Generation method, discrimination method, computer program, generation device, and discrimination device

Abstract

To provide a discrimination method, a discrimination device, and a computer program for discriminating a growth state of a plant.SOLUTION: Provided is a method for generating a discriminant model for discriminating a growth state of a plant, executed by a computer including an arithmetic circuit configured to access a storage device, the storage device storing a chloroplast density image which is an image reflecting a difference between absorption spectra of chlorophyll and carotenoids contained in the plant obtained from a plurality of images of leaves of plants for training, being executed by the arithmetic circuit, reading the chloroplast density image from the storage device, and generating a discriminant model of the growth state of the plant using the chloroplast density image as training data for machine learning.SELECTED DRAWING: Figure 6A

Description

[Technical field]

[0001] The present invention relates to a method and device for generating a discrimination model for discriminating the growth state of a plant, a discrimination method and device for discriminating the growth state of a plant, and a computer program for executing these methods. [Background technology]

[0002] The growth state of a plant may be observed from the living leaves of the plant. For example, Patent Document 1 describes an apparatus for diagnosing whether a plant is infected with a pathogen by irradiating the leaves of the plant with excitation light. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] JP 2013-36889 A Summary of the Invention [Problem to be solved by the invention]

[0004] However, there is still room for improvement in terms of making it easier and more accurate to observe the growing state of plants.

[0005] Therefore, the present disclosure aims to provide a method and device for generating a discrimination model for discriminating the growth state of a plant, which enables easy and / or accurate observation of the growth state of the plant, a discrimination method and device for discriminating the growth state of a plant, and a computer program for executing these methods. [Means for solving the problem]

[0006] A generation method according to the present disclosure generates a discrimination model for discriminating a growth state of a plant, the discrimination model being executed by a computer including an arithmetic circuit that can access a storage device. The storage device stores a chloroplast sparse / dense image, which is an image obtained from images of leaves of a plurality of learning plants and reflects the difference in the absorption spectra of chlorophyll and carotenoid contained in the plant leaves. The generation method is executed by the arithmetic circuit, reads out the chloroplast sparse / dense image from the storage device, and generates a discrimination model for the growth state of a plant by using the chloroplast sparse / dense image as learning data for machine learning.

[0007] Such general and specific aspects may be realized by a system, a method, and a computer program, as well as a combination thereof. Effect of the Invention

[0008] The method and device for generating a discriminant model for discriminating the growth condition of a plant, the discrimination method and discrimination device for discriminating the growth condition of a plant, and the computer program for executing these methods provide easy and / or accurate observation of the growth condition of a plant. [Brief description of the drawings]

[0009] [Figure 1A] A model of the emission spectrum of a white LED is shown. [Figure 1B] A model of the chlorophyll absorption spectrum is shown. [Figure 1C] 1 shows a model of the absorption spectrum of carotenoids. [Figure 1D] A model of the action spectrum of chlorophyll is shown. [Figure 2A] A model of the spectrum of the healthy portion is shown. [Figure 2B] A model of the spectrum of the yellowed area is shown. [Figure 2C] A model of the spectrum of the leaf vein is shown. [Figure 3A] A model of the R spectral sensitivity curve of a digital camera is shown. [Figure 3B]A model of the G spectral sensitivity curve of a digital camera is shown. [Figure 3C] A model of the B spectral sensitivity curve of a digital camera is shown. [Figure 4] The difference between the model spectrum of the yellowed area and the model spectrum of the healthy area is shown. [Figure 5A] This is an image of the surface of a leaf. [Figure 5B] 2B is a binarized image obtained by converting the image in FIG. 2A into a monochrome image. [Figure 5C] This is a binarized image of a chloroplast density distribution image obtained using the R image and B image obtained from the image in Figure 2A. [Figure 6A] FIG. 1 is a schematic diagram illustrating generation of a discriminant model. [Figure 6B] FIG. 1 is a schematic diagram illustrating discrimination using a discriminant model. [Figure 7] FIG. 2 is a block diagram showing an example of a generating device according to the first embodiment. [Figure 8] 1 is a block diagram showing an example of a discrimination device according to a first embodiment. [Figure 9] 5 is a flowchart showing an example of a process for generating a discriminant model executed by the generating device according to the first embodiment. [Figure 10] 4 is a flowchart showing an example of a discrimination process executed by the discrimination device according to the first embodiment. [Figure 11] FIG. 11 is a block diagram showing an example of a generating device according to a second embodiment. [Figure 12A] FIG. 2 is a schematic diagram illustrating clustering data. [Figure 12B] FIG. 13 is a schematic diagram illustrating a clustering result. [Figure 12C] FIG. 13 is a schematic diagram illustrating the acceptance of labels for clustering results. [Figure 13] 13 is a flowchart showing an example of a clustering and discriminant model generation process executed by the generation device according to the second embodiment. [Figure 14] FIG. 2 is a schematic diagram showing an example of data used in Experimental Example 1. [Figure 15]15 is a schematic diagram showing an example of learning data and evaluation data selected from the group in FIG. 14 in the first to ninth times. FIG. [Figure 16] 13 is a graph showing the results of Experimental Example 3. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0010] Hereinafter, the embodiments of the present disclosure will be described with reference to the drawings as appropriate. However, in the detailed description, unnecessary parts of the description of the prior art and substantially the same configuration may be omitted. This is for the purpose of simplifying the description. In addition, the following description and the accompanying drawings are disclosed so that a person skilled in the art can fully understand the present disclosure, and are not intended to limit the subject matter of the claims.

[0011] The present disclosure relates to a method for generating a discrimination model for discriminating the growth state of a plant and a discrimination method thereof. In the following, a discrimination model for discriminating the healthy growth state of a plant as the growth state of the plant is described as an example. More specifically, a case where a disease of a plant appears as yellowing of leaves is described as an example.

[0012] <Plant growth status> The growth state of a plant may be reflected in the leaves of the plant. For example, it may be possible to determine whether a plant is in a healthy state or in a diseased state from the appearance of the leaves. On the other hand, even if a plant is healthy, it may appear to be in a similar state to a diseased state at first glance.

[0013] For example, when a plant is infected with a disease, the leaves of the plant may turn yellow. Therefore, when the leaves of a plant turn yellow, a plant grower may determine that the plant is infected with a disease. On the other hand, even when the plant is not infected with a disease, the leaves of the plant may turn yellow. For example, when a plant is under conditions such as nutrient deficiency or lack of sunlight, the leaves may turn yellow even when the plant is not infected with a disease. In such cases, even for an expert, it may be difficult to determine whether a plant is infected with a disease or not.

[0014] The measures taken for a plant differ depending on whether the plant is infected with a disease and whether the plant is not infected with a disease. Therefore, it is desirable to improve the accuracy of determining whether a plant is infected with a disease.

[0015] The leaves of plants turn yellow when the amount of chlorophyll changes, causing a relative increase in the density of carotenoids. Yellowing due to nutrient deficiency and yellowing due to disease can have differences in shape, distribution, and shade. In some cases, only experienced farmers can recognize the difference in the yellowing.

[0016] In order for plants to carry out photosynthesis, the mesophyll of leaves contains photosynthetic pigments such as chlorophyll and carotenoids, which are found in chloroplasts. Chlorophyll is one of the main photosynthetic pigments, and carotenoids are one of the auxiliary photosynthetic pigments.

[0017] When a plant is not infected with a virus or other disease and is growing well, that is, when the plant is healthy, the mesophyll of the leaf contains sufficient chlorophyll and carotenoids. Hereinafter, such a part containing sufficient chlorophyll and carotenoids is also referred to as a "healthy part."

[0018] When a plant is infected with disease or becomes nutrient deficient, the development of the mesophyll green bodies is impaired, chlorophyll is reduced, and the leaves turn yellow. Hereinafter, such parts are referred to as "yellowing parts." That is, yellowing parts may be caused by "disease" or by "something other than disease."

[0019] Furthermore, the veins of leaves usually contain little or no chlorophyll and carotenoids.

[0020] Therefore, the amount of chlorophyll and carotenoids in each portion is as shown in Table 1 below. (Table 1) TIFF2025085455000002.tif21115

[0021] As described above, the yellowing may be caused by either "disease" or "other causes." The distribution of chlorophyll on the surface of the leaf including the yellowing may differ depending on the cause. In other words, the shape, shading, and distribution of the yellowing caused by a decrease in the amount of chlorophyll may differ. If it is possible to distinguish the distribution of chlorophyll, which cannot be easily distinguished by appearance, it is possible to distinguish whether the cause of the yellowing is "disease" or "other." For example, if it becomes easier to distinguish the shape, distribution, and shading of the yellowing on the leaf, the accuracy of the distinction can be improved.

[0022] <Light transmission through chloroplasts> Next, we will explain the light transmission of chlorophyll and carotenoids. As an example, consider the case where a white LED is used as a light source. Figure 1A shows a model of the emission spectrum of a white LED. Figure 1B shows a model of the absorption spectrum of chlorophyll. Figure 1C shows a model of the absorption spectrum of carotenoids.

[0023] However, the absorbance shown in Figure 1B and Figure 1C is ideally measured in solution, and in reality, light that enters the leaf is internally scattered within the leaf, and even green light can be absorbed by chloroplasts or other substances in the leaf and contribute to photosynthesis. This effect can be seen, for example, in the form of the photosynthetic action spectrum, as shown in Figure 1D.

[0024] As mentioned above, the green light absorption effect is largely due to internal scattering within the leaf, so it is known that thin plants such as Ulva have a photosynthetic action spectrum close to the ideal chlorophyll absorption spectrum. Regardless of the substance that absorbs green light, the green light absorption effect is largely due to the increase in the probability of the incident light meeting the light absorbing substance caused by the internal scattering effect due to the presence of chloroplasts, so if the chloroplasts disappear due to the yellowing of the leaves, the green light absorption effect will also weaken. In that sense, for convenience, the action spectrum in Figure 1D will be adopted here as the action spectrum of chlorophyll. For carotenoids, Figure 1 will be adopted as is.

[0025] According to Table 1, the light from the light source is subjected to the absorption spectrum under each condition. That is, the healthy portion is subjected to the chlorophyll absorption spectrum shown in Figure 1B and the carotenoid absorption spectrum shown in Figure 1C. The yellowed portion is subjected to only the carotenoid absorption spectrum shown in Figure 1C. Neither the chlorophyll absorption spectrum nor the carotenoid absorption spectrum is applied to the vein portion. Figure 2A shows the results for the healthy portion (model of the spectrum for the healthy portion), Figure 2B shows the results for the yellowed portion (model of the spectrum for the yellowed portion), and Figure 2C shows the results for the vein portion (model of the spectrum for the vein portion).

[0026] Next, the digital camera's spectral sensitivity curve is applied to these spectra. In other words, the colors are divided into the three colors RGB, but when they are separated, no wavelength information remains, and only the integrated pixel values ​​remain. The smallest elements that make up an image, arranged in a grid pattern, are called pixels. Each pixel expresses light intensity and color through a numerical value. The value that expresses the light intensity and color of each pixel is called the pixel value. Figure 3A shows a model of the R spectral sensitivity curve. Figure 3B shows a model of the G spectral sensitivity curve. Figure 3C shows a model of the B spectral sensitivity curve.

[0027] Table 2.1 shows an example of pixel values ​​that represent the light intensity of healthy parts, yellowed parts, and leaf veins when R, G, and B lights are applied. Table 2.1 shows that in all cases of R, G, and B, the leaf veins have the highest value, and the yellowed parts are not emphasized. Also, when comparing the pixel values, it can be seen that the leaf veins have high values ​​in all bands of the R, G, and B images, while the B image has lower values ​​than the R and G images in healthy and yellowed parts. (Table 2.1) TIFF2025085455000003.tif26143

[0028] FIG. 4 shows the difference between the model spectrum of the yellowed part and the model spectrum of the healthy part. Specifically, it is a graph showing the value obtained by subtracting the model spectrum of FIG. 2A from the model spectrum of FIG. 2B. In FIG. 4, it can be seen that the difference is larger on the medium wavelength side (e.g., about 500 nm to 600 nm) and the long wavelength side (e.g., about 600 nm to 780 nm) than on the short wavelength side (e.g., wavelengths of about 400 nm to 500 nm). In other words, there is a large difference in the wavelength portion distributed across G (green light: e.g., medium wavelength) and R (red light: e.g., long wavelength), and the difference between yellowed and healthy is relatively small in the wavelength band counted as B (blue light: e.g., short wavelength). This is also clear when the difference in the values ​​of the yellowed part and the healthy part is calculated for each of the R, G, and B pixel values ​​in the above matrix.

[0029] From the above, the following two points can be made. The difference between yellowed and healthy areas appears in the pixel values ​​in the wavelength bands spanning R and G. The leaf veins have high pixel values ​​in all RGB wavelength bands.

[0030] From this, it can be said that formula (1.1) is the optimal calculation formula for maximally emphasizing the "difference between the yellowed parts and the healthy parts," that is, the "density and density of chloroplasts." (Formula for blending R and G images with arbitrary weighting) - (Formula using B image) (1.1)

[0031] Here, (the formula for blending the R image and the G image with any weighting) includes the arithmetic mean, the geometric mean, and formulas that allow weighting of two other elements. (The formula using the B image) may use a coefficient or a correction term, for example, to enhance the effect of removing veins. (The formula for blending the R image and the G image with any weighting) and (The formula using the B image) may be normalized to make the maximum and minimum ranges of both the formulas the same, since the chloroplast density image is ultimately output as a monochrome image.

[0032] The first term in the above formula (1.1) makes the difference between yellowed and healthy parts clear. The second term in formula (1.1) cuts out the information about the leaf veins. Here, the maximum and minimum ranges of the first and second terms in formula (1.1) must be the same. Generally, pixel values ​​for each of RGB range from 0 to 255. Therefore, the first term in formula (1.1) becomes formula (1.2). (R image x + G image y) / (x + y) (1.2) (where 0≦x, 0≦y)

[0033] The reason why the values ​​of x and y cannot be uniquely determined is that the action spectrum of photosynthesis differs depending on the plant leaf. Depending on the type of plant, the weighting of R or G will emphasize the difference between the yellowed and healthy parts more.

[0034] Table 2.2 shows an example of the integrated pixel values ​​of the healthy part, the yellowed part, and the leaf vein part when the x and y values ​​of the following formula (1.3) are (0, 1), (1, 0), and (1, 1), respectively. In each case, the yellowed part has the largest pixel value, and it can be seen that the yellowed part is most emphasized in the monochrome image. (R image x + G image x y) / (x + y) - B image (1.3) (where 0≦x, 0≦y) The image obtained by equation (1.3) is an example of a "chloroplast sparse / dense image" in this disclosure. (Table 2.2) TIFF2025085455000004.tif24131

[0035] <Chloroplast density distribution image> Here, the chloroplast sparse / dense image used in the discrimination model of the present disclosure will be described with reference to Fig. 5A to Fig. 5C. Fig. 5A is an image of the surface of a leaf. In Fig. 5A, a monochrome image is shown, but in reality, it is a color image.

[0036] FIG. 5B is an image in which the monochrome image of FIG. 5A was binarized using a predetermined threshold value. The threshold value was determined while adjusting it so that the yellowed parts could be extracted. In the image of FIG. 5B, the black parts are mainly considered to be the yellowed parts. However, in the image of FIG. 5B, the non-yellowing leaf vein parts and the like are also detected as black.

[0037] Figure 5C shows an example of an image obtained by binarizing a chloroplast sparse / dense image using the R image and B image obtained from the color image in Figure 5A by adjusting the threshold so that the yellowed portion can be extracted. Specifically, the image in Figure 5C is an image obtained by binarizing a chloroplast sparse / dense image when the values ​​of x and y are (1, 0), respectively, in the following formula (1.3). (R image x + G image x y) / (x + y) - B image (1.3) In the image of Figure 5C, the black areas are also mainly yellowed areas. Comparing the binarized image of Figure 5B with the binarized image of the chloroplast sparse / dense image of Figure 5C, the binarized image of the chloroplast sparse / dense image of Figure 5C can reduce the extraction of the vein areas as yellowed. This is due to the vein reduction effect (the effect of cutting out the information of the vein areas) as described above.

[0038] <Discrimination model> A brief description will be given of a discrimination model M for discriminating the growth state of a plant with reference to Figs. 6A and 6B. As shown in Fig. 6A, learning data D13 used for generating the discrimination model includes a plurality of sets of image data and labels attached to the images. Each image data is a "chloroplast sparse / dense image of a plant leaf." The label is the "growth state of the plant" corresponding to each image data. Specifically, the growth state of a plant is at least one of a diseased state, a malnutrition / overnutrition state, a poor growth environment state (inappropriate sunlight, temperature, humidity, watering), and an insect damage state, which may affect the health state and leaf color.

[0039] The learning device learns the relationship between each image data included in the learning data D13 and the growth state of the plant by machine learning, and generates a discrimination model M. When a new image showing the density of the plant is input, the discrimination model M outputs the growth state of the plant as a discrimination result. If the growth state of the plant, which is the label, is "good" and "poor", the discrimination model M outputs, for example, the possibility that the plant is "good" or "poor" for an input image showing the density of the plant. Note that the labels indicating the growth state of the plant are not limited to two types, "good" and "poor". For example, the growth state of the plant may be represented by three or more types of labels. By using the discrimination model M, it becomes possible to easily distinguish plants that are difficult to distinguish by visual inspection alone.

[0040] Each image data (chloroplast sparse / dense image) is generated from image data obtained by photographing the surface of a plant leaf, for example. It is preferable that the image data included in one learning data set is photographed and generated under the same conditions. For example, the image data (chloroplast sparse / dense image) included in one learning data set may be generated from image data photographed under the same or similar light irradiation conditions. More specifically, it may be image data photographed using a transmission optical system incorporating a surface light source whose luminance is approximately uniform across the surface. Furthermore, the image data (chloroplast sparse / dense image) included in one learning data set is generated from the photographed data by the same method (arithmetic formula). In this way, the same or similar learning data and discrimination data are photographed under the same conditions, enabling accurate discrimination.

[0041] If images are captured under multiple shooting conditions with different light irradiation conditions, the light source spectrum and spectral sensitivity curve may differ. According to the above discussion of the model spectrum, when the light irradiation conditions are different, the optimal values ​​of x and y cannot be uniquely determined, which is inefficient. Therefore, more accurate discrimination is possible by using the same or similar shooting conditions for the learning data and the discrimination data.

[0042] Furthermore, the above discussion of the model spectrum is premised on the assumption that the leaves are photographed so that the healthy (mesophyll) parts, the yellowed parts, and the vein parts can be clearly distinguished. The present invention is a technology for further improving the discrimination accuracy of the discrimination model M on the assumption that high-quality plant leaf photographic data is used. For example, by adopting a high-precision photographic environment such as a transmission optical system, it is possible to photograph high-quality images of plant leaves.

[0043] <Configuration of the Generation Device> The configuration of a generating device 1 that generates a discriminant model M will be described with reference to Fig. 7. As shown in Fig. 7, the generating device 1 includes an arithmetic circuit 11, an input device 12, an output device 13, a communication circuit 14, and a storage device 15. For example, the generating device 1 is a personal computer or an information processing device.

[0044] The arithmetic circuit 11 is a controller that controls the entire generation device 1. For example, the arithmetic circuit 11 realizes various processes related to the generation of the discriminant model M by reading and executing the generation program P1 stored in the storage device 15. The arithmetic circuit 11 may be various processors such as a CPU, an MPU, a GPU, an FPGA, a DSP, an ASIC, or a dedicated hardware circuit.

[0045] The input device 12 is used for user operations and data input. The input device 12 can be, for example, an operation button, a keyboard, a mouse, a touch panel, a microphone, etc. The output device 13 is used for outputting processing results and data. The output device 13 can be, for example, output means such as a display, a speaker, etc.

[0046] The communication circuit 14 performs data communication with other devices. The data communication is performed wired and / or wirelessly, for example, according to a known communication standard. For example, wired data communication may be performed by using a communication controller of a semiconductor integrated circuit that operates in compliance with the Ethernet (registered trademark) standard and / or the USB (registered trademark) standard as the communication circuit 14. Wireless data communication may be performed by using a communication controller of a semiconductor integrated circuit that operates in compliance with the IEEE802.11 standard for wireless LAN (Local Area Network) and / or the fourth / fifth / sixth generation mobile communication system, so-called 4G / 5G / 6G, for mobile communication, as the communication circuit 14.

[0047] The storage device 15 is a recording medium for recording various information. The storage device 15 is realized, for example, by a RAM, a ROM, a flash memory, an SSD (Solid State Drive), a hard disk drive, other storage devices, or an appropriate combination of these. The storage device 15 stores the generation program P1, the shooting data D11, the light intensity data D12, the learning data D13, the discrimination model M, etc. The generation program P1 is a computer program executed by the arithmetic circuit 11. By executing the generation program P1, various processes for generating the discrimination model M are performed.

[0048] The photographed data D11 is image data obtained by photographing the leaves of a plant.

[0049] The light intensity data D12 is data of an R image, a G image, and a B image generated from the shooting data D11.

[0050] The learning data D13 includes a plurality of pairs of chloroplast sparse / dense images and labels indicating the corresponding growth states of plants, as described above with reference to Fig. 6A. The chloroplast sparse / dense images are generated from the light intensity data D12.

[0051] As briefly described with reference to FIGS. 6A and 6B, the discriminant model M is generated by the generating device 1 using the learning data D13.

[0052] 7 shows an example in which the generating device 1 is realized by one information processing device. However, the generating device 1 may be realized by multiple information processing devices. For example, at least a part of the multiple processes to be executed may be executed by an arithmetic circuit. Also, for example, at least a part of the multiple data may be stored in different storage devices.

[0053] The generating device 1 learns using a chloroplast sparse / dense image with a label indicating each state, and can generate a discrimination model M that distinguishes between "yellowing caused by disease" and "yellowing caused by factors other than disease" described above in relation to Table 1. That is, "yellowing caused by disease" and "yellowing caused by factors other than disease" are difficult to distinguish by visual inspection of a color image alone. However, by using a chloroplast sparse / dense image indicating the chloroplast sparse / dense determined according to the amount of chlorophyll as learning data, it becomes possible to distinguish between "the degree of yellowing caused by disease" and "the degree of yellowing caused by factors other than disease", which are difficult to distinguish by visual inspection.

[0054] <Configuration of the discrimination device> The configuration of a discrimination device 2 that discriminates the growth state of a plant using a discrimination model M will be described with reference to Fig. 8. As shown in Fig. 8, the discrimination device 2 includes an arithmetic circuit 21, an input device 22, an output device 23, a communication circuit 24, and a storage device 25. For example, the discrimination device 2 is a personal computer or an information processing device.

[0055] The arithmetic circuit 21, the input device 22, the output device 23, the communication circuit 24 and the memory device 25 are realized by specific means similar to those of the arithmetic circuit 11, the input device 12, the output device 13, the communication circuit 14 and the memory device 15 described above with reference to Figure 8, respectively.

[0056] The storage device 25 stores a discrimination program P2, a discrimination model M, shooting data D21, light intensity data D22, discrimination data D23, and the like.

[0057] The discrimination program P2 is a computer program executed by the arithmetic circuit 21. When the discrimination program P2 is executed, various processes for discrimination are performed.

[0058] The discriminant model M is a trained model generated by the generating device 1 described above.

[0059] The photographed data D21 is image data of the leaves of a plant whose growth state is to be determined.

[0060] The light intensity data D22 is data of an R image, a G image, and a B image generated from the shooting data D21.

[0061] The discrimination data D23 is a chloroplast sparse / dense image obtained from the photographic data D21.

[0062] 8 shows an example in which the discrimination device 2 is realized by one information processing device. However, the discrimination device 2 may be realized by a plurality of information processing devices.

[0063] The discrimination device 2 shown in Fig. 8 may be realized by the same device as the generation device 1 described above with reference to Fig. 7. That is, the generation of the discriminant model M and the discrimination of the plant growth state using the discriminant model M may be executed by the same information processing device.

[0064] <Process for generating discriminant models> The process of generating the discriminant model M executed by the generating device 1 will be described with reference to the flowchart shown in FIG.

[0065] The arithmetic circuit 11 first acquires photographic data D11 of leaves of a plurality of plants (S01). The arithmetic circuit 11 stores the acquired photographic data D11 in the storage device 15. Here, each photographic data D11 may be associated in advance with a label indicating the growth state of the plant. For example, each photographic data D11 is associated with a label such as "healthy", "early stage of illness", "middle stage of illness", or "end stage of illness". In addition, each photographic data D11 is associated with a label indicating the growth state of the plant and stored in the storage device 15. Note that when the photographic data D11 acquired in step S01 is not associated with a label indicating the growth state of the plant, the arithmetic circuit 11 executes a process of labeling each photographic data D11.

[0066] The arithmetic circuit 11 generates light intensity data D12 using the photographic data D11 acquired in step S01 (S02). Specifically, the arithmetic circuit 11 decomposes each photographic data D11 into an R component, a G component, and a B component, and generates an R image indicating the intensity of red light, a G image indicating the intensity of green light, and a B image indicating the intensity of blue light, thereby generating a plurality of light intensity data D12. The arithmetic circuit 11 also stores the generated light intensity data D12 in the storage device 15.

[0067] The arithmetic circuit 11 generates a chloroplast sparse / dense image using the light intensity data D12 generated in step S02 (S03). The arithmetic circuit 11 generates a chloroplast sparse / dense image, for example, by applying the following formula (1.3) to each corresponding pixel of the image. The processing from step S02 to step S03 is so-called preprocessing. Here, if other preprocessing is required before or after the processing of step S02 or step S03, the other preprocessing may be executed. For example, one example of the preprocessing may be processing such as image cropping or image rotation. (R image x + G image x y) / (x + y) - B image (1.3) (where 0≦x, 0≦y)

[0068] In one discrimination model, the same arithmetic formula is used to obtain a chloroplast density image. Here, in formula (1.3) or the above formula (1.2), all combinations of x and y that result in the same formula after reduction are defined as being the same. In other words, when the ratio of x to y is the same, they are defined as being the same. For example, combinations where x:y is 1:2, 2:4, and 3:6 are defined as being the same.

[0069] The arithmetic circuit 11 associates a corresponding label with each chloroplast sparse / dense image generated in step S03 to generate learning data D13 (S04). The arithmetic circuit 11 also stores the generated learning data D13 in the storage device 15.

[0070] The arithmetic circuit 11 generates a discrimination model M using the learning data D13 generated in step S04 (S05). Specifically, the arithmetic circuit 11 learns the relationship between the image showing the density of chloroplasts included in the learning data D13 and the growth state of the plant to generate the discrimination model M. The arithmetic circuit 11 also stores the generated discrimination model M in the storage device 15.

[0071] In this manner, the generating device 1 can generate a discriminant model M that discriminates the growth state of a plant leaf from an image of the plant leaf.

[0072] <Discrimination process using discriminant model> The process of discriminating the growth state of a plant using the discrimination model M executed by the discriminator 2 will be described with reference to the flowchart shown in FIG.

[0073] The arithmetic circuit 21 first acquires photographic data of the leaves of the plant whose growth state is to be determined (S11). The arithmetic circuit 21 stores the acquired photographic data D21 in the storage device 25.

[0074] The arithmetic circuit 21 generates light intensity data D22 using the photographing data D21 acquired in step S11 (S12). The light intensity data D22 can be generated by the same process as step S02 described above with reference to Fig. 6. The arithmetic circuit 21 also stores the generated light intensity data D22 in the storage device 25.

[0075] The arithmetic circuit 21 generates a chloroplast sparse / dense image using the light intensity data D22 generated in step S12 (S13). The chloroplast sparse / dense image can be generated by the same process as step S03 described above with reference to Fig. 9. The arithmetic circuit 21 also stores the generated chloroplast sparse / dense image in the storage device 25 as discrimination data D23.

[0076] The arithmetic circuit 21 uses the discrimination data D23 generated in step S13 as an input to the discrimination model M to discriminate the growth state of the target plant (S14).

[0077] The arithmetic circuit 21 outputs the result of the determination in step S14 to the output device 23 (S15).

[0078] In this manner, the discrimination device 2 can use the discrimination model M to discriminate the growth state of a plant leaf from an image of the plant leaf.

[0079] 《Determining x and y to be used in formula (1.3)》 The optimal values ​​of x and y used in formula (1.3) can be determined by feeding back the discrimination accuracy obtained at that time and optimizing it. For example, the determination of the values ​​of x and y used in formula (1.3) may be executed in the generating device 1. At this time, the arithmetic circuit 11 can determine the values ​​of x and y using a method used in general optimization problems. The data used to determine the values ​​of x and y is, for example, data associated with "imaging data" that has been labeled by an expert with "health", "early stage of illness", "middle stage of illness", and "end stage of illness".

[0080] Second Embodiment A generating device 1A according to the second embodiment will be described with reference to FIGS. 11 to 13. In the above-described first embodiment, an example has been described in which the generating device 1 generates a discriminant model M using supervised learning. Specifically, the generating device 1 generates a discriminant model M using learning data D13 including images that have been labeled in advance by preprocessing. In contrast, the generating device 1A according to the second embodiment clusters images that have not been labeled using unsupervised learning. Furthermore, upon receiving the labels of each cluster generated by clustering, the generating device 1A generates learning data to generate a discriminant model M. In the second embodiment, the same configurations and processes as those of the generating device 1 according to the first embodiment may be denoted by the same reference numerals and descriptions thereof may be omitted.

[0081] <Configuration of the Generation Device> 11, in the generating device 1A according to the second embodiment, an arithmetic circuit 11 is realized by an information processing device including an input device 12, an output device 13, a communication circuit 14, and a storage device 15. The storage device 15 of the generating device 1A according to the second embodiment stores a clustering program P3, a generating program P1, photographed data D11, light intensity data D12, clustering data D31, clustering result data D32, learning data D13, and a discriminant model M, as shown in FIG.

[0082] The clustering program P3 is a computer program executed by the arithmetic circuit 11. When the clustering program P3 is executed, a clustering process is performed in the generation device 1A. The clustering program P3 generates clusters by using hard clustering, such as hierarchical clustering and non-hierarchical clustering, or soft clustering.

[0083] The clustering data D31 is data including simple images with no flags attached, as shown in Fig. 12A. The clustering data may be, for example, chloroplast sparse / dense images generated from a plurality of pieces of photography data D11.

[0084] The clustering result data D32 is a result of clustering the clustering data D31. For example, as shown in FIG. 12B, a plurality of clusters are obtained from the clustering data D31 by unsupervised learning, and the clustering result data D32 is obtained. For example, FIG. 12B shows an example in which clusters A to E are obtained using hard clustering in which one piece of data belongs to only one cluster. In addition, when using soft clustering in which one piece of data is allowed to belong to multiple clusters, labels of multiple items such as "disease and nutrient deficiency" may be assigned. Furthermore, the relative distribution of the multiple labels may be evaluated. In general, natural phenomena including yellowing of leaves are caused by multiple factors.

[0085] The learning data D13 is data generated by labeling each of the clusters A to E of the clustering result data D32 by the user. The learning data D13 is used in generating the discriminant model M. FIG. 12C shows an example in which the labels "disease", "health", and "nutritional deficiency" are respectively attached to the clusters A to C obtained in FIG. 12B. Note that it may be impossible to identify the labels of all the clusters. If it is impossible to identify them, the label "unknown" may be attached as shown in FIG. 12C. In addition, there may be cases where there are multiple patterns of disease manifestation, such as citrus greening disease, or where there are multiple gene lineages of the pathogen. In such cases, for example, the cluster with the "disease" label may be subdivided and hierarchized, such as "disease pattern A" and "disease pattern B". Note that the subdivision and hierarchization of the label are not limited to the "disease" label.

[0086] <Processing for generating clustering and discriminant models> The process of generating clustering and discriminant model M executed by generating device 1A will be described with reference to the flowchart shown in Fig. 13. Specifically, the process enclosed by dashed lines in the flowchart shown in Fig. 13 (S21 to S23) is the clustering process.

[0087] The arithmetic circuit 11 first acquires photographic data D11 of leaves of a plurality of plants (S01). The arithmetic circuit 11 stores the acquired photographic data D11 in the storage device 15. Note that in the generation device 1A, labels are assigned after clustering, so no labels are assigned in step S01.

[0088] The arithmetic circuit 11 generates light intensity data D12 using the photographing data D11 acquired in step S01 (S02). The method of generating the light intensity data D12 is the same as that of the generating device 1 according to the first embodiment. The arithmetic circuit 11 also stores the generated light intensity data D12 in the storage device 15.

[0089] The arithmetic circuit 11 generates a chloroplast sparse / dense image using the light intensity data D12 generated in step S02 (S03). The method of generating the chloroplast sparse / dense image is the same as that of the generating device 1 according to the first embodiment. The arithmetic circuit 11 also stores the generated multiple chloroplast sparse / dense images in the storage device 15 as clustering data D31.

[0090] The arithmetic circuit 11 performs clustering by unsupervised learning using the clustering data D31 stored in step S03 (S21). The arithmetic circuit 11 stores the result of the clustering in the storage device 15 as clustering result data D32.

[0091] The arithmetic circuit 11 receives the label of each cluster obtained in step S21 (S22). For example, the label of each cluster is input by the user via the input device 12.

[0092] The arithmetic circuit 11 associates each chloroplast sparse / dense image in the clustering data D31 with the corresponding label received in step S22 to generate training data D13 (S23). The arithmetic circuit 11 also stores the generated training data D13 in the storage device 15.

[0093] The arithmetic circuit 11 generates a discriminant model M using the learning data D13 generated in step S04 (S05). The method of generating the discriminant model M is the same as that of the generating device 1 according to the first embodiment. The arithmetic circuit 11 also stores the generated discriminant model M in the storage device 15.

[0094] In this way, the generation device 1A performs clustering from images of plant leaves. The generation device 1A can also generate a discrimination model M that discriminates the growth state of plant leaves using the results of the clustering.

[0095] In the above description, an example has been described in which the generation device 1A executes all of the generation of the chloroplast sparse / dense image, the clustering, and the generation of the discriminant model M. However, some of the processes may be executed by another information processing device. Furthermore, each process may be executed by a different information processing device. For example, clustering may be executed using a chloroplast sparse / dense image generated by another information processing device. Furthermore, clustering and the generation of the discriminant model M may be executed by different information processing devices.

[0096] The discrimination by the discrimination model M generated as described above in the second embodiment is performed in the same manner as described above in the first embodiment. In FIG. 12C, an example was described in which the labels "disease", "health", "nutrition deficiency", "unknown 1", and "unknown 2" are attached to each cluster. In the discrimination model M generated by the learning data D13 generated in this way, the discrimination result may be "disease", "health", or "nutrition deficiency", or may be "unknown". Furthermore, when the label to be attached to the cluster that was "unknown" is later determined, the correct label may be attached. This enables the discrimination model M to make more accurate discrimination.

[0097] <Variation 1> In the above-mentioned first embodiment, an example was described in which it was determined whether the plant was healthy or sick with yellowing. In addition, the determination was made using a discrimination model trained using labels indicating whether the plant was in the early stage, middle stage, or final stage of illness. However, the discrimination model may of course use learning data with other labels attached. For example, the discrimination model may be generated using learning data with labels such as "nutritional deficiency" and "overwatering" in addition to or instead of the above labels.

[0098] <Variation 2> In the above example, the orientation of the leaves in the photographed data is not mentioned. On the other hand, there are plants such as bamboo leaves whose vein orientation is easy to see. In such cases, the orientation of the leaves can be adjusted to a predetermined orientation in the learning data and the classification data to improve the classification accuracy.

[0099] <Variation 3> In the second modification, the leaf orientation is adjusted for plants in which the direction of the veins is easy to see. However, in the case of leaves in which the direction of the veins is not uniform, one piece of photographed data may be rotated to increase the number of learning data. By learning the rotated image data for one piece of photographed data in this way, discrimination is possible without depending on the photographed direction of the leaves, and it is possible to easily capture the photographed data for discrimination. Even if the image has a gradation, learning can be performed without depending on the direction of the gradation. Therefore, overlearning due to the gradation of the learning data can be prevented, and the accuracy of discrimination can be improved.

[0100] Variation 4 In the above example, the photographing data D11 is described as an example of photographing using a "transmissive optical system with substantially uniform brightness". However, the photographing optical system used to photograph the photographing data D11 is not limited to a transmissive optical system. Specifically, the photographing data D11 may be data capable of separating the yellowed portion, the healthy portion, and the leaf vein portion. In addition, the photographing optical system used to photograph the photographing data D11 may be a photographing optical system capable of photographing each sample under uniform conditions. For example, it may be a transmissive optical system that can be carried outdoors to the field, or a photographing optical system that has a uniform indoor photographing environment.

[0101] Other Modifications In the above example, an example of determining the state of a plant by machine learning has been described, but a generative AI can also be realized using, for example, a generative adversarial network based on the same learning data. Specifically, a chloroplast sparse / dense image is trained, a non-existent chloroplast sparse / dense image is generated under any condition, and an RGB image of the leaf is restored in a pseudo manner by coloring, for example, from high to low pixel values ​​from yellow to green. Such a generative AI can generate images of desired conditions, such as an image of a virtual "healthy" leaf, an image of a virtual "early stage of disease + mid-stage of nutritional deficiency" leaf, etc. By using such a generative AI, for example, in the case of citrus greening disease, it is possible to predict what kind of yellowing will occur when citrus fruits in an area where the disease has not yet spread are affected by the disease, which can be useful for early detection.

[0102] Experimental Example 1 The experimental results of generating the discrimination model M described above are explained below. In the experiment, 841 pieces of photographed data including healthy plant leaves and diseased plant leaves were prepared. Each piece of photographed data was an image photographed using a transmission optical system incorporating a surface light source with a substantially uniform luminance across the surface. The plant was a citrus leaf such as Shikuwasa, and the disease was Citrus Greening Disease. As shown in FIG. 14, each piece of photographed data was labeled as "healthy," "early stage of disease," "middle stage of disease," or "late stage of disease." In addition, the photographed data was divided into group 1 (early stage of photographing), group 2 (middle stage of photographing), and group 3 (late stage of photographing) according to the photographing time, and learning data and evaluation data were selected from each group and evaluated. The groups were created so that images of the same leaf would not be included in different groups.

[0103] As shown in FIG. 15, learning and evaluation were performed three times with different combinations, and this was repeated three times. Specifically, in the first learning, learning was performed based on the photographic data of groups 2 and 3, and evaluation was performed based on the photographic data of group 1. In addition, in the second learning, learning was performed based on the photographic data of groups 1 and 3, and evaluation was performed based on the photographic data of group 2. In the third learning, learning was performed based on the photographic data of groups 1 and 2, and evaluation was performed based on the photographic data of group 3. In the fourth and seventh times, evaluation was performed using the same group as in the first time. In the fifth and eighth times, evaluation was performed using the same group as in the second time. In the sixth and ninth times, evaluation was performed using the same group as in the first time.

[0104] Specifically, the generation and verification of the discrimination model was performed using images obtained as a result of various processes or calculations. Here, in addition to the shooting data itself (Table 3.1), we show the results with high evaluation accuracy. Specifically, we show the results of generating and verifying the discrimination model using an image calculated using the R image and B image (R image - B image) (Table 3.2), an image calculated using the G image and B image (G image - B image) (Table 3.3), an image obtained by (G image + R image) / 2-B image (Table 3.4), and an image obtained by rotating the images in Table 3.3 by 45° each time and increasing the size by 8 times (Table 3.5).

[0105] In the table below, "early stage illness" is referred to simply as "early stage," "middle stage illness" is referred to simply as "middle stage," and "terminal stage illness" is referred to simply as "terminal stage."

[0106] In the following, the "correct answer rate" is the correct answer rate of the classification result classified into each label from the early to middle stages of the disease. In addition, the "disease leaf correct answer rate" is the correct answer rate of the classification result of whether it is healthy or diseased. For example, the boundary between the early and middle stages of the disease is vague. Also, the boundary between the middle and end stages of the disease is vague. Therefore, the correct answer rate is the correct answer rate when the early to end stages of the disease are regarded as one discrimination result of "disease", and if it is classified into any of them, it is regarded as correct. In the following, for the sake of simplicity, it will be described as "improved accuracy of evaluation result" when both the "disease leaf correct answer rate" and the "correct answer rate" tend to increase, focusing on the "disease leaf correct answer rate".

[0107] Furthermore, the results below show the combined results of the first to ninth learning and evaluations performed with the combinations shown in Figure 15.

[0108] (Table 3.1) TIFF2025085455000005.tif41141

[0109] Table 3.2 shows an example of learning and evaluation using images obtained by applying the following formula (2.1) to each pixel of R and B images obtained from photographic data. R image-B image (2.1)

[0110] The accuracy of the evaluation results has been improved compared to the case of the shooting data in Table 3.1. (Table 3.2) TIFF2025085455000006.tif36123

[0111] Table 3.3 shows an example of learning and evaluation using images obtained by applying the following formula (2.2) to each pixel of the G and R images obtained from the photographic data. G image-B image (2.2)

[0112] The accuracy of the evaluation results has been improved compared to the case of the shooting data in Table 3.1. (Table 3.3) TIFF2025085455000007.tif41141

[0113] Table 3.4 shows an example of learning and evaluation using the G and R images obtained from the shooting data, for each pixel, using the images obtained by the following formula (2.3). (G image + R image) / 2-B image (2.3) Equation (2.3) is the case where the values ​​of x and y in the above equation (1.3) are (1, 1).

[0114] The accuracy of the evaluation results has been improved compared to the case of the shooting data in Table 3.1. (Table 3.4) TIFF2025085455000008.tif41141

[0115] Table 3.5 shows an example of learning and evaluation using images obtained by rotating the images obtained by the above formula (2.2) by 45° and increasing the size by 8 times.

[0116] The accuracy of the evaluation results is improved compared to the case of the photographic data in Table 3.1. Also, the evaluation results shown in Table 3.5 show higher accuracy than all other evaluation results in Tables 3.2 to 3.4 above. (Table 3.5) TIFF2025085455000009.tif41141

[0117] Experimental Example 2 In Experimental Example 2, images of hue (H), saturation (S), brightness (V), R, G, and B were used. In Experimental Example 2, the learning and evaluation were repeated a total of nine times for the group shown in FIG. 14, as shown in FIG. 15, and the results were totaled. In addition, Experimental Example 2 was compared with the results of Table 3.5, which had the highest accuracy of the evaluation results in Experiment 1. The evaluation results for the H, S, V, R, G, and B images all had lower accuracy than the results of Table 3.5. As an example, only the evaluation results for the H image are explained below.

[0118] Table 4 shows an example of learning and evaluation using images that were converted from color image data into hue images. The accuracy of the evaluation results is lower than in the case of Table 3.5. Also, the accuracy rate of diseased leaves has worsened overall compared to the results of Table 3.1, which is the original image. (Table 4) TIFF2025085455000010.tif41141

[0119] It is believed that the poor evaluation results were due to the H image treating the information on dead leaf parts as the same as other yellowed and healthy parts. This shows that although yellowing is a difference in color, using a discrimination model that uses H images, which show differences in color, does not necessarily improve the accuracy of discrimination. It also shows that a discrimination model that uses images obtained with formula (1.3) is useful.

[0120] Experimental Example 3 In experimental example 3, the change in discrimination accuracy when the values ​​of x and y in formula (1.3) were changed was evaluated. Experimental example 3 shows the discrimination accuracy when an image obtained by formula (1.3) was generated from the image used in experimental example 1, and a discrimination model was generated. Table 5 shows the results of experimental example 3. Table 5 shows representative values ​​of the diseased leaf correct identification rate when the ratio of R in R:G (x / (x+y)) is each value. The representative value of the diseased leaf correct identification rate is the total numerical value of the diseased leaf correct identification rate. (Table 5) TIFF2025085455000011.tif67146

[0121] The results of Table 5 are shown in Figure 16. From the trend of the graph shown in Figure 16, it can be seen that the smaller the ratio of R, the higher the accuracy of discrimination. It is possible to adopt x = 0, y = 1, which provides the highest accuracy of discrimination. Also, for example, the values ​​of x and y may be x = 0.15, y = 0.85, which are the average values ​​of the ratio of R between 0.0 and 0.3, at which the accuracy of discrimination is 87% or higher.

[0122] (1) A generation method of the present disclosure is a generation method of a discrimination model for discriminating a growth state of a plant, the generation method being executed by a computer including an arithmetic circuit that can access a storage device, the method including: The storage device includes: A chloroplast density image is obtained from images of leaves of a plurality of learning plants, and the image reflects the difference in the absorption spectra of chlorophyll and carotenoid contained in the plants. Executed by the arithmetic circuit, Reading the chloroplast sparse / dense image from the storage device; The chloroplast sparse / dense image is used as learning data for machine learning to generate a discrimination model for the growth state of a plant.

[0123] (2) In the generation method of (1), the chloroplast sparse / dense image is an image obtained by using two or more of an R image, a G image, and a B image obtained from an image of a plant leaf, for each corresponding pixel value, according to the following formula (1): (Formula for combining R and G images with arbitrary weighting) - (Formula using B image) (1) In the formula (1), the value of a pixel of an unused image among the R image, the G image, and the B image may be 0.

[0124] (3) In the production method of (2), the formula (1) may be the following formula (2). (R image x + G image x y) / (x + y) - B image (2) (where 0≦x, 0≦y)

[0125] (4) In the production methods of (1) to (3), the growth state of the plant may include at least one selected from the group consisting of a healthy state, a diseased state, a malnutrition or overnutrition state, a poor growth environment state, and an insect damage state.

[0126] (5) The generating methods of (1) through (4) may include selected levels of varying degrees.

[0127] (6) The generation method of (3) may include a step of optimizing the values ​​of x and y in the formula (2) to a combination that provides the highest discrimination accuracy.

[0128] (7) In the generation methods of (1) to (6), the image of the plant leaf may be acquired using a transmissive optical system incorporating a surface light source having a substantially uniform brightness over a surface.

[0129] (8) In the generation methods of (1) to (7), the machine learning may include at least one of supervised learning and unsupervised learning.

[0130] (9) In the generation method according to (1) to (8), the storage device comprises: storing a label indicating a state of the plant in association with each of the chloroplast sparse / dense images; The arithmetic circuit includes: reading out the label together with the chloroplast sparse / dense image from the storage device as learning data for supervised learning; The relationship between the multiple chloroplast sparse / dense images included in the training data and the corresponding labels may be learned through machine learning, and a discrimination model may be generated that outputs the state of a new plant as a discrimination result for the chloroplast sparse / dense image of the new plant.

[0131] (10) In the generation method according to (1) to (9), the arithmetic circuit reading out a plurality of the chloroplast sparse / dense images from the storage device as learning data for unsupervised learning; generating a plurality of clusters for each characteristic of a plant state based on a result of analyzing the characteristics of the plurality of chloroplast sparse / dense images of the learning data; When each cluster is assigned a label indicating the state of the plant, the multiple chloroplast sparse / dense images contained in the cluster and the labels assigned to the cluster may be used as training data for supervised learning, and a discrimination model may be generated that learns the relationship between the multiple chloroplast sparse / dense images and the corresponding labels, and outputs the state of the new plant as a discrimination result for the chloroplast sparse / dense image of the new plant.

[0132] (11) A method for determining a state of a plant according to the present disclosure is executed by a computer including an arithmetic circuit that can access a storage device, the method comprising: Executed by the arithmetic circuit, Reading out an image of a leaf of a plant for identification from the storage device; A chloroplast density image showing the density of the chloroplasts of the plant is calculated from the image; The calculated chloroplast sparse / dense image is used as an input to the discrimination model generated by the generation method of claim 1 to discriminate the state of the plant.

[0133] (12) The computer program disclosed herein executes any one of the generation methods (1) to (10).

[0134] (13) The computer program disclosed herein executes the determination method of (11).

[0135] (14) A generation device according to the present disclosure includes an arithmetic circuit that can access a storage device, and is a generation device for a discrimination model that discriminates a state of a plant, The storage device includes: A chloroplast density image is obtained from images of leaves of a plurality of learning plants, and the image reflects the difference in the absorption spectra of chlorophyll and carotenoid contained in the plants. The arithmetic circuit Reading the chloroplast sparse / dense image from the storage device; The chloroplast sparse / dense image is used as learning data for machine learning to generate a discrimination model for the growth state of a plant. [Explanation of symbols]

[0136] 1 generator 2 Discrimination device 11,21 Arithmetic circuit 12,22 Input devices 13,23 Output device 14,24 Communication circuits 15,25 Storage device D11 Shooting data D12 Light Intensity Data D13 Learning data D21 shooting data D22 Light Intensity Data D23 Discrimination data M Discriminant Model P1 Generator P2 discrimination program

Claims

1. A method for generating a discrimination model for discriminating a growth state of a plant, the method being executed by a computer including an arithmetic circuit that can access a storage device, the method comprising: The storage device includes: A chloroplast density image is obtained from images of leaves of a plurality of learning plants, and the image reflects the difference in the absorption spectra of chlorophyll and carotenoid contained in the plants. Executed by the arithmetic circuit, Reading the chloroplast sparse / dense image from the storage device; generating a discrimination model for a plant growth state using the chloroplast sparse / dense image as learning data for machine learning; Generation method.

2. The chloroplast sparse / dense image is an image obtained by using two or more of an R image, a G image, and a B image obtained from an image of a plant leaf, for each corresponding pixel value, according to the following formula (1): (Formula for blending R and G images with arbitrary weighting) - (Formula using B image) (1) In the formula (1), the value of a pixel of an unused image among the R image, the G image, and the B image is 0. The method of claim 1 .

3. The above formula (1) is the following formula (2): (R image x x + G image x y) / (x + y) - B image (2) (where 0≦x, 0≦y) The method of claim 2.

4. The growth state of the plant includes at least one selected from the group consisting of a healthy state, a diseased state, a malnutrition or overnutrition state, a poor growth environment state, and an insect damage state; The method of claim 1 .

5. The growth state includes a selected level among different levels of severity. The method of claim 1 .

6. Optimizing the values ​​of x and y in the formula (2) to a combination that maximizes the discrimination accuracy. The method of claim 3.

7. To obtain the image of the plant leaf, a transmission optical system incorporating a surface light source with a substantially uniform brightness over the surface is used. The method of claim 1 .

8. The machine learning includes at least one of supervised learning and unsupervised learning, The method of claim 1 .

9. The storage device includes: storing a label indicating a growth state of the plant in association with each of the chloroplast sparse / dense images; The arithmetic circuit includes: reading out the label together with the chloroplast sparse / dense image from the storage device as learning data for supervised learning; a discrimination model that learns a relationship between the plurality of chloroplast sparse / dense images included in the learning data and the corresponding labels by machine learning and outputs a growth state of the new plant as a discrimination result for the chloroplast sparse / dense image of the new plant; The method of claim 1 .

10. The arithmetic circuit includes: reading out a plurality of the chloroplast sparse / dense images from the storage device as learning data for unsupervised learning; generating a plurality of clusters for each characteristic of a plant growth state based on a result of analyzing the characteristics of the plurality of chloroplast sparse / dense images of the learning data; When a label indicating the growth state of the plant is assigned to each cluster, the plurality of chloroplast sparse / dense images included in the cluster and the label assigned to the cluster are used as learning data for supervised learning, and a relationship between the plurality of chloroplast sparse / dense images and the corresponding labels is learned, and a discrimination model is generated that outputs the growth state of the new plant as a discrimination result for the chloroplast sparse / dense image of the new plant. The method of claim 1 .

11. A method for determining a growth state of a plant, the method being executed by a computer including an arithmetic circuit that can access a storage device, the method comprising: Executed by the arithmetic circuit, Reading out an image of a leaf of a plant for identification from the storage device; A chloroplast density image showing the density of the chloroplasts of the plant is calculated from the image; The calculated chloroplast density image is input to the discrimination model generated by the generation method of claim 1 to discriminate the growth state of the plant. Discrimination method.

12. A computer program for carrying out the method according to any one of claims 1 to 10.

13. A computer program product for executing the determination method according to claim 11.

14. A device for generating a discrimination model for discriminating a growth state of a plant, the device including an arithmetic circuit capable of accessing a storage device, the device comprising: The storage device includes: A chloroplast density image is obtained from images of leaves of a plurality of learning plants, and the image reflects the difference in the absorption spectra of chlorophyll and carotenoid contained in the plants. The arithmetic circuit Reading the chloroplast sparse / dense image from the storage device; generating a discrimination model for a plant growth state using the chloroplast sparse / dense image as learning data for machine learning; generator.

Images