Method and apparatus for diagnosing efficacy and mechanism of action of herbicides based on plant image analysis

Abstract

The present invention relates to a method and an apparatus for diagnosing the efficacy and mechanism of action of herbicides based on plant image analysis. Since the efficacy and mechanism of action of herbicides can be quickly diagnosed within several hours to several days after herbicide treatment without destroying plants, the present invention can be advantageously used for rapid efficacy evaluation and diagnosis of mechanism of action according to high-throughput screening of novel herbicide candidate materials.

Description

Method and diagnostic device for diagnosing the efficacy and mechanism of action of herbicides based on plant image analysis

[0001] The present invention relates to a method and device for diagnosing the efficacy and mechanism of action of herbicides based on plant image analysis.

[0002] This invention was made with the support of the Korean government under Project No. RS-2024-0039786 of the Rural Development Administration, "Development of Crop Protection Agent Efficacy Diagnosis Technology and Screening Technology."

[0003] Recently, plant imaging analysis has been utilized as a core technology for plant phenotype analysis and diagnosis in plant and agricultural sciences. One of the advantages of plant imaging analysis is that it enables non-destructive, high-throughput screening of plant responses. Recent advancements in imaging technology have made it possible to acquire spectral images of plants, allowing for the investigation of a large number of plant phenotypes in a more efficient and practical manner.

[0004] RGB images are commonly used to analyze vegetation indices related to plant growth and leaf length development. In particular, they are highly accessible and have the potential for diverse applications because they provide useful and high-resolution results even with commercial cameras installed on mobile devices. RGB images are known to provide color-based features and can be used to evaluate plant growth, health, and nitrogen status of cultivated land using vegetation indices such as NDI and ExG.

[0005] Plants absorb carbon dioxide necessary for photosynthesis through the stomata of their leaves, release oxygen—a product of photosynthesis—and regulate their body temperature through transpiration. Consequently, plant stomata respond very sensitively to the surrounding environment to regulate their opening and closing, so stomatal conductance responds very sensitively to the plant's physiological state. When stomatal conductance decreases, transpiration decreases, causing the plant's body temperature to increase. It is known that the plant's stomatal conductance or photosynthetic response can be indirectly diagnosed by measuring this plant temperature using an IR (infrared) thermal imaging camera.

[0006] Fluorescence emitted from plant chlorophyll is the re-emission of some light energy that was not used in the initial photochemical reactions of photosynthesis. It exhibits an inverse pattern, where fluorescence values ​​increase when photochemical reactions in plants decrease and decrease when they do. Therefore, it is known that changes in the structure and function of the photosynthetic apparatus according to the physiological state of plants can be sensitively detected through the measurement and analysis of fluorescence values ​​using a chlorophyll fluorescence camera.

[0007] Meanwhile, it is widely known that herbicides with diverse mechanisms of action induce various visual symptoms in plants. According to previous reports, while the use of multi-hyperspectral imaging analysis has been proven useful for investigating herbicide stress, it has not been suitable for the large-scale screening of herbicide candidates because acquiring and analyzing such spectral images remains technically difficult and costly.

[0008] Therefore, studies on plant responses to herbicides have been conducted using RGB imaging, chlorophyll fluorescence imaging, or IR thermal imaging, which are easier to acquire and analyze. However, despite such research, there have been almost no reported results regarding the accurate diagnosis of herbicide responses based on image analysis, and there is no known research at all on early diagnosis of herbicide responses by analyzing RGB imaging, chlorophyll fluorescence imaging, and IR thermal imaging together. In particular, there has been no research at all on the ability to diagnose the mode of action (MoA) of herbicide candidate substances through such image analysis.

[0009] Accordingly, the inventors completed the present invention by conducting research to diagnose the efficacy and mechanism of action of herbicides early based on plant image analysis.

[0010] One objective of the present invention is to provide a method for diagnosing the efficacy and mechanism of action of a herbicide on a plant, comprising the steps of: treating a plurality of herbicides with known mechanisms of action on a plurality of plant individuals according to their respective mechanisms of action; generating image data from the plant individuals; generating spectral parameter data from the image data; training a statistical algorithm using the parameter data to generate a model for predicting the efficacy and mechanism of action of the herbicide on the plant; treating a herbicide candidate substance with an unknown mechanism of action on a plant to generate image data of the plant treated with the candidate substance, and inputting the spectral parameter data generated from the image data of the plant treated with the candidate substance into the prediction model to diagnose the efficacy and mechanism of action of the herbicide candidate substance on the plant.

[0011] Another objective of the present invention is to provide an apparatus for diagnosing the efficacy and mechanism of action of herbicides on plants, comprising: a herbicide treatment unit that treats a plurality of herbicides with known mechanisms of action to a plurality of plant individuals according to their respective mechanisms of action; an image data generation unit that generates image data from the plant individuals; a spectrum parameter data generation unit that generates spectrum parameter data from the image data; a learning unit that generates a model for predicting the efficacy and mechanism of action of herbicides on plants using the parameter data; and a diagnosis unit that treats a herbicide candidate substance with an unknown mechanism of action to plants to generate image data of the plants treated with the candidate substance, and inputs the image data of the plants treated with the candidate substance into the prediction model to diagnose the efficacy and mechanism of action of the herbicide candidate substance on plants.

[0012] One aspect of the present invention provides a method for diagnosing the efficacy and mechanism of action of a herbicide on a plant, comprising the steps of: treating a plurality of herbicides with known mechanisms of action on a plurality of plant individuals according to their respective mechanisms of action; generating image data from the plant individuals; generating spectral parameter data from the image data; training a statistical algorithm using the parameter data to generate a model for predicting the efficacy and mechanism of action of the herbicide on the plant; treating a herbicide candidate substance with an unknown mechanism of action on a plant to generate image data of the plant treated with the candidate substance, and inputting the spectral parameter data generated from the image data of the plant treated with the candidate substance into the prediction model to diagnose the efficacy and mechanism of action of the herbicide candidate substance on the plant.

[0013] According to one embodiment of the present invention, the mechanism of action includes inhibition of acetyl CoA carboxylase, inhibition of branched amino acid biosynthesis (ALS inhibition), inhibition of aromatic amino acid biosynthesis (EPSP inhibition), inhibition of glutamine synthase, inhibition of photosystem II (D1 Serine 264 binders), inhibition of photosystem II (D1 Histidine 215 binders), inhibition of photosystem I electron transfer (Bipyridylium system), inhibition of chlorophyll biosynthesis (PPO inhibition), inhibition of carotenoid biosynthesis (PDS inhibition), inhibition of carotenoid biosynthesis (HPPD inhibition), inhibition of carotenoid biosynthesis (Lycopene Cyclase), inhibition of DOXPS (1-deoxy-D-xylulose 5-phosphate synthase), inhibition of folic acid biosynthesis (Asulam), inhibition of microtubule assembly, inhibition of mitosis / microtubule formation, and long chain It may be one or more mechanisms of action selected from the group consisting of inhibition of fatty acid (VLCFA) synthesis, inhibition of cell wall (cellulose) synthesis, inhibition of fatty acid thioacetase (TE), inhibition of lipid synthesis, inhibition of pyrimidine biosynthesis (Dihydroorotate dehydrogenase, DHODH), membrane disruption, auxin (indoleacetic acid) analogue, and inhibition of auxin transport.

[0014] According to one embodiment of the present invention, the image data may be generated from one or more images selected from the group consisting of RGB images, chlorophyll fluorescence images and IR thermal images.

[0015] According to one embodiment of the present invention, the spectral parameters are mNDI (modified Normalized Difference Index), NDI (Normalized Difference Index), ExG (Excess Green Index), F v / F m , ΦPSⅡ, F d / F m It may be one or more parameters selected from a group consisting of temperature difference and temperature index.

[0016] According to one embodiment of the present invention, the mNDI, NDI, and ExG can be obtained from the following mathematical formula 1.

[0017]

[0018] [Mathematical Formula 1]

[0019]

[0020] R: Red RGB value

[0021] G: Green RGB value

[0022] B: Blue RGB values

[0023] NDI h NDI of herbicide-treated plants

[0024] NDI0: NDI of plants not treated with herbicides.

[0025]

[0026] According to one embodiment of the present invention, the F v / F m , ΦPSⅡ and F d / F m It can be obtained from the following mathematical formula 2.

[0027]

[0028] [Mathematical Formula 2]

[0029]

[0030] F0: Baseline chlorophyll fluorescence value in dark adaptation state

[0031] F m : Chlorophyll fluorescence value at the point (1 second) when the plant emits maximum fluorescence after exposure to light

[0032] F s : Chlorophyll fluorescence value at the point when the plant returns to standard conditions (60 seconds).

[0033]

[0034] According to one embodiment of the present invention, the temperature difference can be obtained from the following mathematical formula 3.

[0035]

[0036] [Mathematical Formula 3]

[0037]

[0038] T h : Leaf temperature of plants treated with herbicides

[0039] T0: Leaf temperature of plants not treated with herbicides.

[0040]

[0041] According to one embodiment of the present invention, the temperature index can be obtained from the following mathematical formula 4.

[0042]

[0043] [Mathematical Formula 4]

[0044]

[0045] T h : Leaf temperature of plants treated with herbicides

[0046] T0: Leaf temperature of plants not treated with herbicides.

[0047]

[0048] According to one embodiment of the present invention, the plant may be barnyard grass or rapeseed.

[0049] According to one embodiment of the present invention, the barnyard grass may be grown in a multi-well plate.

[0050] According to one embodiment of the present invention, the herbicide is cyhalofop-butyl, metamifop, fenoxaprop-ethyl, fluazifop-P-butyl, sethoxydim, clethodim, benfuresate, mefenacet, S-metolachlor, alachlor, butachlor, napropamide, fenoxasulfone, imazapyr, imazaquin, flucetosulfuron, flazasulfuron, pyrazosulfuron, Nicosulfuron, Imazosulfuron, Pyrimisulfan, Bensulfuron-methyl, Propyrisulfuron, Azimsulfuron, Trifloxysulfuron-sodium, Triafamone, Pyriminobac-methyl, Pyribenzoxim, Bispyribac-sodium, Penoxsulam, Paraquat, Atrazine, Bentazone, Linuron, Dimethamethrine, Propanil, Pyraclonil, Tiafenacil, Bifenox, Oxyfluorfen, Carfentrazone-ethyl, Pentoxazone, Pyraflufen-ethyl, Oxadiargyl,Fluthiacet-methyl, Dicamba, Florpyrauxifen-benzyl, 2,4-D, MCPP, MCPA, Fluroxypyr-meptyl, Triclopyr TEA, Tetflupyrolimet, Pendimethalin, Ethalfluralin, Oryzalin, Dithiopyr, Isoxaben, Indaziflam, Clomazone), Isoxaflutole, Mesotrione, Tefuryltrione, Benzobicyclon, Pyrazolate, It may be one or more herbicides selected from the group consisting of tolpyralate, glyphosate, glufosinate-ammonium, and oxaziclomefone.

[0051] According to one embodiment of the present invention, the statistical algorithm may be a principal component analysis or a machine learning algorithm.

[0052] According to one embodiment of the present invention, the machine learning algorithm may be a Subspace Discriminant or Bagged Trees.

[0053] According to one embodiment of the present invention, the step of generating the image data may be performed at two or more times between 3 hours and 144 hours after the step of processing each of the plurality of plant individuals.

[0054] Another aspect of the present invention provides an apparatus for diagnosing the efficacy and mechanism of action of herbicides on plants, comprising: a herbicide treatment unit for treating a plurality of herbicides with known mechanisms of action to a plurality of plant individuals according to their respective mechanisms of action; an image data generation unit for generating image data from the plant individuals; a spectrum parameter data generation unit for generating spectrum parameter data from the image data; a learning unit for generating a model for predicting the efficacy and mechanism of action of herbicides on plants using the parameter data; and a diagnosis unit for treating a herbicide candidate substance with an unknown mechanism of action to plants to generate image data of the plants treated with the candidate substance, and inputting the image data of the plants treated with the candidate substance into the prediction model to diagnose the efficacy and mechanism of action of the herbicide candidate substance on plants.

[0055] According to the method and device for diagnosing the efficacy and mechanism of action of herbicides based on plant image analysis, the efficacy and mechanism of action of herbicides can be rapidly diagnosed within hours to a few days after herbicide treatment without destroying the plant body; therefore, it can be usefully utilized for rapid efficacy evaluation and mechanism of action diagnosis following high-throughput screening of novel herbicide candidate substances.

[0056] FIG. 1 is a schematic diagram illustrating the overall process of multi-well plate analysis combined with spectral image analysis: (A) preparation of barnyard grass and herbicide treatment, (B) acquisition of spectral images, (C) image analysis process, and (D) segmentation of the acquired images.

[0057] Figure 2 shows (A) RGB, (B) CF, and (C) IR thermal images of crabgrass after 3, 6, 24, 48, 72, and 120 hours after herbicide treatment: (a) glyphosate, (b) paraquat, (c) thiafenacil, (d) isoxaflutol, (e) phenoxulam, (f) glufosinate, and (g) untreated control.

[0058] Figure 3 is a graph showing the changes in (A) mNDI and (B) ExG values ​​over time after treatment with herbicides of various mechanisms of action.

[0059] Figure 4 is a graph showing the change in chlorophyll fluorescence parameters over time after treatment with herbicides of various mechanisms of action: (A) F v / F m , (B) ΦPSⅡ and (C) F d / F m .

[0060] Figure 5 is a graph showing the change in temperature difference over time after treatment with herbicides of various mechanisms of action.

[0061] Figure 6 shows the PCA results for six spectral parameters of crabgrass observed at 3, 6, 24, 48, 72, and 120 hours after treatment with herbicides of various mechanisms of action. Each symbol indicates reproducibility.

[0062] Figure 7 shows the PCA results for six spectral parameters of barnyard grass, synthesized from all data including 3, 6, 24, 48, 72, and 120 hours after treatment with herbicides of various mechanisms of action. Each symbol indicates reproducibility.

[0063] Figure 8 shows the results of principal component analysis using the PC1, PC2, and PC3 axes for six spectral parameters of crabgrass after 3, 6, 24, 48, 72, and 120 hours following treatment with herbicides of various mechanisms of action. Each symbol indicates reproducibility.

[0064] Figure 9 shows the results of principal component analysis using the PC1, PC2, and PC3 axes for six spectral parameters of barnyard grass, based on aggregating all data including 3, 6, 24, 48, 72, and 120 hours after treatment with herbicides of various mechanisms of action. Each symbol indicates reproducibility.

[0065] Figure 10 shows the untreated control group and NDI, ExG, and F after herbicide treatment on rapeseed. d / F m This is a figure comparing the and temperature indices. ns indicates insignificance, and *, **, ***, and **** indicate significance at P<0.05, P<0.01, P<0.001, and P<0.0001, respectively.

[0066] Figure 11 is an RGB image of rapeseed after treatment with propanyl, oxyfluorphen, mesotrione, and glyphosate.

[0067] Figure 12 is a graph showing the changes in NDI and ExG of rapeseed after treatment with propanyl, oxyfluorphen, mesotrione, and glyphosate.

[0068] Figure 13 shows chlorophyll fluorescence images of rapeseed after treatment with propanyl, oxyfluorphen, mesotrione, and glyphosate. The color gradient bar and the scale on the right are F d / F m It represents a value, with black representing the lowest value (0) and white representing the highest value (1).

[0069] Fig. 14 shows F of rapeseed after treatment with profanyl, oxyfluorphen, mesotrione, and glyphosate. d / F m This is a graph showing the change.

[0070] Figure 15 shows IR thermal images of rapeseed after treatment with propanyl, oxyfluorphen, mesotrione, and glyphosate. The color gradient bar and the scale on the right represent the leaf temperature of the rapeseed, with black representing the lowest value (20°C) and white representing the highest value (36°C).

[0071] Figure 16 is a graph showing the change in temperature difference of rapeseed after treatment with propanyl, oxyfluorphen, mesotrione, and glyphosate.

[0072] Figure 17 is a graph showing the verification and test accuracy of the herbicide mechanism of action estimation model according to the learning algorithm.

[0073] Figure 18 is a graph showing verification accuracy according to (A) image acquisition time, (B) image acquisition period, and (C) spectral index. TI and CF are the temperature index and F, respectively. d / F m It represents.

[0074] Figure 19 shows the confusion matrix representing the test accuracy for each herbicide mechanism of action when tested with the Subspace Discriminant algorithm. Treatments of propanyl (PS II), oxyfluorophene (PPO), mesotrione (HPPD), and glyphosate (EPSPS) were used in the training dataset (Test 1), and treatments of bentazone (PS II), thiafenacil (PPO), isoxaflutol (HPPD), and glyphosate (EPSPS) were used in the test dataset (Test 2). Blue squares represent the True Positive Rate (TPR), and red squares represent the False Negative Rate (FNR).

[0075] Figure 20 shows RGB, IR thermal images and CF images confirming the overall changes in spectral response to the inhibition of barnyard grass growth according to herbicide treatment. Each group represents the untreated group (A), cetoxydim (ACCase, B), benfuresate (lipid synthesis, C), S-metholachlor (VLCFA, D), pendimethalin (microtubules, E), indaziflam (cellulose, F), pyribenzoxime (ALS, G), glyphosate (EPSPS, H), glufosinate (GS, I), tetuflupyrrolimet (DHODH, J), dicamba (Auxin, K), oxaziclomefon (unknown, L), paraquat (PSI, M), bentazone (PSII, N), thiafenacil (PPO, O), clomazone (DOXPS, P), and isoxaflutol (HPPD, Q).

[0076] Figure 21 is a graph showing the time-dependent change in the spectral response to the inhibition of barnyard grass growth by herbicide treatment.

[0077] Figure 22 is a graph showing the clustering based on the principal component analysis of the spectral response to the herbicide action mechanism of barnyard grass.

[0078] Figures 23 to 25 are graphs showing the spectral response of 69 types of herbicides according to their mechanisms of action applied to barnyard grass over time.

[0079] Figure 26 is a graph showing clustering by mechanism of action based on the analysis of the active ingredients of 69 types of herbicides treated on barnyard grass.

[0080] Figure 27 is a graph showing the verification and test accuracy of 69 types of herbicide mechanism of action estimation models according to a learning algorithm.

[0081] Figure 28 shows a confusion matrix representing the test accuracy for each herbicide mechanism of action when tested with the Bagged Trees algorithm. The blue squares represent the True Positive Rate (TPR), and the red squares represent the False Negative Rate (FNR).

[0082] One aspect of the present invention provides a method for diagnosing the efficacy and mechanism of action of a herbicide on a plant, comprising the steps of: treating a plurality of herbicides with known mechanisms of action on a plurality of plant individuals according to their respective mechanisms of action; generating image data from the plant individuals; generating spectral parameter data from the image data; training a statistical algorithm using the parameter data to generate a model for predicting the efficacy and mechanism of action of the herbicide on the plant; treating a herbicide candidate substance with an unknown mechanism of action on a plant to generate image data of the plant treated with the candidate substance, and inputting the spectral parameter data generated from the image data of the plant treated with the candidate substance into the prediction model to diagnose the efficacy and mechanism of action of the herbicide candidate substance on the plant.

[0083]

[0084] The method for diagnosing the efficacy and mechanism of action of a herbicide on plants according to the present invention can be provided in the following steps.

[0085]

[0086] The first step is to apply multiple herbicides with known mechanisms of action to multiple plant individuals according to their respective mechanisms of action.

[0087] In this stage, various herbicides with known conventional mechanisms of action are applied, one type per plant individual according to their mechanism of action.

[0088] For example, if the mechanism of action of a herbicide candidate substance to be determined through the diagnostic method of the present invention is lipid synthesis inhibition, a herbicide having a lipid synthesis inhibition mechanism may be selected as the herbicide to be treated in the first step of the present invention. In this case, the mechanism of action may be diagnosed by finally classifying it into two types, such as a herbicide having a lipid synthesis inhibition mechanism and a herbicide not having such a mechanism, through the diagnostic method of the present invention. Alternatively, in addition to the herbicide having a lipid synthesis inhibition mechanism, one or more other mechanisms of action, for example, a herbicide having an amino acid biosynthesis inhibition mechanism, may be further selected. In this case, in the second step, image data may be generated from plant individuals treated with the herbicide having a lipid synthesis inhibition mechanism and plant individuals treated with the herbicide having an amino acid biosynthesis inhibition mechanism, respectively. In this case, through the diagnostic method of the present invention, the mechanism of action can be diagnosed by classifying it into three types: herbicides having a mechanism of action that inhibits lipid synthesis, herbicides having a mechanism of action that inhibits amino acid biosynthesis, and herbicides that do not fall under these categories. That is, if herbicides having n different types of mechanisms of action are selected in the first step of the present invention, the number of mechanisms of action of the candidate substances finally diagnosed becomes n+1. Therefore, if as many different types of herbicides as possible are selected, the mechanism of action of the herbicide candidate substances can be accurately diagnosed, and if herbicides for all currently known mechanisms of action are selected, candidate substances having novel mechanisms of action can also be diagnosed.

[0089] According to one embodiment of the present invention, the mechanism of action may be one or more mechanisms of action selected from the group consisting of inhibition of lipid (fatty acid) synthesis, inhibition of amino acid biosynthesis, inhibition of photosynthesis, inhibition of pigment biosynthesis, inhibition of folic acid biosynthesis, inhibition of cell division, inhibition of cell wall synthesis, inhibition of energy metabolism, and inhibition or disturbance of auxin action.

[0090] Specifically, inhibition of lipid (fatty acid) synthesis may be inhibition of acetyl CoA carboxylase, inhibition of branched amino acid biosynthesis (ALS inhibition), inhibition of aromatic amino acid biosynthesis (EPSP inhibition), and inhibition of glutamine synthase; inhibition of photosynthesis may be inhibition of photosystem II (D1 seine 264 binders), inhibition of photosystem II (D1 histidine 215 binders), and inhibition of photosystem I electron transfer (bipyridylium system); and inhibition of pigment biosynthesis may be inhibition of chlorophyll biosynthesis (PPO inhibition), inhibition of carotenoid biosynthesis (PDS inhibition), inhibition of carotenoid biosynthesis (HPPD inhibition), inhibition of carotenoid biosynthesis (Lycopene Cyclase), and inhibition of DOXPS (1-deoxy-D-xylulose 5-phosphate synthase); Inhibition of folic acid biosynthesis may be inhibition of folic acid biosynthesis (asulam); inhibition of cell division may be inhibition of microtubule assembly, inhibition of mitosis / microtubule formation and inhibition of long-chain fatty acid (VLCFA) synthesis; inhibition of cell wall synthesis may be inhibition of cell wall (cellulose) synthesis and inhibition of fatty acid thioacetase (TE); inhibition of energy metabolism may be membrane disruption; inhibition and disturbance of auxin action may be inhibition of auxin (indoleacetic acid) mimicry and inhibition of auxin transport.

[0091] According to one embodiment of the present invention, the mechanism of action includes inhibition of acetyl CoA carboxylase, inhibition of branched amino acid biosynthesis (ALS inhibition), inhibition of aromatic amino acid biosynthesis (EPSP inhibition), inhibition of glutamine synthase, inhibition of photosystem II (D1 Serine 264 binders), inhibition of photosystem II (D1 Histidine 215 binders), inhibition of photosystem I electron transfer (Bipyridylium system), inhibition of chlorophyll biosynthesis (PPO inhibition), inhibition of carotenoid biosynthesis (PDS inhibition), inhibition of carotenoid biosynthesis (HPPD inhibition), inhibition of carotenoid biosynthesis (Lycopene Cyclase), inhibition of DOXPS (1-deoxy-D-xylulose 5-phosphate synthase), inhibition of folic acid biosynthesis (Asulam), inhibition of microtubule assembly, inhibition of mitosis / microtubule formation, and long chain It may be one or more mechanisms of action selected from the group consisting of inhibition of fatty acid (VLCFA) synthesis, inhibition of cell wall (cellulose) synthesis, inhibition of fatty acid thioacetase (TE), inhibition of lipid synthesis, inhibition of pyrimidine biosynthesis (Dihydroorotate dehydrogenase, DHODH), membrane disruption, auxin (indoleacetic acid) analogue, and inhibition of auxin transport.

[0092] According to one embodiment of the present invention, the herbicide is cyhalofop-butyl, metamifop, fenoxaprop-ethyl, fluazifop-P-butyl, sethoxydim, clethodim, benfuresate, mefenacet, S-metolachlor, alachlor, butachlor, napropamide, fenoxasulfone, imazapyr, imazaquin, flucetosulfuron, flazasulfuron, pyrazosulfuron, Nicosulfuron, Imazosulfuron, Pyrimisulfan, Bensulfuron-methyl, Propyrisulfuron, Azimsulfuron, Trifloxysulfuron-sodium, Triafamone, Pyriminobac-methyl, Pyribenzoxim, Bispyribac-sodium, Penoxsulam, Paraquat, Atrazine, Bentazone, Linuron, Dimethamethrine, Propanil, Pyraclonil, Tiafenacil, Bifenox, Oxyfluorfen, Carfentrazone-ethyl, Pentoxazone, Pyraflufen-ethyl, Oxadiargyl,Fluthiacet-methyl, Dicamba, Florpyrauxifen-benzyl, 2,4-D, MCPP, MCPA, Fluroxypyr-meptyl, Triclopyr TEA, Tetflupyrolimet, Pendimethalin, Ethalfluralin, Oryzalin, Dithiopyr, Isoxaben, Indaziflam, Clomazone), Isoxaflutole, Mesotrione, Tefuryltrione, Benzobicyclon, Pyrazolate, It may be one or more herbicides selected from the group consisting of tolpyralate, glyphosate, glufosinate-ammonium, and oxaziclomefone.

[0093] In the present invention, 4, 6, 16, and 69 types of herbicides with different mechanisms of action selected from the group consisting of paraquat, thiafenacil, phenoxulam, isoxaflutol, glufosinate, glyphosate, propanyl, oxyfluorphen, mesotrione, and bentazone were selected and used for plant image analysis, and through this, it was confirmed that each type of herbicide can be separated and distinguished according to efficacy and mechanism of action by PCA and machine learning models.

[0094] According to one embodiment of the present invention, the plant may be barnyard grass or rapeseed.

[0095] In this invention, it was confirmed that the efficacy and mechanism of action of herbicides based on plant image analysis are possible for barnyard grass and rapeseed. In this case, one-leaf seedlings were used for barnyard grass, and three-leaf seedlings were used for rapeseed. That is, since this invention has verified that the efficacy and mechanism of action of herbicides can be diagnosed early using plants at the seedling stage, it is useful for the rapid and large-scale verification of herbicide candidate substances. Furthermore, the method of this invention is suitable for evaluating expensive or small-quantity candidate substances, as it allows for the diagnosis of the efficacy and mechanism of action of herbicides using only one-quarter or less of the amount of commercially used herbicides.

[0096] According to one embodiment of the present invention, the barnyard grass can be grown in a multi-well plate.

[0097] This invention is the first to collect and analyze three different types of spectral images using a multi-well plate to diagnose the mechanism of action of herbicides. A synergistic effect was confirmed in the entire experimental process, including plant preparation, image collection, and analysis, through multi-well plate analysis combined with spectral image analysis. Compared to conventional herbicide analysis targeting the entire plant, the method of this invention can be applied to testing a large number of herbicide candidate substances within a limited space and time; therefore, it can be effectively utilized as an automated herbicide bioassay and High-Throughput Screening (HTS) method in the discovery and development of novel herbicide substances.

[0098]

[0099] The second step is to generate image data from the above plant individuals.

[0100] In this step, for plant image analysis, RGB images, chlorophyll fluorescence (CF) images and / or IR (Infrared Red) thermal images, preferably these three types of images, are acquired to generate data.

[0101] According to one embodiment of the present invention, the image data may be generated from one or more images selected from the group consisting of RGB images, chlorophyll fluorescence images and IR thermal images.

[0102] The present invention relates to an analysis method combined with various plant spectrum image analysis to rapidly diagnose the efficacy and mechanism of action of herbicides. By using three different sensors, RGB, CF, and IR thermal sensors, spectral responses are obtained at various points in time after herbicide treatment, and the efficacy and mechanism of action of herbicides can be diagnosed based on various spectral responses according to the mechanism of action of the herbicide and the time at which the spectral images were acquired.

[0103] According to one embodiment of the present invention, the step of generating the image data may be performed at two or more times between 3 hours and 144 hours after the step of processing each of the plurality of plant individuals.

[0104] For example, by generating image data according to this step at two or more time points, preferably four or more time points, and more preferably six time points, among 3 hours, 6 hours, 24 hours, 48 ​​hours, 72 hours, 96 hours, 120 hours, and 144 hours after herbicide treatment, the efficacy and mechanism of action of the herbicide can be diagnosed through the analysis of changes over time after herbicide treatment. In particular, according to the machine learning diagnostic model of the present invention, it is preferable to include image data at the time point of 6 hours after herbicide treatment, as it has been confirmed that spectral parameters obtained at 6 hours after herbicide treatment are important for achieving higher accuracy of the machine learning algorithm.

[0105]

[0106] The third step is to generate spectrum parameter data from the above image data.

[0107] According to one embodiment of the present invention, the spectral parameters are mNDI (modified Normalized Difference Index), NDI (Normalized Difference Index), ExG (Excess Green Index), F v / F m , ΦPSⅡ, F d / F m It may be one or more parameters selected from a group consisting of temperature difference and temperature index.

[0108] It can be determined that the greater the efficacy of the herbicide on the plant, the more rapidly the values ​​of each spectral parameter obtained in this step decrease as the time elapses for generating image data after herbicide treatment. However, it was confirmed that this pattern varies depending on the type of plant and the type of herbicide's mechanism of action, and in some cases, the decreased values ​​recover. Furthermore, it was confirmed in this invention that such patterns of change differ depending on the type of spectral parameter. Finally, this invention demonstrates that the efficacy and mechanism of action of the herbicide can be diagnosed based on the changes in various spectral parameters over time derived through such image analysis.

[0109] According to one embodiment of the present invention, the mNDI, NDI, and ExG can be obtained from the following Equation 1:

[0110]

[0111] [Mathematical Formula 1]

[0112]

[0113] R: Red RGB value

[0114] G: Green RGB value

[0115] B: Blue RGB values

[0116] NDI h NDI of herbicide-treated plants

[0117] NDI0: NDI of plants not treated with herbicides.

[0118] According to one embodiment of the present invention, the F v / F m , ΦPSⅡ and F d / F m It can be obtained from the following mathematical formula 2:

[0119]

[0120] [Mathematical Formula 2]

[0121]

[0122] F0: Baseline chlorophyll fluorescence value in dark adaptation state

[0123] F m : Chlorophyll fluorescence value at the point (1 second) when the plant emits maximum fluorescence after exposure to light

[0124] F s : Chlorophyll fluorescence value at the point when the plant returns to standard conditions (60 seconds).

[0125] According to one embodiment of the present invention, the temperature difference can be obtained from the following mathematical formula 3:

[0126]

[0127] [Mathematical Formula 3]

[0128]

[0129] T h : Leaf temperature of plants treated with herbicides

[0130] T0: Leaf temperature of plants not treated with herbicides.

[0131] According to one embodiment of the present invention, the temperature index can be obtained from the following mathematical formula 4:

[0132]

[0133] [Mathematical Formula 4]

[0134]

[0135] T h : Leaf temperature of plants treated with herbicides

[0136] T0: Leaf temperature of plants not treated with herbicides.

[0137]

[0138] The fourth step is to generate a model for predicting the efficacy and mechanism of action of herbicides on plants by training a statistical algorithm using the aforementioned parameter data.

[0139] In the present invention, as a result of performing principal component analysis or machine learning on the spectral parameter data obtained in the third step, it was confirmed that herbicides are clustered or distinguished according to their efficacy and mechanism of action on plants, and can be distinguished according to the same efficacy and mechanism of action.

[0140] According to one embodiment of the present invention, the statistical algorithm may be Principal Component Analysis (PCA) or a machine learning algorithm.

[0141] In the present invention, 2D PCA and 3D PCA were performed, and it was confirmed that the results obtained from 3D PCA were consistent with the results obtained from 2D PCA, and it was confirmed that the efficacy and mechanism of action of herbicides on plants can be diagnosed based on more distinct differences according to the herbicide mechanism of action.

[0142] Meanwhile, a machine learning model for predicting the efficacy and mechanism of action of herbicides on plants can be generated by training multiple machine learning algorithms using the spectral parameter data generated in the third step to create multiple preliminary machine learning models, and then, for example, deriving 10-fold cross-validation and test data performance indicators for each model, and selecting the machine learning model with the best performance indicator when the two results (training and testing) are combined. The performance evaluation of the preliminary machine learning models can be performed, for example, using 10-fold cross-validation, and a portion of the data regarding the herbicide used in the step of generating the spectral parameter data can be used as training data to train the machine learning model, and the remaining portion can be used as test data, or image data resulting from a separate herbicide treatment separate from the herbicide can be collected and utilized as test data.

[0143] According to one embodiment of the present invention, the machine learning algorithm may be a Subspace Discriminant or Bagged Trees.

[0144] In this invention, the spectral parameter data selected in the third step were trained on Fine Tree, Medium Tree, Coarse Tree, Linear Discriminant, Quadratic Discriminant, Gaussian Naive Bayes, Kernel Naive Bayes, Linear SVM, Quadratic SVM, Cubic SVM, Fine Gaussian SVM, Medium Gaussian SVM, Coarse Gaussian SVM, Fine KNN, Medium KNN, Coarse KNN, Cosine KNN, Cubic KNN, Weighted KNN, SVM Kernel, Logistic Regression Kernel, Boosted Trees, Bagged Trees, Subspace Discriminant, Subspace KNN, Boosted Trees, Narrow Neural Network, Medium Neural Network, Wide Neural Network, Bilayered Neural Network, and Trilayered Neural Network machine learning algorithms to analyze the performance indicators for diagnosing herbicide efficacy and mechanism of action on plants. The machine learning algorithm exhibiting the best predictive performance was Subspace Discriminant (validation accuracy 100% and test accuracy It was confirmed to be 87.5%) and Bagged Trees (test accuracy 86.1%).

[0145]

[0146] The final step is to treat plants with a herbicide candidate substance whose mechanism of action is unknown to generate image data of the plants treated with the candidate substance, and to input spectral parameter data generated from the image data of the plants treated with the candidate substance into the prediction model to diagnose the efficacy and mechanism of action of the herbicide candidate substance on the plants.

[0147] A herbicide candidate substance with an unknown mechanism of action is applied to plants to generate image data of the treated plants. Subsequently, spectral parameter data is generated from the image data of the treated plants and input into a model for predicting the efficacy and mechanism of action of the herbicide candidate substance to diagnose its efficacy and mechanism of action on plants. Through this diagnosis, not only can the herbicidal activity of the herbicide candidate substance be analyzed, but its mechanism of action can also be rapidly diagnosed; thus, this method can be effectively and usefully applied to the development of novel herbicide candidate substances.

[0148]

[0149] Another aspect of the present invention provides an apparatus for diagnosing the efficacy and mechanism of action of a herbicide on a plant, comprising: a herbicide treatment unit for treating a plurality of herbicides with known mechanisms of action to a plurality of plant individuals according to their respective mechanisms of action; an image data generation unit for generating image data from the plant individuals; a spectrum parameter data generation unit for generating spectrum parameter data from the image data; a learning unit for generating a model for predicting the efficacy and mechanism of action of the herbicide on the plant using the parameter data; and a diagnosis unit for treating a herbicide candidate substance with an unknown mechanism of action to a plant to generate image data of the plant treated with the candidate substance, and inputting the spectrum parameter data generated from the image data of the plant treated with the candidate substance into the prediction model to diagnose the efficacy and mechanism of action of the herbicide candidate substance on the plant.

[0150] To avoid excessive complexity caused by unnecessary repetition in this specification, common details are omitted.

[0151] A device for diagnosing the efficacy and mechanism of action of herbicides on plants may include, for example, personal computers such as desktops and laptops, as well as mobile terminals capable of wired and wireless communication. The mobile terminal is a wireless communication device that ensures portability and mobility, and may include smartphones, tablet PCs, and wearable devices, as well as various devices equipped with communication modules such as Bluetooth Low Energy, NFC, RFID, ultrasonic wave, infrared, Wi-Fi, and Li-Fi.

[0152] The herbicide treatment unit and the image data generation unit may be equipped in a facility or device where plants are growing, and in this case, plant growth, herbicide treatment, and plant growth and image data acquisition may take place in the same location.

[0153] The image data generation unit can generate data by acquiring images of plants at multiple points in time after herbicide treatment.

[0154] The spectral parameter data generation unit, based on the acquired image data, generates mNDI, NDI, ExG, and F v / F m , ΦPSⅡ, F d / F m , can be derived by calculating parameters corresponding to the temperature difference and / or temperature index.

[0155] The learning unit can train a machine learning model or perform principal component analysis using spectral parameter data. For example, the learning unit can train a machine learning model to diagnose the efficacy and mechanism of action of herbicides for each parameter data by performing supervised learning based on selected parameter data.

[0156] The diagnostic unit can diagnose the efficacy and mechanism of action of herbicide candidate substances by inputting spectral parameters derived from image data over time after herbicide treatment into a trained model for predicting the efficacy and mechanism of action of herbicides on plants. For example, the diagnostic unit can diagnose the mechanism of action of a candidate substance based on the type of mechanism of action, which is the output value of a machine learning model. Alternatively, it can diagnose the mechanism of action of a candidate substance based on the clustering of mechanisms of action of a principal component analysis model. In other words, the diagnostic unit can diagnose the efficacy of or the mechanism of action of a herbicide candidate substance based on the output value of the training model.

[0157] The present invention will be explained in more detail below through one or more embodiments. However, these embodiments are intended to illustrate the invention and the scope of the invention is not limited to these embodiments.

[0158]

[0159] Example 1. Mechanism of action of herbicide and time-dependent biological assays after treatment based on principal component analysis of crabgrass spectral imaging using a multi-well plate.

[0160]

[0161] <Experimental Method>

[0162] 1. Selection of plant materials and cultivation of plants in multi-well plates

[0163] Crabgrass (Digitaria ciliaris), a monocotyledonous weed, was selected as a model plant to diagnose plant response to herbicides in multi-well plates. In 2021, seeds of crabgrass were collected, dried, and stored at 4°C until use. After optimizing the weed cultivation conditions in a multi-well plate (SPL Life Sciences, Pocheon, South Korea) consisting of 24 wells (4×6) with a diameter of 15.5 mm, 15 crabgrass seeds were sown per well in the multi-well plate and cultured in a growth chamber (Hanbaek Science, Bucheon, South Korea) maintaining 30°C / 20°C (day / night) with a 16-hour photoperiod until the first leaf stage.

[0164]

[0165] 2. Herbicide application method and application volume

[0166] When crabgrass reached the 1-leaf stage, six types of herbicides with different modes of action were applied at a spray rate of 300 L / ha using a CO2 pressure belt-driven sprayer (R&D sprayers, Opelousas, LA, USA) equipped with a flat fan nozzle 8001 (Spraying Systems Co., Glendale Heights, IL, USA) (Table 1). The treatment rate was set to 1 / 4 of the recommended amount based on the growth stage of crabgrass in a multi-well plate and preliminary tests.

[0167]

[0168] Ingredient Name | Mechanism of Action | Peony Dosage (g ai / ha) | Product Name | Formulation | Manufacturer | Recommended Dose | Experimental Dose Paraquat | PSI Inhibitor | 500 | 125 | Gramoxone Liquid (SL) | Farm Hannong, Seoul, South Korea Tiafenacil | PPO Inhibitor | 160 | 40 | Teradomy Emulsion (ME) | Farm Hannong, Seoul, South Korea Penoxulam | ALS Inhibitor | 120 | 30 | Salchodaechup Liquid Wettable Powder (SC) | Hankook Samgong, Seoul, South Korea Isoxaflutole | HPPD Inhibitor | 200 | 50 | Merlin Granular Wettable Powder (WG) | BASF, Lutwigshafen, Germany Glufosinate | GS Inhibitor | 1440 | 360 | Basta Liquid (SL) | Bayer CropScience, Seoul, South Korea Glyphosate | EPSPS Inhibitor | 3690 | 922.5 | Geunsami Liquid (SL) | Farm Hannong Seoul, South KoreaPSI: photosystem ⅠPPO: protoporphyrinogen oxidaseALS: acetolactate synthaseHPPD: 4-hydroxyphenylpyruvate dioxygenaseGS: glutamine synthetaseEPSPS: 5-enolpyruvylshikimate-3-phosphate synthase

[0169]

[0170] The herbicide was applied to only 2 of the 6 rows of the multi-well plate, the herbicide-treated multi-well plate was transferred to a growth chamber maintained at 30°C / 20°C (day / night) with a 16-hour photoperiod, and all treatment groups were arranged in the growth chamber according to a fully randomized design of 8 replicates.

[0171]

[0172] 3. RGB Image Acquisition and Analysis

[0173] RGB images of crabgrass grown in multi-well plates were acquired using a CMOS camera (EOS-600D, Canon, Tokyo, Japan) at 3, 6, 24, 48, 72, and 120 hours after herbicide treatment, and analyzed using MATLAB R2023b (TheMathWorks Inc., Natick, MA, USA). The acquired RGB images were analyzed after adjusting the white balance, and the crabgrass was segmented within the images by converting the RGB images to the CIE Lab color space. Subsequently, the background was removed by applying the Otsu threshold to the a* channel in the Lab color space (Fig. 1). To measure the plant's visual response and green level, the red (R), green (G), and blue (B) values ​​of the crabgrass were normalized using Equation 1 to calculate the modified Normalized Difference Index (mNDI) and Excess Green Index (ExG). In Equation 1, NDI h and NDI0 represent the NDI values ​​of herbicide-treated barnyard grass and the untreated control group, respectively.

[0174]

[0175] [Mathematical Formula 1]

[0176]

[0177]

[0178] 4. Chlorophyll Fluorescence Imaging Acquisition and Analysis

[0179] Chlorophyll fluorescence (CF) images of *Crassula ovata* grown in multi-well plates were acquired at 3, 6, 24, 48, 72, and 120 hours after herbicide treatment using the SNU-KIST 2 imaging system (Seoul National University / KIST, Seoul / Gangneung, South Korea). This system consists of a machine vision camera (NRI GigE Camera, Basler, Germany) equipped with a band-pass filter (High Performance Longpass Filter, Edmund Optics, NJ, USA). Each CF image was represented by red pixels based on the acquired chlorophyll fluorescence index value; the CF value was derived by applying an Otsu threshold to each CF image to remove the background and then averaging the index values ​​of the *Crassula ovata* pixels. Finally, CF image parameters were calculated using Equation 2. F v / F m is the maximum quantum yield, ΦPSⅡ is the photosystem II quantum yield, F d / F m represents the fluorescence attenuation rate. F0 is the baseline CF in the dark-adapted state, and F m is the CF value at the point (1 second) when the plant emits maximum fluorescence after exposure to light, and F s is the CF value at the point in time (60 seconds) when the plant adapted to light. The CF parameter was derived by normalizing the value of the herbicide-treated barnyard grass by dividing it by the value of the control group that was not treated with herbicide.

[0180]

[0181] [Mathematical Formula 2]

[0182]

[0183]

[0184] 5. IR Thermal Imaging Acquisition and Analysis

[0185] IR thermal images of crabgrass grown in a multi-well plate were acquired at 3, 6, 24, 48, 72, and 120 hours after herbicide treatment using an A65sc infrared camera (FLIR, Wilsonville, OR, USA) mounted in a plant growth chamber (Korea Institute of Science and Technology, Suwon, South Korea) maintained at 33°C. An image registration algorithm was used to derive the leaf temperature of the crabgrass by identifying the plants using the corresponding RGB images segmented in Examples 1-3. The temperature difference was calculated using Equation 3, utilizing the leaf temperature data obtained from IR thermal image analysis. T h T0 and T0 represent the leaf temperatures of herbicide-treated barnyard grass and the control group not treated with herbicide, respectively.

[0186]

[0187] [Mathematical Formula 3]

[0188]

[0189]

[0190] 6. Statistical Analysis

[0191] For all plant spectral image data, Analysis of Variance (ANOVA) was first performed, and then the statistical significance of the effects of two factors (herbicide and time) on the spectral response of crabgrass following herbicide treatment was investigated. Next, Principal Component Analysis (PCA) was performed on normalized spectral parameters to evaluate whether spectral responses could be separated and clustered according to the herbicide mechanism of action. Initially, correlation-based PCA was performed separately for each time point (3, 6, 24, 48, 72, and 120 HAT (hours after treatment)), and then on the pooled data with default settings. All statistical analyses were performed using R 3.2.3 (R Foundation for Statistical Computing, Vienna, Austria).

[0192]

[0193] <Experimental Results>

[0194] 1. Establishment of Optimal Cultivation Conditions for Crabgrass in Multi-Well Plates

[0195] Considering rapid and uniform germination and compact plant growth, barnyard grass was selected as the model plant. Barnyard grass seeds were sown in the wells of 24-well and 96-well plates, and seedling growth was monitored. It was found that the well size of the 96-well plate was too small for barnyard grass to grow. On the other hand, it was found that the well size of the 24-well plate was sufficiently large for barnyard grass growth, so the 24-well plate was selected for the cultivation of barnyard grass. Meanwhile, to optimize the germination of barnyard grass seeds and seedling growth, various media including water, agarose (0.4%, w / v), and MS (Murashige & Skoog) medium were tested, and it was confirmed that a medium consisting only of agarose (Inclonebiotech Co., Seongnam, South Korea) was suitable.

[0196] Next, to optimize growth conditions in a 24-well plate, the number of seeds per well and the growth temperature from germination to seedling were analyzed, and it was confirmed that 15 seeds per well was appropriate. Meanwhile, it was confirmed that the optimal growth conditions in the 24-well plate were 30℃ / 20℃ (day / night) and the photoperiod until the first leaf stage was 16 hours. At the first leaf stage, crabgrass was found to have a sufficient leaf area to exhibit a spectral response to herbicides, confirming that crabgrass grown in 24-well plates can be effectively utilized for herbicide bioassays.

[0197] Finally, the optimal conditions for cultivating crabgrass in a multi-well plate were set by sowing 15 crabgrass seeds in 0.8 mL of agar medium per well and culturing them until the 1-leaf stage (5 days) in a plant growth chamber maintained at a 16-hour photoperiod and 30℃ / 20℃ (day / night), and images were collected in the plant growth chamber after herbicide treatment (Fig. 1).

[0198]

[0199] 2. Confirmation of changes in spectral response of crabgrass according to herbicide mechanism of action and elapsed time after treatment

[0200] As a result of analyzing plant spectral images acquired with various sensors (RGB, CF, IR thermal sensors) according to the methods of Example 1 and Experimental Methods 3 to 5, it was confirmed that clear differences appeared depending on the mechanism of action of the herbicide and the time after herbicide treatment (Fig. 2). Specifically, it was found that the spectral changes due to herbicide treatment became clearer as time passed after herbicide treatment. Among all spectral images, the fastest change was observed in the paraquat treatment, followed by thiafenacil and glufosinate. The slowest spectral change was observed in phenoxulam, followed by isoxaflutol and glyphosate.

[0201] Spectral changes in RGB images included dehydration, chlorophyll deficiency, and necrosis, which were confirmed to vary according to the herbicide mechanism of action and time after herbicide treatment (Fig. 2A). For CF images, the most noticeable and rapid changes were observed in the paraquat treatment at 3 HAT, while distinct differences between herbicide mechanisms of action were observed at 48 HAT (Fig. 2B). In IR thermal images as well, although the changes were relatively slow, clear differences between herbicide mechanisms of action were still confirmed (Fig. 2C). According to the analysis of variance (ANOVA) of all parameters estimated from spectral image analysis, very high statistical significance was found for herbicides, time, and their interactions (Table 2). Considering the F-values ​​of the ANOVA, it was found that herbicides had a greater influence on RGB and CF images than time, while having a smaller influence on IR thermal images than time. In particular, the F-value between herbicides and time v / F m and F d / F m The difference was found to be the largest, confirming that the CF parameter is useful for diagnosing the mechanism of action of herbicides.

[0202]

[0203] Spectral Parameters Dispersion Component F Value p Value p Value Summary RGB m NDI Herbicide 1063.089<0.0001****Hour 606.495<0.0001****Herbicide × Hour 95.708<0.0001****ExG Herbicide 1816.526<0.0001****Hour 1185.414<0.0001****Herbicide × Hour 166.365<0.0001****CFF v / F m Herbicide 8354.295<0.0001****Hour 1894.120<0.0001****Herbicide × Hour 341.633<0.0001****ΦPSⅡHerbicide 859.989<0.0001****Hour 653.637<0.0001****Herbicide × Hour 73.445<0.0001****F d / F m Herbicide 5966.657<0.0001**** Hour 661.369<0.0001**** Herbicide × Hour 238.078<0.0001**** IR Temperature Difference Herbicide 11.923<0.0001**** Hour 37.174<0.0001**** Herbicide × Hour 15.847<0.0001****

[0204]

[0205] 3. Confirmation of changes in RGB spectrum response of crabgrass according to herbicide mechanism of action and elapsed time after treatment

[0206] Since leaf discoloration such as yellowing and bleaching is the most common symptom after herbicide treatment, RGB image analysis was performed using the method of Example 1 and Experimental Method 3 to analyze the change in RGB color, particularly the green level of the leaves, for the herbicide.

[0207] As a result, it was confirmed that mNDI (Fig. 3A) and ExG (Fig. 3B) values ​​decreased significantly as visual symptoms became more pronounced. Specifically, mNDI and ExG showed a trend similar to the visual symptoms and decreased continuously over time. Paraquat showed the fastest and largest decrease in mNDI and ExG values, followed by thiafenacil and glufosinate. At 6 HAT, the mNDI and ExG values ​​of paraquat were reduced to 53.9% and 73.5%, respectively, compared to the untreated control group; this trend was consistent with the rapid weeding symptoms of paraquat demonstrated in RGB imaging. Crabgrass was confirmed to have almost completely died off at 24 HAT for paraquat and 48 HAT for thiafenacil. Glufosinate and glyphosate began to show color changes at 24 HAT and 72 HAT, respectively, which was confirmed to be consistent with the decrease in mNDI and ExG values. Isoxaflutol and fenoxulam showed minimal herbicidal activity based on visual observation, and no significant changes in mNDI and ExG were observed up to 120 HAT after treatment. These differential changes in mNDI and ExG after herbicide treatment indicate that herbicidal activity and mechanism of action can be diagnosed through plant image analysis using mNDI and ExG values.

[0208]

[0209] 4. Confirmation of changes in CF spectrum responses of crabgrass according to herbicide mechanism of action and elapsed time after treatment

[0210] Considering that photosynthesis requires plant pigments, antioxidant systems, various enzymes, and proteins linked to carbohydrates, it is highly likely that all stress experienced by plants will affect photosynthesis. Furthermore, herbicides act on different targets depending on their mechanism of action, consequently affecting photosynthesis in various ways. Therefore, chlorophyll fluorescence imaging analysis was performed using the method of Example 1 and Experimental Method 4, and F, a parameter of chlorophyll fluorescence for herbicidesv / F m , ΦPSⅡ, F d / F m The changes in were analyzed.

[0211] As a result, it was confirmed that the three chlorophyll fluorescence parameters showed differential changes over time following herbicide treatment, depending on the herbicide's mechanism of action (Fig. 4). Specifically, with the exception of fenoxulam, which showed no herbicidal activity against crabgrass, all other herbicides—namely paraquat, glufosinate, thiafenacil, glyphosate, and isoxaflutol—showed a significant decrease in CF parameters compared to the untreated control group. Herbicide treatment primarily resulted in a decrease in all CF parameters, although this varied depending on the herbicide's mechanism of action. With the exception of fenoxulam, the CF spectral response to the herbicides appeared more immediate and pronounced compared to other spectral responses, such as RGB and IR sequences. Among the herbicides, paraquat showed the fastest and largest decrease in all CF parameters, and even in 3 HAT, F of the untreated control group v / F m , ΦPSⅡ and F d / F m These decreased significantly by up to 45.9%, 56.7%, and 31.8%, respectively, and reached 0% at 6 HAT. Tiafenacil showed the second fastest and largest significant decrease in all CF parameters, reaching nearly 0% from 24 HAT to 72 HAT. For glufosinate and glyphosate, ΦPSII decreased abruptly at 24 HAT, but compared to ΦPSII, F v / F m and F d / F m It was confirmed that the decrease was not very significant. Isoxaflutol did not show any significant changes in all CF parameters up to 48 HAT, but a sharp decrease in ΦPSⅡ was observed at 120 HAT, at which time no change was found in the RGB and IR thermal spectral responses.

[0212]

[0213] 5. Confirmation of changes in IR thermal spectral response of crabgrass according to herbicide mechanism of action and elapsed time after treatment

[0214] Leaf temperature has a high correlation with the overall physiological state of plants, particularly with stomatal conductivity and photosynthetic rate. Additionally, a symptom of plant stress is the closing of leaf stomata, which occurs along with a decrease in carbon dioxide absorption during photosynthesis. This ultimately leads to a decrease in transpiration, and consequently, the temperature of stressed leaves rises as transpiration decreases. Leaf temperature is known to be a reliable indicator for detecting numerous types of plant stress that affect photosynthesis, such as drought, salinity stress, and herbicides. Therefore, IR thermal imaging analysis was performed using the method of Example 1 and Experimental Method 5 to analyze the temperature difference estimated based on leaf temperature for herbicides.

[0215] As a result, the temperature difference estimated based on the leaf temperature of crabgrass determined by IR thermal imaging analysis was found to show distinct differences among the treated herbicides (Fig. 5). Specifically, distinct temperature changes were observed in the herbicide-treated groups compared to the untreated control group. For paraquat, thiafenacil, glufosinate, and glyphosate, the temperature difference was found to increase initially and then decrease from a certain point as the plants became dehydrated. As observed in other spectral responses, paraquat showed the largest change in temperature difference, increasing up to 6 HAT and then suddenly decreasing at 24 HAT. Glufosinate showed a continuous increase in temperature difference up to 24 HAT, whereas thiafenacil and glyphosate showed an increase up to 48 HAT before decreasing. Isoxaflutol showed a significant increase even at 120 HAT, whereas phenoxulam was found to show no noticeable change even at 120 HAT.

[0216]

[0217] 6. Confirmation of herbicide mechanism of action and community formation based on principal component analysis and elapsed time after treatment

[0218] According to two-dimensional principal component analysis of all spectral imaging parameters observed at different times after herbicide treatment, it was confirmed that the herbicides gradually clustered according to their mechanisms of action over time (Fig. 6, Table 3). Specifically, paraquat was noticeably separated from other herbicides at 3 HAT. At 24 HAT, thiafenacil clustered independently from other herbicides. At 48 HAT, glufosinate and glyphosate clustered independently. Isoxaflutol was finally separated from other herbicides at 120 HAT. However, fenoxulam did not cluster independently even at 120 HAT. Principal component analysis using pooled data from 3 HAT to 120 HAT confirmed that clear clustering of herbicides based on their mechanisms of action is possible, with the exception of fenoxulam, which was not separated from the untreated control group due to its very low efficacy against crabgrass (Fig. 7, Table 4). Similar clustering according to the herbicide mechanism of action was also confirmed in the three-dimensional principal component analysis of observed spectral imaging parameters for pooled data at each time point (Figs. 8 and 9). Fast-acting herbicides such as paraquat, thiafenacil, and glufosinate required only 2 days for clustering according to the herbicide mechanism of action, whereas slow-acting herbicides such as glyphosate and isoxaflutol required a longer time after herbicide treatment.

[0219]

[0220] Time (HAT) Variance Ratio (%) of Total Variance PC1 PC2 PC3 3 6 2.7 19.7 8 2.5 6 7 7.5 16.6 9 4.1 2 4 8 5.6 10.0 9 5.6 4 8 7 5.8 18.2 9 4.0 7 2 8 9.9 7.0 9 6.9 1 2 0 8 4.4 9.9 9 4.3 Pooling 6 9.6 14.7 8 4.3

[0221]

[0222] Time (HAT) Variance Ratio (%) of Total Variance PC1 PC2 PC3 Cumulative Variance Ratio 3 6 2.7 19.7 11.3 9 3.8 6 7 7.5 16.6 5.3 9 9.4 2 4 8 5.6 10.0 3.0 9 8.5 4 8 7 5.8 18.2 3.6 9 7.6 7 2 8 9.9 7.0 2.4 9 9.2 1 2 0 8 4.4 9.9 4.6 9 8.8 Pooling 6 9.6 14.9 4.1 8 8.4

[0223]

[0224] Example 2. Mechanism of action of herbicide and biological assay at time intervals after treatment based on machine learning analysis of rapeseed spectrum images

[0225]

[0226] <Experimental Method>

[0227] 1. Plant materials and cultivation conditions

[0228] Commercial rapeseed (Brassica napuscv. Tammi) was used as a model plant. Rapeseed seedlings were grown until the 3-leaf stage in a glass greenhouse at the Seoul National University Experimental Farm in Suwon, Korea, at 25±5℃ under a 14-hour photoperiod during the experiment period.

[0229]

[0230] 2. Herbicide treatment

[0231] Six types of herbicides with different modes of action were applied to rapeseed seedlings grown in Experimental Method 1 at a spraying rate of 300 L / ha using a CO2-pressurized belt-driven sprayer (R&D sprayers, Opelousas, LA, USA) equipped with a flat fan nozzle 8001 (Spraying Systems Co., Glendale Heights, IL, USA) (Table 5). After herbicide treatment, the rapeseed seedlings were immediately transferred to a glass greenhouse and arranged according to a randomized block design of three replications.

[0232]

[0233] Data Set Herbicide Ingredient Mechanism of Action Dosage (g ai / ha) Product Name Manufacturer Test 1 (Training Data Set) Propanil PSⅡ No. 2100 Pronil Korea Samgong, Seoul, South Korea Oxyfluorfen PPO Inhibitor 1175 Golkyung Nong, Seoul, South Korea Mesotrione HPPD Inhibitor 152 Tenacity Syngenta, South Korea Glyphosate EPSPS Inhibitor 1845 Geunsami Pharm Hannong, Seoul, South Korea Test 2 (Experimental Data Set) Bentazon PSⅡ Inhibitor 1600 Basagran Sungbo Chemical, South Korea Tiafenacil PPO Inhibitor 80 Teradopharm Hannong, Seoul, South Korea Isoxaflutole HPPD Inhibitor 100 Merlin BASF Lutwigshafen, Germany Glyphosate EPSS inhibitor 1845 Geunsami Pharm Hannong, Seoul, South Korea

[0234]

[0235] 3. RGB Image Acquisition and Analysis

[0236] RGB images of rapeseed were acquired using a CMOS camera (EOS-600D, Canon, Tokyo, Japan) at 6 hours, 1 day, 2 days, 3 days, 4 days, 5 days, and 6 days after herbicide treatment, and analyzed using MATLAB R2023b (TheMathWorks Inc., Natick, MA, USA). The acquired RGB images were converted to the Lab color space, the background was removed, and a threshold was set. To measure the visual response and green level of the plants, the red (R), green (G), and blue (B) values ​​of rapeseed were normalized using Equation 1 of Example 1 and Experimental Example 3, and the Normalized Difference Index (NDI) and Excess Green Index (ExG) were calculated.

[0237]

[0238] 4. Chlorophyll Fluorescence Imaging Acquisition and Analysis

[0239] CF images of rapeseed were acquired using the SNU-KIST 2 imaging system (Seoul National University / KIST, Seoul / Gangneung, South Korea) at 6 hours, 1 day, 2 days, 3 days, 4 days, 5 days, and 6 days after herbicide treatment. This device consists of a CMOS camera (EOS-700D, Canon, Japan) equipped with a band-pass filter (XNiteBPB, LDP LLC, USA). Light and image acquisition were controlled by a Raspberry Pi 3 (Raspberry Pi, UK) board. CF images were acquired at 1 second and 60 seconds under blue LED illumination (470 nm) after 30 minutes of dark adaptation. After background removal, the red pixel intensity of the plants was extracted using MATLAB 2021b and normalized to the CF of the rapeseed. Based on CF image analysis, the F of each leaf d / F m It was calculated using Equation 2 of Example 1 and Experimental Example 4.

[0240]

[0241] 5. IR Thermal Imaging Acquisition and Analysis

[0242] IR thermal images of rapeseed were captured in a glass greenhouse using an IR camera (A65sc, FLIR, Sweden) at 6 hours, 1 day, 2 days, 3 days, 4 days, 5 days, and 6 days after herbicide treatment. The temperature at the time of IR thermal image acquisition was recorded. The acquired IR thermal images were analyzed using MATLAB 2021b. For data extraction, the background was removed from the corresponding RGB images using image registration, and the plant leaf temperatures were extracted using MATLAB 2021b. Subsequently, the temperature index was calculated using Equation 4 based on the leaf temperature data estimated from IR thermal image analysis. T h T0 and T0 represent the leaf temperatures (°C) of rapeseed treated with herbicide and rapeseed not treated, respectively.

[0243]

[0244] [Mathematical Formula 4]

[0245]

[0246]

[0247] 6. Statistical Analysis

[0248] NDI, ExG, and F d / F m The treated sample values ​​were normalized by dividing them by the untreated control values. For all normalized spectral image data, a two-way analysis of variance (ANOVA) was first performed, followed by an examination of the interactions between herbicide, time, and herbicide and time. Once statistical significance was confirmed, Fisher's LSD test was performed for multiple comparisons with the untreated control group. All statistical analyses were performed using R 3.2.3 (R Foundation for Statistical Computing, Vienna, Austria).

[0249]

[0250] 7. Training and Testing of Herbicide Mechanism of Action Classification Models

[0251] Spectral indices obtained from RGB, CF, and IR thermal image analysis in Test 1 were used as input data to train a herbicide mechanism of action classification model. The dataset was divided into five folds for cross-validation to prevent overfitting. All training algorithms provided by the Classifier Learner Application in MATLAB 2021b (Table 6) were used to calculate the validation accuracy for the mechanism of action of each herbicide according to Equation 5, where TP, TN, FP, and FN represent true positive, true negative, false positive, and false negative, respectively.

[0252]

[0253] [Mathematical Formula 5]

[0254]

[0255]

[0256] 구분머신러닝 알고리즘Decision treesFine TreeMedium TreeCoarse TreeDiscriminant analysisLinear DiscriminantQuadratic DiscriminantNaive Bayes classifiersGaussian Naive BayesKernel Naive BayesSupport vector machinesLinear SVMQuadratic SVMCubic SVMFine Gaussian SVMMedium Gaussian SVMCoarse Gaussian SVMNearest neighborFine KNNMedium KNNCoarse KNNCosine KNNCubic KNNWeighted KNNKernel approximation classifiersSVM KernelLogistic Regression KernelEnsemble classifiersBoosted TreesBagged TreesSubspace DiscriminantSubspace KNNRUSBoosted TreesNeural network classifiersNarrow Neural NetworkMedium Neural NetworkWide Neural NetworkBilayered Neural NetworkTrilayered Neural Network

[0257]

[0258] After identifying the most accurate algorithm, the effective timing of image acquisition and spectral indices for estimating the herbicide mechanism of action was investigated. Validation accuracy was calculated by 1) spectral indices collected at each time point, 2) continuously adding spectral indices 6 hours after treatment and subsequent image acquisition times, and 3) learning all possible combinations of each spectral indices. The spectral indices obtained in Test 2 were used to test the performance of the classification model trained on the entire training data obtained in Test 1. The analytical accuracy for the mechanism of action of each herbicide was calculated according to Equation 5.

[0259]

[0260] <Experimental Results>

[0261] 1. Overall changes in the spectral response of rapeseed to herbicides

[0262] After treatment with the herbicides profanyl, oxyfluorophene, mesotrione, and glyphosate, RGB, CF, and IR thermal images were captured and analyzed daily according to the methods of Examples 2 and Experimental Methods 3 to 5 to obtain NDI, ExG, and F d / F m and temperature indices were derived.

[0263] As a result, both herbicides and time were found to have a significant effect on all tested spectral indices, and the interaction between herbicides and time was also found to be statistically significant (Table 7).

[0264]

[0265] Spectral Parameters Dispersion Component %P Value of Total Dispersion P Value Summary NDI Herbicide 50.41<0.0001****Hour 13.07<0.0001****Hour × Herbicide 16.44<0.0001****ExG Herbicide 50.13<0.0001****Hour 9.851<0.0001****Hour × Herbicide 15.71<0.0001****F d / F mHerbicide 44.64<0.0001****Hour 1.2430.0030**Hour × Herbicide 30.44<0.0001****Temperature Index Herbicide 49.2<0.0001****Hour 17.33<0.0001****Hour × Herbicide 17.86<0.0001****

[0266]

[0267] In addition, NDI, ExG, and F between herbicide-treated rapeseed and an untreated control group d / F m As a result of comparing the and temperature index values, it was confirmed that there were differences in the pattern of change of the spectral index depending on the herbicide mechanism of action (Fig. 10). Specifically, NDI, ExG, and the temperature index changed significantly starting from 1 DAT (days after treatment), and F d / F m It changed starting from 6 HAT (hours after treatment). All indices except ExG were found to show significant differences in all herbicide treatments.

[0268]

[0269] 2. Changes in RGB spectral response to herbicides

[0270] RGB image analysis was performed using the method of Example 2 and Experimental Method 3 to analyze changes in RGB colors for herbicides, particularly the green level of leaves.

[0271] As a result, the herbicides caused visual symptoms and changes in leaf color in rapeseed, and it was confirmed that these patterns varied depending on the mechanism of action (Fig. 11). Specifically, while visible changes were found to be the fastest and clearest in rapeseed treated with oxyfluorphen, there were almost no visible changes in rapeseed treated with profanyl even after 6 days (DAT), and in the case of rapeseed treated with mesotrione and glyphosate, visible changes were observed only after 3 DAT.

[0272] However, spectral indices extracted from the same RGB images, particularly NDI, were found to indicate the response of rapeseed to herbicides more rapidly and distinctly than visible symptoms. Specifically, it was confirmed that both NDI and ExG values ​​in rapeseed decreased significantly after herbicide treatment (Fig. 12). In the case of propanil treatment, there was a slight decrease from 6 hours after treatment (HAT) to 6 DAT. NDI reached 75.7% of the untreated control group at 1 DAT, maintained similar values ​​until 3 DAT, and then continuously decreased to 53.9% by 6 DAT. Meanwhile, ExG remained relatively stable until 4 DAT and then decreased by 85.5% at 6 DAT. After oxyfluorophene treatment, NDI and ExG continuously decreased over time. NDI began to decrease sharply to 81.8% at 6 HAT and then significantly decreased to 2.6% of the untreated control group by 3 DAT. At the same time, ExG decreased to 83.3% of the untreated control group by 6 HAT and reached 6.0% at 6 DAT. Mesotrione slightly reduced NDI and ExG after herbicide treatment. NDI decreased to 82.4% at 3 DAT and further decreased to 65.6% at 6 DAT. ExG began to decrease at 3 DAT and reached 90.0% of the untreated control group by 6 DAT. Glyphosate induced a gradual and sustained decrease in NDI, reaching 92.3%, 82.6%, 80.8%, 65.6%, 64.3%, 52.9%, and 36.8% of the untreated control group at each measurement, while a visible decrease in ExG was observed, reaching 73.7% at 5 DAT.

[0273] The fastest and most significant decrease in the RGB-based index was observed with oxyfluorphen treatment, and it was confirmed that rapeseed was completely wilted by 4 DAT. Glyphosate was found to be the second fastest and most potent in terms of effect. Meanwhile, propanyl showed a gradual decrease in the index, whereas mesotrione was found to show the slowest decrease in the index.

[0274]

[0275] 3. Changes in CF spectrum response to herbicides

[0276] Chlorophyll fluorescence imaging analysis was performed using the method of Example 2 and Experimental Method 4, and F for herbicides d / F m The changes in were analyzed.

[0277] As a result, differences in CF images between the herbicide treatment groups were detected starting at 6 HAT (Fig. 13). At 6 HAT, the most significant change was observed in the propanil treatment, followed by oxyfluorphen. Notably, plants treated with propanil began to recover starting at 3 DAT. In contrast, no such recovery was observed in the CF images of the oxyfluorphen treatment group, and they withered completely by 4 DAT. Mesotrione treatment began showing symptoms at 1 DAT, and as confirmed in the images at 3, 4, and 5 DAT, more severe damage was observed in new leaves compared to old leaves; the effects of the herbicide treatment were confirmed in all leaves by 6 DAT. For glyphosate treatment, the pseudo-color changed from yellow to red by 3 DAT. After 4 DAT, the pseudo-color remained relatively stable, but wilting symptoms were observed in the leaves.

[0278] Even if visual symptoms did not appear in RGB images, differences between the control group and the herbicide-treated group were confirmed in CF images. CF images showed that the herbicidal activity of propanyl and oxyfluorphen could be detected as early as 6 HAT. On the other hand, mesotrione and glyphosate showed a somewhat slower spectral response, and it was confirmed that herbicide damage could be detected as early as 2 DAT using CF images.

[0279] Regarding spectral indices extracted from CF images, F in rapeseed after herbicide treatment d / F m A decrease in the value was observed (Fig. 14). The propanyl-treated group reached 12.2% of the untreated control group at 6 HAT and maintained a similar level until 2 DAT; however, the value began to recover from 3 DAT, reaching 38.9%, and steadily increased to 71.9% by 6 DAT. The oxyfluorphen-treated group F at 6 HAT d / F m This decreased rapidly to 55.4% of the untreated control, then gradually decreased to 39.6% of the untreated control by 6 DAT. The mesotrione-treated group showed a trend very similar to that of the oxyfluorophen-treated group, and F d / F m After recording 86.3% of the untreated control group at 6 HAT, it continuously decreased to 36.6% at 6 DAT. The glyphosate-treated group showed no noticeable effect on the index at 6 HAT, but a sharp decrease was observed at 1 DAT (80.6%) and continued to decrease to 3 DAT (43.6%), after which it showed no significant change.

[0280] Propanyl is F d / F mIt showed the most immediate and noticeable decrease, but was confirmed to show a recovery thereafter. Oxyfluorphen was found to be the second fastest and strongest in terms of effect. Mesotrione and glyphosate were found to show a gradual decrease over time.

[0281]

[0282] 4. Changes in IR spectral response to herbicides

[0283] IR thermal imaging analysis was performed using the method of Example 2 and Experimental Method 5 to analyze the temperature index estimated based on the leaf temperature for the herbicide.

[0284] As a result, it was confirmed that all herbicides affected the IR thermal imaging of rapeseed (Fig. 15), and differences between the untreated control group and the herbicide-treated groups were observed in the IR thermal imaging even when no visual symptoms appeared in the RGB images. Specifically, subtle changes were observed in the IR thermal imaging of the propanyl-treated group, indicating a slight increase in leaf temperature. Conversely, in the oxyfluorophene-treated group, rapid and significant changes were observed in the IR thermal imaging at 1 DAT, indicating a sharp increase in leaf temperature. The false color of the mesotrione-treated group did not differ significantly from the untreated control group up to 4 DAT. In the glyphosate-treated group, gradual changes were observed in the IR thermal imaging after 3 DAT.

[0285] In relation to spectral indices extracted from IR thermal imaging, it was confirmed that the temperature index increased after herbicide treatment (Fig. 16). Specifically, a slight increase in the temperature index was observed after treating rapeseed with propanyl; at 1 DAT, the index recorded 103.8, indicating a slight increase in leaf temperature compared to the untreated control group, and the temperature index remained relatively stable until 6 DAT, reaching 105.1. In the case of the mesotrione-treated group, the temperature index showed no change until 4 DAT, but began to show changes from 5 DAT, recording 105.0 and 106.2 at 6 DAT. The glyphosate-treated group showed a slight increase to 104.0 at 1 DAT, followed by a significant increase to 117.2 at 3 DAT. The fastest and most significant response in the temperature index was observed in the oxyfluorophen-treated group, while glyphosate was found to exhibit the second fastest speed and severity in terms of efficacy. It was confirmed that the propanyl and mesotrione treatment groups showed the smallest change in temperature index after herbicide treatment.

[0286]

[0287] 5. Estimation of Herbicide Mechanism of Action Using Machine Learning of Spectral Indices

[0288] 5-1. Selecting a Machine Learning Algorithm

[0289] Overall, spectral indices were found to differ depending on the herbicide mechanism of action. RGB-based indices tended to decrease after treatment with all herbicides. Oxyfluorophene showed the fastest decrease, followed by glyphosate, while profanyl and mesotrione were found to induce only a slight decrease. F, the CF-based index d / F mIn this case, it was confirmed that propanyl showed the fastest initial response in the first measurement and then gradually increased, whereas oxyfluorphene showed a significant decrease after the first measurement. Glyphosate and mesotrione were confirmed to show a gradual decrease. In the case of the temperature index, which is an IR-based index, oxyfluorphene showed the fastest increase, followed by glyphosate. Both propanyl and mesotrione were confirmed to show a relatively slow response. Based on these various spectral indices, the herbicide activity and mechanism of action were estimated by performing an integrated machine learning analysis of the entire spectral index using the method of Example 2 and Experimental Method 7.

[0290] Machine learning (Test 1) results on spectral indices obtained after treatment with propanyl, oxyfluorophene, mesotrione, and glyphosate confirmed that different machine learning algorithms exhibited different validation accuracy (20 to 100%) in identifying the herbicide mechanism of action (Fig. 17). Among the machine learning algorithms exhibiting 100% validation accuracy, the Subspace Discriminant algorithm was selected for further analysis.

[0291]

[0292] 5-2. Learning the Subspace Discriminant Algorithm

[0293] To find the effective timing of image acquisition and spectral indices for estimating the mechanism of action of herbicides, the collected dataset was trained with the Subspace Discriminant algorithm for individual acquisition times.

[0294] As a result, the highest validation accuracy of 89.6% was achieved when training was performed with data obtained from 6 HATs, and an accuracy of over 80% was confirmed in data from 2, 3, and 6 DATs (Fig. 18A). As more data collection times were included in the dataset, the accuracy increased, and it reached 100% when data from 3 DATs was added to 6 HATs (Fig. 18B). Through this, it was confirmed that the mechanism of action of herbicides can be identified within 3 DATs.

[0295] Meanwhile, differences in validation accuracy were observed depending on the spectral indices used in the training dataset (Fig. 18C). When each spectral index was trained alone, F d / F m [It] showed the highest accuracy at 91.7%, followed by the temperature index (89.6%), NDI (75.0%), and ExG (64.6%). Accuracy increased when two or more spectral indices were combined, and F d / F m It was confirmed that 100% was reached when the temperature index was included in the training set, so the pooled data (training data set) of Test 1 was finally trained using the Subspace Discriminant algorithm to construct a model for estimating herbicide activity and mechanism of action.

[0296]

[0297] 5-3. Verification of the Effectiveness of Identifying Herbicide Mechanisms of Action Based on Experimental Data Set Testing

[0298] The model constructed in Example 2, Experimental Result 5-2 was tested using a new independent spectral index (Test 2; experimental data set) obtained after treatment with bentazone, thiafenacil, isooxaflutol, and glyphosate, which are herbicides representing PSⅡ, PPO, HPPD, and EPSPS inhibitors, respectively.

[0299] As a result, the test accuracy was found to be 87.5% (Fig. 19). Specifically, the test accuracy of PSⅡ and PPO inhibitors was 93.8%, and the test accuracy of HPPD and EPSPS inhibitors was found to be 81.3%.

[0300] Through these results, it was confirmed that the machine learning model constructed in this invention can be effectively applied to the estimation of herbicide activity and mechanism of action.

[0301]

[0302] Example 3. Mechanism of action of herbicide and biological assay at time intervals after treatment based on machine learning analysis of crabgrass spectrum images

[0303]

[0304] <Experimental Method>

[0305] 1. Selection of Plant Materials and Cultivation Conditions

[0306] Digitaria ciliaris was used as a model plant. Seedlings were cultivated until the two-leaf stage in a glass greenhouse at the Seoul National University Experimental Farm in Suwon, Korea, maintaining a 14-hour photoperiod at 33 / 25°C during the experiment. 150 gm of seeds were sown in 50-cell pots filled with horticultural potting soil (Barokor, Seoul Bio, South Korea). -1 Sowing was performed at a density of [value]. Afterwards, the plants were cultivated until the 2-leaf stage in a plant growth chamber (Hanbaek Science, South Korea) maintained at 33℃ / 25℃ (day / night).

[0307]

[0308] 2. Herbicide treatment

[0309] 2-1. Herbicide treatment to determine image acquisition period

[0310] Sixteen herbicides representing each mechanism of action were treated to determine the image acquisition period required to estimate the herbicide mechanism of action (Table 8).

[0311]

[0312] No. Classification Mechanism of Action Herbicide Dosage (g ai / ha) 1 Cellular metabolism ACCase Sethoxydim 50 2 Lipid synthesis Benfuresate 500 3 VLC FAS-metolachlor 750 4 ALS Pyribenzoxim 30 5 Cellulose Indaziflam 47.5 6 EPSPS Glyphosate 184 57 Light activation of ROS Photosystem I Paraquat 200 8 Photosystem II Bentazone 1200 9 PPO Tiafenacil 2010 DOXPS Clomazone 600 11 HPPD Isoxaflutole 100 12 Glutamine Synthetase (GS) Inhibition Glufosinate-ammonium 72013 Cell division & growth Dicamba 144614 Microtubule assembly Pendimethalin 158515DHODH Tetflupyrolimet 20016- Unknown Oxaziclomefone 60

[0313]

[0314] 2-2. Herbicide Treatment for Generating a Mechanism of Action Estimation Model

[0315] Spectral information obtained from treating a total of 69 types of herbicides with 16 different mechanisms of action was applied to a machine learning algorithm to create a mechanism of action estimation model (Table 9).

[0316]

[0317] Mechanism of Action Chemical Structure Herbicide Formulation Dose (g ai ha -1

[0318]

[0319] 2-3. Herbicide Application Method

[0320] For the active ingredient treatment in experimental methods 2-1 and 2-2, acetone and water were mixed in a 2:1 (v:v) ratio to dissolve the active ingredient, and then Tween 20 (Sigma Aldrich, Germany) was added at a concentration of 1000 ppm. A flat-pan nozzle 8001 (Spraying Systems Co., Glendale Heights, IL, USA) was mounted on a CO2-pressurized belt-driven sprayer (R&D sprayers, Opelousas, LA, USA) to treat the crabgrass plants prepared in experimental method 1 with the herbicide at a spraying rate of 300 L / ha. After herbicide treatment, the plants were immediately transferred to a growing bed, and the samples were arranged according to a randomized block design with 4 replications for experimental method 2-1 and 5 replications for experimental method 2-2.

[0321]

[0322] 3. RGB Image Acquisition and Analysis

[0323] RGB images of plants were acquired using a CMOS camera (EOS-600D, Canon, Tokyo, Japan) at 3 hours, 6 hours, 1 day, 2 days, 3 days, 4 days, 5 days, 6 days, 7 days, 8 days, 9 days, and 10 days after herbicide treatment, and analyzed using MATLAB R2024b (TheMathWorks Inc., Natick, MA, USA). The acquired RGB images were converted to the L*a*b color space, a threshold was set, and the background was removed. To measure the visual response and greenness level of the plants, the red (R), green (G), and blue (B) values ​​of the plants were normalized using Equation 1 of Example 1, Experimental Method 3, and the NDI (Normalized Difference Index) and ExG (Excess Green Index) were calculated.

[0324]

[0325] 4. Chlorophyll Fluorescence Imaging Acquisition and Analysis

[0326] Plant CF images were acquired using the SNU-KIST 2 imaging system (Seoul National University / KIST, Seoul / Gangneung, South Korea) at 3 hours, 6 hours, 1 day, 2 days, 3 days, 4 days, 5 days, 6 days, 7 days, 8 days, 9 days, and 10 days after herbicide treatment. This device consists of a machine vision camera (NRI GigE Camera, Basler, Germany) equipped with a band-pass filter (High Performance Longpass Filter, Edmund Optics, NJ, USA). Each CF image was represented by red pixels according to the acquired chlorophyll fluorescence value, and the CF value was derived by applying an Otsu threshold to each CF image to remove the background and then averaging the values ​​of the plant pixels. Finally, CF image parameters were calculated using Equation 2 of Example 2 and Experimental Method 4.

[0327]

[0328] 5. IR Thermal Imaging Acquisition and Analysis

[0329] 33 at 3 hours, 6 hours, 1 day, 2 days, 3 days, 4 days, 5 days, 6 days, 7 days, 8 days, 9 days, and 10 days after herbicide treatment o IR thermal images of the plants were acquired using an A65sc IR thermal imaging camera (FLIR, Wilsonville, OR, USA) mounted on a plant growth chamber maintained at C. The leaf temperatures of the plants were accurately determined by identifying the plants through alignment with the IR thermal images based on the previously segmented RGB images.

[0330]

[0331] 6. Statistical Analysis

[0332] For all spectral image data, a two-way analysis of variance (2-way ANOVA) was first performed, followed by an examination of the interactions between herbicides, time, and herbicides and time. Subsequently, the spectral parameters acquired from the herbicide treatment groups in Experimental Method 2-1 were evaluated using Principal Component Analysis (PCA) to assess whether spectral responses could be separated and clustered according to the herbicide mechanism of action. Initially, PCA was performed individually for each image acquisition time point (3, 6, 24, 48, 72, 96, and 120 hours), and then the data from all acquisition times points were integrated and re-analyzed. Meanwhile, the similarity of spectral parameters between herbicides was measured using Manhattan distances calculated based on the spectral parameters, and UPGMA analysis was conducted to generate clusters of herbicide mechanisms of action. All statistical analyses were performed using R 4.3.2 (R Foundation for Statistical Computing, Vienna, Austria).

[0333]

[0334] 7. Training and Testing of Herbicide Mechanism of Action Classification Models

[0335] Spectral parameters obtained from RGB, CF, and IR thermal image analysis of the herbicide treatment groups in Experimental Method 2-2 were used as input data to train a herbicide mechanism of action classification model. The dataset was randomly divided into five groups for cross-validation to prevent the possibility of overfitting. All learning algorithms provided by the Classifier Learner Application in MATLAB 2024b (Table 6) were used to calculate the validation accuracy for the mechanism of action of each herbicide according to Equation 5 of Example 2.

[0336] Based on the generated classification model, test accuracy was calculated using the dataset obtained from the herbicide treatment group in Experimental Method 2-1. The herbicide mechanism of action was estimated using the spectral parameters of the herbicide treatment group dataset in Experimental Method 2-1 and then compared with the mechanism of action of the herbicide actually applied. Test accuracy was calculated according to the same method as described in Equation 5 of Example 2.

[0337]

[0338] <Experimental Results>

[0339] 1. Confirmation of overall changes in spectral response to crabgrass growth inhibition by herbicide treatment

[0340] RGB, IR thermal images, and CF images obtained in experimental methods 3 to 5 showed the gradual changes in visual herbicide efficacy and growth inhibition of crabgrass over time (Fig. 20). Each letter represents the untreated group (A), cetoxydim (ACCase, B), benfuresate (lipid synthesis, C), S-metholachlor (VLCFA, D), pendimethalin (microtubules, E), indaziflam (cellulose, F), pyribenzoxime (ALS, G), glyphosate (EPSPS, H), glufosinate (GS, I), tetuflupyrrolimet (DHODH, J), dicamba (Auxin, K), oxaziclomefon (unknown, L), paraquat (PSI, M), bentazone (PSII, N), thiafenacil (PPO, O), clomazone (DOXPS, P), and isoxaflutol (HPPD, Q). Visual changes were observed in RGB images when barnyard grass was treated with cetoxydim (ACCase), benfuresate (lipid synthesis), S-metholachlor (VLCFA), glyphosate (EPSPS), glufosinate (GS), paraquat (PSI), thiafenacil (PPO), clomazone (DOXPS), and isoxaflutol (HPPD). In IR thermal imaging, leaf temperature was found to increase with pendimethalin (microtubules), glyphosate (EPSPS), glufosinate (GS), thiafenacil (PPO), clomazone (DOXPS), and isoxaflutol (HPPD). In the CF video, it was found that CF values ​​decreased for cetoxydim (ACCase), pendimethalin (microtubules), glyphosate (EPSPS), glufosinate (GS), paraquat (PSI), bentazone (PSII), tiafenacil (PPO), clomazone (DOXPS), and isoxaflutol (HPPD).

[0341] In the untreated group, the normalized g value increased over time, while the b value decreased. Meanwhile, the normalized r value remained almost the same regardless of the measurement time point. After herbicide treatment, the normalized r value increased, with the exception of some herbicides, and distinct patterns were observed depending on the herbicide's mechanism of action (Fig. 20). When comparing the normalized r, g, and b values, the first time point at which the parameter was found to be statistically significant varied depending on the herbicide. At 48 hours, cetoxydim (ACCase), at 24 hours, benfuresate (lipid synthesis), at 24 hours, S-metholachlor (VLCFA), at 3 hours, pendimethalin (microtubules), at 48 hours, pyribenzoxime (ALS), at 24 hours, glufosinate (GS), at 72 hours, tetflupyrrolimet (DHODH), at 3 hours, paraquat (PSI), at 24 hours, tiaphenacil (PPO), at 48 hours, clomazone (DOXPS), at 48 hours, isooxaflutol (HPPD), at 72 hours, indaziflam (cellulose), at 48 hours, dicamba (auxin), and at 72 hours, oxaziclomephon (unknown mechanism of action); however, bentazone (PSII) was found to have no effect.

[0342] Leaf temperature increased after herbicide treatment, with the exception of some herbicides, and distinct patterns were observed depending on the herbicide's mechanism of action (Fig. 20). As a result of comparing temperatures, the first time point at which spectral parameters showed a statistically significant difference varied depending on the herbicide, and was not significant for cetoxydim (ACCase), benfuresate (lipid synthesis), S-metholachlor (VLCFA), pyribenzoxime (ALS), tetflupyrrolimet (DHODH), indaziflam (cellulose), and oxaziclomefon (unknown mechanism of action). Pendimethalin (microtubules) appeared at 72 hours, glyphosate (EPSPS) at 24 hours, gluphosate (GS) at 24 hours, paraquat (PSI) at 3 hours, bentazone (PSII) at 3 hours, tiphenacil (PPO) at 24 hours, clomazone (DOXPS) at 72 hours, isoxaflutol (HPPD) at 48 hours, and dicamba (auxin) at 216 hours.

[0343] Except for some herbicides, CF parameters decreased after herbicide treatment, and distinct patterns were observed according to the herbicide mechanism of action (Fig. 20). The first time point at which the CF parameter showed a statistically significant difference in the comparison results varied depending on the herbicide, and was not significant for cethoxydim (ACCase), benfurate (lipid synthesis), S-metholachlor (VLCFA), pyribenzoxime (ALS), tetflupyrrolimet (DHODH), indaziflam (cellulose), and oxaziclomefon; it was at 72 hours for pendimethalin (microtubules), 24 hours for glyphosate (EPSPS), 24 hours for glufosinate (GS), 3 hours for paraquat (PSI), 3 hours for bentazone (PSII), 24 hours for tiafenacil (PPO), 72 hours for clomazone (DOXPS), 48 hours for isoxaflutol (HPPD), and 216 hours for dicamba (auxin).

[0344]

[0345] 2. Confirmation of time-dependent changes in spectral response to crabgrass growth inhibition by herbicide treatment

[0346] Analysis of RGB, IR thermal images and CF images showed a gradual change in the spectral index of barnyard grass over time (Fig. 21).

[0347] Specifically, thiafenacil (PPO, B) was shown to cause a rapid decrease in ExG and NDI after herbicide treatment. After 24 hours, compared to the untreated group (A), leaf temperature increased and CF parameters decreased significantly. Unlike thiafenacil (PPO, B), isoxaflutol (HPPD, C) showed a slow decrease in NDI, ExG, and CF parameters. Leaf temperature also increased slowly, peaking at 72 hours. Unlike the untreated group, ExG remained stable even after treatment with cetoxydim (ACCase, D). A slight increase in leaf temperature was observed at 48 and 72 hours. Pyribenzoxime (ALS, E) and dicamba (auxin, F) did not show significant changes after herbicide treatment.

[0348]

[0349] 3. Confirmation of clustering of spectral responses to the herbicide mechanism of action of crabgrass

[0350] PCA was performed independently using spectral parameters acquired at each time point. Three herbicides, paraquat (PSI), bentazone (PSII), and pendimethalin (microtubules), were the first to be isolated from the untreated group. At 24 hours after herbicide treatment, glufosinate (GS), glyphosate (EPSPS), and thiafenacil (PPO) began to be isolated from the untreated group. At 72 hours, clomazone (DOXPS), isoxaflutol (HPPD), and cetoxydim (ACCase) began to be isolated, while bentazone (PSII) and pendimethalin (microtubules) approached the untreated group. At 72 hours, cetoxydim (ACCase) began to be isolated, and after 120 hours, no further herbicides were isolated from the untreated group until the end of the measurement, with the exception of benfuresate (lipid synthesis). PCA was performed by integrating spectral parameters obtained 3 to 120 hours after herbicide treatment, and it was confirmed that the herbicide mechanism of action could be differentiated and clustered with the untreated control group (Fig. 22).

[0351]

[0352] 4. Confirmation of community formation of crabgrass according to the mechanism of action of each herbicide

[0353] Since the spectral response up to 120 hours of treatment was significant in Experimental Method 2-1, 69 types of herbicides were applied to the weed crabgrass, and the spectral response was observed from 0 to 120 hours after herbicide treatment (Figs. 23 to 25). Except for ALS inhibitors, lines of the same color indicate the same mechanism of action of the herbicides. In the case of ALS inhibitors, lines of the same color indicate the same chemical structure. The measurement results showed that herbicides sharing the same mechanism of action exhibited similar spectral responses to one another, with some exceptions.

[0354] Hierarchical clustering phylogenetic analysis was performed to cluster herbicide mechanisms of action based on integrated spectral parameters and the UPGMA algorithm. Most ALS inhibitors, with the exception of imazapyr (ALS), were clustered into the same group. HPPD, ACCase, and ALS inhibitors formed different groups. GS, DOXPS inhibitors, and oxaziclomefon (unidentified mechanism of action) were located close to each other. Lipid synthesis and VLCFA inhibitors were located close to the control group. PPO, PSII, and PSI inhibitors were also located close to each other. Microtubule and auxin herbicides formed separate groups (Fig. 26).

[0355]

[0356] 5. Construction of a Diagnostic Model for Herbicide Mechanism of Action Based on Spectral Response and Machine Learning

[0357] According to previous experimental results, spectral parameters were confirmed to differ depending on the herbicide's mechanism of action. RGB-based parameters showed a decreasing trend after herbicide treatment. F, a CF-based parameter v / F m In the case of [specific element], it also showed a decreasing trend after herbicide treatment, while leaf temperature analyzed in IR thermal imaging increased after herbicide treatment. Based on these various spectral parameters, the herbicide activity and mechanism of action were estimated by performing an integrated machine learning analysis of all spectral parameters using the method of Example 2 and Experimental Method 7. As a result of measuring the learning and testing accuracy for the nine mechanisms of action in which the herbicide effect was observed, it was confirmed that the Bagged Trees algorithm had the highest testing accuracy of 86.1% among the machine learning algorithms (Figs. 27 and 28).

[0358]

[0359] The present invention has been described above with reference to its embodiments. Those skilled in the art will understand that the present invention may be embodied in modified forms without departing from the essential characteristics of the invention. Therefore, the disclosed embodiments should be considered in an illustrative rather than a restrictive sense. The scope of the invention is defined by the claims, not by the foregoing description, and all variations within the scope of equivalents should be interpreted as being included in the invention.

Claims

1. A step of treating multiple plant individuals with multiple herbicides having known mechanisms of action, each according to their respective mechanisms of action; A step of generating image data from the above plant entities; A step of generating spectrum parameter data from the above image data; A step of generating a model for predicting the efficacy and mechanism of action of herbicides on plants by training a statistical algorithm using the above parameter data; A step of treating a plant with a herbicide candidate substance whose mechanism of action is unknown to generate image data of the plant treated with the candidate substance, and inputting spectral parameter data generated from the image data of the plant treated with the candidate substance into the prediction model to diagnose the efficacy and mechanism of action of the herbicide candidate substance on the plant. A method for diagnosing the efficacy and mechanism of action of herbicides on plants including 2. In claim 1, the mechanism of action is acetyl CoA carboxylase inhibition, branched amino acid biosynthesis inhibition (ALS inhibition), aromatic amino acid biosynthesis inhibition (EPSP inhibition), glutamine synthase inhibition, photosystem II inhibition (D1 Serine 264 binders), photosystem II inhibition (D1 Histidine 215 binders), photosystem I electron transfer inhibition (bipyridylium system), chlorophyll biosynthesis inhibition (PPO inhibition), carotenoid biosynthesis inhibition (PDS inhibition), carotenoid biosynthesis inhibition (HPPD inhibition), carotenoid biosynthesis inhibition (Lycopene Cyclase), DOXPS (1-deoxy-D-xylulose 5-phosphate synthase) inhibition, folate biosynthesis inhibition (Asulam), microtubule assembly inhibition, mitosis / microtubule formation inhibition, long chain A method for diagnosing the efficacy and mechanism of action of a herbicide on plants, wherein the mechanism of action is one or more selected from the group consisting of inhibition of very low-grade fatty acid (VLCFA) synthesis, inhibition of cell wall (cellulose) synthesis, inhibition of fatty acid thioacetase (TE), inhibition of lipid synthesis, inhibition of pyrimidine biosynthesis (Dihydroorotate dehydrogenase, DHODH), membrane disruption, auxin (indoleacetic acid) analogue, and inhibition of auxin transport.

3. A method for diagnosing the efficacy and mechanism of action of a herbicide on plants according to claim 1, wherein the image data is generated from one or more images selected from the group consisting of RGB images, chlorophyll fluorescence images, and IR thermal images.

4. In claim 1, the spectral parameters are mNDI (modified Normalized Difference Index), NDI (Normalized Difference Index), ExG (Excess Green Index), F v / F m , ΦPSⅡ, F d / F m A method for diagnosing the efficacy and mechanism of action of a herbicide on plants, which is one or more parameters selected from a group consisting of temperature difference and temperature index.

5. A method for diagnosing the efficacy and mechanism of action of a herbicide according to claim 4, wherein the mNDI, NDI, and ExG are obtained from the following mathematical formula 1: [Mathematical Formula 1] R: Red RGB value G: Green RGB value B: Blue RGB values NDI h NDI of herbicide-treated plants NDI0: NDI of plants not treated with herbicides.

6. In Clause 4, the above F v / F m , ΦPSⅡ and F d / F m A method for diagnosing the efficacy and mechanism of action of a herbicide obtained from the following mathematical formula 2: [Mathematical Formula 2] F0: Baseline chlorophyll fluorescence value in dark adaptation state F m : Chlorophyll fluorescence value at the point (1 second) when the plant emits maximum fluorescence after exposure to light F s : Chlorophyll fluorescence value at the point when the plant returns to standard conditions (60 seconds).

7. A method for diagnosing the efficacy and mechanism of action of a herbicide, wherein the temperature difference in Clause 4 is obtained from the following mathematical formula 3: [Mathematical Formula 3] T h : Leaf temperature of plants treated with herbicides T0: Leaf temperature of plants not treated with herbicides.

8. A method for diagnosing the efficacy and mechanism of action of a herbicide, wherein the temperature index in Paragraph 4 is obtained from the following mathematical formula 4: [Mathematical Formula 4] T h : Leaf temperature of plants treated with herbicides T0: Leaf temperature of plants not treated with herbicides.

9. A method for diagnosing the efficacy and mechanism of action of a herbicide on a plant, wherein the plant is barnyard grass or rapeseed, in accordance with claim 1.

10. A method for diagnosing the efficacy and mechanism of action of a herbicide on plants, wherein the above-mentioned barnyard grass is grown in a multi-well plate.

11. In claim 1, the herbicide is Cyhalofop-butyl, Metamifop, Fenoxaprop-ethyl, Fluazifop-P-butyl, Sethoxydim, Clethodim, Benfuresate, Mefenacet, S-metolachlor, Alachlor, Butachlor, Napropamide, Fenoxasulfone, Imazapyr, Imazaquin, Flucetosulfuron, Flazasulfuron, Pyrazosulfuron, Nicosulfuron, Imazosulfuron, Pyrimisulfan, Bensulfuron-methyl, Propyrisulfuron, Azimsulfuron, Trifloxysulfuron-sodium, Triafamone, Pyriminobac-methyl, Pyribenzoxim, Bispyribac-sodium, Penoxsulam, Paraquat, Atrazine, Bentazone, Linuron, Dimethamethrine, Propanil, Pyraclonil, Tiafenacil, Bifenox, Oxyfluorfen, Carfentrazone-ethyl, Pentoxazone, Pyraflufen-ethyl, Oxadiargyl,Fluthiacet-methyl, Dicamba, Florpyrauxifen-benzyl, 2,4-D, MCPP, MCPA, Fluroxypyr-meptyl, Triclopyr TEA, Tetflupyrolimet, Pendimethalin, Ethalfluralin, Oryzalin, Dithiopyr, Isoxaben, Indaziflam, Clomazone), Isoxaflutole, Mesotrione, Tefuryltrione, Benzobicyclon, Pyrazolate, A method for diagnosing the efficacy and mechanism of action of a herbicide on plants, wherein the herbicide is one or more selected from the group consisting of tolpyralate, glyphosate, glufosinate-ammonium, and oxaziclomefone.

12. A method for diagnosing the efficacy and mechanism of action of a herbicide on plants, wherein the statistical algorithm in claim 1 is a principal component analysis or machine learning algorithm.

13. A method for diagnosing the efficacy and mechanism of action of a herbicide on plants, wherein the machine learning algorithm in claim 12 is Subspace Discriminant or Bagged Trees.

14. A method for diagnosing the efficacy and mechanism of action of a herbicide on a plant, wherein the step of generating the image data in claim 1 is performed at two or more points in time between 3 hours and 144 hours after the step of processing each of the plurality of plant individuals.

15. A herbicide treatment unit that treats multiple herbicides with known mechanisms of action on multiple plant individuals according to their respective mechanisms of action; Image data generation unit that generates image data from the above plant entities; Spectrum parameter data generation unit that generates spectrum parameter data from the above image data; A learning unit that generates a model for predicting the efficacy and mechanism of action of herbicides on plants using the above parameter data; and A diagnostic unit that diagnoses the efficacy and mechanism of action of a herbicide candidate substance on plants by treating plants with a herbicide candidate substance whose mechanism of action is unknown to generate image data of the plants treated with the candidate substance, and inputting the image data of the plants treated with the candidate substance into the prediction model. A device for diagnosing the efficacy and mechanism of action of herbicides on plants, including

Images