum of the absorbance of the received light and performs calculations to calculate the weight of the specific part of the organism by irradiating a sample organism with near-infrared light in advance, receiving reflected or transmitted light from the sample organism, and measuring the weight of the specific part of the sample organism, and a weight calculation unit that calculates the weight of the specific part of the organism to be weighed from the absorbance of the light received by the light receiving unit and the calculation model stored in the calculation model storage unit.

[0019] In this device, it is desirable that the light source section has a dustproof and waterproof mechanism, and that it is disposed so as not to come into contact with the living body.

[0020] Since this device is a device for realizing the method for measuring a specific part of a living body according to the present invention, the description of the living body is the same as that described above and will be omitted. As a result, this inspection device irradiates near-infrared light onto the living organism to be weighed from the light source unit, receives diffusely reflected or transmitted light from the living organism in the light receiving unit, measures the absorbance of the received light, and calculates the weight of a specific part of the living organism based on the absorbance and a computational model in the control unit. This allows for non-destructive weighing of a specific part attached to the living organism that is separable from its surroundings, and allows for weighing the living organism while it is still alive, which is useful and causes minimal damage to the living organism. For example, in the case of a living organism that is edible and distributed as a whole without removing the specific part, it is useful to know the weight of the required specific part in advance and evaluate its quality. This is particularly useful when there are many such living organisms, as it allows for quality sorting of the living organisms. Furthermore, because it causes minimal damage to the living organism and allows for weighing while it is still alive, it can be applied, for example, in the case of seafood, to understanding the growth conditions and resource management in aquaculture farms and fishing grounds.

[0021] and, as necessary, the computational model is an analytical model constructed using statistical analysis or a trained model constructed using machine learning; The statistical analysis is selected from discriminant analysis, multiple regression analysis, PLS regression analysis, and support vector machine (SVM), The above machine learning is configured to be selected from tree analysis such as decision trees (CART), regression trees, random forests, and gradient boosting trees; neural networks such as perceptrons, convolutional neural networks (CNNs), recurrent neural networks (RNNs), residual networks (ResNets), and generative adversarial networks (GANs); deep learning (DNNs); and ensemble analysis consisting of a combination of these.

[0022] In this device, it is also effective to store two or more different types of calculation models as the calculation model in the calculation model storage unit, configure the weight calculation unit so that it can calculate the weight of a specific part of the living body to be weighed using any of the calculation models stored in the calculation model storage unit, and configure the control unit to have a calculation model designation means for designating the calculation model to be used by the weight calculation unit.

[0023] The computational model consists of an analytical model constructed using statistical analysis or a trained model constructed using machine learning, which can improve weighing accuracy. In particular, when deep learning (DNN) is used, weighing accuracy can be significantly improved. The wavelengths in the computational model can be selected and combined from statistically significant wavelengths in the near-infrared region ranging from 700 nm to 2500 nm. This allows for accurate weighing of specific parts of a living body that can be separated from the surrounding area.

[0024] Furthermore, if necessary, the apparatus may include a total weight measuring means for measuring the total weight of the living body to be weighed, and the control unit may include a weight ratio calculation unit for calculating the weight ratio of the specific part calculated by the weight calculation unit to the total weight of the living body to be weighed measured by the total weight measuring means. Since the weight ratio calculation unit calculates the weight ratio of the specific part, it becomes possible to evaluate the weight of the specific part relative to the whole living body. Therefore, this is useful, for example, in the case of a living body in which the specific part is attached to the living body and used for food, and the whole living body is distributed without removing the specific part.

[0025] In the present invention, it is effective if the transport unit has a sorting function for sorting living organisms based on the weight of the specific portion calculated by the weight calculation unit, as needed. More specifically, as necessary, the device comprises a base, a transport unit mounted on the base and transporting the living bodies to be measured one by one at a predetermined interval, a light detection unit having the light source unit and a light receiving unit, which irradiates the living bodies to be measured transported by the transport unit with light in the near-infrared region from the light source unit and receives diffuse reflected light or diffuse transmitted light from the living bodies to be weighed by the light receiving unit, and a receiving unit mounted downstream of the light detection unit in the transport unit and receives the living bodies to be measured, the receiving unit being configured to have a plurality of divided areas corresponding to weight ratio divisions set by dividing the weight ratio of a specific part to the total weight of the living body into a plurality of predetermined ranges, and the transport unit is configured to have a transfer means for transferring the living bodies to be weighed to the corresponding divided areas according to the weight ratio calculated by the weight ratio calculation unit of the control unit.

[0026] As a result, the living organism to be weighed is delivered to the corresponding sorting area by the delivery means in accordance with the weight ratio calculated by the weight ratio calculation unit of the control unit. Therefore, since the living organism is even sorted into weight ratio categories corresponding to the calculated weight ratio of the specific part, when there are a large number of living organisms, the quality of the living organisms can be sorted more reliably and sorting efficiency can be improved.

[0027] Furthermore, if necessary, the living body has a shell covering the specific part, and the shell is formed so that the near-infrared rays to be irradiated can pass through it, and / or is formed so that it has a transmissive portion through which the near-infrared rays to be irradiated can pass. In this case, it is effective that the shell is formed to have a transmissive portion through which the irradiated near-infrared rays can pass, and that the near-infrared rays to be irradiated are irradiated toward the transmissive portion. In this configuration, if necessary, the living organism is a sea urchin having a shell with a mouth opening as the transmissible part with the mouth exposed at the center at the bottom and having gonads inside the shell, and it is effective to set the specific part to be measured as the gonad and irradiate the near-infrared light related to the irradiation toward the mouth opening as the transmissible part.

[0028] More specifically, the living organism is a sea urchin having a shell with a mouth opening as the permeable portion, with the mouth exposed at the center of the lower part, and having gonads inside the shell, and the specific part to be measured is the gonads, The transport unit is configured as a conveyor having a plurality of trays arranged at predetermined intervals, each tray holding the living sea urchin with its mouth opening facing up, and the light detection unit is configured such that the light source unit is irradiated with near-infrared light related to the irradiation toward the mouth opening as the transmissive portion, and the light receiving unit is configured to receive reflected light from the mouth opening as the transmissive portion, Each of the above-mentioned division areas is constituted by a box placed below the above-mentioned tray, and the above-mentioned delivery means is constituted so as to drop the living sea urchin from the above-mentioned tray into the corresponding box.

[0029] Sea galls are echinoderms, and the gonads inside their shells have traditionally been eaten. In recent years, however, they have increasingly been distributed with the shells intact. Therefore, it is useful to weigh the gonads inside the shells. In this case, the inventors discovered that while near-infrared light can be irradiated onto the general portion of the shell, since the shells of sea galls are relatively thin and have many small holes, it is particularly effective to irradiate near-infrared light into the interior through the mouth opening, where the mouth is exposed in the center of the lower part of the shell, and then receive the light from this mouth opening and measure the absorbance. This allows for accurate gonad weighing for a large number of sea galls.

[0030] Furthermore, in order to achieve the object of the present invention, the trained model of the present invention is a trained model for causing a computer to function to calculate and output the weight of a specific part attached to a living body that is separable from its surroundings, and is a trained model constructed using the above-mentioned machine learning. It has the same functions and effects as those described above. [Effects of the Invention]

[0031] According to the present invention, near-infrared light is irradiated onto a living organism to be weighed, diffusely reflected or transmitted light from the organism is received, the absorbance of the received light is measured, and the weight of a specific part of the organism is calculated based on the absorbance and a computational model. This allows the weight of a specific part attached to the organism and separable from its surroundings to be measured non-destructively, allowing the weight of the specific part to be known and evaluated without removing it, and allowing the organism to be weighed alive, which is useful and causes little damage to the organism. For example, in the case of an organism whose specific part is attached to the organism and used for food, and which is distributed as a whole without removing the specific part, it is useful to know the weight of the necessary specific part in advance and evaluate its quality. This is particularly useful when there are many such organisms, as it allows the quality of the organisms to be sorted, which is extremely useful. Furthermore, because the method causes little damage to the organism and allows weighing while alive, it can be applied, for example, in the case of seafood such as sea urchin, to understanding the growth status and resource management in aquaculture farms and fishing grounds. [Brief explanation of the drawings]

[0032] [Figure 1] 1 is a perspective view showing a device for measuring a specific portion of a living body according to an embodiment of the present invention; [Figure 2] 1 is a front view showing a device for measuring a specific portion of a living body according to an embodiment of the present invention; [Figure 3] 1 is a side view showing a device for measuring a specific portion of a living body according to an embodiment of the present invention; [Figure 4] 1 is a plan view showing a device for measuring a specific portion of a living body according to an embodiment of the present invention; [Figure 5] 1A and 1B show a light detection unit in a device for measuring a specific portion of a living body according to an embodiment of the present invention, where FIG. 1A is a perspective view and FIG. 1B is a front view. [Figure 6] 1 is a diagram showing a state in which near-infrared light is irradiated and received by a light detection unit on sea urinary bladder, which is a living body to be weighed, in a specific portion weighing device for a living body according to an embodiment of the present invention. FIG. [Figure 7] 2 is a block diagram showing the configuration of a control unit in the device for measuring a specific part of a living body according to the embodiment of the present invention. FIG. [Figure 8] 1 is a diagram schematically illustrating a trained model according to an embodiment of the present invention used by a control unit in a specific part weighing device for a living body according to an embodiment of the present invention. FIG. [Figure 9] 4 is a flowchart showing a control flow in a control unit of the device for measuring a specific portion of a living body according to the embodiment of the present invention. [Figure 10] 4 is a flowchart showing the control flow of a calculation / separation routine in a control unit of a device for measuring a specific portion of a living body according to an embodiment of the present invention. [Figure 11] 10 is a flowchart showing a control flow of a control unit of a device for measuring a specific portion of a living body according to an embodiment of the present invention, illustrating a control flow of another calculation and separation routine. [Figure 12] FIG. 3 is a graph showing the correlation between actual measured values and predicted values in the calculation model (multiple regression analysis) according to the first embodiment of the present invention. [Figure 13]FIG. 10 is a graph showing the correlation between actual measurements and predicted values in the computational model (ensemble trained model combining decision trees and random forests) according to the second embodiment of the present invention. [Figure 14] FIG. 10 is a graph showing the state of the learning process for the computational model (trained model based on the DNN method) according to the third embodiment of the present invention. [Figure 15] FIG. 11 is a graph showing the correlation between the actual measurement value and the predicted value in the computation model (trained model based on the DNN method) of the third embodiment of the present invention. [Figure 16] FIG. 10 is a table showing the top 50 wavelengths with the highest contribution to prediction in order according to the computational model (trained model based on the DNN method) of Example 3 of the present invention. [Figure 17] FIG. 10 is a graph showing the correlation between actual measurements and predicted values in the computation model (trained model based on the CNN method) of the fourth embodiment of the present invention. [Figure 18] FIG. 11 is another graph showing the correlation between the actual measurement value and the predicted value in the computation model (trained model based on the CNN method) according to the fourth embodiment of the present invention. [Figure 19] FIG. 10 is a table showing the top 50 wavelengths with the highest contributions to prediction in order according to the computation model (trained model based on the CNN method) of the fourth embodiment of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0033] A method for weighing a specific part of a living body, a weighing device for a specific part of a living body, and a trained model according to an embodiment of the present invention will be described below with reference to the accompanying drawings. The method for weighing a specific part of a living body according to the embodiment is realized by the weighing device for a specific part of a living body and the trained model according to the embodiment, and will be described in the description of the weighing device for a specific part of a living body and the trained model. As shown in Figures 1 to 8, the weighing device S for a specific part of a living body according to the embodiment of the present invention non-destructively weighs the weight of a specific part Wa attached to a living body W and separable from its surroundings.

[0034] The living organism W may be a plant or animal, and may be alive or dead. The specific portion Wa is a specific portion of the tissue mass that is attached to the living organism W and is separable from the surrounding tissue. The specific portion Wa may be entirely contained within the living organism W, or a portion of the specific portion may be exposed from the living organism W. The specific portion Wa of a living organism weighing device S according to the embodiment is a device for a living organism W, which has a shell covering the specific portion Wa. The shell itself is formed to be transparent to the near-infrared light irradiated thereto and / or has a permeable portion through which the near-infrared light irradiated thereto can pass. Specifically, the living organism W is a living sea urchin 1 having a shell 2 with a mouth opening 3 as a permeable portion in the lower center, exposing the mouth, and having gonads 4 within the shell 2, as shown in FIG. 6 . In this sea urchin 1, the specific portion Wa to be weighed is the edible gonads 4.

[0035] Sea urchins 1 are echinoderms, and there are over 100 species found in the waters around Japan alone, with representative species including the pulcherrimus, the northern sea urchin, the purple sea urchin, the northern purple sea urchin, and the red sea urchin. The specific portion weighing device S of the living organism according to the embodiment can accommodate any of these representative sea urchins 1 by constructing a computational model, which will be described later.

[0036] The device S for weighing a specific portion of a living organism according to the embodiment comprises a base 10, a transport unit 11 mounted on the base 10 and transporting the living organisms W to be weighed one by one at a predetermined interval, a light detector 20 equipped with a light source 21 for irradiating the living organisms W to be weighed with near-infrared light during transport by the transport unit 11 and a light receiver 22 for receiving diffusely reflected or transmitted light from the living organisms W to be weighed, a receiver 30 mounted downstream of the light detector 20 on the transport unit 11 and receiving the living organisms W to be weighed, and a control unit 50 including a CPU or other functional unit for calculating the weight of the specific portion Wa of the living organism W to be weighed based on the absorbance of the light received by the light receiver 22 and for performing various controls. In the figure, reference numeral 51 denotes a display unit such as a CRT.

[0037] The transport unit 11 includes an endless conveyor 12, e.g., a timing belt type, driven by a motor (not shown). The conveyor 12 is provided with a plurality of trays 13, arranged at predetermined intervals along its moving direction, each holding one living organism W to be weighed. As shown in Figures 5(b) and 6, the upper surface of the tray 13 is formed in a truncated cone shape with a concave center, and supports and holds the sea urchin 1 (living organism W) with its mouth opening 3 facing upward. Each tray 13 is provided with a weighing scale (not shown) as total weight measuring means 14 (Figure 7) for measuring the total weight of the living organism W to be weighed, and the measured weight is transmitted as data to the control unit 50. Reference numeral 15 denotes a loading stage provided next to the transport unit 11 for loading the sea urchin 1 onto the tray.

[0038] As shown in FIGS. 5 and 6 , the light detection unit 20 includes a light source unit 21 and a light receiving unit 22. The light source unit 21 is composed of a light source lamp 24 covered with a shade 23 that emits light in the near-infrared region. The light receiving unit 22 is composed of a condensing fiber 25, which is composed of a laminate of optical fibers and receives light at its end face and transmits it to the control unit 50. The light source lamp 24 and the condensing fiber 25 are supported on a support base 26. The light source lamp 24, which is the light source unit 21, is mounted on the support base 26 so as to radiate near-infrared light toward the mouth opening 3, which serves as the transmissive portion of the sea urinary tract 1, in a non-contact manner. The end face of the condensing fiber 25, which serves as the light receiving unit 22, is mounted on the support base 26 so as to receive reflected light from the mouth opening 3, which serves as the transmissive portion, in a non-contact manner. The support base 26 also includes a plate-shaped quartz glass 27, which is located in front of the light source lamp 24 in the direction of irradiation and transmits near-infrared light. The support base 26 is also provided with a blower fan 28 that blows air to the front of the quartz glass 27 to prevent dust from entering between the living body W and the light source lamp 24 .

[0039] In this configuration, the shell 2 of the sea urchin 1 is relatively thin and has many small holes, so near-infrared rays can be irradiated onto the general part of the shell 2, but in particular, near-infrared rays are irradiated into the interior from the mouth opening 3, where the mouth is exposed in the center of the lower part of the shell of the sea urchin 1, and the light from this mouth opening 3 is received and the absorbance is measured, so that accurate gonad measurement can be performed for a large number of sea urchins.

[0040] As shown in Fig. 4, the receiving unit 30 is configured with a plurality of (four in this embodiment) sectional areas E1 to E4 corresponding to weight ratio ranges set by dividing the weight ratio ρ of the specific portion Wa to the total weight of the living body W into a plurality of predetermined ranges. In this embodiment, each of the sectional areas E1 to E4 is formed by a box 31 and is installed on an installation surface below the tray 13 of the base 10. For example, the sectional area E1 accepts sea urchin 1 with 0 < ρ < 0.1, the sectional area E2 accepts sea urchin 1 with 0.1 < ρ < 0.15, the sectional area E3 accepts sea urchin 1 with 0.18 < ρ < 0.25, and the sectional area E4 accepts sea urchin 1 with 0.25 < ρ. The number of sectional areas and the value of the weight ratio ρ are not limited to these. It is also possible to classify only by the weight of a specific part. For example, the area E1 could be 90g or less, E2 91-110g, E3 111-120g, and E4 121g or more. The areas are not limited to these weight values.

[0041] 3, the transport unit 11 is provided with a transfer means 40 that transfers the living organism W to be weighed to the corresponding sorting area E1 to E4 according to a weight ratio calculated by a weight ratio calculation unit 56 of the control unit 50 (described later). The trays 13 are tiltably supported by the conveyor 12, and the transfer means 40 is provided for each tray 13 and is composed of, for example, an electric motor that moves the tray 13 between a position for holding the living organism W and a tilted position. The electric motor tilts the tray 13 in response to a command from the control unit 50, causing the living organism W on the tray 13 to drop into a box 31 in the corresponding sorting area E1 to E4.

[0042] 7, the control unit 50 includes a calculation processing unit 52 that performs overall control. The calculation processing unit 52 includes a calculation model storage unit 53 that stores a calculation model that is constructed based on the second derivative spectrum of the absorbance of the received light by irradiating the sample living body W with near-infrared light, receiving the reflected light or transmitted light from the sample living body W, and measuring the weight of a specific portion Wa of the sample living body W, and performs calculations to calculate the weight of the specific portion Wa of the living body W.

[0043] The calculation processing unit 52 also includes a weight calculation unit 54 that calculates the weight of a specific part Wa of the living body W to be weighed from the absorbance of the light received by the light receiving unit 22 and the calculation model stored in the calculation model memory unit 53.

[0044] The computational model is an analytical model constructed using statistical analysis or a trained model constructed using machine learning, Statistical analysis was selected from discriminant analysis, multiple regression analysis, PLS regression analysis, and support vector machine (SVM). Machine learning is selected from tree analysis such as decision trees (CART), regression trees, random forests, and gradient boosted trees; neural networks such as perceptrons, convolutional neural networks (CNNs), recurrent neural networks (RNNs), residual networks (ResNets), and generative adversarial networks (GANs); deep learning (DNNs); and ensemble analysis consisting of a combination of these.

[0045] Specifically, in multiple regression analysis and PLS regression analysis, near-infrared light is irradiated onto a sample living organism W, the diffusely reflected light or diffusely transmitted light from the sample living organism W is received, and the absorbance of the received light is measured. Then, the shell 2 of the sea urchin 1 serving as the living organism W is broken, the gonad 4 serving as the specific portion Wa is removed from inside, and the weight of the gonad 4 serving as the specific portion Wa is measured. An analytical model (computation model) relating to the wavelengths attributable to the weight of the gonad 4 serving as the specific portion Wa is then identified through statistical analysis of the second-derivative spectrum of the absorbance. Next, near-infrared light is irradiated onto the living organism W to be weighed, the diffusely reflected light or diffusely transmitted light from the living organism W to be weighed is received, the absorbance of the received light is measured, and the weight of the specific portion Wa of the living organism W to be weighed is calculated from these absorbances and the analytical model.

[0046] <Multiple regression analysis> The analytical model is determined by multiple regression analysis, where the absorbance spectrum of light received from the biological sample W is subjected to MSC processing and then second-order differentiation. MSC (Multiplicative Scatter Correction) is a preprocessing method for removing multiplicative factors (such as light scattering) and additive factors that occur in the measured spectrum (hereinafter the same).

[0047] The analytical model was specified by an equation (discriminant) that satisfies the relationship of the following general formula (B) using the absorbances of wavelengths 1 to n that have high correlation coefficients as variables.

[0048]

number

[0049] In the general formula (B), when selecting the first wavelength (λ1) to the nth wavelength (λn), first, the absorbance of the sample living body W is measured in advance. The weight of the specific part Wa of the living body W is then separately measured by actually taking it out from the living body W. The wavelength range of near-infrared light of the first wavelength (λ1) with the highest correlation coefficient between the absorbance of the sample living body W and the weight of the reproductive gonad 4 as the specific part Wa is selected. Next, by performing multiple regression analysis on the near-infrared wavelength range of the first wavelength (λ1) and the wavelength range of 700 nm to 2500 nm, the wavelength range of the second wavelength (λ2) that belongs to the weight of the reproductive gonad 4 as the specific part Wa and has a correlation coefficient higher than that of the near-infrared wavelength range of the first wavelength (λ1) is selected. Then, by performing multiple regression analysis on the near-infrared wavelength ranges of the first wavelength (λ1) and the second wavelength (λ2) and the wavelength range of 700 nm to 2500 nm, the wavelength range of the third wavelength (λ3) that belongs to the weight of the reproductive gonad 4 as the specific part Wa and has a correlation coefficient higher than that of the near-infrared wavelength ranges of the first wavelength (λ1) and the second wavelength (λ2) is selected. In this way, by performing multiple regression analysis on the near-infrared wavelength ranges of the first wavelength (λ1) to the (n - 1)th wavelength (λn - 1) and the wavelength range of 700 nm to 2500 nm, the wavelength range of the nth wavelength (λn) that belongs to the weight of the reproductive gonad 4 as the specific part Wa and has a correlation coefficient not less than that of the near-infrared wavelength range of the first wavelength (λ1) is selected.

[0050] <PLS regression analysis> The above analysis model was calculated and identified by the PLS (Partial least square projection to Latent Structure) analysis method after secondarily differentiating the absorbance spectrum of the light received from the sample living body W after MSC processing.

[0051] <Support Vector Machine (SVM)> The above trained model was identified by the SVM (Saport Vecter Machine) method after secondarily differentiating the absorbance spectrum of the light received from the sample living body W after MSC processing.

[0052] Next, an example of the deep learning method is given. <Deep learning method (1)> For example, the trained model may be a trained model obtained by a so-called ensemble method using a decision tree of a deep learning method and a random forest method, after MSC processing of the absorbance spectrum of light received from the sample living body W. A statistically superior trained model is obtained by the ensemble method. The combination is not limited to this, and may be selected and combined as appropriate.

[0053] <Deep Learning Method (2)> As the trained model, the absorbance spectrum of the light received from the sample living body W is subjected to MSC processing and then second-order differentiation, and a trained model using the deep learning method DNN (Deep Neural Network) is used.

[0054] <Deep Learning Method (3)> As the trained model, the absorbance spectrum of the light received from the sample living body W is subjected to second-order differentiation after MSC processing, and the waveform is visualized to use a trained model using the deep learning method CNN (Convolutional Neural Network).

[0055] <Deep Learning Method (4)> We use a trained model that uses a generative adversarial network (GAN), a type of AI that generates new data based on input data. Its structure consists of two networks, a generator and a discriminator, which compete with each other to create data that is closer to the real thing. One network generates new data by taking samples of the input data and modifying them as much as possible, while the other network predicts whether the generated data output belongs to the original dataset. The fake data values are improved, increasing accuracy and approaching the real thing, until the predictive network can no longer distinguish between fake and original data.

[0056] In the embodiment, the trained model is a trained model that causes a computer to calculate and output the weight of a specific part Wa that is attached to a living organism W and is separable from its surroundings, and refers to a trained model constructed using the above-mentioned machine learning. Figure 8 shows an example configuration of the trained model. In the embodiment, for example, 550 pieces of absorbance data at wavelength intervals of 2 nm are input to the input layer.

[0057] More specifically, as shown in Figure 8, this is a trained model for causing a computer to function to output a quantified value for the weight of the gonads 4 based on near-infrared spectroscopic wavelength absorbance data relating to the weight of the gonads 4 of the sea urinary bladder 1, and is composed of, for example, a first neural network and a second neural network connected so that the output from the first neural network is input.

[0058] In this trained model, the first neural network is composed of the input layer to the intermediate layer of a feature extraction neural network in which the number of neurons in at least one intermediate layer is smaller than the number of neurons in the input layer, the number of neurons in the input layer and the output layer are the same, and weighting coefficients are trained so that the input value to each input layer is equal to the output value from each output layer corresponding to each input layer; the weighting coefficients of the second neural network are trained without changing the weighting coefficients of the first neural network; and the computer is caused to function so that a calculation is performed based on the trained weighting coefficients in the first and second neural networks for a specific absorbance obtained from absorbance data at near-infrared spectral wavelengths related to the weight of gonads 4 input to the input layer of the first neural network, and a quantified value of the weight of gonads 4 is output from the output layer of the second neural network.

[0059] The trained model according to the embodiment is for causing a computer to function and sort the gonads 4, which are the edible part of the sea urchin 1, by outputting a quantified value for the weight of the gonads 4 based on near-infrared spectroscopic wavelength absorbance data relating to the measurement of the weight of the gonads 4. The trained model comprises a first neural network and a second neural network connected to receive the output from the first neural network. The trained model operates in such a way that, in accordance with instructions from the trained model stored in memory, the CPU of the computer performs calculations based on the trained weighting coefficients and response functions of the first and second neural networks on input data input to the input layer of the first neural network (the absorbance data relating to the weight of the gonads 4, and the absorbance corresponding to the weight obtained by dissecting the sea urchin 1 and measuring the weight of the gonads 4). The trained model then outputs the result (a quantified value for the weight, such as 10 g, 20 g, 84 g, etc.) from the output layer of the second neural network.

[0060] The first neural network is composed of the input layer to the hidden layer of the feature extraction neural network. This feature extraction neural network is generally called an autoencoder, and the number of neurons in the hidden layer is smaller than the number of neurons in the input layer, and the number of neurons in the output layer is set to one. The response function of each neuron in the input and output layers is a linear function, and the response function of each neuron other than that is a ReLU (ramp function) or a sigmoid function (1 / (l+exp(-x))). The feature extraction neural network is trained using the well-known backpropagation method, and the weighting coefficients between neurons are updated.

[0061] The neural network is trained using the well-known backpropagation method, and the weighting coefficients between neurons are updated. In this embodiment, near-infrared spectroscopic wavelength absorbance data relating to the weight of the gonads 4 of the sea urinary bladder 1 is input to the input layer, and training is performed to minimize the mean square error for all input data, which is calculated from the input data and output from the output layer.

[0062] As mentioned above, since nonlinear functions such as ReLU (ramp function) and sigmoid functions are used as neuron response functions, the weighting coefficients between neurons are not symmetric across the intermediate layer. As the feature extraction neural network learns, features that represent the properties of each input data can be acquired in the intermediate layer. The features that appear in the intermediate layer do not necessarily have a clear physical meaning, but can be thought of as compressed to the extent that the information input to the input layer can be restored to the information output in the output layer. Therefore, regardless of the input features to the input layer, the features that appear in the intermediate layer will be approximately the same, eliminating the need to appropriately set the input features to the input layer in advance.

[0063] In this device, the portion of the feature extraction neural network in which the weighting coefficients have been trained, from the input layer to the intermediate layer, is connected to the second neural network as the first neural network. The weighting coefficients of the second neural network are updated through training without changing the weighting coefficients of the first neural network. As with the above, this training is also performed using the well-known backpropagation method. Because the trained model of the present invention is composed of the first and second neural networks described above, it is possible to accurately quantify the weight of the gonads 4 without presetting input features such as wavelength selection.

[0064] In the embodiment, a plurality of different types of computation models are provided as the computation model, and these can be selectively used in the control unit 50. Fig. 11 shows a case where a multiple regression analysis model, a support vector machine (SVM), and a plurality of AI-trained models are used.

[0065] That is, in the control unit 50, the calculation model memory unit 53 stores two or more different types of calculation models as calculation models, and the weight calculation unit 54 is configured to be able to calculate the weight of a specific part Wa of the living body W to be weighed using any of the calculation models stored in the calculation model memory unit 53, and the calculation model to be used can be specified to the weight calculation unit 54 from outside via the operation command unit 64 described below.

[0066] The control unit 50 also includes a weight ratio calculation unit 56 that calculates the weight ratio of the specific part Wa calculated by the weight calculation unit 54 to the total weight of the living body W to be weighed, measured by the total weight measurement means 14 (a weighing scale not shown).

[0067] The calculation results of all the calculation models can be displayed on the display unit 51. In this embodiment, the total weight of the living body W, the weight of the specific part Wa of the living body W, and the weight ratio of the specific part Wa of the living body W to the total weight are output and displayed as final results.

[0068] 7, in the control unit 50, the arithmetic processing unit 52 has an operation control unit 60, which has a processing history storage function, a result display function on the display unit 51, an internet communication function, a light amount control function, etc. The control unit 50 is also configured to have a communication unit 61 controlled by the internet communication function, a light control unit 62 which has a spectrometer and controls the spectroscopic control circuit thereof and performs control related to the light detection unit 20, a drive control unit 63 which has a transport unit control circuit which controls the transport unit 11 and a delivery means control circuit which controls the delivery means 40, etc., and an operation command unit 64 such as a keyboard (not shown) and a setting panel (not shown).

[0069] The processing history storage function of the operation control unit 60 chronologically accumulates and stores the cumulative number of measurements of the living organism W to be weighed, the total weight of the living organism W, the weight of the specific part Wa of the living organism W, and the calculation results of the weight ratio of the specific part Wa of the living organism W to the total weight. The Internet communication function has a function for understanding the discrimination situation and system status when used in a remote location, and for remotely performing maintenance such as updating the application (discrimination method). By using this Internet communication function, the living organism W to be weighed can be weighed regardless of location, thereby improving the versatility of the device. The light intensity control function adjusts the light intensity of the light source and the sensitivity of the sensor in the spectroscopic unit of the light control unit 62 so that the light detection sensor in the spectroscopic unit can perform measurements at an appropriate level.

[0070] In addition, the operation control unit 60 sends a command to the drive control unit 63 to drive the electric motor of the transfer means 40 at a predetermined timing so that the living body W to be weighed is dropped and transferred into the box 31 of the corresponding division area E1 to E4 in accordance with the weight ratio calculated by the weight ratio calculation unit 56.

[0071] Various data are displayed on the display unit 51. The display on the display unit 51 is operated by the screen operation unit (operation command unit 64), and can be switched appropriately to display an input setting screen, a discrimination result display screen, an error display screen, etc. The discrimination results may be displayed on an LCD panel. The discrimination results may also be output as voice. Furthermore, an external data output interface may be provided.

[0072] Therefore, when weighing a specific part of a living body W using the weighing device S according to this embodiment, the process is as follows: This will be explained using the flow charts shown in Figs. As shown in Figure 9, when the main power supply switch (not shown) is turned on, the control unit 50 starts up and checks each mechanism, such as the spectrometer, sensor, and lamp (1-1). At this time, a screen indicating that the system is being initialized appears on the display unit 51, and the indicator light turns yellow (1-2). After the check operation for each mechanism is completed, a reference waveform that serves as the basis for calculating absorbance is measured (1-3), and this is stored in the memory unit inside the control unit 50 (1-4). The main screen is then displayed, and the indicator light turns green (1-8). If a setting value is called up from the setting panel (not shown) during this time (1-5), a confirmation screen for the accumulated lamp lighting time, date and time, etc. is displayed (1-6). To perform a measurement, operate the setting panel to return to the main screen (1-7).

[0073] The sea urinary tract 1 as the living organism W to be weighed is placed on a tray 13 on a loading stage 15 provided next to the transport unit 11. When the indicator light on the main screen is green, pressing the start switch of the control button (1-9) executes a calculation and separation routine for measuring the total weight of the living organism W to be weighed and calculating the weight of a specific portion Wa of the living organism W to be weighed (1-10). In this calculation and separation routine, as shown in Figure 10, the control unit 50 first issues an instruction to operate the conveyor 123 of the transport unit 11 (2-1), and the living organism W to be weighed is transported. During this transport process, when the position detection sensor detects that the living body W has been positioned at a predetermined position in the light detection unit 20 (2-2), light is irradiated from the light source unit 21 onto the living body W to be weighed, and the diffusely reflected light or diffusely transmitted light is received by the light receiving unit 22 and transmitted via optical fiber to the spectroscopic unit of the light control unit 62, where the intensity of the near-infrared wavelength is converted into voltage for each wavelength range and extracted and transmitted to the calculation processing unit 52 (2-3).Then, the intensity spectrum of the diffusely reflected light or diffusely transmitted light is obtained, and the weight of the specific portion Wa of the living body W is calculated using a calculation and classification routine (2-4).

[0074] In another routine, as shown in FIG. 11, noise reduction processing is performed on the diffused light intensity (diffuse light spectrum) transmitted from the spectrometer (3-1, 3-2), the reference waveform stored during initialization is read (3-3), and the near-infrared absorbance of the living body W to be weighed is calculated (3-4). The absorbance is standardized (3-5), and then subjected to second-order differentiation (3-6). The second-order derivative spectrum is then centered and scaled (3-7). Using this spectrum, the weight of a specific portion Wa of the living body W is calculated (3-12) using a computation model pre-stored in the computation model storage unit 53. This calculation is performed using multiple regression analysis (3-8), AI-trained model (1), (2), and (3) type computation (3-9), AI-trained model (4) type computation (3-10), or support vector machine (SVM) computation (3-11). Next, the weight ratio calculation unit 56 calculates the weight ratio of the specific portion Wa calculated by the weight calculation unit 54 to the total weight of the living body W to be weighed measured by the total weight measurement means 14 (3-13).

[0075] Then, returning to FIG. 10, the transfer means 40 is operated at a predetermined timing based on the weight ratio of the specific portion Wa of the living body W, and the sea urinary tract 1 as living body W is dropped from the tray 13 into the boxes 31 of the receiving section 30 corresponding to the sorting areas E1 to E4 for sorting (2-5). Also, returning to FIG. 9, the data up to that point is displayed (1-13). If an error occurs in any of the mechanisms or the emergency stop button is pressed (1-11), an error screen is displayed and the indicator light turns red (1-12). To restore operation, check the lamp, spectrometer, sensor, and other mechanisms. Each piece of data and the diffused light spectrum during measurement of the living body W to be weighed are stored in the memory function of the control unit 50. This process is repeated until the stop button is pressed (1-15).

[0076] This allows the weight of the gonads 4, a specific part of the sea urchin 1, to be measured nondestructively and without contact, allowing for evaluation without removing the gonads 4. Furthermore, since the organism can be weighed alive, it is less damaging to the organism and is useful. This is particularly useful for sea urchin 1, which is distributed with the gonads 4 intact and in their shells. Furthermore, the weight ratio of the gonads 4 to the total weight of the sea urchin 1 is calculated, and the sea urchin 1 is sorted into multiple (four in this embodiment) sorting areas E1-E4 corresponding to the weight ratio ranges set for each range. This allows for reliable quality sorting when there are a large number of sea urchins 1, improving sorting efficiency. Furthermore, this method is useful not only for sorting for consumption, but also for growth management at aquaculture farms and fishing grounds. [Example]

[0077] Next, an example of a calculation model will be shown, along with the results of verification carried out by actually measuring the calculation model. <Example 1: Multiple regression analysis> For the sea urchin 1 of sample W, a total of 8,232 pieces were used: 7,032 live purple sea urchins harvested in Hachinohe City, Aomori Prefecture, and 1,200 live red sea urchins harvested in Hirono Town, Iwate Prefecture. The data for creating the multiple regression analysis model was obtained by performing near-infrared spectroscopy on the sea urchin 1 of all living organism samples W, then splitting their shells 2 to extract the gonads 4 as specific parts Wa, and correlating the weight data of these gonads 4 with the absorbance data of the near-infrared spectroscopic wavelengths of the living organism samples W. Using this creation data, the necessary wavelengths were searched for in order of highest correlation, and a multiple regression analysis model was created.

[0078] The accuracy of calculating the weight of the gonad 4 as the specific part Wa was verified using the calculation model (multiple regression model) according to Example 1. As the living organisms W to be weighed, 840 live purple sea urchins harvested in Hachinohe City, Aomori Prefecture were used. The results are shown in Figure 12. The coefficient of determination obtained from the actual measured values and the regression equation was R2 = 0.9338, the predicted value RMSE (Root Mean Square Error) = 1.0671, and the p-value < 0.0001. From this, it can be said that this model has statistically significant predictions and is an extremely effective regression equation.

[0079] Example 2: Ensemble trained model combining decision trees and random forests For the sea urchin 1 of sample W, a total of 8,232 pieces were used: 7,032 live purple sea urchins harvested in Hachinohe City, Aomori Prefecture, and 1,200 live red sea urchins harvested in Hirono Town, Iwate Prefecture. Here, we present an overview of the algorithm used in creating software for predicting the weight of gonads 4 from the absorbance data of sea urinary bladder 1 in this trained model (the same applies to the training models shown below). In this process, we created a program that was divided into three main steps: data preprocessing, model training, and prediction. 1. Data Preprocessing Multivariate scatter correction (MSC processing) was applied to the absorbance spectral data to minimize the influence of scattering between different samples. Second-order derivatives were applied to the MSC-processed spectral data. This data preprocessing was used to build all models, including multiple regression analysis. 2. Training the model An ensemble learning model was constructed in Python, combining decision trees and random forests as an application of ensemble methods. Decision trees are a method of forming decision rules based on data features and using them to predict weight. They were programmed to understand which wavelength features contribute most to the prediction. Random forests are models that combine multiple decision trees, each learning from a different random subset of the dataset. Random forests can suppress overfitting of decision trees and improve overall prediction accuracy. 3. Prediction The preprocessed new absorbance data is used as input to predict weight using the trained ensemble model. The predicted value obtained from the model is interpreted as the weight of the specific part, the gonad 4 of the sea urchin 1 (the edible part of raw sea urchin). This algorithm was constructed and implemented using the Python machine learning library TensorFlow.

[0080] Then, the accuracy of calculating the weight of the gonad 4 as the specific part Wa was verified using the trained model according to Example 2. As the living organisms W to be weighed, 242 live purple sea urchins harvested in Hachinohe City, Aomori Prefecture were used. The results are shown in Figure 13. In this result, Error center: -0.086671036 Median error rate: -1.036989615 Average error: -0.147371493 RMSE: 0.850170461 Coefficient of determination: 0.965231152 This is what happened.

[0081] Overall, these results indicate that the model captures the variability of the data very well (high coefficient of determination). Although the predicted values are systematically lower than the actual values (negative median and mean errors), the RMSE is relatively low, suggesting that the errors in individual predictions are not large. Therefore, the model is highly effective at predicting weight.

[0082] <Example 3: Trained model based on DNN method> For the sea urchin 1 of sample W, a total of 8,232 pieces were used: 7,032 live purple sea urchins harvested in Hachinohe City, Aomori Prefecture, and 1,200 live red sea urchins harvested in Hirono Town, Iwate Prefecture. This program uses a network that combines two fully connected layers with batch normalization, activation functions, and dropout. The input data is sea urchin 1 (sample W) irradiated with near-infrared light, and the resulting diffused light is treated as 'x', which has the feature value of 550 absorbance data sets at 2nm intervals from 1100nm to 1998nm. This 'x' is then processed as follows to estimate the weight 'h' of gonad 4 (edible part of raw sea urchin) of sea urchin 1 (specific part).

[0083] As a data preprocessing step, multivariate scatter correction (MSC processing) was applied to the absorbance spectral data to minimize the effect of scattering between different samples. Second-order derivatives were then applied to the MSC-processed spectral data. The program structure is as follows: 1. Linear transformation (Affine): Transform the input data with a weight matrix and bias. 2. BatchNormalization: Normalize the output data. 3. Activation function (Swish): Apply the Swish function to the output data. 4. Dropout: Randomly invalidate input data. 5. BatchNormalization: Normalizes the output data. 6. Linear transformation (Affine): Transforms the input data using a weight matrix and bias. 7. BatchNormalization: Normalizes the output data. 8. Loss function (SquaredError): Calculates the error between the output data and the target data.

[0084] During training, the network parameters are updated to minimize the loss function 'h' using input data 'x' and target data 'y'. During inference, the loss function 'h' is calculated using only input data 'x'. Adam was used as the optimal algorithm. The learning rate update coefficients were beta1 = 0.9, beta2 = 0.999, and the learning rate decay rate hadecay = 0.001. Training was performed for 10,000 epochs. The training results indicated that the optimal convergence value was a loss function value of 0.341.

[0085] The learning process is shown in Figure 14. The learning was performed up to 10,000 epochs, but this figure shows the process from 1 to 100.

[0086] Then, the accuracy of calculating the weight of the gonad 4 as the specific part Wa was verified using the trained model according to Example 3. As the living organisms W to be weighed, 242 live purple sea urchins harvested in Hachinohe City, Aomori Prefecture were used. The results are shown in Figure 15. In this result, Median error: -0.2697 Median error rate: -2.9389 Average error: -0.2473 PMSE: 7.1553 Coefficient of determination: 0.9615 This is what happened.

[0087] Overall, these results indicate that the model captures the variability of the data very well (high coefficient of determination). Although the predicted values are systematically lower than the actual values (negative median and mean errors), the RMSE is relatively low, suggesting that the errors in individual predictions are not large. Therefore, the model is highly effective at predicting weight.

[0088] Furthermore, in Example 3, the contribution of wavelengths can be shown. The above 550 wavelengths contribute to a certain extent, but the order of the degree of contribution of the wavelengths follows the wavelength contribution (SHAP2 values.csv) as shown in FIG. 16. The SHAP (SHapley Additive exPlanations) value is an index used to explain the contribution of each feature amount in the prediction of a machine learning model. FIG. 16 shows the top 50 wavelengths with high contribution.

[0089] <Example 4: Trained model based on CNN method> For the sea urchin 1 of sample W, a total of 8,232 pieces were used: 7,032 live purple sea urchins harvested in Hachinohe City, Aomori Prefecture, and 1,200 live red sea urchins harvested in Hirono Town, Iwate Prefecture. This model performs MSC processing on the absorbance spectrum of light received from the living organism W, performs second-order differentiation, and converts this data into image format before inputting it into a convolutional neural network. For input data, near-infrared light is irradiated onto the sea urinary tract 1 (the living organism W), and the resulting diffused light is absorbance data with 550 features at 2-nm intervals from 1100 nm to 1998 nm. This data is then converted into 1x1x550 3D data and input into the convolutional neural network. The network passes through multiple units, ultimately estimating the weight of the gonad 4 of the sea urinary tract 1 as a specific region.

[0090] Then, the accuracy of calculating the weight of the gonad 4 as the specific part Wa was verified using the trained model according to Example 4. As the living organisms W to be weighed, 588 live purple sea urchins harvested in Hiranai Town, Aomori Prefecture were used. The results are shown in Figure 17. In this result, Median error: -0.01518 Median error rate: 0.16934 RMSE: 5.68493 Coefficient of determination: 0.9631 This is what happened.

[0091] Furthermore, using the trained model according to Example 4, the accuracy of calculating the weight of the gonad 4 as a specific part Wa of another sea urchin 1 was verified. As the living organisms W to be weighed, 251 live purple sea urchins harvested in Hiranai Town, Aomori Prefecture were used. The results are shown in Figure 18. In this result, Median error: 0.29398 Median error rate: -2.81877 RMSE: 5.83108 Coefficient of determination (R-squared): 0.96861

[0092] These results show that this model has high prediction accuracy, and in particular, the coefficient of determination is very high, which means that it effectively explains most of the variation in the target data. As with the above examples, this model can be said to be extremely effective in predicting weight.

[0093] Furthermore, in Example 4, as in Example 3, the degree of wavelength contribution can be shown. The 550 wavelengths mentioned above all contribute to a certain extent, but the order of the degree of contribution of the wavelengths follows the wavelength contribution (SHAP2 values.csv), as shown in Figure 19. Figure 19 shows the top 50 wavelengths with the highest contribution. The order of wavelength contribution and SHAP values may vary slightly depending on the sample, but these wavelengths are essential when estimating the weight of the gonad 4 as a specific part Wa of the sea urinary bladder 1. At least 50 to 120 wavelengths are considered important.

[0094] In the above embodiment, the example of the sea urinary bladder 1 has been described as the living body W, but it is needless to say that the present invention is not limited to this and can be applied to various living bodies W. In short, those skilled in the art can easily make many modifications to these exemplary embodiments without substantially departing from the novel teachings and effects of the present invention, and these many modifications are included in the scope of the present invention. [Explanation of symbols]

[0095] S. Device for measuring specific parts of living organisms W Living Body Wa specific part 1 Sea urinary tract (live) 2. Shell 3 Mouth opening (transmissible part) 4. Gonads (specific parts) 10 Foundations 11 Conveyor 12 Conveyor 13 Tray 14 Total weight measurement means 15 Mounting stage 20 Light detection unit 21 Light source section 22 Light receiving part 23 Umbrella 24 Light source lamp 25 Optical fiber 26 Support stand 27 Quartz glass 28 Blower fan 30 Receiving section 31 boxes E1~E4 division area 40 Delivery Method 50 control section 51 Display section 52 Processing unit 53 Calculation model storage unit 54 Weight calculation section 56 Weight ratio calculation section 60 Operation control section 61 Communications Department 62 Optical control section 63 Drive control unit 64 Operation command part

Claims

1. A method for weighing a specific part of a living body, which non-destructively measures the weight of a specific part attached to the living body and separable from its surroundings, comprising: a calculation model is constructed in advance, which performs calculations to calculate the weight of a specific part of a living organism based on a second derivative spectrum of the absorbance of the received light, by irradiating a sample living organism with near-infrared light, receiving reflected light from or transmitted through the sample living organism, and measuring the weight of the specific part of the living organism; A method for weighing a specific part of a living body, characterized by irradiating near-infrared light onto the living body to be weighed, receiving diffusely reflected light or diffusely transmitted light from the living body to be weighed, measuring the absorbance of the received light, and calculating the weight of the specific part of the living body to be weighed from the absorbance and the above-mentioned calculation model.

2. The computational model is an analytical model constructed using statistical analysis or a trained model constructed using machine learning, The statistical analysis is selected from discriminant analysis, multiple regression analysis, PLS regression analysis, and support vector machine (SVM); The method for quantifying a specific part of a living organism according to claim 1, characterized in that the machine learning is selected from tree analysis such as decision trees (CART), regression trees, random forests, and gradient boosting trees, neural networks such as perceptrons, convolutional neural networks (CNN), recurrent neural networks (RNN), residual networks (ResNet), and generative adversarial networks (GAN), deep learning (DNN), and ensemble analysis consisting of a combination of these.

3. 3. The method for weighing a specific part of a living body according to claim 2, further comprising measuring the total weight of the living body to be weighed and calculating the weight ratio of the calculated specific part to the total weight of the living body to be weighed.

4. A method for weighing a specific part of a living organism as described in claim 3, characterized in that a plurality of weight ratio categories are set in which the weight ratio of the specific part to the total weight of the living organism is divided into predetermined ranges, and the living organism to be weighed is separated into weight ratio categories corresponding to the calculated weight ratio of the living organism.

5. A method for weighing a specific part of a living body described in any one of claims 1 to 4, characterized in that the living body has a shell covering the specific part, and the shell is formed so that the irradiated near-infrared light can pass through it, and / or is formed so that it has a transmissive portion through which the irradiated near-infrared light can pass.

6. The method for weighing a living organism according to claim 5, characterized in that the shell is formed with a transmissive portion through which the irradiated near-infrared rays can pass, and the near-infrared rays are irradiated toward the transmissive portion.

7. The method for weighing a specific part of a living organism as described in claim 6, characterized in that the living organism is a sea urchin having a shell with an oral opening as the transmissible part, with the oral opening exposed in the center at the bottom, and having gonads inside the shell, the specific part to be weighed is the gonad, and the near-infrared light related to the irradiation is irradiated toward the oral opening as the transmissible part.

8. A device for measuring a specific part of a living body that nondestructively measures the weight of a specific part that is attached to a living body and can be separated from its surroundings, a light source unit that irradiates a living body to be weighed with light in the near-infrared region; a light receiving unit that receives diffuse reflected light or diffuse transmitted light from the living body to be weighed; and a control unit that calculates the weight of a specific part of the living body to be weighed based on the absorbance of the light received by the light receiving unit, The control unit is configured to irradiate a sample organism with near-infrared light, receive reflected or transmitted light from the sample organism, and measure the weight of a specific part of the sample organism, thereby storing a calculation model constructed based on the second derivative spectrum of the absorbance of the received light and performing calculations to calculate the weight of the specific part of the organism, and a weight calculation unit to calculate the weight of the specific part of the organism to be weighed from the absorbance of the light received by the light receiving unit and the calculation model stored in the calculation model storage unit.

9. The computational model is an analytical model constructed using statistical analysis or a trained model constructed using machine learning, The statistical analysis is selected from discriminant analysis, multiple regression analysis, PLS regression analysis, and support vector machine (SVM); The device for quantifying a specific part of a living body according to claim 8, characterized in that the machine learning is selected from tree analysis such as decision trees (CART), regression trees, random forests, gradient boosting trees, etc., neural networks such as perceptrons, convolutional neural networks (CNN), recurrent neural networks (RNN), residual networks (ResNet), generative adversarial networks (GAN), etc., deep learning (DNN), and ensemble analysis consisting of a combination of these.

10. 10. The apparatus for weighing a specific portion of a living organism according to claim 9, wherein the transport unit has a sorting function for sorting the living organism based on the weight of the specific portion calculated by the weight calculation unit.

11. The device for weighing a specific part of a living body as described in claim 9, further comprising a total weight measuring means for measuring the total weight of the living body to be weighed, and the control unit further comprising a weight ratio calculation unit for calculating the weight ratio of the specific part calculated by the weight calculation unit to the total weight of the living body to be weighed measured by the total weight measuring means.

12. 12. A measuring device for a specific portion of a living body as described in claim 11, comprising: a base; a transport unit mounted on the base and transporting the living body to be measured one by one at a predetermined interval; a light detection unit having the light source unit and a light receiving unit, which irradiates the living body to be measured transported by the transport unit with light in the near-infrared region from the light source unit and receives diffuse reflected light or diffuse transmitted light from the living body to be weighed by the light receiving unit; and a receiving unit mounted downstream of the light detection unit in the transport unit and receives the living body to be measured, wherein the receiving unit is configured to have a plurality of divided areas corresponding to weight ratio divisions set by dividing the weight ratio of the specific portion to the total weight of the living body into a plurality of predetermined ranges, and the transport unit is provided with a transfer means for transferring the living body to be weighed to the corresponding divided area according to the weight ratio calculated by the weight ratio calculation unit of the control unit.

13. A device for measuring a specific part of a living body as described in any one of claims 8 to 12, characterized in that the living body has a shell covering the specific part, and the shell is formed so that the irradiated near-infrared light can pass through it, and / or is formed so that it has a transmissive portion through which the irradiated near-infrared light can pass.

14. The device for measuring a specific part of a living body as described in claim 13, characterized in that the shell is formed with a transmissive portion through which the irradiated near-infrared rays can pass, and the near-infrared rays related to the irradiation are irradiated toward the transmissive portion.

15. The device for measuring a specific part of a living organism as described in claim 14, characterized in that the living organism is a sea urinary tract having a shell with an oral opening as the transmissible part, with the oral opening exposed at the center of the lower part, and having gonads inside the shell, the specific part to be measured is the gonad, and the near-infrared light related to the irradiation is irradiated toward the oral opening as the transmissible part.

16. the living organism is a sea urinary urinary bladder having a shell with a mouth opening as the permeable portion through which the irradiated near-infrared light can pass, the shell having a mouth exposed at the center of the lower part, and having gonads inside the shell, the specific part to be measured being the gonads, The transport unit is configured as a conveyor having a plurality of trays arranged at predetermined intervals, each tray holding the living sea urchin with its mouth opening facing up, and the light detection unit is configured such that the light source unit is irradiated with near-infrared light related to the irradiation toward the mouth opening as the transmissive portion, and the light receiving unit is configured to receive reflected light from the mouth opening as the transmissive portion, A specific portion weighing device for a living organism as described in claim 13, characterized in that each of the division areas is constituted by a box installed below the tray, and the transfer means is configured to drop the living organism sea urchin from the tray into the corresponding box.

17. A trained model for causing a computer to function to calculate and output the weight of a specific part attached to a living body and separable from its surroundings, A trained model constructed using the machine learning method described in claim 9.

Images