Information processing device, information processing method, and control program

Abstract

To more easily and at lower cost estimate the progress of a plant disease whose symptoms appear externally. [Solution] The information processing device (10) comprises an acquisition unit (12) that acquires a target image of a target plant, and an estimation unit (13) that uses a learning model that takes the target image as input and outputs target classification information together with at least one of (1) target probability information including at least one of a first probability that the target plant will be classified into a first class and a second probability that the target plant will be classified into a second class, and (2) target accuracy information including at least one of a first accuracy rate when the target plant is classified into the first class and a second accuracy rate when the target plant is classified into the second class, and the estimation unit (13) estimates the progression of a specified plant disease in the target plant based on at least one of the target probability information and the target accuracy information.

Description

[Technical Field]

[0001] One aspect of the present invention relates to an information processing device, an information processing method, and a control program. [Background technology]

[0002] There are known techniques for estimating whether a plant included in an image is infected with a specific plant disease and for estimating the progression of the plant disease in the plant. Patent Document 1 discloses a plant disease diagnosis system that can easily diagnose plant diseases without the need to manually extract image feature data of the plant disease from the image. Patent Document 2 discloses an optimal pest control recipe providing device that analyzes image data of crops and environmental information about the cultivation area to determine the type and progression of pest damage. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Application Laid-Open No. 2016-168046 [Patent Document 2] Japanese Patent Publication No. 2022-41805 Summary of the Invention [Problem to be solved by the invention]

[0004] However, the above-described conventional techniques require hyperspectral images and environmental information measured using multiple devices to estimate the progression of plant diseases, which poses problems such as the need for expensive equipment and complicated operations to acquire the data, and the inability to estimate the progression of plant diseases that are apparent from external appearances, such as the degree of wilting of the plant.

[0005] One aspect of the present invention has been made in view of the above-mentioned problems, and aims to make it possible to more easily and at lower cost to estimate the progression of a plant disease that manifests symptoms externally. [Means for solving the problem]

[0006] In order to solve the above problem, an information processing device according to one embodiment of the present invention includes an acquisition unit that acquires a target image of a target plant, and an estimation unit that uses a learning model to input the target image and output target classification information indicating whether the target plant is classified into a first class indicating that it is infected with a specified plant disease or a second class indicating that it is not infected with the specified plant disease, together with at least one of: (1) target probability information including at least one of a first probability that the target plant will be classified into the first class and a second probability that the target plant will be classified into the second class; and (2) target accuracy information including at least one of a first accuracy rate when the target plant is classified into the first class and a second accuracy rate when the target plant is classified into the second class, and the estimation unit estimates the progression of the specified plant disease in the target plant based on at least one of the target probability information and the target accuracy information.

[0007] In order to solve the above-mentioned problem, an information processing method according to one embodiment of the present invention is an information processing method executed by an apparatus, and includes: an acquisition step of acquiring a target image of a target plant; and an estimation step using a learning model that takes the target image as an input and outputs target classification information indicating whether the target plant is classified into a first class indicating that it is infected with a predetermined plant disease or a second class indicating that it is not infected with the predetermined plant disease, together with at least one of (1) target probability information including at least one of a first probability that the target plant will be classified into the first class and a second probability that the target plant will be classified into the second class, and (2) target accuracy information including at least one of a first accuracy rate when the target plant is classified into the first class and a second accuracy rate when the target plant is classified into the second class, and estimating the progression of the predetermined plant disease in the target plant based on at least one of the target probability information and the target accuracy information.

[0008] In order to solve the above problem, an information processing device according to one embodiment of the present invention includes an acquisition unit that acquires an image of an object, and an estimation unit that uses a learning model to input the object image and output object classification information indicating whether the object is classified into a first class or a second class into which the object is qualitatively classified, together with at least one of (1) object probability information including at least one of a first probability that the object will be classified into the first class and a second probability that the object will be classified into the second class, and (2) object accuracy information including at least one of a first accuracy rate when the object is classified into the first class and a second accuracy rate when the object is classified into the second class, and that estimates a predetermined quantitative value of the object based on at least one of the object probability information and the object accuracy information.

[0009] The information processing device according to each aspect of the present invention may be realized by a computer. In this case, the control program of the information processing device that causes the computer to operate as each part (software element) of the information processing device to realize the information processing device on the computer, and the computer-readable recording medium on which the control program is recorded, also fall within the scope of the present invention. [Effects of the Invention]

[0010] According to one aspect of the present invention, it is possible to more easily and at lower cost estimate the progression of a plant disease whose symptoms appear externally. [Brief explanation of the drawings]

[0011] [Figure 1] FIG. 1 is an example of a functional block diagram of an information processing system. [Figure 2] 1 is a diagram illustrating an example of a feature space in which feature amounts extracted from an image including a plant are mapped; [Figure 3] 10 is an example of a flowchart illustrating a flow of processing executed by the information processing system. [Figure 4] FIG. 1 shows examples of plant images of diseased individuals exhibiting mosaic symptoms at each stage. [Figure 5] 10 is an example of a table in which the values ​​of the first probability and the first accuracy rate are organized by stage according to the progression of mosaic symptoms in a target plant. [Figure 6] 1A to 1C are diagrams showing examples of images of diseased plant individuals exhibiting wilting symptoms at various stages. [Figure 7] 10 is an example of a table in which the values ​​of the first probability and the first accuracy rate are organized by stage according to the progression of wilt symptoms in a target plant. DETAILED DESCRIPTION OF THE INVENTION

[0012] Hereinafter, one embodiment of the present invention will be described in detail.

[0013] [1. Example of information processing system configuration] An information processing system 1 according to this embodiment will be described. FIG. 1 is an example of a functional block diagram of the information processing system 1 according to this embodiment. The information processing system 1 is a system for estimating the presence or absence and progression of a plant disease in a plant included in an image, and includes an information processing device 10, an imaging device 21, and a display device 22. Note that each component included in the information processing system 1 is not limited to being singular, and may be multiple. Furthermore, the function of a single component included in the information processing system 1 may be realized by multiple other components, and the functions of multiple components included in the information processing system 1 may be realized by a single other component.

[0014] The information processing device 10 is a device realized as a personal computer, a tablet, a smartphone, or the like, and includes a control unit 11 and a storage unit 19.

[0015] The control unit 11 is a control device such as a CPU that controls the entire information processing device 10, and also operates as an acquisition unit 12, an estimation unit 13, and a learning unit .

[0016] The acquisition unit 12 acquires a target image of a target plant from the imaging device 21 or the storage unit 19. Here, the target plant is a plant for which the presence or absence of a plant disease and the degree of its progression are to be estimated. In many cases, the target plant is any plant with leaves, but this is not necessarily limited to this.

[0017] The estimation unit 13 receives a target image as input and estimates the progression of a predetermined plant disease in a target plant based on the target probability information and target accuracy information output by a learning model that outputs target classification information together with at least one of target probability information and target accuracy information. The predetermined plant disease includes plant diseases caused by plant pathogens such as viruses, bacteria, fungi, viroids, mycoplasma, or bacteria-like microorganisms, and whose symptoms appear externally as mosaic symptoms or wilting. Here, the disease symptoms refer to the characteristics of individual plants affected by the plant disease.

[0018] Furthermore, the target classification information is information indicating whether the target plant is classified (classified) into a diseased class indicating that the plant is affected by a plant disease, or into a healthy class indicating that the plant is not affected by a plant disease. Here, the diseased class is an example of a first class in the present disclosure, and the healthy class is an example of a second class in the present disclosure. Furthermore, a plant affected by a plant disease refers to a plant whose disease progression is equal to or greater than a predetermined threshold. If the progression is expressed as a percentage, the threshold may be set to any value at least greater than 0%, such as 50% or 100%.

[0019] The aforementioned target probability information is information including at least one of a first probability that the target plant will be classified into the diseased class and a second probability that the target plant will be classified into the healthy class. The target accuracy information is information including at least one of a first accuracy rate when the target plant is classified into the diseased class and a second accuracy rate when the target plant is classified into the healthy class. In the case of two-class classification, the value of either the first probability or the second probability is determined based on the value of the other, so estimation based on the first probability and estimation based on the second probability are essentially synonymous. The first accuracy rate and second accuracy rate are also described in a similar manner.

[0020] FIG. 2 is a diagram for supplementing the classification process performed by the estimation unit 13 with an example, and is an example of a diagram showing a feature amount space in which feature amounts extracted from a target image are mapped.

[0021] In the example of FIG. 2 , a symptom of plant disease is a yellow color on the surface of a plant, while the surface of a healthy plant, which is not affected by the plant disease, is green. Furthermore, feature point 31a is a feature point extracted from an image of a diseased plant, which is an affected plant, and feature point 31b is a feature point extracted from an image of a plant that is not affected by the plant disease. Furthermore, boundary curve 32 is a boundary curve used for two-class classification to classify a plant corresponding to each feature point into either a diseased class or a healthy class. Specifically, the upper left side of boundary curve 32 corresponds to the diseased class, and the lower right side corresponds to the healthy class. Note that when a feature point is located on boundary curve 32, the plant corresponding to the feature point may be classified into any of the predefined classes. Furthermore, in the example of FIG. 2 , a feature point located on boundary curve 32 indicates that the degree of disease progression in the plant corresponding to the feature point is 50%.

[0022] Furthermore, the farther a feature point is mapped from the boundary curve, the more the plant individual corresponding to that feature point exhibits disease symptoms or characteristics of a healthy individual in its appearance, and the higher the reliability of the classification process. Conversely, the closer a feature point is mapped to the boundary curve, the more the plant individual corresponding to that feature point exhibits a mixture of disease symptoms and characteristics of a healthy individual in its appearance, and the lower the reliability of the classification process.

[0023] 2, the more a feature point is mapped to the upper left, the higher the first probability and the first correct answer rate that the plant individual corresponding to the feature point will be classified into the diseased class, and the lower the second probability and the second correct answer rate that the plant individual will be classified into the healthy class. The first probability can also be expressed as the probability that the feature point will be mapped to the upper left of the boundary curve 32. On the other hand, the more a feature point is mapped to the lower right, the lower the first probability and the first correct answer rate that the plant individual corresponding to the feature point will be classified into the diseased class, and the higher the second probability and the second correct answer rate that the plant individual will be classified into the healthy class. The second probability can also be expressed as the probability that the feature point will be mapped to the lower right of the boundary curve 32.

[0024] In addition, the target accuracy information of the plant individual corresponding to the feature point may be calculated by the estimation unit 13 based on the accuracy rate when classifying the plant individual corresponding to the feature point that was previously mapped near the feature point in the feature space.

[0025] The learning unit 14 trains the learning model using pairs of specimen images and classification information indicating whether the specimen plants contained in the images are infected with a predetermined plant disease as training data. For example, the training data includes multiple specimen images as explanatory variables and classification information corresponding to each of the multiple specimen images as a target variable. In other words, the specimen images are images used for training or verifying the learning model. Hereinafter, when there is no particular distinction between target images and specimen images, images containing plants may also be simply referred to as plant images.

[0026] The specimen image is an image of a specimen plant captured in the same manner as the target image of a target plant. For example, if the target image is an image captured with an optical camera, the specimen image may also be an image captured with an optical camera. Alternatively, if the target image is an image captured with an infrared camera, the specimen image may also be an image captured with an infrared camera. The specimen plant may also be, but is not limited to, a plant of the same species as the target plant. For example, the specimen plant may be a plant whose change between its state before and after being infected with a specified plant disease is similar to the change between the state of the target plant before and after being infected with the specified plant disease.

[0027] The storage unit 19 is a storage device such as a memory that stores various information at least temporarily, and stores, for example, parameter sets that define one or more learning models, and images captured by the imaging device 21.

[0028] The imaging device 21 is realized as a digital camera or the like and is a device that captures plant images or the like. However, for example, when the information processing device 10 is realized as a smartphone, the imaging device 21 is usually a camera provided in the smartphone. Similarly, the display device 22 may also be configured as an integral part of the information processing device 10. The plant images captured by the imaging device 21 are supplied to the information processing device 10 via a wired or wireless connection.

[0029] When the imaging device 21 images a plant, it is desirable to perform imaging with a spatial resolution that allows disease symptoms to be confirmed, and imaging may be performed using a sunshade to prevent overexposure due to shadows or direct light reflection. Furthermore, when imaging indoors, tracing paper or the like may be placed over the light source to use diffused light.

[0030] The display device 22 is a display that displays images, text, etc. For example, the display device 22 displays the target image supplied from the information processing device 10 and information indicating the progress of a plant disease in the plant.

[0031] The above describes an example of the configuration of the information processing system 1. Additionally, each unit included in the information processing system 1 has a function to execute the processes described below.

[0032] [2. Examples of processing by information processing systems] Next, a description will be given of the flow of processing executed by the information processing system 1. Fig. 3 is an example of a flowchart showing the flow of the processing.

[0033] In S1 (step S1), the imaging device 21 captures a target image. The acquisition unit 12 acquires the image captured by the imaging device 21.

[0034] In S2, the estimation unit 13 inputs the target image acquired by the acquisition unit 12 into the learning model, and estimates the value of the first probability output by the learning model as the progress of the plant disease in the target plant.

[0035] The estimation unit 13 may estimate the stage of the plant disease based on at least one of the object probability information and the object classification information output by the learning model. For example, the estimation unit 13 may estimate the stage of the plant disease based on the value of the first accuracy rate or the average or weighted sum of the first probability and the first accuracy rate, or may perform processing such as adding a predetermined shift value to either value.

[0036] In S3, the estimation unit 13 estimates whether the target plant is affected by a plant disease based on the result of comparing the progression level of the plant disease in the target plant with a threshold value. Typically, the estimation unit 13 estimates that the target plant is affected by a plant disease if the progression level is equal to or greater than the threshold value.

[0037] In S4, the control unit 11 causes the display device 22 to display the target image, the estimation result as to whether or not the plant is affected by a plant disease, the value of the first probability, and the like.

[0038] As described above, the information processing method executed by the information processing device 10 includes an acquisition step of acquiring a target image, and an estimation step of estimating the progression of a predetermined plant disease in a target plant based on at least one of the target probability information and the target accuracy information output by the learning model, etc. According to the configuration of this example, it is possible to more easily and at lower cost estimate the progression of a plant disease whose symptoms appear externally.

[0039] [3. Additional Notes] The machine learning method executed by the estimation unit 13 and the learning unit 14 is not limited to a specific method, and may be, for example, a method using a convolutional neural network (CNN) or a recurrent neural network (RNN). Also, either a classification or regression model may be used. When using a neural network such as a CNN or an RNN, input data may be pre-processed for input to the neural network. Such processing may include, in addition to two-dimensional or multidimensional data arrangement, various data augmentation techniques, adjustment of brightness, color tone, image quality, plant angle, etc., object detection for extracting plant regions, segmentation, etc.

[0040] Furthermore, when using CNN, a convolutional layer that performs convolutional operations may be provided as one or more layers included in the neural network, and a filter operation (product-sum operation) may be performed on input data input to the layer. When performing the filter operation, processing such as padding may be used in combination, or an appropriately set stride width may be adopted.

[0041] Alternatively, for example, any of the following machine learning techniques or a combination thereof may be used. Support Vector Machine (SVM) Clustering Inductive Logic Programming (ILP) Genetic Algorithm (GP: Genetic Programming) Bayesian Network (BN) Autoencoder The information processing system 1 may also be configured to estimate and display the degree of progression of a plant disease for all of a plurality of individual plants included in one or more target images. In this configuration, the acquisition unit 12 acquires one or more target images including a plurality of individual target plants. The estimation unit 13 then estimates the value obtained by dividing the sum of the first probabilities or first accuracy rates estimated for each of the target plants by the number of individual plants included in one or more target images as the degree of progression of the plant disease for all of the target plants included in the one or more target images.

[0042] The information processing system 1 may also be configured to estimate and display the proportion of target plants that are affected by a plant disease among multiple target plants included in one or more images. In this configuration, the acquisition unit 12 acquires one or more target images that include multiple target plants. The estimation unit 13 then estimates whether each of the target plants is affected by a plant disease based on the progression of the plant disease in each target plant, and estimates the proportion of target plants that are affected by a plant disease by dividing the number of target plants that are affected by the plant disease by the number of target plants included in one or more target images.

[0043] Furthermore, the images used by the learning unit 14 as training data may include a first specimen image captured of a specimen plant with a disease progression level equal to or greater than a first reference value, and a second specimen image captured of a specimen plant with a disease progression level equal to or less than a second reference value that is smaller than the first reference value. This configuration also includes cases where "equal to or greater than the first reference value" is interpreted as "greater than the first reference value" and "equal to or less than the second reference value" is interpreted as "less than the second reference value." Even in this configuration, the estimation unit 13 may estimate the disease progression level as a value greater than the second reference value and less than the first reference value. This means that the specimen images used by the learning unit 14 as training data may be specimen images including a specimen plant with a very high disease progression and clearly a diseased individual, and specimen images including a specimen plant with a very low disease progression and clearly a healthy individual. Even in this case, the estimation unit 13 according to the present disclosure can also perform estimation for target plants with an intermediate level of plant disease progression, depending on at least one of the target probability information and the target accuracy information. As an example, the images used as training data by the learning unit 14 may be a first specimen image including a specimen plant with a plant disease progression level of 90% or more and a second specimen image including a specimen plant with a plant disease progression level of 10% or less. Even if specimen images of specimen plants with a plant disease progression level of more than 10% and less than 90% are not used as training data, the estimation unit 13 can estimate that the plant disease progression level in the target plant is either more than 10% and less than 90%. This allows specimen images including specimen plants that can be clearly estimated as diseased or healthy individuals by human visual inspection to be used as training data, which also contributes to improving the accuracy of estimating the plant disease progression level.

[0044] The configuration of the information processing system 1 of the present disclosure is also applicable to cases where the target is something other than a plant. In a generalized version of the configuration, the acquisition unit 12 acquires a target image obtained by capturing an image of the target. The estimation unit 13 receives the target image as input and estimates a predetermined quantitative value of the target based on at least one of the target probability information and the target accuracy information output by a learning model that outputs target classification information indicating whether the target is classified into a first class or a second class, which are qualitatively classified, together with at least one of the target probability information and the target accuracy information. The progression of the plant disease described above is an example of the predetermined quantitative value. The target probability information is information including at least one of a first probability that the target is classified into the first class and a second probability that the target is classified into the second class. The target accuracy information is information including at least one of a first accuracy rate when the target is classified into the first class and a second accuracy rate when the target is classified into the second class. The diseased class and healthy class are examples of the first and second qualitatively classified classes. The learning unit 14 uses, as training data, a pair of a specimen image and classification information indicating whether an object (specimen) included in the specimen image is classified into the first or second class. The training data used by the learning unit 14 may include specimen images containing objects whose quantitative values ​​are equal to or greater than a first reference value and specimen images containing objects whose quantitative values ​​are equal to or less than a second reference value. Even in this case, the estimation unit 13 may estimate the quantitative value of the object as a value greater than the second reference value and less than the first reference value. The object is not limited to a plant, but may be any organic or inorganic substance. The quantitative value may be, in addition to the progression of a disease or the like, the degree of withering of the plant, the degree of disease resistance, the degree of pesticide control effect, the degree of pest damage, the degree of aging, or the degree of contamination.

[0045] Furthermore, the configuration of the present disclosure that estimates the progress level based on the first probability or the first accuracy rate is also applicable to object detection processing and segmentation processing that may include class classification as part of the processing.

[0046] Furthermore, the information processing system 1 may estimate and display the degree of progression of multiple types of plant diseases in parallel, or may classify plants into three or more classes. For example, the estimation unit 13 may estimate the degree of progression for each of mosaic symptoms and wilting of the leaves of the same plant based on at least one of the target probability information and the target classification information. In this case, the estimation unit 13 may classify plants into three classes: a healthy class, a mosaic symptom class, and a wilting class, or into four classes: a healthy class, a mosaic symptom class, a wilting class, and a class with both mosaic symptoms and wilting. In the former three-class classification, plants showing both mosaic symptoms and wilting symptoms are classified into the class corresponding to the more severe symptom.

[0047] [4. Software implementation example] The functions of the information processing device 10 (hereinafter referred to as the "device") can be realized by a program that causes a computer to function as the device, and a program that causes a computer to function as each control block of the device (particularly each part included in the control unit 11).

[0048] In this case, the device includes a computer having at least one control device (e.g., a processor) and at least one storage device (e.g., a memory) as hardware for executing the program. The control device and storage device execute the program, thereby realizing the functions described in each of the above embodiments.

[0049] The program may be non-transitory and may be recorded on one or more computer-readable recording media. The recording media may or may not be included in the device. In the latter case, the program may be supplied to the device via any wired or wireless transmission medium.

[0050] Furthermore, some or all of the functions of the control blocks can be realized by logic circuits. For example, an integrated circuit in which a logic circuit that functions as each of the control blocks is formed is also included in the scope of the present invention. In addition, the functions of the control blocks can also be realized by, for example, a quantum computer.

[0051] Furthermore, each process described in each of the above embodiments may be executed by AI (Artificial Intelligence). In this case, the AI ​​may run on the control device or on another device (for example, an edge computer or a cloud server).

[0052] The present invention is not limited to the above-described embodiments, and various modifications are possible within the scope of the claims. Embodiments obtained by appropriately combining the technical means disclosed in different embodiments are also included in the technical scope of the present invention. [Example]

[0053] A first embodiment of the present invention will be described below. In this embodiment, a process will be described in which a practitioner uses a system similar to the information processing system 1 to estimate the progression of a mosaic symptom caused by a certain plant disease.

[0054] First, the implementer defined the following five levels of condition for the target plants according to the progression of mosaic symptoms. 80% below is an example of the first standard value mentioned above, and 20% is an example of the second standard value. Mosaic symptoms (strong): Progression is less than 100% and more than 80%. Mosaic symptoms (medium): Progression is less than 80% and more than 60%. Mosaic symptoms (weak): Progression is less than 60% and more than 40%. Healthy (weak): Progression is less than 40% and more than 20%. · Healthy (Strong): Progression is below 20%.

[0055] The researchers also defined that plant individuals with strong, medium, or weak mosaic symptoms should be classified as diseased, and that plant individuals with weak and strong healthy symptoms should be classified as healthy. It was clear to the naked eye that plant individuals with strong mosaic symptoms were diseased, and that plant individuals with strong healthy symptoms were healthy. In contrast, it was not clear to the naked eye whether plant individuals in other states, especially those with weak mosaic symptoms, were diseased or healthy.

[0056] Figure 4 shows examples of plant images of diseased individuals exhibiting mosaic symptoms at various stages. In Figure 4, the numbers below the plant images indicate the percentage of progression of the mosaic symptoms. That is, the four plant images corresponding to the numbers 100 and 90 are images of plants with strong mosaic symptoms, the four plant images corresponding to the numbers 80 and 70 are images of plants with medium mosaic symptoms, and the four plant images corresponding to the numbers 60 and 50 are images of plants with weak mosaic symptoms.

[0057] The researchers then trained the learning model using random plant images that were visually distinguishable, specifically images of plants with strong mosaic symptoms and images of healthy plants with strong symptoms, as sample images, i.e., training data and validation data. Images of plants with medium and weak mosaic symptoms and healthy plants with weak symptoms were not used as training data or validation data.

[0058] Next, the inventors used the system with a fully trained learning model to perform two-class classification of target plants contained in target images randomly selected as test data into either a diseased class exhibiting mosaic symptoms or a healthy class. As test data, plant images of the five stages mentioned above, specifically, plant images of (strong), (medium), and (weak) mosaic symptoms and healthy (weak) and (strong) symptoms, were used.

[0059] Figure 5 shows an example of a table in which the values ​​of the first accuracy rate and the first probability are organized by stage according to the progression of mosaic symptoms in the target plants when the above-mentioned two-class classification is performed. The average value for the first probability is also shown. For example, the average value of the first probability for target plants with strong mosaic symptoms to be classified as a diseased class was 91.7%. Furthermore, the system estimates that a target plant is diseased if the value of the first accuracy rate or the first probability is over 40%, and that a plant is healthy if it is 40% or less.

[0060] As shown in the results in Figure 5, regardless of whether the first probability or the first accuracy rate is used to estimate the progression of mosaic symptoms, it was shown that it is possible to quantify and estimate the progression corresponding to plant images of mosaic symptoms (moderate) and (weak) and healthy (weak) that were not used as training data or validation data. [Example]

[0061] A second embodiment of the present invention will be described below. In this embodiment, a process will be described in which a practitioner uses a system similar to the information processing system 1 to estimate the progress of wilting symptoms caused by a certain plant disease.

[0062] First, the implementer defined the following five levels of condition for the target plants according to the progression of wilt symptoms. 80% below is an example of the first standard value mentioned above, and 20% is an example of the second standard value. Wilt symptoms (strong): Progression is less than 100% and more than 80%. Wilt symptoms (medium): Progression is less than 80% and more than 60%. Wilt symptoms (weak): Progression is less than 60% and more than 40%. Healthy (weak): Progression is less than 40% and more than 20%. · Healthy (Strong): Progression is below 20%.

[0063] The researchers also defined that plant individuals with strong, medium, or weak wilting symptoms should be classified as diseased individuals, and that plant individuals with weak and strong healthy symptoms should be classified as healthy individuals. Here, it was clear to the naked eye that plant individuals with strong wilting symptoms were diseased, and that plant individuals with strong healthy symptoms were healthy. In contrast, it was not clear to the naked eye whether plant individuals in other states, especially those with weak wilting symptoms, were diseased or healthy.

[0064] Figure 6 shows examples of plant images of diseased individuals exhibiting various stages of wilting symptoms. In Figure 6, the numbers below the plant images indicate the percentage of progression of the wilting symptoms. That is, the four plant images corresponding to the numbers 100 and 90 are images of plants exhibiting strong wilting symptoms, the four plant images corresponding to the numbers 80 and 70 are images of plants exhibiting medium wilting symptoms, and the four plant images corresponding to the numbers 60 and 50 are images of plants exhibiting weak wilting symptoms.

[0065] The researchers then trained the learning model using random plant images that were visually distinguishable, specifically images of plants with strong wilt symptoms and images of healthy plants, as sample images, i.e., training data and validation data. Images of plants with medium and weak wilt symptoms and healthy plants with weak wilt symptoms were not used as training data or validation data.

[0066] Next, the implementer used the system with a fully trained learning model to perform two-class classification of target plants contained in target images randomly selected as test data into either a diseased class exhibiting wilting symptoms or a healthy class. As test data, plant images of the five stages mentioned above, specifically plant images of wilting symptoms (strong), (medium), and (weak) and healthy (weak) and (strong), were used.

[0067] Figure 7 shows an example of a table in which the values ​​of the first accuracy rate and the first probability are organized by stage according to the progression of wilt symptoms in the target plants when the above-mentioned two-class classification is performed. The average value for the first probability is also shown. For example, the average value of the first probability for target plants with strong wilt symptoms to be classified into the diseased class was 93.4%. Furthermore, the system estimates that a target plant is diseased if the value of the first accuracy rate or the first probability is over 40%, and that a target plant is healthy if it is 40% or less.

[0068] As shown in the results in Figure 7, regardless of whether the first probability or the first accuracy rate is used to estimate the progression of wilt symptoms, it was shown that it is possible to quantify and estimate the progression corresponding to plant images with wilt symptoms (moderate), (weak), and healthy (weak), which were not used as training data or validation data. [Explanation of symbols]

[0069] 1. Information Processing Systems 10. Information processing equipment 11 Control section 12 Acquisition Department 13 Estimation part 14 Learning Department 19 Memory section 21 Imaging device 22 Display device

Claims

1. an acquisition unit that acquires a target image of a target plant; an estimation unit using a learning model that receives the target image as an input and outputs target classification information indicating whether the target plant is classified into a first class indicating that the target plant is infected with a predetermined plant disease or a second class indicating that the target plant is not infected with the predetermined plant disease, together with at least one of (1) target probability information including at least one of a first probability that the target plant is classified into the first class and a second probability that the target plant is classified into the second class, and (2) target accuracy information including at least one of a first accuracy rate when the target plant is classified into the first class and a second accuracy rate when the target plant is classified into the second class, and that estimates the progress of the predetermined plant disease in the target plant based on at least one of the target probability information and the target accuracy information; Equipped with The estimation unit and estimating whether or not the target plant is affected by the plant disease based on the result of comparing the progression level with a threshold value. Information processing device.

2. The learning model is the training data includes a first specimen image obtained by capturing an image of a specimen plant in which the degree of progression of the predetermined plant disease is equal to or greater than a first reference value, and a second specimen image obtained by capturing an image of the specimen plant in which the degree of progression of the predetermined plant disease is equal to or less than a second reference value that is smaller than the first reference value, wherein classification information indicating the first class is associated with the first specimen image and classification information indicating the second class is associated with the second specimen image, and the training data is generated by machine learning using the training data; The information processing device according to claim 1 .

3. The acquisition unit acquiring one or more target images including a plurality of target plants; The estimation unit 3. The information processing device according to claim 1, wherein the total value of the first probability or the first accuracy rate estimated for each of the target plants is divided by the number of the target plants included in the one or more target images, and the resulting value is estimated as the progression of the plant disease for all of the target plants included in the one or more target images.

4. The acquisition unit acquiring one or more target images including a plurality of target plants; The estimation unit 2. The information processing device according to claim 1, wherein the information processing device estimates whether each of the target plants is infected with the plant disease, and estimates the proportion of the target plants infected with the plant disease by dividing the number of the target plants infected with the plant disease by the number of the target plants included in the one or more target images.

5. 1. An information processing method performed by an apparatus, comprising: an acquisition step of acquiring a target image of a target plant; an estimation step using a learning model that receives the target image as an input and outputs target classification information indicating whether the target plant is classified into a first class indicating that the target plant is infected with a predetermined plant disease or a second class indicating that the target plant is not infected with the predetermined plant disease, together with at least one of (1) target probability information including at least one of a first probability that the target plant is classified into the first class and a second probability that the target plant is classified into the second class, and (2) target accuracy information including at least one of a first accuracy rate when the target plant is classified into the first class and a second accuracy rate when the target plant is classified into the second class, and an estimation step that estimates the progress of the predetermined plant disease in the target plant based on at least one of the target probability information and the target accuracy information; Including, In the estimation step, and estimating whether or not the target plant is affected by the plant disease based on the result of comparing the progression level with a threshold value. Information processing methods.

6. A control program for causing a computer to function as the information processing device according to claim 1, the control program causing the computer to function as the acquisition unit and the estimation unit.

7. an acquisition unit that acquires a target image obtained by capturing an image of a target; an estimation unit that uses a learning model and receives the target image as an input and outputs object classification information indicating whether the object is classified into a first class or a second class into which the object is qualitatively classified, together with at least one of (1) object probability information including at least one of a first probability that the object is classified into the first class and a second probability that the object is classified into the second class, and (2) object accuracy information including at least one of a first accuracy rate when the object is classified into the first class and a second accuracy rate when the object is classified into the second class, and that estimates a predetermined quantitative value of the object based on at least one of the object probability information and the object accuracy information; Equipped with The estimation unit and estimating whether the object is classified into the first class or the second class depending on a result of comparing the quantitative value with a threshold. Information processing device.

Images