JP2023146757A - Biological testing equipment and trained models - Google Patents

Abstract

To improve inspection efficiency and versatility by enabling non-destructive inspection of a living body of an object to be inspected.SOLUTION: An inspection device of a living body includes: a calculation model storage unit and a determination unit. The calculation model storage unit stores a calculation model that is constructed based on a quadratic differential spectrum in absorbance about a wavelength of a near infrared region of diffuse reflected light or diffuse transmitted light obtained by irradiating a sample living body infected with a parasite and a sample living body not infected with a parasite with light of a near infrared region, and performs calculation related to presence or absence of infection with a parasite of a living body and/or a numerical damage level from no onset to maximum onset due to infection. The determination unit determines presence or absence of infection and/or a degree of damage by calculation results calculated from absorbance of light received by a light receiving unit and the calculation model stored in the calculation model storage unit.SELECTED DRAWING: Figure 9

Description

TECHNICAL FIELD The present invention relates to a living body testing device that tests living organisms such as plants and animals for infection with parasites, and a trained model that can be used in this device.

In general, living organisms consisting of, for example, edible plants, ornamental plants, or edible animals (fish and meat) may be infected with parasites and cannot be used for that purpose. Conduct tests related to infection. As an example of an apparatus for inspecting a living body, for example, in the case of a nematode as a parasite, the following example of garlic as a living body can be given.

In recent years, garlic as a living organism has become infected with a parasitic nematode, Imogusare nematode, and various countermeasures have been taken. In this case, it is essential to first test the harvested garlic for infection with Imogusare nematode. Conventionally, the method described in Patent Document 1 (Japanese Unexamined Patent Publication No. 2017-219423), for example, is known as a method for inspecting this nematode. This is done by injecting and immunizing domestic rabbits with the internal fluid obtained by crushing and centrifuging the nematode nematode, obtaining antiserum, and removing reactants that show non-specific reactions from this antibody serum. Obtain serum. Then, the garlic to be tested is crushed and brought into contact with the anti-nematode serum for antigen-antibody reaction, and then the antibodies in the anti-nematode serum are labeled. If a reaction complex is formed, it is detected. As a labeling treatment for detecting a reaction complex, a conventionally known treatment method using an enzyme, a fluorescent substance, etc. is used.

In addition, as a method for extracting nematodes, for example, the Bellman method is known, in which the scales of infected garlic whose surface has been sterilized are finely chopped and treated with chemicals on a funnel to obtain a nematode suspension (see the above patent document). 1 (Japanese Patent Application Publication No. 2017-219423), Patent Document 2 (Japanese Patent Application Publication No. 2010-100568), etc.).

JP 2017-219423 Publication Japanese Patent Application Publication No. 2010-100568

However, in this conventional living body testing method, each living body must be crushed and treated with anti-nematode serum and other necessary chemicals, which is time-consuming and makes the testing work extremely complicated. Therefore, in the case of products such as garlic that are shipped as scales, there is a problem in that it is not possible to inspect all products.

The present invention has been made in view of these problems, and aims to improve testing efficiency by making it possible to non-destructively test living organisms even if they are infected with parasites. The purpose of the present invention is to provide a biological testing device and a trained model with improved versatility.

In order to achieve such an object, the living body testing device of the present invention is a living body testing device that performs a test for infection of a living body as a subject with parasites, and includes:
A light source section that irradiates light in the near-infrared region to the test object, a light receiving section that receives diffusely reflected light or diffusely transmitted light from the test object, and a light receiving section that receives light based on the absorbance of the light received by the light receiving section. and a control unit for determining infection with the parasite to be tested,
The above-mentioned control unit is configured to generate diffuse reflected light or diffused light obtained from the sample living body by irradiating the sample living body infected with parasites and the sample living body not infected with parasites with light in the near-infrared region in advance. Constructed based on the second-order differential spectrum of absorbance for wavelengths in the near-infrared region of transmitted light, the presence or absence of infection with parasites in living organisms and/or the degree of onset from a predetermined maximum of no onset to onset due to infection a calculation model storage unit that stores a calculation model for performing calculations related to the degree of damage quantified;
The present invention is configured to include a determining unit that determines the presence or absence of infection and/or the degree of damage based on the calculation result calculated from the absorbance of the light received by the light receiving unit and the calculation model stored in the calculation model storage unit.

Here, the living body consists of a plant or animal infected with a parasite, and may be in a living state or a dead state. Examples include edible plants such as vegetables and fruits, ornamental plants, fish, and meat. Examples of the parasite include protozoa, fleshy flagellates, sporozoites, ciliates, helminths, nematodes (nematodes), cestodes, trematodes, and talcocephalians.

In addition, infection refers to a state in which parasites are present, and onset refers to the appearance of some symptoms due to parasite infestation, such as changes in the tissues of the living body and the appearance of lesions in the living body. . The degree of damage can be determined not only on the surface but also on internal pathologies. Since it is possible to identify not only the surface but also the inside, it is possible to detect the very early stages of disease, making it more suitable for diagnosing early stages of disease than visual inspection.

As a result, according to the present inspection device, the light source unit irradiates near-infrared rays onto the living body to be tested, the light receiving unit receives diffuse reflected light or diffusely transmitted light from the living body to be tested, and the control unit: Based on the calculation result calculated from the absorbance of the received light and the calculation model, it is determined whether the living body is infected by parasites and/or the degree of damage. Therefore, it is possible to non-destructively test the living body to be examined, and it is possible to improve the test efficiency and the versatility. In addition, since a computational model is used, even if the parasite has invaded the internal body or the infection is slight, it is possible to reliably determine whether the living body is infected by the parasite and determine the degree of damage caused by the disease. This can improve inspection accuracy. Furthermore, when determining the degree of damage, the degree of spread of disease caused by parasites can be known, making it easy to evaluate the quality of living organisms. Furthermore, since the degree of damage is a value that quantifies the degree of onset, it is possible to reliably grasp the degree of spread of onset. Therefore, it is possible to sort the quality of living organisms, and the versatility can be further improved.

Then, if necessary, the above-mentioned degree of damage is also ranked for subjects who are not infected. In this way, by determining the degree of damage, it is also possible to determine whether there is an infection or not, resulting in extremely high efficiency.

In this case, it is effective to determine the degree of damage based on the degree of discoloration appearing on the subject. Although it depends on the parasite, it is suitable for living organisms where lesions appear on the surface of the living body when infected with the parasite.

In this configuration, the degree of damage can be configured as an index with 0 indicating no onset and 100 representing the maximum occurrence of onset.
In addition, in this configuration, the degree of damage is expressed as an index in which the degree of onset is set to 0 for no onset and 100 for the maximum of onset, where no onset is 0 and the side with onset is a predetermined index range. It can be composed of ranks that increase in order. When configured in terms of classes, it is possible to easily classify the quality of living organisms, and the versatility can be further improved.

Also, if necessary, the test subjects are separated according to the presence or absence of infection, and/or the test subjects are separated according to the degree of damage, based on the discrimination result of the discrimination section of the control section. The structure is equipped with a separation means. When classifying test subjects according to the presence or absence of infection, it is possible to divide the test subjects into two, those without infection and those with infection.In other words, the good and bad can be automatically sorted, which is extremely effective. It becomes convenient. Furthermore, when classifying the test objects according to the degree of damage, the test objects can be automatically separated in stages, which is extremely convenient.

Then, as necessary, the calculation model consists of 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, logistic regression analysis, support vector machine (SVM),
The above machine learning includes tree analysis such as decision tree (CART), regression tree, random forest, gradient boosting tree, perceptron, convolutional neural network (CNN), recurrent neural network (RNN), residual network (ResNet), etc. The configuration is selected from neural networks, deep learning (DNN), and ensemble analysis consisting of a combination of these.

Since the calculation model is composed of an analytical model constructed using statistical analysis or a learned model constructed using machine learning, inspection accuracy can be improved. In particular, when deep learning (DNN) is used, inspection accuracy can be significantly improved. As for the wavelength in the calculation model, statistically superior wavelengths can be selected from wavelengths in the near-infrared region in the range of 700 nm to 2500 nm and used in combination. Thereby, it is possible to accurately determine whether the living body is infected with a parasite and the degree of the disease caused by the infection.

Further, if necessary, the arithmetic processing unit may include a plurality of different types of arithmetic models as the arithmetic models, and may include a determining unit that determines whether the judgment result of each arithmetic model is acceptable. Optimal results can now be obtained from multiple calculation models, and inspection accuracy can be improved.

In this case, when a plurality of calculation models of different types for determining the presence or absence of infection are used as the calculation model, any one of the determination results using each calculation model determines that there is infection. In this case, it is effective to have a configuration that gives priority to the determination result. This makes it possible to accurately determine whether a living organism is infected with a parasite.

Alternatively, when three or more calculation models of different types for determining the presence or absence of infection are used as the calculation models, the determination unit is configured to determine the determination result of the presence or absence of infection of each calculation model by majority vote. This is effective. This also makes it possible to accurately determine whether the living body is infected with parasites.

The parasite is a nematode (nematode) that infects living plants. The living body may be any plant that can infect nematodes, such as garlic, potato, sweet potato, or iris, which are classified as plants.
As nematodes, there are Japanese nematode, Negus nematode (Sweet potato nematode, Javanese cat nematode, Arenalia cat nematode, Northern cat nematode), Negusare nematode (Minami Negusare nematode, Channel nematode, Northern Negusare nematode, Walnut Negusare nematode), Hagare nematode ( Examples include the nematode, the nematode, the nematode, the nematode, the nematode, the cyst nematode, the nematode, the nematode, the nematode, the nematode, the nematode, the nematode, and the nematode.

In this case, the living organism is selected from garlic, potato, sweet potato, and iris, and the parasite is a nematode.

Furthermore, in this case, the object to be tested is composed of garlic bulbs in which scale pieces are gathered. Inspection efficiency is high because each garlic bulb is inspected. In addition, since garlic is mainly shipped in the form of garlic bulbs, inspection efficiency is high in this respect as well.

If the subject to be tested is made up of garlic bulbs with aggregated scales, the above-mentioned degree of damage is determined as necessary based on the degree of discoloration that appears on the subject. It is composed of the damage level shown below, which is an index with the maximum value being 100.

Damage degree (index) = ((number of scales at t = 1) + (number of scales at t = 2 x 2) + (number of scales at t = 3 x 3) + (number of scales at t = 4 x 4))/(Total number of surveyed scales x 4) x 100
here,
t is a numerical value corresponding to the size of the discolored area due to parasite infection per garlic bulb, and is determined so that the larger the value, the larger the discolored area, t = 1 , t=2, t=3, and t=4.

In addition, if the subject to be tested is composed of garlic bulbs with aggregated scales, the degree of damage described above may be determined based on the degree of discoloration that appears on the subject, and the degree of onset may be determined based on the degree of discoloration that appears on the subject. When expressed as an index with no onset as 0 and the maximum for onset as 100, it is expressed by the degree of onset as shown below, with no onset as 0, and the side with onset as increasing in order for each predetermined index range. ing.

Degree of onset (index) = ((number of scales at t = 1) + (number of scales at t = 2 x 2) + (number of scales at t = 3 x 3) + (number of scales at t = 4 ×4))/(Total number of surveyed scales ×4) ×100
here,
t is a numerical value corresponding to the size of the discolored area due to parasite infection per garlic bulb, and is determined so that the larger the value, the larger the discolored area, t = 1 , t=2, t=3, and t=4.

Furthermore, if necessary, a base, a transport unit provided on the base and for sequentially transporting the test objects one by one at predetermined intervals, a robot installed near the base, and a robot arm of the robot. A grasping hand that grips the object to be inspected during the transportation process of the transport unit and positions the object to be inspected at the tip of the robot; an inspection section for inspecting an object to be inspected, which is provided near the gripping hand and positioned at an inspection position by the gripping hand; and an inspection section that is moved by a robot after inspection by the inspection section and is gripped by the gripping hand. and a receiving section for receiving the information. Since the inspection part and the gripping hand are provided on the robot arm, the inspection object can be inspected while being transported, and the subsequent sorting work can be carried out extremely efficiently.

In addition, if necessary, an inspection section that has the light source section and the light receiving section and inspects the object to be inspected is provided, and the inspection section is provided with a detection section that has the light source section and the light receiving section exposed on one end surface. The present invention includes an inspection unit, and the inspection unit is provided with a dustproof mechanism that injects air along the detection section to form an air curtain covering the detection section. Since the light source section and the light receiving section of the detection section can be kept clean at all times, inspections can be performed reliably and maintenance can be facilitated.

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 output a calculation result related to the presence or absence of infection with parasites in a living body,
The absorbance of the sample living body is measured in advance, and the presence or absence of parasites in the living body is then determined by another method such as a culture method. The absorbance data of the specimen and the absorbance data of a sample living body infected with parasites were made to correspond to the presence or absence of parasites, and this corresponding data was used as training data, and learning was performed using the training data. , irradiates near-infrared rays onto the living body to be tested, receives diffusely reflected light or diffusely transmitted light from the living body to be tested, measures the absorbance of the received light, and uses this absorbance data to determine if the living body is infected with parasites. The computer is configured to function so as to output a quantified value indicating whether or not the It has the same action and effect as above.

In addition, another trained model of the present invention allows a computer to output a calculation result related to the degree of damage that quantifies the degree of onset from a predetermined maximum of no onset to the maximum of onset due to infection with parasites of living organisms. A trained model for functioning,
The absorbance of the sample organism is measured in advance, and the degree of damage caused by the infection of the organism to the parasite is determined by another means such as visual inspection.The degree of damage to the organism that has been determined is used as an indicator to identify samples that have not developed symptoms. The absorbance data of the living body and the absorbance data of the sample living body that has developed the disease are respectively made to correspond to the degree of damage to the living body, the corresponding data is used as teacher data, and learning is performed using the teacher data, Near-infrared rays are irradiated onto the living body to be tested, diffuse reflected light or diffused transmitted light is received from the living body to be tested, and the absorbance of the received light is measured. From this absorbance data, it is possible to determine whether the living body is infected with parasites. The computer is configured to function in such a way that it outputs a value that quantifies the degree of damage caused by the disease. This also produces the same effects and effects as described above.

According to the present invention, a living body to be examined is irradiated with near-infrared rays, diffusely reflected light or diffusely transmitted light from the living body to be examined is received, the absorbance of the received light is measured, and this absorbance and a calculation model are used. The presence or absence of infection and/or the degree of damage caused by parasites in a living body is determined based on the calculation results calculated from the above, so that the living body to be tested can be tested non-destructively, thereby improving testing efficiency. In addition, it is possible to improve versatility. In addition, since a computational model is used, even if the parasite has invaded the internal body or the infection is slight, it is possible to reliably determine whether the living body is infected by the parasite and determine the degree of damage caused by the disease. This can improve inspection accuracy. Furthermore, when determining the degree of damage, the degree of spread of disease caused by parasites can be known, making it easy to evaluate the quality of living organisms. Furthermore, since the degree of damage is a value that quantifies the degree of onset, it is possible to reliably grasp the degree of spread of onset. Therefore, it is possible to sort the quality of living organisms, and the versatility can be further improved.

1 is a perspective view showing a living body inspection device according to an embodiment of the present invention. FIG. 2 is a diagram showing garlic, which is a living body targeted by the living body inspection device according to the embodiment of the present invention, together with a holding part. 1 is a front view showing a living body inspection device according to an embodiment of the present invention. 1 is a plan view showing a living body inspection apparatus according to an embodiment of the present invention. FIG. 1 is a side view showing a living body inspection device according to an embodiment of the present invention. FIG. 2 is a diagram showing the relationship between the grasping hand of the robot and the inspection section together with the operation steps of the grasping hand in the living body inspection apparatus according to the embodiment of the present invention. 1A and 1B illustrate an inspection unit in a living body inspection apparatus according to an embodiment of the present invention, FIG. 1A being a perspective view of the inspection unit, and FIG. FIG. 2 is a cross-sectional view showing the testing section of the living body testing device according to the embodiment of the present invention in a state where garlic as a living body is being tested. FIG. 2 is a block diagram showing the configuration of a control section in the biological testing apparatus according to the embodiment of the present invention. FIG. 2 is a diagram schematically showing a trained model according to an embodiment of the present invention used by a control unit in a living body inspection apparatus according to an embodiment of the present invention. It is a flowchart which shows the control flow in the control part of the biological examination apparatus based on embodiment of this invention. 2 is a flowchart showing a control flow of a sample measurement and discrimination routine in a control unit of a living body testing apparatus according to an embodiment of the present invention. 2 is a flowchart showing a control flow of a parasite discrimination routine, which is a control flow related to a control unit of a living body inspection apparatus according to an embodiment of the present invention. Regarding the calculation model (multiple regression analysis (determining the presence or absence of infection)) of Example 1 of the present invention, (a) is a graph showing the correlation between actual measured values and estimated values, and (b) is a graph showing the underlying numerical values. FIG. An example of the verification of the calculation model (multiple regression analysis (determination of the presence or absence of infection)) of Example 1 of the present invention is shown, (a) is a graph diagram showing the correlation between the actual measured value and the estimated value, and (b) is a graph diagram thereof. It is a table diagram showing the base numerical values. Regarding the calculation model (multiple regression analysis (damage level (index) determination)) of Example 2 of the present invention, (a) is a graph diagram showing the correlation between actual measured values and estimated values, and (b) is a graph showing the verification results. FIG. Regarding the calculation model (logistic regression model (determination of presence or absence of infection)) of Example 3 of the present invention, (a) is a table showing the correlation between actual measured values and estimated values, and (b) is a table showing the verification results. It is a diagram. Regarding the calculation model (logistic regression model (damage level (6 classes) discrimination) of Example 4 of the present invention, (a) is a table showing the correlation between actual measured values and estimated values, and (b) is the verification result. FIG. Regarding the calculation model of Example 5 of the present invention (AI trained model A (determination of presence/absence of infection)), (a) is a table showing the correlation between actual measured values and estimated values, and (b) is a distribution of estimated probabilities. FIG. It is a table which shows the verification result regarding the calculation model (AI learned model A (infection presence/absence determination)) of Example 5 of this invention. Regarding the calculation model of Example 6 of the present invention (AI trained model B (damage level (6 classes) discrimination)), (a) is a table showing the correlation between actual measured values and estimated values, and (b) is a table thereof. It is a table figure showing a verification result. Regarding the calculation model (AI trained model C (damage degree (index) determination)) of Example 7 of the present invention, (a) is a graph showing the correlation between actual measured values and estimated values, and (b) is its verification It is a graph diagram showing the results. Regarding the calculation model of Example 8 of the present invention (AI-trained model D (determination of presence/absence of infection)), (a) is a table showing the correlation between actual measured values and estimated values, and (b) is a table showing the verification results. FIG. Regarding the calculation model of Example 9 of the present invention (AI trained model E (damage level (6 classes) discrimination)), (a) is a table showing the correlation between actual measured values and estimated values, and (b) is a table showing the correlation between actual measured values and estimated values. It is a table figure showing a verification result. Regarding the calculation model (AI trained model F (damage level (index) discrimination)) of Example 10 of the present invention, (a) is a graph diagram showing the correlation between actual measured values and estimated values, and (b) is its verification It is a graph diagram showing the results. Regarding the calculation model of Example 11 of the present invention (AI trained model G (determination of presence/absence of infection)), (a) is a table showing the correlation between actual measured values and estimated values, and (b) is a table showing the verification results. FIG. Regarding the calculation model of Example 12 of the present invention (AI trained model H (damage degree (6 class) discrimination)), (a) is a table showing the correlation between the actual measured value and the estimated value, and (b) is its It is a table figure showing a verification result. Regarding the calculation model (AI trained model I (damage degree (index) determination)) of Example 13 of the present invention, (a) is a graph diagram showing the correlation between actual measured values and estimated values, (b) is its verification It is a graph diagram showing the results.

DESCRIPTION OF THE PREFERRED EMBODIMENTS A living body inspection apparatus according to an embodiment of the present invention will be described below based on the accompanying drawings. As shown in FIGS. 1 to 9, a living body inspection apparatus S according to an embodiment of the present invention is used to test a living body, which is a subject W, for infection with parasites, and as shown in FIG. In this case, the living organism is harvested garlic, and the parasite is a caterpillar nematode. The test object W is composed of a garlic bulb in which scale pieces Wa are gathered. The garlic bulb (test object W) has a flower diameter Wb projecting several centimeters upwards, a disc Wc below, is covered with an outer skin, and has been washed and dried. Infection by the parasite occurs from the disc Wc side. F indicates a lesion infected by the parasite (indicated by a dotted line). Generally, the lesion F is hidden by the outer skin and is invisible or difficult to see. When the onset is severe, the outer skin is affected and it becomes noticeable on the surface.

The living body inspection device S according to the embodiment includes a base 1, a transport unit 2 provided on the base 1 and configured to sequentially transport test objects W one by one at predetermined intervals, and a transport unit 2 installed near the base 1. A robot 10 , a gripping hand 12 at the tip of a robot arm 11 of the robot 10 that grips the object W to be inspected during the transportation process of the transfer unit 2 and positions the object W to be inspected at an inspection position Y, and a robot arm of the robot 10 . 11, which is provided near the gripping hand 12 and which inspects the object W to be examined which is positioned at the inspection position Y by the gripping hand 12; The receiving unit 50 receives the subject W held by the holding hand 12, and the control unit 60 realizes an inspection function and controls the transport unit 2, the robot 10, the inspection unit 30, etc. ing. In the figure, 70 is a display unit made of a CRT or the like.

In the robot 10, as shown in FIG. 6, the gripping hand 12 is provided at the tip of the robot arm 11 via an opening/closing actuator 13 that opens and closes it, and the opening/closing actuator 13 is swingable. By swinging the opening/closing actuator 13, the gripping hand 12 is moved to two positions: a gripping position X where the object to be inspected W on the transport section 2 can be gripped, and an inspection position Y where the inspection section 30 performs the inspection. .

The conveyance unit 2 includes an endless belt conveyor 3 driven by a motor (not shown), and a plurality of holding units 20 that are provided at predetermined intervals on the belt conveyor 3 along the moving direction and hold the test object W. Configured with the necessary features. As shown in FIG. 2, the holding section 20 supports and holds the garlic bulb to be tested W with its stem Wc facing upward, and has a cylindrical side wall 21 whose lower surface is fixed to the surface of the belt conveyor 3. and a cup-shaped holding body 23 having an upper wall 22; an insertion hole 24 formed in the upper wall 22 of the holding body 23 and having the axis of the holding body 23 as the center and through which the flower diameter Wb of the garlic bulb is inserted; 24 (in the embodiment, the entire insertion hole 24), and a conical support recess 25 that supports the periphery of the garlic bulb on the flower diameter Wb side. Thereby, the disk Wc side of the garlic bulb of the subject W held by the gripping hand 12 faces the inspection section 30.

The base 1 is provided with a position detection sensor (not shown) that detects when the test object W transported by the belt conveyor 3 of the transport unit 2 is located at a predetermined position to be gripped by the gripping hand 12 of the robot 10. Based on the detection by the position detection sensor, the object W to be inspected is gripped and taken out by the gripping hand 12 of the robot 10. A camera 26 is provided above the belt conveyor 3 of the conveyance unit 2 to take an image of the subject W being conveyed. The size of the object W to be examined can be known by capturing the image with the camera 26.

As shown in FIGS. 6 to 8, the inspection section 30 includes a cylindrical inspection unit 31, a support rod 32 that supports the inspection unit 31 at the tip of the robot arm 11 and near the gripping hand 12, and an inspection unit 31. A cover 33 is provided on the distal end side of the unit 31 and covers the subject W positioned at the inspection position Y. Since the inspection unit 30 and the gripping hand 12 are provided on the robot arm 11, the inspection object W can be inspected while being transported, and the subsequent sorting operation can be carried out extremely efficiently.

As shown in FIGS. 7 and 8, the inspection unit 31 includes a cylindrical unit main body 34 and a light source section 35 that includes a plurality of optical fiber tips and irradiates the test object W with light in the near-infrared region. , and a detection section 37 that includes a light receiving section 36 that is configured of the tip of an optical fiber and receives diffusely reflected light or diffusely transmitted light from the object W to be inspected. The optical fibers are collected and held in a solid cylindrical holder 38, and the end face of each optical fiber is exposed to one end face of the holder 38 on the side of the object W to be tested, and this one end face is detected. Section 37. The holder 38 is held by a solid cylindrical unit body 34 that is coaxial with the axis of the holder 38 . The detection section 37, which is one end surface of the holder 38, is located rearward of the one end surface of the unit main body 34 on the side of the object W to be tested, and a recess 39 is formed in the front side of the detection section 37.

The inspection unit 31 also includes a dustproof mechanism 40 that protects the detection section 37 from dust. The dustproof mechanism 40 includes a protrusion 41 that protrudes from a part of the outer circumferential side of the end surface of the unit body 34, and injects air into the protrusion 41 along one end surface of the unit body 34 to form a recess. An air injection port 42 is formed to form an air curtain covering the area 39. An air passage 43 is formed in the unit body 34 and is parallel to the axis of the unit body 34 and guides air to the air injection port 42 . 44 is an air supply pipe connected to the air passage 43 at the other end of the unit main body 34. Since the light source section 35 and the light receiving section 36 of the detection section 37 can be kept clean at all times, inspections can be performed reliably and maintenance can be facilitated.

The receiving unit 50 includes six shooters 51 for receiving the test object W. Shooters 51 are provided corresponding to determined damage degrees (0 to 5), which will be described later, and the robot 10 receives a command from the control unit 60 and after the inspection, the robot 10 picks up the object W held by the gripping hand 12. is thrown into the corresponding chute 51. That is, the present apparatus includes a sorting means 52 that sorts the test object W according to the presence or absence of parasite infection, and in the embodiment, according to the degree of damage. It is composed of a robot 10 that is controlled by the robot 10 and a receiving section 50, and the separation is performed by these functions.

As shown in FIG. 9, the control section 60 includes an arithmetic processing section 61 that performs comprehensive control. The arithmetic processing unit 61 preliminarily irradiates a sample living body infected with parasites and a sample living body not infected with parasites with light in the near-infrared region to generate diffuse reflected light or light obtained from the sample living body. It is constructed based on the second-order differential spectrum of the absorbance of diffusely transmitted light at wavelengths in the near-infrared region, and is used to determine the presence or absence of infection with parasites in living organisms and/or the onset of infection from a predetermined maximum range of no onset to onset of infection. It is provided with a calculation model storage unit 62 that stores a calculation model for performing calculations related to the degree of damage that is expressed as a numerical value.

Further, the arithmetic processing unit 61 is a discriminating unit that determines the presence or absence of infection and/or the degree of damage based on the arithmetic results calculated from the absorbance of the light received by the light receiving unit 36 and the arithmetic model stored in the arithmetic model storage unit 62. It is equipped with 63. That is, the determining unit 63 has an infection determining function that determines whether or not the living body is infected with a parasite based on the calculation result calculated from the absorbance of the light received by the light receiving unit 36 and the calculation model. Further, the determination unit 63 determines the degree of onset from a predetermined maximum of no onset to onset due to infection with parasites of the living body, based on the calculation result calculated from the absorbance of the light received by the light receiving unit 36 and the calculation model. It is equipped with a damage level determination function that determines the degree of damage expressed as a numerical value.

In the embodiment, the degree of damage is determined based on the degree of discoloration appearing on the subject W, as shown in FIG. In terms of damage level, it is also possible to rank test subjects W without infection. In the embodiment, no infection is included in no onset of symptoms. If you want to closely examine the degree of damage and the presence or absence of infection, you can also use each trained model. Two types of damage levels are used. On the other hand, the degree of damage is composed of an index with 0 indicating no symptoms and 100 representing the maximum number of symptoms occurring. The details are determined according to the degree of damage shown below.

Damage degree (index) = ((number of scales at t = 1) + (number of scales at t = 2 x 2) + (number of scales at t = 3 x 3) + (number of scales at t = 4 x 4))/(Total number of surveyed scales x 4) x 100
here,
t is a numerical value corresponding to the size of the discolored area due to parasite infection per garlic bulb, and is determined so that the larger the value, the larger the discolored area, t = 1 , t=2, t=3, and t=4.

The degree of damage on the other hand is expressed as an index with no onset as 0 and the maximum of onset as 100, with no onset as 0 and the side with onset increasing in order for each predetermined index range. It is made up of classes. In the embodiment, the degree of damage is set to the following six classes.

Damage degree 0 = degree of onset 0
Degree of onset 0<Damage level 1≦Degree of onset 20
Degree of onset 20 <Damage level 2 ≦ Degree of onset 40
Degree of onset: 40 <Damage level: 3 ≦ Degree of onset: 60
Degree of onset: 60 <Damage level: 4 ≦ Degree of onset: 80
Degree of onset: 80 <Damage level: 5 ≦ Degree of onset: 100

The degree of onset is determined as follows.
Degree of onset (index) = ((number of scales at t = 1) + (number of scales at t = 2 x 2) + (number of scales at t = 3 x 3) + (number of scales at t = 4 ×4))/(Total number of surveyed scales ×4) ×100
here,
t is a numerical value corresponding to the size of the discolored area due to parasite infection per garlic bulb, and is determined so that the larger the value, the larger the discolored area, t = 1 , t=2, t=3, and t=4.

The computational model consists of an analytical model constructed using statistical analysis or a trained model constructed using machine learning,
The statistical analysis was selected from discriminant analysis, multiple regression analysis, PLS regression analysis, logistic regression analysis, support vector machine (SVM),
Machine learning includes tree analysis such as decision trees (CART), regression trees, random forests, gradient boosting trees, etc., perceptrons, convolutional neural networks (CNN), recurrent neural networks (RNN), residual networks (ResNet), etc. The method is selected from neural networks, deep learning (DNN), and ensemble analysis consisting of a combination thereof.

Specifically, in multiple regression analysis, PLS regression analysis, and logistic regression analysis, near-infrared rays are irradiated to sample organisms infected with parasites and sample organisms not infected with parasites, and An analytical model (calculation model) related to the attributable wavelength caused by the above-mentioned parasite by receiving diffusely reflected light or diffusely transmitted light, measuring the absorbance of the received light, and statistically analyzing the second derivative spectrum of the absorbance. , irradiate the object W to be tested with near-infrared rays, receive the diffusely reflected light or diffusely transmitted light from the living body of the test object W, measure the absorbance of the received light, and measure these absorbances. Based on the calculation results calculated from the above analysis model and the above analysis model, the degree of damage is determined by quantifying the presence or absence of infection of the organism with the parasite and/or the degree of onset from a predetermined maximum level of no onset due to infection to the maximum of onset.

<Multiple regression analysis>
The analytical model is specified by multiple regression analysis by second-order differentiation of the absorbance spectrum of light received from the sample living body after MSC processing. 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 (the same applies hereinafter).

The above analysis model was specified using a formula (discriminant formula) that satisfies the relationship of the following general formula (B) in which the absorbances of the first to nth wavelengths having high correlation coefficients are used as variables.

In general formula (B), when selecting the first wavelength (λ1) to the nth wavelength (λn), first measure the absorbance of the sample living body in advance to determine whether the living body is infected with parasites. This is then carried out using another method such as a culturing method using Bellman's method, and the absorbance of the sample living body infected with the parasite and the parasite Select the near-infrared wavelength range of the first wavelength (λ1) that has the highest correlation coefficient attributed to the parasite, which was determined by multiple regression analysis with the absorbance of a sample living body that is not infected with the parasite, and then By multiple regression analysis between the near-infrared wavelength range of the first wavelength (λ1) and the wavelength range of 700 nm to 2500 nm, the correlation coefficient of the near-infrared wavelength range of the first wavelength (λ1) or more is determined to be attributable to the parasite. Select the wavelength range of the second wavelength (λ2) that has a high correlation coefficient of According to multiple regression analysis, the third wavelength (λ3) which is attributed to the parasite and has a correlation coefficient higher than the correlation coefficient of the near-infrared wavelength range of the first wavelength (λ1) and the second wavelength (λ2) By selecting a wavelength range and performing a multiple regression analysis of the near-infrared wavelength range of the first wavelength (λ1) to the (n-1)th wavelength (λn-1) and the wavelength range of 700 nm to 2500 nm, the above The wavelength range of the n-th wavelength (λn) that belongs to the parasite and has a correlation coefficient greater than or equal to the near-infrared wavelength range of the first wavelength (λ1) is selected.

<PLS regression analysis>
The analytical model was identified by calculating the absorbance spectrum of the light received from the sample object using a PLS (Partial least square projection to latent structure) analysis method by performing second-order differentiation after MSC processing.

<Logistic regression analysis method>
The trained model was identified by calculating the absorbance spectrum of the light received from the sample object using the MSC process, second-order differentiation, and calculating the result using a logistic regression analysis method.

<Support Vector Machine (SVM)>
The trained model was identified by the SVM (Saport Vector Machine) method by second-order differentiation of the absorbance spectrum of light received from the sample object after MSC processing.

Next, we will give an example of deep learning method.
<Deep learning method (1)>
For example, as the trained model, a trained model based on a so-called ensemble method using a decision tree of a deep learning method and a random forest method is used, which is obtained by second-order differentiation of the absorbance spectrum of light received from the sample living body after MSC processing. Create a statistically superior trained model using the ensemble method. The combinations are not limited to these, and may be selected and combined as appropriate. These are the types of AI trained models A to C described later.

<Deep learning method (2)>
As the trained model, a trained model based on the DNN (Deep Neural Network) method, which is a deep learning method, is used, in which the absorbance spectrum of light received from the sample living body is subjected to MSC processing and second-order differentiated. These are the types of AI trained models D to F, which will be described later.

<Deep learning method (3)>
As the trained model, the absorbance spectrum of the light received from the sample living body is subjected to second order differentiation after MSC processing, and the waveform is imaged, and a trained model based on a deep learning CNN (Convolutional Neural Network) method is used. These are the types of AI trained models G to I, which will be described later.

In the embodiment, the trained model regarding the presence or absence of infection is a trained model for causing a computer to function to output a calculation result regarding the presence or absence of infection of a living organism (garlic) with a parasite (Imogusare nematode),
The absorbance of the sample living body is measured in advance, and the presence or absence of parasites in the living body is then determined by another method such as a culture method. The absorbance data of the specimen and the absorbance data of a sample living body infected with parasites were made to correspond to the presence or absence of parasites, and this corresponding data was used as training data, and learning was performed using the training data. , irradiates near-infrared rays onto the living body to be tested, receives diffusely reflected light or diffusely transmitted light from the living body to be tested, measures the absorbance of the received light, and uses this absorbance data to determine if the living body is infected with parasites. The computer functions to output a value that quantifies whether or not it is true.

In addition, the trained model regarding the degree of damage related to the onset of disease is based on the degree of damage that quantifies the degree of onset from a predetermined maximum of no onset to the maximum of onset due to infection of a living organism (garlic) with a parasite (Imogusare nematode). A trained model for making a computer function to output calculation results,
The absorbance of the sample organism is measured in advance, and the degree of damage caused by the infection of the organism to the parasite is determined by another means such as visual inspection.The degree of damage to the organism that has been determined is used as an indicator to identify samples that have not developed symptoms. The absorbance data of the living body and the absorbance data of the sample living body that has developed the disease are respectively made to correspond to the degree of damage to the living body, the corresponding data is used as teacher data, and learning is performed using the teacher data, Near-infrared rays are irradiated onto the living body to be tested, diffuse reflected light or diffused transmitted light is received from the living body to be tested, and the absorbance of the received light is measured. From this absorbance data, it is possible to determine whether the living body is infected with parasites. The computer is operated to output a value that quantifies the degree of damage caused by the disease. FIG. 10 shows a configuration example of a learned model. In the embodiment, for example, 550 pieces of absorbance data at wavelength intervals of 2 nm are input to the input layer.

In the embodiment, a plurality of different types of calculation models are provided as calculation models, and these can be selectively used. In Figures 9 and 13, types A to F are shown as multiple regression analysis models, logistic regression analysis models, support vector machines (SVM), and AI trained models, and types G to I are shown as other AI trained models. The case is shown below. The specific contents of these will be shown in the examples described later.

The calculation processing unit 61 includes a plurality of calculation models of different types, and includes a determination unit 64 that determines whether or not the determination result of each calculation model is acceptable. When selecting a plurality of calculation models with different types for determining the presence or absence of infection, and when any one of the determination results by the calculation models determines that there is infection, the determination unit 64 prioritizes and adopts the determination result. It can be configured to do this. Further, the determining unit 64 may be configured to use three or more calculation models of different types for determining the presence or absence of infection, and to determine and adopt the determination result of the presence or absence of infection of each calculation model by majority vote. Furthermore, when determining the degree of damage, any one of the calculation models related to the degree of damage can be selected and used. In this case, it is possible to perform calculations on all calculation models and use the results selectively. The determining unit 64 employs the calculation result of the selected calculation model. Note that the display unit 70 can display the calculation results of all calculation models. In the embodiment, six classes of damage levels are output as the final result.

Further, as shown in FIG. 9, in the control section 60, the arithmetic processing section 61 has an operation control section 65. This includes a processing history storage function, a result display function on the display unit 70, a robot 10 operation function, an Internet communication function, and a light amount control function. Further, the control unit 60 controls a communication unit 66 that is controlled by an Internet communication function, an inspection control unit 67 that has a spectroscopic unit and performs control related to the inspection unit 30 such as controlling its spectroscopic control circuit, and a transport unit 2. It is configured to include a drive control section 68 including a transport section control circuit and a robot control circuit for controlling the robot 10, and an operation command section 69 such as a keyboard (not shown) and a setting panel (not shown).

The processing history storage function in the operation control unit 65 accumulates and stores the cumulative number of measurements of the test object W, the discrimination results, and the cumulative results in chronological order. The Internet communication function is a function for understanding the discrimination status and system status when used in a remote location, and for performing maintenance such as updating the version of the application (discrimination method) remotely. . By using this Internet communication function, the test object W can be tested regardless of the location, so the versatility of the device can be improved. The light amount control function adjusts the light amount of the light source and the sensitivity of the sensor in the spectroscopic section so that the photodetection sensor in the spectroscopic section can perform measurement at an appropriate level.

Various data are displayed on the display section 70. The display on the display section 70 is operated by a screen operation section (operation command section 69), and can be switched to display an input setting screen, a discrimination result display screen, an error display screen, etc. as appropriate. Note that the determination results may be displayed on an LCD panel. Further, the determination result may be outputted as voice. Furthermore, a data output interface to the outside may be provided.

In the embodiment, when the determining unit 63 determines the degree of damage, the arithmetic processing unit 61 causes the sorting means 52 to classify the subject W according to the degree of damage. That is, when the discrimination section 63 discriminates the degree of damage, the robot operation function of the operation control section 65 and the robot control circuit of the drive control section 68 are used to sort the object W to be inspected according to the determined degree of damage. The robot 10 is operated to cause the object W to be inspected held by the holding hand 12 to be thrown into the corresponding chute 51. In addition, the control unit 60 performs various controls.

Therefore, when testing the subject W using the biological testing apparatus S according to this embodiment, the procedure is as follows. This will be explained using the flowcharts shown in FIGS. 11 to 13.
As shown in FIG. 11, when the main power supply switch (not shown) is turned on, the control section 60 is activated and checks each mechanism such as the spectroscopic section, sensor, lamp, etc. (1-1). At this time, a screen indicating that the system is being initialized is displayed on the display unit 70, and the indicator light turns on in yellow (1-2). After the check operation of each mechanism is completed, a reference waveform serving as a standard for absorbance calculation is measured (1-3), and this is stored in the storage section inside the control section 60 (1-4). After that, the main screen is displayed and the indicator light turns green (1-8). During this time, if there is a setting value recall operation (1-5) from the setting panel (not shown), a confirmation screen for the cumulative lighting time of the lamp, date and time, etc. is displayed (1-6). To perform an inspection, operate the setting panel and return to the main screen (1-7).

A garlic bulb as a test object W is placed on the holding part 20 with its disk Wc facing upward. When the start switch of the control button is pressed when the indicator light on the main screen is green (1-9), a routine for measuring and determining the object W to be tested is executed (1-10). In this routine for measuring and identifying the object W to be inspected, as shown in FIG. do. During this transportation process, when the position detection sensor detects that the subject W is located at a predetermined position to be grasped by the grasping hand 12 of the robot 10 (2-2), an image is taken by the camera (2-3). At the same time, after that, the object W to be inspected is grasped and taken out by the grasping hand 12 of the robot 10 at a predetermined position, the object W to be inspected is positioned at the inspection position Y, and the inspection section 30 measures the object W to be inspected. . In the inspection section 30, light is irradiated onto the test object W from the light source section 35, and its diffusely reflected light or diffusely transmitted light is received by the light receiving section 36 and transmitted to the spectroscopic section of the inspection control section 67 via an optical fiber. Then, the intensity of the near-infrared wavelength is converted into a voltage for each wavelength range in the spectroscopy section, extracted, and transmitted to the arithmetic processing section 61 (2-4). Then, the intensity spectrum of the diffusely reflected light or the diffusely transmitted light is obtained, and discrimination is performed using a separate routine for parasite discrimination (2-5).

In the discrimination routine, as shown in Fig. 13, noise removal processing is performed on the diffused light intensity (diffuse light spectrum) transmitted from the spectrometer (3-1, 3-2), and the reference waveform stored at the time of initialization is used. Read (3-3), and calculate near-infrared absorbance of the measurement target W (3-4). Standardization processing is performed on the absorbance (3-5), second-order differential processing is further performed (3-6), and centering and scaling processing is performed on the second-order differential spectrum (3-7). Using this spectrum, discrimination is performed using a calculation model stored in advance in the calculation model storage unit 61. Discrimination is performed using multiple regression analysis (3-8), logistic regression analysis (3-9), calculations of types A to F of AI trained models (1) and (2) (3-10), and AI trained models (3-10). 3) G to I type operations (3-11) and support vector machine (SVM) operations (3-12).

When a plurality of calculation models with different types for determining the presence or absence of infection are selected as calculation models, when any one of the determination results by each calculation model determines that infection is present, the judgment unit 64 determines that the corresponding The discrimination results are adopted with priority (3-13), and the final discrimination results are calculated (3-14). Alternatively, when three or more calculation models of different types for determining the presence or absence of infection are used as calculation models, the determination unit 64 determines and adopts the infection determination result of each calculation model by majority vote. (3-13), and calculate the final discrimination result (3-14). In the embodiment, a computation model for determining the degree of damage is employed as the computation model, and the determination unit 64 employs the computation result of the selected computation model (3-13) and calculates the final determination result. (3-14). In the embodiment, six classes of damage levels are output as the final result. Further, all judgment results can be viewed on the display section 70.

When the measurement of the test object W is completed, returning to FIG. 12, the robot 10 is operated based on the result of the degree of damage, and the test object W is put into the corresponding chute 51 of the receiving section 50 and separated (2 -6). Also, returning to FIG. 11, the discrimination result and the cumulative total of the discrimination results thus far are displayed (1-13). During this time, if an error occurs in each mechanism or the emergency stop button is pressed (1-11), an error screen will be displayed and the indicator light will turn on in red (1-12). To recover, check the mechanisms of the lamp, spectrometer, sensor, etc. The determination results and the diffused light spectrum at the time of measurement of the test object W are stored in a storage function within the control unit 60. This flow is repeated until the stop button is pressed (1-15).

Next, an example of a calculation model will be shown, as well as verification results that were actually tested and verified using the calculation model.
<Example 1: Multiple regression analysis (determination of presence or absence of infection)>
The garlic bulbs to be tested W were produced at an experimental farm in Rokunohe Town, Aomori Prefecture, and 12,480 garlic bulbs prepared for shipping were used.
As shown in Figure 14, the data for creating the multiple regression analysis model is obtained by performing near-infrared spectroscopy on all the test subjects W, and then culturing Imogusare nematode using the Bellman method for the test subjects W to confirm the presence or absence of infection. The presence or absence of infection is associated with absorbance data at near-infrared wavelengths of the sample. Using the creation data, we searched for the wavelengths necessary to determine the presence or absence of infection in descending order of correlation, and created a multiple regression analysis model for the presence or absence of infection.

As the wavelengths necessary for determining the presence or absence of infection using a multiple regression analysis model, 191 wavelengths with a logarithmic value of 0.069 or more were used. As shown in FIG. 14, the correlation coefficient of the multiple regression model for the presence or absence of infection is 0.6898. Although a correlation is recognized, it cannot be said that there is a high correlation. However, when the estimated value of 0.5 or more of the multiple regression analysis model is set as a discriminant value to indicate "infection", 99.6% of "no infection" and 99.6% of "infection". Ta. This multiple regression analysis model can determine whether or not garlic is infected with Imogusare nematode by setting discriminant values.

The discrimination rate and accuracy were verified using the calculation model (multiple regression model for the presence or absence of infection) according to Example 1. As the test subject W, 200 garlic bulbs produced in an experimental field in Rokunohe Town, Aomori Prefecture and prepared for shipping were used. The results are shown in FIG. From this result, in the multiple regression model of the presence or absence of infection, the multiple correlation coefficient was 0.6836 (FIG. 15(a)). The accuracy of this multiple regression model for determining the presence or absence of infection is 97.1% for "no infection" and 90% for "infection" with a precision of 0.9422 when the discriminant value of the estimated value is set to 0.5 or higher. Discrimination was possible with an accuracy of .7% (Fig. 15(b)).

<Example 2: Multiple regression analysis (damage level (index) determination)>
As the garlic bulbs to be tested W, 12,480 garlic bulbs from the same production area as above and prepared for shipping were used. After near-infrared spectroscopic examination of the garlic bulbs to be tested, the test garlic bulbs were cultured for Imogusare nematode using the Bellman method, and the degree of damage was calculated using the damage level (index) described above based on the degree of lesions on the garlic. A multiple regression model of the degree of damage (index) yielded 196 wavelengths (not shown) selected with logarithmic values of 0.018 or higher. A multiple regression equation using these wavelengths was used as a damage level (index) estimation multiple regression model. As shown in FIG. 16(a), the multiple regression model had a correlation coefficient of 0.7815, and the degree of damage (index) could be estimated.

The discrimination rate and accuracy were verified using the calculation model (multiple regression model of damage level (index)) according to Example 2. 200 garlic bulbs prepared for shipping from the same production area as above were used. The results are shown in FIG. 16(b). From this result, the correlation coefficient of the multiple regression model was 0.7965. The results showed that it was possible to estimate the degree of damage (index).

<Example 3: Logistic regression model (determination of presence/absence of infection)>
As the garlic bulbs to be tested W, 12,480 garlic bulbs from the same production area as above and prepared for shipping were used. After applying near-infrared spectroscopy on the garlic bulbs of the test subject W, we culture the nematode nematode using the Bellman method, test to confirm the presence or absence of infection, and use the test data to correlate the wavelengths necessary to determine the presence or absence of infection. We created a logistic regression model for the presence or absence of infection by searching in descending order of . As the wavelengths necessary for determining the presence or absence of infection using a logistic regression model, 164 wavelengths (not shown) with a logarithmic value of 0.039 or more were used. As shown in Figure 17(a), regarding the discrimination accuracy of the logistic regression model for the presence or absence of Imogusare nematode infection, the average Precision is 0.9574, which is 95.4% for "no infection" and 96.0% for "infection". . In other words, it is an extremely accurate model.

The classification rate and accuracy were verified using the calculation model (logistic regression model for determining the presence or absence of infection) according to Example 3. 200 garlic bulbs prepared for shipping from the same production area as above were used. The results are shown in FIG. 17(b). From this result, the accuracy of determining the presence or absence of infection using the logistic regression model was 97.7% for "no infection" and 92.9% for "infected" with an average precision of 0.9674. Ta.

<Example 4: Logistic regression model (damage degree (6 classes) discrimination)>
As the garlic bulbs to be tested W, 3,120 garlic bulbs from the same production area as above and prepared for shipping were used as data for preparation. Near-infrared spectroscopy was performed on all test subjects W, and then the tested garlic bulbs were cultured for Imogusare nematode using the Bellman method. Calculated. The logistic regression analysis model for the degree of damage (6 classes) uses near-infrared spectral absorbance data corresponding to the actual measurement data for the degree of damage (6 classes) as training data, is analyzed by the logistic regression analysis method, and the regression equation is calculated. This is what I created.

As shown in FIG. 18(a), the logistic regression analysis model is a statistical analysis of the measured values and estimated values of the degree of damage (6 classes), and the accuracy is 0.7583, the average Precision is 0.7149, and the average Recall is 0. 4044, the average F-value is 0.4295, and the recall rate tends to be low, but this logistic regression analysis model is a model that can classify the degree of damage (6 classes) of Garlic nematode infestation.

The classification rate and accuracy were verified using the calculation model (logistic regression model for damage level (6 classes) classification) according to Example 4. Using 870 garlic bulbs prepared for shipping from the same production area as above, we verified the discrimination rate and accuracy of this logistic regression model for determining the degree of damage (6 classes) for infection or non-infection. The results are shown in FIG. 18(b). From this result, it can be seen that the classification accuracy of the degree of damage (6 classes) of the ordered logistic regression model varies from 0.000 to 0.8709 depending on the degree of damage, but it is possible to change the classification class.

<Example 5: AI trained model A (determining presence/absence of infection)>
As the garlic bulbs to be tested W, 15,075 garlic bulbs from the same production area as above and ready for shipping were used as supervised data. This training data is obtained by performing a near-infrared spectroscopy test on all the test subjects W, and then culturing the test garlic bulbs using the Bellman method to confirm the presence or absence of infection. This corresponds to absorbance data at infrared spectroscopic wavelengths. The basic structure of the learned model according to the example is composed of an input layer (550 pieces of absorbance data at wavelength intervals of 2 nm are input), an intermediate layer, and an output layer (the same in the following examples).

This trained model A uses the near-infrared spectral absorbance data corresponding to the actual measurement data of the presence or absence of infection as training data, and uses the deep learning (DNN method) intersection 5-fold verification method to detect 3 to 5 types and gradients. This model is constructed using an ensemble method in which 3 to 10 types of models are created using the 5-division method of crossing the boosting tree, and the optimal model is selected from the correct answer rate, incorrect answer rate, recall rate, and F value of each model. be.

As shown in FIG. 19(a), the trained model A has extremely high discrimination with AUC 0.9999, Accuracy 0.9992, Precision 0.9992, Recall 0.9992, and F value 0.9992 through statistical analysis. It has precision. Moreover, as shown in FIG. 19(b), the distribution of the teacher data used to establish the learned model A is distributed in a state that can be separated very clearly.

The classification rate and accuracy were verified using the calculation model according to Example 5 (AI trained model A for determining presence/absence of infection). 3,873 garlic bulbs from the same production area as above and ready for shipping were used. As shown in FIG. 20, in this trained model A, the multiple correlation coefficient was 0.9979, and the accuracy of determining the presence or absence of infection was 100.0% for no infection and 99.9% for infection. In other words, it can be seen that this model is capable of determining with high accuracy whether or not garlic is infected with Imogusare nematode.

<Example 6: AI trained model B (damage degree (6 classes) discrimination)>
As the garlic bulbs to be tested W, 12,480 garlic bulbs from the same production area as above and ready for shipping were used as supervised data. Near-infrared spectroscopy was performed on all test subjects W, and then the tested garlic bulbs were cultured for Imogusare nematode using the Bellman method. Calculated. The learned model B for damage level (6 classes) uses the near-infrared spectral absorbance data corresponding to the actual measurement value data for the damage level (6 classes) as training data, and performs deep learning (DNN method) cross-5-fold verification. Create 3 to 5 types of models using the method and 3 to 10 types of models using the gradient boosting tree intersection 5-division method, and then select the optimal model based on the correct answer rate, incorrect answer rate, recall rate, and F value of each model. It was constructed using an ensemble method to select.

As shown in FIG. 21(a), the trained model B has an Accuracy of 0.9965, an average Precision of 0.9986, and an average Recall of 0.9965, based on a statistical analysis of the measured and estimated damage levels (6 classes). 9902, with an average F value of 0.9943, and has extremely high discrimination accuracy, precision, and recall. In other words, it is a model that can classify the degree of damage (6 classes) of garlic nematode with high accuracy.

The classification rate and accuracy were verified using the calculation model (damage level (6 class) AI trained model B) according to Example 6. 3,120 garlic bulbs from the same production area as above and ready for shipping were used. The results are shown in FIG. 21(b). From this result, trained model B for damage level (6 classes) was found to have extremely high discrimination accuracy and fitness, with an accuracy of 0.9999 for actual and estimated values and an average precision of 0.9979. In other words, it can be seen that the model is capable of classifying the degree of damage (6 classes) of garlic nematode with high accuracy.

<Example 7: AI trained model C (damage level (index) determination)>
As the garlic bulbs to be tested W, 12,480 garlic bulbs from the same production area as above and ready for shipping were used as data for supervised learning. The actual value of the degree of damage (index) is determined by culturing the garlic bulbs for Imogusare nematode using the Bellman method after near-infrared spectroscopic examination of all the test objects W, and measuring the extent of lesions on the garlic according to the degree of damage (index) described above. ) was calculated. The trained model C for damage level (index) is created by using the above-mentioned actual damage level data as supervised data, and using the cross-validation method (5 divisions) of deep learning (DNN method) to create 3 to 5 types of learned model C. , create 3 to 10 types of models using the gradient boosting tree cross-validation method (5 divisions), and find the most optimal one based on the statistical results of the coefficient of determination, median error, average error, and PMSE of each model. The model was set as the learned model.

As shown in FIG. 22(a), the learned model C of the degree of damage (index) had a determination coefficient of 0.9809 between the actual measured value and the estimated value. Extremely high estimation accuracy was obtained with a median error of 0.303, mean error of 2.36, RMSE of 5.98, and coefficient of determination of 0.9809. Therefore, the trained model C is a model that can quickly, highly accurately, and non-destructively estimate the degree of damage (index) caused by Garlic nematode infection.

The method and accuracy were verified using the calculation model (damage degree (index) discrimination AI trained model C) according to Example 7. 3,120 garlic bulbs from the same production area as above and ready for shipping were used. The results are shown in FIG. 22(b). From this result, trained model C showed extremely high estimation accuracy with a coefficient of determination of 0.9936, a median error of 0.1646, an average error of 0.9788, and an RMSE of 3.5269. In other words, it can be seen that the model is capable of estimating the degree of damage (index) of garlic nematode with high accuracy.

<Example 8: AI trained model D (determining presence/absence of infection)>
As the garlic bulbs to be tested W, 18,960 garlic bulbs from the same production area as above and ready for shipping were used as teacher data. The training data for this trained model D is obtained by performing a near-infrared spectroscopy test on all test subjects W, culturing the test garlic using the Bellman method to confirm the presence or absence of infection, and confirming the presence or absence of infection. This corresponds to the absorbance data of the near-infrared spectral wavelength of the sample.

The trained model D has a basic structure based on the DNN (deep neural network) method of deep learning, and is composed of an input layer, five intermediate layers (consisting of activation functions such as drop and Relu functions), and an output layer. The optimization algorithm was Adam (Adaptive moment estimation), and the Learning Rate was set to 0.1, the Momentum was set to 0.9, and the Weight Decay was set to 0.0001, and the Leaning Rate Scheduler was set to Exponential, Multipier to 0.1, and Interval to 600,540 iterations. This is a trained model created with consideration to overfitting.

As shown in FIG. 23(a), this trained model D has an AUC (Area Under the Curve) of 0.9998, Accuracy of 0.9994, Precision of 0.9996, Recall of 0.9972, and F value of 9.994, which shows extremely high discrimination. It is a highly accurate model that is extremely effective in determining the presence or absence of infection with Imogusare nematode.

Using trained model D for determining the presence or absence of infection according to Example 8, its classification rate and accuracy were verified. 3,914 garlic bulbs prepared for shipment from the same production area as above were used. As shown in FIG. 23(b), in this trained model D, the precision of determining whether or not an infection was present was 100.0% at Precision 1.0. In other words, it can be seen that this model is capable of determining with high accuracy whether or not garlic is infected with Imogusare nematode.

<Example 9: AI trained model E (damage degree (6 classes) discrimination)>
As the garlic bulbs to be tested W, 15,600 garlic bulbs from the same production area as above and ready for shipping were used as supervised data. Near-infrared spectroscopy was performed on all test subjects W, and then the tested garlic bulbs were cultured for Imogusare nematode using the Bellman method. Calculated. The trained model E was created by deep learning (DNN: deep neural network) using near-infrared spectral absorbance data corresponding to the actual measurement data of the damage degree (six classes) described above as training data. The learned model E consists of an input layer, six intermediate layers (activation functions such as drop and Relu functions), and an output layer, and the optimization algorithm is Adams, with a weight decay of 0.0001. , this model was constructed with consideration to overfitting.

As shown in FIG. 24(a), the trained model E has an accuracy of 0.9976 between the measured value and the estimated value, an average Precision of 0.9991, an average Recall of 0.9945, and an average F value of 0.9968, showing extremely high discrimination accuracy. It is a trained model that has a precision and a recall rate, and can be applied as an extremely effective method for discriminating and classifying damage levels (6 classes).

The classification rate and accuracy were verified using the calculation model (damage level (6 class) AI trained model E) according to Example 9. 3,873 garlic bulbs prepared for shipment from the same production area as above were used. The results are shown in FIG. 24(b). From this result, the learned model E for the degree of damage (6 classes) has an average precision of 0.9994, exhibits an extremely high degree of fitness, and can be used for classifying the degree of damage. That is, it can be seen that the trained model E is a model that can quickly, highly accurately, and non-destructively estimate the degree of damage (6 classes) caused by the infection of Garlic nematode.

<Example 10: AI trained model F (damage level (index) determination)>
As the garlic bulbs to be tested W, 12,480 garlic bulbs from the same production area as above and ready for shipping were used as data for supervised learning. The actual value of the degree of damage (index) is determined by culturing the garlic bulbs for Imogusare nematode using the Bellman method after near-infrared spectroscopic examination of all the test objects W, and measuring the extent of lesions on the garlic according to the degree of damage (index) described above. ) was calculated. The learned model F was created by deep learning (DNN: deep neural network) using near-infrared spectroscopy absorbance data corresponding to the actual measurement value data of the above-mentioned damage level (index) as training data.

The trained model F has the basic structure of the DNN method, and has been developed through repeated basic studies and efforts to obtain the optimal conditions for the input layer, hidden layer, and output layer. The output layer consists of activation functions such as Relu function), the optimization algorithm is Adam (Adaptive moment estimation), LearningRate is 0.1, Momentum is 0.9, Weight Decay is 0.0001, and the Learning Rate Scheduler is Exponential. , Multipier0.1, Interval, and 600540 iterations. This model is characterized by being created with consideration to overfitting.

As shown in FIG. 25(a), the learned model F of the degree of damage (index) is determined by statistical data analysis between the estimated value and the actual measured value, with a coefficient of determination of 0.9987 and a median error of 0.2737. , the average error is 0.2593, and the RMSE is 1.6037, making it an extremely accurate model. This is a model that can quickly, accurately, and non-destructively estimate the degree of damage caused by Garlic nematode infection.

Using the calculation model (damage level (index) discrimination AI trained model F) according to Example 10, the discrimination rate and accuracy were verified. 2,851 garlic bulbs prepared for shipment from the same production area as above were used. The results are shown in FIG. 25(b). From this result, trained model F has a multiple correlation coefficient of 0.999, an error average of 0.5260, an error median of 0.2743, and an RMSE of 1. We have obtained an inspection result of 3025. That is, it can be seen that the trained model F is a model that can quickly, highly accurately, and non-destructively estimate the degree of damage (index) caused by the infection of garlic by Imogusare nematode.

<Example 11: AI trained model G (determining presence/absence of infection)>
As the garlic bulbs to be tested W, 12,480 garlic bulbs from the same production area as above and prepared for shipping were used as data for the teacher. The training data for this trained model G is obtained by performing a near-infrared spectroscopy test on all test subjects W, culturing the test garlic using the Bellman method to confirm the presence or absence of infection, and confirming the presence or absence of infection. This corresponds to the absorbance data of the near-infrared spectral wavelength of the sample.

The trained model G was created using deep learning (CNN: convolutional neural network) using near-infrared spectral absorbance data corresponding to actual measurement data on the presence or absence of infection as training data. The trained model G converts the waveform of near-infrared spectral absorbance data (1 row and 550 columns) into an image, and uses an input layer, 6 intermediate layers (activation functions such as drop and Relu functions), and an output layer. This is a model created using the Adams optimization algorithm, with a weight decay of 0.0001, and with consideration given to overfitting.

As shown in FIG. 26(a), this trained model G has an accuracy of 1.0000 between the measured value and the estimated value, an average Precision of 1.0000, an average Recall of 1.0000, and an average F value of 1.0000, which is extremely high. This model has high discrimination accuracy, precision, and recall, and can be applied as an extremely effective method for determining the presence or absence of infection with Imogusare nematode.

Using the learned model G for determining the presence or absence of infection according to Example 11, its classification rate and accuracy were verified. 3,121 garlic bulbs prepared for shipping from the same production area as above were used. As shown in FIG. 26(b), the learned model G had an accuracy of 100.0% in determining the presence or absence of infection. In other words, it can be seen that this model is capable of determining with high accuracy whether or not garlic is infected with Imogusare nematode.

<Example 12: AI trained model H (damage degree (6 classes) discrimination)>
As the garlic bulbs to be tested W, 12,480 garlic bulbs from the same production area as above and ready for shipping were used as supervised data. Near-infrared spectroscopy was performed on all test subjects W, and then the tested garlic bulbs were cultured for Imogusare nematode using the Bellman method. Calculated. The learned model H was created by deep learning (CNN: convolutional neural network) using near-infrared spectral absorbance data corresponding to the actual measurement data of the damage degree (six classes) described above as training data. The trained model H converts the waveform of near-infrared spectrophotometric absorbance measurement data (1 row and 550 columns) into an image, and creates an input layer, 6 intermediate layers (activation functions such as drop and Relu functions), and an output layer. This model was constructed using Adams optimization algorithm, with a weight decay of 0.0001, and with consideration given to overfitting.

As shown in FIG. 27(a), the trained model H has an accuracy of 1.0000 between the measured value and the estimated value, an average Precision of 1.0000, an average Recall of 1.0000, and an average F value of 1.0000, indicating extremely high discrimination. It is a model that has accuracy, precision, and recall, and can be applied as an extremely effective method for discriminating and classifying damage levels (6 classes).

The classification rate and accuracy were verified using the calculation model (damage level (6 class) AI trained model H) according to Example 12. 3,120 garlic bulbs from the same production area as above and ready for shipping were used. The results are shown in FIG. 27(b). From this result, it became clear that the trained model H had a fitness of 1.000 and could be used for classification of damage levels (6 classes). That is, it can be seen that the trained model H is a model that can quickly, highly accurately, and non-destructively estimate the degree of damage (6 classes) caused by the infection of Garlic nematode.

<Example 13: AI trained model I (damage level (index) determination)>
As the garlic bulbs to be tested W, 12,480 garlic bulbs from the same production area as above and ready for shipping were used as data for supervised learning. The actual value of the degree of damage (index) is determined by culturing the garlic bulbs for Imogusare nematode using the Bellman method after near-infrared spectroscopic examination of all the test objects W, and measuring the extent of lesions on the garlic according to the degree of damage (index) described above. ) was calculated. The trained model I was created by deep learning (CNN: convolutional neural network) using near-infrared spectral absorbance data corresponding to the actual measurement data of the damage degree (index) described above as training data. The trained model I is constructed by converting the waveform of near-infrared spectral absorbance data (1 row and 550 columns) into an image using the CNN method, and includes an input layer and 6 intermediate layers (activation functions such as drop and Relu functions). ), and an output layer, and the model was constructed using the Adams optimization algorithm, with a weight decay of 0.0001, and with consideration given to overfitting.

As shown in FIG. 28(a), the trained model I has a coefficient of determination of 0.9831, a median error of 1.2806, and an average error of 4.28% by statistical data analysis of the estimated value and the actual measured value. 2906, RMSE5.8131, it is a highly accurate model. This is a model that can quickly, highly accurately, and non-destructively estimate the degree of damage (index) caused by Garlic nematode infection.

The classification rate and accuracy were verified using the calculation model (damage level (index) discrimination AI trained model I) according to Example 13. 2,851 garlic bulbs prepared for shipment from the same production area as above were used. The results are shown in FIG. 28(b). From this result, we obtained test results with a statistically multiple correlation coefficient of 0.999 between the estimated value and the actual value, an error average of 4.2590, an error median of 1.2513, and an RMSE of 5.72955. That is, it can be seen that the trained model I is a model that can quickly, highly accurately, and non-destructively estimate the degree of damage (index) caused by the nematode infection of garlic.

In the above embodiment, the discrimination was performed by combining five types of discrimination methods: discriminant analysis method, multiple regression analysis method, PLS analysis method, back propagation method of neural network method, and SVM method. The method is not limited, and only one type of discrimination method may be used, or several types of discrimination methods may be combined, and may be changed as appropriate. Further, in the embodiment described above, class discrimination is performed in six stages, but it is not necessarily limited to this, and the number of stages may be determined as appropriate. Further, in the embodiment, the subject to be tested is a garlic plant, but the subject is not necessarily limited to this, and individual scale pieces may be tested, and changes may be made as appropriate.

Further, in the embodiment, garlic is used as a living organism and nematode nematode is used as a parasite. However, the present invention is not limited to this, and the present invention can be applied to various plants such as nematodes that parasitize these. can. Furthermore, it goes without saying that the present invention can be applied not only to plants and nematodes, but also to various living organisms and various parasites. In short, those skilled in the art will readily be able to make many changes to these illustrative embodiments without departing substantially from the novel teachings and advantages of the present invention, and these many changes will be apparent to those skilled in the art. Included in the range.

S Living body inspection device W Test object (garlic bulb)
Wa Scale piece Wb Flower diameter Wc Stem 1 Base 2 Conveyance part 3 Belt conveyor 10 Robot 11 Robot arm 12 Gripping hand 13 Opening/closing actuator Support recess 26 Camera 30 Inspection part 31 Inspection unit 32 Support rod 33 Cover 34 Unit body 35 Light source part 36 Light receiving part 37 Detection part 38 Holder 39 Recess 40 Dust-proof mechanism 41 Projection part 42 Air injection port 43 Air passage 50 Receiving part 51 Shooter 52 Sorting means 60 Control section 61 Arithmetic processing section 62 Computation model storage section 63 Discrimination section 64 Judgment section 65 Operation control section 66 Communication section 67 Inspection control section 68 Drive control section 69 Operation instruction section 70 Display section

Claims (19)

A living body testing device that performs a test for parasite infection in a living body to be examined,
A light source section that irradiates light in the near-infrared region to the test object, a light receiving section that receives diffusely reflected light or diffusely transmitted light from the test object, and a light receiving section that receives light based on the absorbance of the light received by the light receiving section. and a control unit for determining infection with the parasite to be tested,
The above-mentioned control unit is configured to generate diffuse reflected light or diffused light obtained from the sample living body by irradiating the sample living body infected with parasites and the sample living body not infected with parasites with light in the near-infrared region in advance. Constructed based on the second-order differential spectrum of absorbance for wavelengths in the near-infrared region of transmitted light, the presence or absence of infection with parasites in living organisms and/or the degree of onset from a predetermined maximum of no onset to onset due to infection a calculation model storage unit that stores a calculation model for performing calculations related to the degree of damage quantified;
The invention is characterized by comprising a determining unit that determines the presence or absence of infection and/or the degree of damage based on the calculation results calculated from the absorbance of the light received by the light receiving unit and the calculation model stored in the calculation model storage unit. A living body testing device. 2. The living body testing device according to claim 1, wherein the degree of damage is classified even for test subjects without infection. 3. The living body inspection apparatus according to claim 1, wherein the degree of damage is determined based on the degree of discoloration appearing on the subject. 4. The biological examination apparatus according to claim 3, wherein the degree of damage is constituted by an index with 0 indicating no onset and 100 indicating a maximum occurrence of onset. The above damage level is expressed as an index with no onset as 0 and the maximum of onset as 100, with no onset as 0 and the side of onset as increasing in order for each predetermined index range. 4. The biological examination apparatus according to claim 3, characterized in that the apparatus comprises: A sorting means is provided for sorting the test subjects according to the presence or absence of infection and/or classifying the test subjects according to the degree of damage based on the discrimination result of the discrimination section of the control section. 6. A living body testing device according to claim 1, characterized in that: The calculation model consists of an analytical model constructed using statistical analysis or a learned model constructed using machine learning,
The statistical analysis is selected from discriminant analysis, multiple regression analysis, PLS regression analysis, logistic regression analysis, support vector machine (SVM),
The above machine learning includes tree analysis such as decision tree (CART), regression tree, random forest, gradient boosting tree, perceptron, convolutional neural network (CNN), recurrent neural network (RNN), residual network (ResNet), etc. 7. The living body inspection method according to claim 1, wherein the method is selected from a neural network, deep learning (DNN), and an ensemble analysis consisting of a combination thereof. 8. Inspection of a living body according to claim 7, wherein the arithmetic processing unit includes a plurality of different types of arithmetic models as the arithmetic models, and a judgment unit that determines acceptance or rejection of judgment results of each arithmetic model. Device. When a plurality of calculation models of different types for determining the presence or absence of infection are used as the calculation model, and any one of the determination results using each calculation model determines that there is infection, 9. The biological examination apparatus according to claim 8, characterized in that said determination result is given priority. The determination unit is characterized in that when three or more calculation models of different types for determining the presence or absence of infection are used as the calculation models, the determination result of the presence or absence of infection of each calculation model is determined by majority vote. The living body inspection device according to claim 8. 11. The living body inspection device according to claim 1, wherein the parasite is a nematode that infects living plants. 12. The living body testing device according to claim 11, wherein the living body is garlic, potato, sweet potato, or iris, and the parasite is a Japanese nematode. 13. The living body testing device according to claim 12, wherein the object to be tested is composed of a garlic bulb in which scale pieces are assembled. The above-mentioned damage level is determined based on the degree of discoloration that appears in the subject, and is characterized by the following damage level, which is an index with 0 indicating no onset and 100 representing the maximum occurrence of onset. The living body testing device according to claim 11.
Damage degree (index) = ((number of scales at t = 1) + (number of scales at t = 2 x 2) + (number of scales at t = 3 x 3) + (number of scales at t = 4 x 4))/(Total number of surveyed scales x 4) x 100
here,
t is a numerical value corresponding to the size of the discolored area due to parasite infection per garlic bulb, and is determined so that the larger the value, the larger the discolored area, t = 1 , t=2, t=3, and t=4. The degree of damage mentioned above is determined based on the degree of discoloration that appears in the subject, and the degree of onset is expressed as an index with no onset as 0 and the maximum of onset as 100, and the degree of onset is as follows: 12. The living body inspection apparatus according to claim 11, wherein the no-onset side is set as 0, and the on-onset side is arranged in ascending ranks in order for each predetermined index range.
Degree of onset (index) = ((number of scales at t = 1) + (number of scales at t = 2 x 2) + (number of scales at t = 3 x 3) + (number of scales at t = 4 ×4))/(Total number of surveyed scales ×4) ×100
here,
t is a numerical value corresponding to the size of the discolored area due to parasite infection per garlic bulb, and is determined so that the larger the value, the larger the discolored area, t = 1 , t=2, t=3, and t=4. a base; a transport unit provided on the base for sequentially transporting test objects one by one at predetermined intervals; a robot installed near the base; and a transport unit at the tip of the robot arm of the robot. a gripping hand that grips the object to be inspected during the transportation process and positions the object to be inspected; and a gripping hand that includes the light source section and the light receiving section, and is located at the tip of the robot arm of the robot near the grasping hand. an inspection section for inspecting an object to be inspected positioned at an inspection position by the gripping hand; and a receiving section for receiving the object moved by the robot and gripped by the gripping hand after the inspection by the inspection section. 16. The living body inspection apparatus according to claim 1, wherein the apparatus is configured as follows. The inspection section includes the above-mentioned light source section and the light-receiving section and inspects the object to be inspected, and the inspection section includes an inspection unit including a detection section with the above-mentioned light source section and the light-receiving section exposed on one end surface. 17. A living body according to claim 1, wherein the inspection unit is provided with a dustproof mechanism that injects air along the detection section to form an air curtain that covers the detection section. Inspection equipment. A trained model for operating a computer to output calculation results regarding the presence or absence of infection with parasites of living organisms,
The absorbance of the sample living body is measured in advance, and the presence or absence of parasites in the living body is then determined by another method such as a culture method. The absorbance data of the specimen and the absorbance data of a sample living body infected with parasites were made to correspond to the presence or absence of parasites, and this corresponding data was used as training data, and learning was performed using the training data. , irradiates near-infrared rays onto the living body to be tested, receives diffusely reflected light or diffusely transmitted light from the living body to be tested, measures the absorbance of the received light, and uses this absorbance data to determine if the living body is infected with parasites. A trained model that operates a computer to output a value that quantifies whether or not the system is working. A trained model for operating a computer to output a calculation result related to the degree of damage that quantifies the degree of onset from a predetermined maximum of no onset to onset of onset due to infection with a parasite of a living body,
The absorbance of the sample organism is measured in advance, and the degree of damage caused by the infection of the organism to the parasite is determined by another means such as visual inspection.The degree of damage to the organism that has been determined is used as an indicator to identify samples that have not developed symptoms. The absorbance data of the living body and the absorbance data of the sample living body that has developed the disease are respectively made to correspond to the degree of damage to the living body, the corresponding data is used as teacher data, and learning is performed using the teacher data, Near-infrared rays are irradiated onto the living body to be tested, diffuse reflected light or diffused transmitted light is received from the living body to be tested, and the absorbance of the received light is measured. From this absorbance data, it is possible to determine whether the living body is infected with parasites. A trained model that operates a computer to output a value that quantifies the degree of damage caused by the disease.

Images