Systems, devices, and methods for determining refined petiole nutrient values based on leaf spectral data

Through portable spectrophotometers and machine learning models, real-time and non-destructive measurement of plant petiole nutrient values ​​is achieved, solving the accuracy and effectiveness of traditional methods, and providing efficient remedial measures.

CN119023610APending Publication Date: 2024-11-26DALHOUSIE UNIVERSITY
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410656656.2
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Priority Date
2024-04-05
Filing Date
2024-05-24
Publication Date
2024-11-26

AI Technical Summary

Technical Problem

The prior art is difficult to determine the nutrient values ​​of plant petioles in real time and non-destructively, resulting in the effectiveness of remedial measures being affected, and traditional methods are disturbed by the atmosphere and soil, and the measurement results are inaccurate.

Method used

Non-destructive spectral data are collected from plant leaves using a portable spectrophotometer, petiole nutrient values ​​are calculated through machine learning models, and results are provided in real time to avoid atmospheric and soil interference, and accurate measurement of petiole nutrients are achieved.

Benefits of technology

Near real-time, non-destructive measurement of plant petiole nutrient values ​​is achieved, improving the effectiveness of remedial measures and the accuracy of measurement results, and reducing damage to plants.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119023610A_ABST
    Figure CN119023610A_ABST
Patent Text Reader

Abstract

A system, apparatus, and method are provided for determining refined petiole nutrient values based on leaf spectral data. The method includes storing leaf spectral data in a memory, converting, using a processor, the leaf spectral data into preliminary petiole nutrient values by inputting the leaf spectral data into one or more linear machine learning models trained to provide a petiole nutrient value output based on the leaf spectral data input, and refining, using a processor, the preliminary petiole nutrient values by inputting the preliminary petiole nutrient values into one or more non-linear machine learning models. One or more non-linear machine learning models are trained to provide a refined petiole nutrient value output based on the preliminary petiole nutrient value input.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This document relates to a system, apparatus and method for determining petiole nutrient values ​​based on leaf spectral data from plant leaves. Background Art

[0002] Plants require many nutrients to grow and maintain life. For example, plants require large amounts of macronutrients like nitrogen (N), phosphorus (P), potassium (K), calcium (Ca), magnesium (Mg), and sulfur (S). Plants also require smaller amounts of micronutrients like manganese (Mn), zinc (Zn), iron (Fe), sodium (Na), copper (Cu), aluminum (Al), and boron (B).

[0003] In agricultural applications, if the crop does not receive the appropriate amounts of various required nutrients, the crop yield and / or quality of the harvested crop may be affected. For example, if the plant does not receive enough nutrients, the crop yield and / or quality may be reduced. In some cases, an oversupply of nutrients may also reduce crop yield and / or quality.

[0004] Determining the nutrient value status of a plant can enable remedial measures to improve crop yield and / or quality. For example, the composition and amount of fertilizer can be customized for one or more specific nutrients based on the determined nutrient value status of the plant. Summary of the invention

[0005] The following summary is provided to introduce the reader to the more detailed discussion below. The summary is not intended to limit or define any claimed or unclaimed invention. One or more inventions may reside in any combination or subcombination of elements or process steps disclosed in any part of this document, including its claims and drawings.

[0006] According to some aspects, a method for non-destructively determining petiole nutrient values ​​from plant leaves in a crop field in near real time using a portable spectrophotometer is provided. The method may include: (a) taking non-destructive spectral measurements of leaves of plants in a crop field using a portable spectrophotometer to generate leaf spectral data, (b) storing the leaf spectral data in a memory; (c) calculating, by a processor, a petiole nutrient value based on the stored leaf spectral data; and (d) providing near real-time results indicating the petiole nutrient value.

[0007] According to some aspects, a device for non-destructively determining petiole nutrient values ​​from plant leaves in a crop field in near real time is provided. The device may include: a memory storing one or more machine learning models trained to provide petiole nutrient value outputs based on leaf spectral data input; and at least one processor communicatively coupled to the memory. The at least one processor may be configured to collectively: receive leaf spectral data generated by a portable spectrophotometer based on non-destructive spectral measurements of leaves of plants in a crop field; store the leaf spectral data in the memory; input the leaf spectral data into one or more machine learning models; receive outputs including petiole nutrient values ​​determined by one or more machine learning models based on the leaf spectral data; and provide near real-time results to a mobile device, the near real-time results indicating the petiole nutrient value.

[0008] According to some aspects, a non-transitory computer-readable medium is provided, which includes instructions that can be executed by a processor. When the instructions are executed, the processor can be configured to: receive leaf spectral data generated by a portable spectrophotometer based on non-destructive spectral measurements of leaves of plants in a crop field; store the leaf spectral data in a memory; input the leaf spectral data into one or more machine learning models trained to provide a petiole nutrient value output based on the leaf spectral data input; receive an output including a petiole nutrient value determined by the one or more machine learning models based on the leaf spectral data; and provide a near real-time result indicating the petiole nutrient value.

[0009] According to some aspects, a system for non-destructively determining petiole nutrient values ​​from plant leaves in a crop field in near real time is provided. The system may include a portable spectrophotometer and a remote server. The portable spectrophotometer may be configured to: generate leaf spectral data based on non-destructive spectral measurements of leaves of plants in a crop field; and transmit the leaf spectral data via a network to be received by a remote server. The remote server may include a processor and a memory, the memory storing one or more machine learning models trained to provide petiole nutrient value outputs based on leaf spectral data input. The remote server may be configured to: receive leaf spectral data; store the leaf spectral data in a memory; input the leaf spectral data into one or more machine learning models; receive petiole nutrient values ​​determined by one or more machine learning models based on the leaf spectral data; and provide near real-time results indicating petiole nutrient values ​​to a mobile device.

[0010] According to some aspects, a method for determining a refined petiole nutrient value based on leaf spectral data from a plant leaf is provided. The method may include: (a) storing the leaf spectral data in a memory, (b) using a processor, converting the leaf spectral data into a preliminary petiole nutrient value by inputting the leaf spectral data into one or more linear machine learning models trained to provide a petiole nutrient value output based on the leaf spectral data input, the one or more linear machine learning models outputting a preliminary petiole nutrient value; (c) using a processor, refining the preliminary petiole nutrient value by inputting the preliminary petiole nutrient value into one or more nonlinear machine learning models trained to provide a refined petiole nutrient value output based on the preliminary petiole nutrient value input, the one or more nonlinear machine learning models outputting a refined petiole nutrient value; and (d) outputting a plant nutrient report indicating the refined petiole nutrient value.

[0011] According to some aspects, a non-transitory computer-readable medium is provided, which includes instructions that can be executed by a processor. When the instructions are executed, the processor can be configured to: store leaf spectral data from plant leaves in a memory; use the processor to convert the leaf spectral data into a preliminary petiole nutrient value by inputting the leaf spectral data into one or more linear machine learning models trained to provide a petiole nutrient value output based on the leaf spectral data input; receive outputs including the preliminary petiole nutrient value from the one or more linear machine learning models; use the processor to refine the preliminary petiole nutrient value by inputting the preliminary petiole nutrient value into one or more nonlinear machine learning models trained to provide a refined petiole nutrient value output based on the preliminary petiole nutrient value input; receive outputs including the refined petiole nutrient value from the one or more nonlinear machine learning models; and output a plant nutrient report indicating the refined petiole nutrient value.

[0012] According to some aspects, a system for determining a refined petiole nutrient value based on leaf spectral data from a plant leaf is provided. The system may include a memory and a processor. The memory may be configured to store leaf spectral data from a plant leaf. The processor may be configured to: convert the leaf spectral data into a preliminary petiole nutrient value by inputting the leaf spectral data into one or more linear machine learning models trained to provide a petiole nutrient value output based on the leaf spectral data input; receive outputs including preliminary petiole nutrient values ​​from one or more linear machine learning models; refine the preliminary petiole nutrient value by inputting the preliminary petiole nutrient value into one or more nonlinear machine learning models trained to provide a refined petiole nutrient value output based on the preliminary petiole nutrient value input; receive outputs including refined petiole nutrient values ​​from one or more nonlinear machine learning models; and output a plant nutrient report indicating a refined petiole nutrient value. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] The drawings included herein are for illustrating various examples of the articles, methods, and apparatuses of the present specification and are not intended to limit the scope of the teachings in any way. In the drawings:

[0014] Figure 1 is a diagram of a leaf of a plant for which a nutrient value is to be determined according to an embodiment;

[0015] Figure 2 is a schematic diagram of a system for non-destructively determining petiole nutrient values ​​from plant leaves in a crop field in near real time according to an embodiment;

[0016] Figure 3 Is Figure 2 A schematic diagram of the results indicating petiole nutrient values ​​provided by the system and graphically displayed on the mobile device;

[0017] Figure 4 is a schematic diagram of an apparatus for non-destructively determining petiole nutrient values ​​from plant leaves in a crop field in near real time according to an embodiment;

[0018] Figure 5 is a flow chart illustrating an example method for non-destructively determining petiole nutrient values ​​from plant leaves in a crop field in near real time using a portable spectrophotometer;

[0019] Figure 6 is a table illustrating correlations of nutrient absorption characteristics of potato plants according to an embodiment; and

[0020] Figure 7 is a flow chart illustrating an example method for determining refined petiole nutrient values ​​based on leaf spectral data from plant leaves. DETAILED DESCRIPTION

[0021] Many embodiments are described in this application, and these embodiments are presented for illustrative purposes only. The described embodiments are not intended to be limiting in any sense. The present invention is widely applicable to many embodiments, as is apparent from the disclosure herein. Those skilled in the art will recognize that the present invention can be practiced by modification and change without departing from the teachings disclosed herein. Although the specific features of the present invention may be described with reference to one or more specific embodiments or the accompanying drawings, it should be understood that such features are not limited to the use in one or more specific embodiments or the accompanying drawings to which they are described with reference.

[0022] The terms "embodiment," "one or more embodiments," "some embodiments," and "one embodiment" mean "one or more (but not all) embodiments of the invention," unless expressly stated otherwise.

[0023] The terms "include," "comprising," and variations thereof mean "including but not limited to," unless expressly stated otherwise. A listing of items does not imply that any or all of the items are mutually exclusive, unless expressly stated otherwise. The terms "a," "an," and "the" mean "one or more," unless expressly stated otherwise.

[0024] As used herein and in the claims, two or more parts are referred to as being "coupled," "connected," "attached," "engaged," "attached," or "fastened," wherein the parts are joined or work together directly or indirectly (i.e., through one or more intermediate parts) as long as there is a link. As used herein and in the claims, two or more parts are referred to as being "directly coupled," "directly connected," "directly attached," "directly engaged," "directly attached," or "directly fastened," wherein the parts are connected in physical contact with one another. As used herein, two or more parts are referred to as being "rigidly coupled," "rigidly connected," "rigidly attached," "rigidly engaged," "rigidly attached," or "rigidly fastened," wherein the parts are coupled so as to move as a unit while maintaining a constant orientation relative to one another. The terms "coupled," "connected," "attached," "engaged," "attached," and "fastened" do not distinguish the manner in which two or more parts are joined together.

[0025] Further, although the method steps may be described in sequence (in the disclosure and / or claims), such methods may be configured to work in an alternating order. In other words, any order or sequence of steps that may be described does not necessarily imply a requirement to perform the steps in that order. The steps of the methods described herein may be performed in any feasible order. Further, some steps may be performed simultaneously.

[0026] As used herein and in the claims, a group of elements is referred to as 'jointly' performing an action, wherein the action is performed by any one of the elements in the group or is performed cooperatively by two or more (or all) elements in the group.

[0027] As used herein and in the claims, a first element is referred to as being “received” in a second element, wherein at least a portion of the first element is received in the second element, unless specifically stated otherwise.

[0028] Some elements herein may be identified by a part number consisting of a base number followed by an alphabetic or subscript numeric suffix (e.g., 112a or 1121). Multiple elements herein may be identified by part numbers that share the same base number but have different suffixes (e.g., 1121, 1122, and 1123). All elements having a common base number may be collectively or generally referred to using the base number without a suffix (e.g., 112).

[0029] As used herein and in the claims, "upward," "downward," "above," "below," "upward," "vertical," "height," "up," "down," and similar terms refer to a directionality that is generally aligned with (e.g., parallel to) gravity. The terms "distal end," "proximal end," and similar terms refer to a directionality that is generally transverse to (e.g., perpendicular to) gravity. However, none of the terms mentioned in this paragraph imply any particular alignment between elements. For example, a first element may be referred to as being "vertically above" a second element, wherein the first element is at a higher height than the second element, and regardless of whether the first element is vertically aligned with the second element.

[0030] Typically, plant tissue tests used to determine the status of plant nutrient values ​​are destructive and do not provide real-time results. For example, farmers can harvest petioles from plants in a crop field for measurement. The harvested petioles can be transported to a laboratory for testing. Depending on the test to be performed, a large number of petioles may need to be destructively removed.

[0031] The laboratory can then perform plant tissue testing and provide farmers with petiole nutrient values. The process of petiole removal, shipping, testing, and notification of results can take weeks to complete. This delay undermines the effectiveness of any remedial actions taken based on the results (e.g., adjusting fertilizer composition and amount) because the nutrient composition of the measured plant may have changed in the meantime.

[0032] The disclosed systems, devices, and methods can provide non-destructive and near real-time determination of the state of plant nutrient values ​​using a portable spectrophotometer. The frequency at which a molecule absorbs light is a characteristic of its structure. The portable spectrophotometer can be used to take non-destructive spectral measurements to indicate the presence and concentration of different molecules (nutrients) in a probe area (e.g., a leaf of a plant). The portable spectrophotometer can generate leaf spectral data (based on non-destructive spectral measurements) that can be used to determine the state of the nutrient values ​​of the plant. There is no need to remove plant leaves or petioles from the plant for testing. The disclosed systems, devices, and methods can provide the state of plant nutrient values ​​without destroying any plant used for testing.

[0033] Further, the disclosed systems, apparatus, and methods can perform analysis and provide results in near real time, e.g., within 10-20 seconds. This can enable any required remedial actions (based on the determined state of the plant nutrient value) to be immediately performed. As used herein and in the claims, a measure is said to be performed in "near real time" if it is performed in less than 1 minute from the time of initiation.

[0034] In canopy reflectance measurements, a ground or aerial platform is used to determine the ratio of the amount of light leaving the canopy to the amount of incident light. Canopy reflectance measurements can be used to calculate crop nutrient status. However, canopy reflectance measurements are often subject to atmospheric and soil interference. The disclosed systems, devices, and methods can use a portable spectrophotometer to probe plant leaves, which avoids atmospheric interference and / or soil interference associated with using a ground or aerial platform for canopy reflectance measurements.

[0035] Reference now Figure 1 , which shows a diagram of a plant 10 whose nutrient value status can be determined using the disclosed system, device or method. In the illustrated example, the plant 10 includes a stem 14, a leaf 18, and a petiole 22 that attaches the leaf 18 to the stem 14. The leaf 18 includes a blade 26 and a plurality of veins 30a, 30b.

[0036] The surface area of ​​the leaf blade 26 may be much larger than the petiole 22, so the spectral reflectance measurements may be taken from the leaf blade 26 rather than the petiole 22. Thus, the measured nutrient values ​​may indicate nutrient values ​​corresponding to the leaf blade rather than the petiole nutrient values.

[0037] Nutrient values ​​in the petiole are not the same as those measured in other parts of the plant (e.g., leaves). Farmers, now and in the past, send petioles (specifically, the fourth petiole from the top of the plant) to the lab, and the lab returns the petiole nutrient values. Thus, farmers often know what remedial measures need to be taken based on the lab petiole nutrient values ​​rather than the nutrient values ​​associated with other parts of the plant.

[0038] The disclosed systems, devices and methods can take spectral measurements from leaves of plants to generate leaf spectral data and calculate petiole nutrient values ​​based on the leaf spectral data. Providing petiole nutrient values ​​(rather than nutrient values ​​associated with leaves) can enable users to select and apply well-known and understood remedies.

[0039] Reference now Figure 2 , which shows a schematic diagram of a system 100 for non-destructively determining petiole nutrient values ​​in near real time from plant leaves 18 in a crop field 34 according to an embodiment. As shown, the system 100 may include a portable spectrophotometer 104 and a remote server 108.

[0040] In the illustrated example, the portable spectrophotometer 104 and the remote server 108 are communicatively coupled using a network 112. The network 112 may include a communication network, such as the Internet, a wide area network (WAN), a local area network (LAN), or other types of networks. In other examples, the portable spectrophotometer 104 and the remote server 108 may not be connected to a common network, and may communicate using an intermediate device and / or communication network.

[0041] The crop field 34 may include a plurality of plants 10 (e.g., at least 500 plants). Each plant 10 may have one or more leaves 18a, 18b, 18c. The plants 10 may include any suitable plants grown in agricultural applications. For example, the plants 10 may include wheat, corn, soybeans, potatoes, etc. The crop field 34 may be located in an open air environment or in a greenhouse.

[0042] The portable spectrophotometer 104 can have any suitable design as long as it is portable and capable of taking non-destructive spectral measurements of leaves of the plant 10. The portable spectrophotometer 104 can be sized and designed to be carried by an operator while the operator uses the portable spectrophotometer to take spectral measurements. For example, the portable spectrophotometer 104 can be battery powered and designed to be carried by the operator 38 in a backpack.

[0043] A light source within the portable spectrophotometer 104 may generate incident light onto the leaves, and a detector within the portable spectrophotometer 104 may measure the light reflected by the leaves of the plant 10. The portable spectrophotometer 104 may generate leaf spectral data based on the reflected spectral measurements.

[0044] In some embodiments, the portable spectrophotometer 104 can take spectral measurements and generate leaf spectral data within a visible and / or near-infrared wavelength range. For example, the portable spectrophotometer 104 can generate leaf spectral data with a wavelength range of 400nm-2500nm. In other examples, the portable spectrophotometer 104 can generate leaf spectral data within a smaller (e.g., 400nm-2000nm) or larger (e.g., 350nm-2500nm) wavelength range. A smaller wavelength range can enable spectral measurements to be completed faster and / or require fewer resources to calculate petiole nutrient values. A larger wavelength range can enable petiole nutrient values ​​to be determined with higher accuracy and / or detect a larger amount of nutrients. The wavelength range can be adjusted manually or automatically based on the nutrient value being measured. For example, the operator 38 can manually adjust the wavelength range. In other examples, the wavelength range can be adjusted automatically, such as by a remote server 108.

[0045] The portable spectrophotometer 104 can take spectral measurements at discrete wavelength intervals within the measured wavelength range. For example, the portable spectrophotometer 104 can take spectral measurements at 0.5nm wavelength intervals within the measured wavelength range. The wavelength intervals used during the measurement can be manually or automatically adjusted based on the nutrient value being determined. For example, the operator 38 can manually adjust the wavelength intervals. In other examples, the wavelength intervals can be automatically adjusted, such as by the remote server 108. Different wavelength intervals ranging from 0.1nm to 10nm can be used. Smaller wavelength intervals can make it possible to determine nutrient values ​​with higher accuracy and / or determine a larger amount of nutrients. Larger wavelength intervals can make it possible to determine nutrient values ​​faster and / or require less computing resources to determine nutrient values.

[0046] In some embodiments, the portable spectrophotometer 104 may be a NIRS TM DS2500Analyzer or ASD TM FieldSpec4 Spectroradiometer. In other embodiments, the portable spectrophotometer 104 may be any other suitable spectrophotometer.

[0047] The portable spectrophotometer 104 may be configured to transmit the generated leaf spectral data to be received by the remote server 108. In the illustrated example, the portable spectrophotometer 104 may transmit the generated leaf spectral data to the remote server 108 via the network 112. In other examples, the portable spectrophotometer 104 may transmit the generated leaf spectral data to an intermediate device, such as the mobile device 42 of the operator 38. The portable spectrophotometer 104 may use point-to-point communication or a local area network (e.g., Bluetooth TM ) transmits the generated leaf spectrum data to the mobile device 42, and then the mobile device 42 can transmit the leaf spectrum data to the remote server 108 using a large area network / Internet.

[0048] The mobile device 42 may be any suitable device as long as it is portable and capable of sending and receiving data via a communication network (e.g., a wireless communication network, such as a cell phone network and / or a Wi-Fi network). For example, the mobile device 42 may be a laptop, a tablet device, or a smartphone with wireless communication capabilities. The mobile device 42 may be a dedicated companion device designed to be used with the portable spectrophotometer 104. In some embodiments, the mobile device 42 may be a general-purpose device that includes proprietary software for communicating with the portable spectrophotometer 104 and / or analyzing data received from the portable spectrophotometer 104. In some embodiments, the functionality of the portable spectrophotometer 104 and the mobile device 42 may be provided by a single device.

[0049] In the illustrated example, the remote server 108 includes a processor 116 and a memory 120. The processor 116 can control the operation of the remote server 108. The processor 116 can be any suitable processor, controller, or digital signal processor that can provide sufficient processing power depending on the configuration, usage, and requirements of the system 100, as known to those skilled in the art. For example, the processor 116 can be a high-performance general-purpose processor. For example, the processor 116 can include a standard processor such as Processor or Processor. Alternatively, processor 116 may include more than one processor, each of which is configured to perform different dedicated tasks. Alternatively, some of the functions provided by processor 116 may be provided using specialized hardware (e.g., a graphics processing unit (GPU)).

[0050] The memory 120 may include one or more of a random access memory (RAM), a read-only memory (ROM), a hard disk drive, and a flash memory (e.g., a solid-state drive). The remote server 108 may store the received leaf spectrum data in the memory 120. For example, the remote server 108 may receive the leaf spectrum data generated by the portable spectrophotometer 104 and store the received leaf spectrum data in the memory 120.

[0051] The memory 120 may also store a machine learning (ML) model trained to provide a petiole nutrient value output based on leaf spectral data input. The machine learning model may be generated and trained by the remote server 108. In some embodiments, the remote server 108 may receive a machine learning model generated by an external device. The remote server 108 may train the received machine learning model and store the trained machine learning model in the memory 120. In other embodiments, the remote server 108 may receive a trained machine learning model from an external device. The remote server 108 may store the received trained machine learning model in the memory 120.

[0052] In various embodiments, any one or more (or all) of the memories described herein for storing machine learning models (e.g., memory 120) may store a machine learning pipeline that includes the machine learning model in addition to associated data preprocessing code (e.g., for building, training, and / or evaluating the machine learning model).

[0053] The leaf spectral data input provided to the machine learning model may include, for example, leaf spectral data generated by the portable spectrophotometer 104 by taking spectral measurements from leaves of the plant. The petiole nutrient value output provided by the machine learning model may include, for example, petiole nutrient values ​​determined for one or more nutrients such as N, P, K, Ca, Mg, S, Mn, Zn, Fe, B, Cu, Al, and Na. The petiole nutrient values ​​for different nutrients may be expressed as percentage values ​​(of total mass), g / kg, and / or parts per million (ppm).

[0054] The machine learning model can be any suitable machine learning model based on the input data to be analyzed, the output data requirements, the available training data and / or the available computing resources. In some embodiments, the machine learning model can be composed of one or more univariate or multivariate multiple linear regression (MLR) models. Lasso or ridge regularization can be used in the MLR model to generate a lasso or ridge regression model. In other embodiments, the machine learning model can be a different model, such as a polynomial regression model. In other embodiments, other data preprocessing steps can also be used, such as dimensionality reduction by the maximum relevance minimum redundancy (MRMR) method.

[0055] Training data for a machine learning model can be generated by taking spectral measurements of a plant (e.g., reflectance spectral measurements of a leaf portion) and generating leaf spectral data, where the leaf spectral data serves as a predictor (x) for the machine learning model. The corresponding petiole of the plant can be removed and used for laboratory tissue analysis. The results of the laboratory tissue analysis of the petiole can be used as a response (y) for the machine learning model. A correlation model can be established between the response (y) and the predictor (x). The following equation (1) provides the correlation coefficient β using i An example of an MLR model:

[0056]

[0057] The number of predictors (x) being greater than the number of responses (y) can lead to overfitting of the MLR model. The accuracy of the MLR model can be improved by using a subset selection method to select a subset of predictors. In some embodiments, lassoMLR modeling can be used as a subset selection method. In other embodiments, other subset selection methods can also be used, such as stepwise subset selection or Bayesian model averaging.

[0058] Lasso MLR modeling can be used to identify the most informative and least redundant predictors (wavelengths) using a complexity parameter (λ) that controls the amount of shrinkage: the larger the value of λ, the greater the penalty for non-zero coefficients in the model, and therefore the greater the shrinkage that can be applied to the coefficient values. A lasso regression model can be generated by selecting a value of λ that minimizes the root mean square error (RMSE). The selected λ parameter can determine the number of coefficients that will make up the lasso regression model, which are selected as the coefficients with the greatest explanatory power related to the target variable to be predicted. For example, model training and performance evaluation can be performed using K-fold cross-validation, which uses different integer values ​​of K, such as 5, where the value of λ is selected based on the minimum RMSE. Table 1 below shows the number, range, and values ​​of the top four significant bands obtained from lasso MLR modeling of different elements, such as leaf spectral data obtained from potato plants.

[0059] Table 1 Number, range and values ​​of the first four significant bands obtained from Lasso MLR modeling of different elements, for example, leaf spectral data obtained from potato plants

[0060]

[0061]

[0062] The processor 116 can input the received leaf spectral data (e.g., from the portable spectrophotometer 104) and / or the stored leaf spectral data (e.g., in the memory 120) into the trained machine learning model. The machine learning model can estimate the petiole nutrient value based on the leaf spectral data. For example, the machine learning model can estimate the petiole nutrient value for N, P, K, Ca, Mg, S, Mn, Zn, Fe, B, Cu, Al, and Na based on the leaf spectral data generated by taking spectral measurements of leaves of potato plants. The processor 116 can receive the petiole nutrient value determined by the machine learning model.

[0063] The processor 116 can provide near real-time results indicative of petiole nutrient values ​​to the mobile device. For example, the processor 116 can provide near real-time results to the mobile device 42 of the operator 38. For example, the processor 116 can provide near real-time results to the mobile device 42 in less than 1 minute (e.g., within 0.01 to 30 seconds) from when the spectral measurement is sent by the portable spectrophotometer. This can enable the operator 38 to non-destructively collect near real-time results indicative of petiole nutrient values ​​from multiple plants in a crop field.

[0064] Reference now Figure 2 and Figure 3 . Figure 3A schematic diagram of the results of indicating petiole nutrient values ​​provided by the system 100 and displayed graphically on a mobile device 42 is shown. The mobile device 42 may include a touch screen display 46 that provides an interactive display to the operator 38. The touch screen display 46 may provide a plurality of display portions 50a-50c. For example, the display portion 50a may provide a graphical display of the determined petiole nutrient values. In the illustrated example, the display portion 50a provides a bar graph showing the determined petiole nutrient values ​​of six nutrients. In other embodiments, the display 46 may not be configured to detect touch input.

[0065] Display portion 50b may provide additional information about the determined petiole nutrient value, such as the number of leaf spectral measurements performed, the confidence level associated with the determined petiole nutrient value, etc. Display portion 50c may provide recommended remedial measures based on the determined petiole nutrient value. In some examples, one or more of display portions 50a, 50b, 50c may include other information unrelated to the results indicating the petiole nutrient value.

[0066] Reference now Figure 4 , which shows a schematic diagram of an apparatus 200 for non-destructively determining petiole nutrient values ​​from plant leaves in a crop field in near real time. In some embodiments, the apparatus 200 can be implemented as a server device (e.g., Figure 2 In other embodiments, the apparatus 200 may be implemented using any other suitable hardware components. For example, the apparatus 200 may be implemented as a mobile device 42 ( Figure 2 and Figure 3 ), which is combined with the portable spectrophotometer 104 ( Figure 2 ) or is implemented to be integrated with the portable spectrophotometer 104 ( Figure 2 ) is a matching device used together with the

[0067] for Figure 4 In the illustrated example embodiment, the apparatus 200 includes a memory 208, an application 212, an output device 216, a display device 220, an auxiliary storage device 224, a processor 228, an input device 232, and a communication device 236. One or more (or all) of the memory 208, the application 212, the output device 216, the display device 220, the auxiliary storage device 224, the processor 228, the input device 232, and the communication device 236 may be communicatively coupled via wire and / or wirelessly.

[0068] In some embodiments, apparatus 200 includes multiple of any one or more of memory 208, application 212, output device 216, display device 220, secondary storage device 224, processor 228, input device 232, and communication device 236. In some embodiments, apparatus 200 does not include one or more of application 212, secondary storage device 224, network connection, input device 232, output device 216, display device 220, and communication device 236.

[0069] In at least one embodiment, the device 200 includes a connection to a network 204, such as a wired or wireless connection to the Internet or a private network. In some cases, the network 204 includes other types of computer or telecommunication networks. The device 200 can receive data from a portable spectrophotometer (e.g., Figure 2 The portable spectrophotometer 104) is based on the crop field (e.g., Figure 2 Leaf spectral data generated by non-destructive spectral measurements of leaves of plants (e.g., potato plants) in a crop field 34).

[0070] In some embodiments, the apparatus 200 may receive the leaf spectral data directly from the portable spectrophotometer, either wired or wirelessly. In other embodiments, the apparatus 200 may receive the leaf spectral data from an intermediate device (e.g., Figure 2 A mobile device 42) receives leaf spectral data generated by a portable spectrophotometer.

[0071] The memory 208 may include one or more of a random access memory (RAM) and a read-only memory (ROM). In some embodiments, the memory 208 stores one or more applications 212 executed by the processor 228. The application 212 corresponds to a software module that includes computer executable instructions to perform processing for the functions and methods described herein.

[0072] The memory 208 may store a machine learning model trained to provide a petiole nutrient value output based on leaf spectral data input. The machine learning model may be generated and trained by the apparatus 200. In some embodiments, the apparatus 200 may receive the generated machine learning model from an external device (e.g., via the network 204), and may train the received machine learning model. In some embodiments, the apparatus 200 may receive a trained machine learning model from an external device (e.g., via the network 204), and store the trained machine learning model in the memory 208. The machine learning model may be any suitable model, such as a lasso regression model.

[0073] In some embodiments, memory 208 may also store received leaf spectral data, determined petiole nutrient values, and / or results derived from petiole nutrient values ​​(eg, graphical representations of petiole nutrient values). The stored data may be retrieved by a user for historical data analysis.

[0074] The secondary storage device 224 may include any suitable non-transitory computer-readable medium that includes instructions that can be executed by a processor (e.g., processor 228). For example, the secondary storage device 224 may include a hard disk drive, a floppy disk drive, a CD drive, a DVD drive, a Blu-ray drive, a solid-state drive, a flash memory, or other type of non-volatile data storage device. The processor 228 may execute the instructions included on the secondary storage device 224 to perform processing for the functions and methods described herein.

[0075] In some embodiments, apparatus 200 stores information in a remote storage device (such as cloud storage) accessible via a network (such as network 204 or another network). In some embodiments, apparatus 200 stores information distributed across multiple storage devices (such as memory 208 and auxiliary storage device 224) (i.e., each of the multiple storage devices stores a portion of the information, and the multiple storage devices collectively store all of the information). Thus, as used herein and in the claims, storing data on a storage device means storing the data in a local storage device, storing the data in a remote storage device, or distributing the data across multiple storage devices, each of which may be local or remote.

[0076] Input device 232 may include any device for inputting information into apparatus 200. For example, input device 232 may be a keyboard, a keypad, a cursor device, a touch screen, a camera, or a microphone. Input device 232 may also include input ports and wireless radios (e.g., or 802.11x).

[0077] Display device 220 may include any type of device for presenting visual information. For example, display device 220 may be a computer monitor, a flat screen display, a projector, or a display panel.

[0078] Output device 216 may include any type of device for presenting a hard copy of information, such as, for example, a printer. Output device 216 may also include other types of output devices, such as speakers. In at least one embodiment, output device 216 includes an output port and a wireless radio (e.g., or 802.11x).

[0079] The communication device 236 may be of any design suitable for receiving analog and / or digital inputs therefrom and providing analog and / or digital outputs. In some embodiments, the communication device 236 may include separate modules for analog and digital signals.

[0080] Processor 228 may be any device capable of executing an application, computer-readable instruction, or program. The application, computer-readable instruction, or program may be stored in memory 208 or secondary storage device 224, or may be received from a remote storage device accessible via network 204. Processor 228 may be a high-performance general-purpose processor, a standard processor (e.g., Processor or processor), specialized hardware (e.g., a GPU), or multiple processing devices that collectively perform the functions provided by processor 228.

[0081] Processor 228 can input leaf spectral data (e.g., leaf spectral data stored in memory 208) into a trained machine learning model (e.g., a trained machine learning model stored in memory 208). The machine learning model can determine a petiole nutrient value based on the leaf spectral data. Processor 228 can receive an output from the machine learning model including a petiole nutrient value determined by the machine learning model.

[0082] In some embodiments, processor 228 may send a signal to a mobile device (e.g., Figure 2 The processor 228 may provide the results to the display device 220, and the petiole nutrient value may be displayed graphically on the display device 220.

[0083] Figure 4 An example hardware schematic diagram of the apparatus 200 is shown. In alternative embodiments, the apparatus 200 includes fewer, additional, or different components. Further, although various aspects of the embodiments of the apparatus 200 are described as being stored in a memory, those skilled in the art will appreciate that these aspects may also be stored on or read from other types of computer program products or computer-readable media (such as auxiliary storage devices, including hard disks, floppy disks, CDs, or DVDs), carrier waves from the Internet or other networks, or other forms of RAM or ROM. For example, the apparatus 200 may include a non-transitory computer-readable medium storing computer-readable instructions that, when executed by the processor 228, configure the processor 228 to perform the methods described herein.

[0084] Reference now Figure 5, which shows a flowchart illustrating an example method 300 for non-destructively determining petiole nutrient values ​​from plant leaves in a crop field in near real time using a portable spectrophotometer. The method 300 can be performed, for example, using the system 100 ( Figure 2 ) or device 200( Figure 4 ) to execute, and also refer to Figure 2 and Figure 4 .

[0085] The method 300 may be performed while carrying a portable spectrophotometer in a crop field. For example, the portable spectrophotometer 104 may be carried in the crop field 34 by a person or a vehicle.

[0086] At 308, non-destructive spectral measurements of leaves of plants in a crop field may be taken using the carried spectrophotometer to generate leaf spectral data. For example, non-destructive spectral measurements of leaves of potato plants may be taken using the portable spectrophotometer 104. The portable spectrophotometer 104 may generate the leaf spectral data based on the spectral measurements.

[0087] At 312, the leaf spectral data may be stored in a memory. In some embodiments, the generated leaf spectral data may be sent to a remote server via a network and stored in a memory of the remote server. For example, the portable spectrophotometer 104 may send the generated leaf spectral data to the remote server 108 via the network 112, and the remote server 108 may store the leaf spectral data in the memory 120.

[0088] In other embodiments, the generated leaf spectral data may be sent to a mobile device, and the mobile device may send the leaf spectral data to a remote server via a network. The leaf spectral data may be stored in a memory of the remote server. For example, the portable spectrophotometer 104 may send the generated leaf spectral data to the mobile device 42, and the mobile device 42 may send the leaf spectral data to the remote server 108 via the network 112. The remote server 108 may store the leaf spectral data in the memory 120.

[0089] At 316, the processor may calculate the petiole nutrient value based on the stored leaf spectral data. The processor may calculate the petiole nutrient value by inputting the leaf spectral data into a machine learning model that is trained to provide a petiole nutrient value output based on the leaf spectral data input. In response, the machine learning model may output the petiole nutrient value to the processor. For example, the processor 116 (or processor 228) may input the leaf spectral data into a trained machine learning model stored in the memory 120 (or memory 208). The machine learning model may be a lasso regression model.

[0090] At 320, near real-time results indicating petiole nutrient values ​​may be provided. In some embodiments, near real-time results may be provided by sending the petiole nutrient values ​​to a mobile device. The petiole nutrient values ​​may then be graphically displayed on the mobile device. For example, the remote server 108 or the device 200 may provide near real-time results by sending the petiole nutrient values ​​to the mobile device 42. The mobile device 42 may graphically display the petiole nutrient values.

[0091] In some embodiments, if there are additional leaves for measurement, method 300 may proceed to 308. For example, method 300 may loop through steps 308 to 320 for at least five other plant leaves in a crop field over a ten-minute period. This may enable near real-time results indicating petiole nutrient values ​​corresponding to non-destructive spectral measurements from multiple leaves over a short period of time. This in turn would allow the farm to simultaneously react to the returned petiole nutrient values ​​(e.g., by adjusting the composition and / or amount of fertilizer) before changes in the measured nutrient composition of the plant occur. This may increase the effectiveness of remedial measures, which may increase the yield and / or amount of the crop.

[0092] In other embodiments, processing leaf spectral data using machine learning may follow a two-tier approach to provide a more accurate estimate of petiole nutrient values ​​(referred to herein as "refined petiole nutrient values"). The first ML layer may involve one or more first machine learning models (e.g., Figure 1-Figure 5 The first ML layer may involve one or more independent instances of the machine learning models described herein. The second ML layer may involve one or more second machine learning models. The petiole nutrient values ​​determined by the first machine learning model in the first layer may be refined to produce statistically more accurate values ​​(i.e., the values ​​of one or more petiole nutrients are closer to the actual values ​​of the one or more nutrients of the petiole as obtained by removing the petiole and destructively measuring it in a laboratory according to traditional methods). In these embodiments, the petiole nutrient values ​​determined by the first layer machine learning model may be referred to as preliminary petiole nutrient values.

[0093] The first-tier machine learning model can be trained and stored in the manner described above. In some embodiments, only a small subset of the most explanatory features (i.e., intensity values ​​associated with specific wavelengths or wavelength intervals of the collected leaf spectral data) are input into the first-tier machine learning model. This solves the problem of the data set including a large number of features (i.e., the high dimensionality of the feature space). Before placing the features into the first-tier machine learning model, a feature reduction step (e.g., minimum redundancy maximum relevance (MRMR) feature selection) can be performed independently for each target variable to extract a subset of features to be input into the first-tier machine learning model. Preselecting features can improve the estimation performance of most nutrients. A model regularization step, such as the lasso regression model explained above, can also be performed to remove any features that are not considered to contribute significantly to the predictive ability of the model.

[0094] The memory 120 (or memory 208) may store a second-layer machine learning model that is trained to provide a refined petiole nutrient value output based on the preliminary petiole nutrient value output by the first-layer machine learning model. The second-layer machine learning model may be generated and trained by the remote server 108. In some embodiments, the remote server 108 may receive a second-layer machine learning model generated by an external device. The remote server 108 may train the received second-layer machine learning model and store the trained second-layer machine learning model in the memory 120 (or memory 208). In other embodiments, the trained second-layer machine learning model may be received by the remote server 108 from an external device. The remote server 108 may store the received trained second-layer machine learning model in the memory 120 (or memory 208).

[0095] The second layer of machine learning models can be composed of any suitable type of machine learning models based on the input data to be analyzed, the output data requirements, the available training data, and / or the available computing resources. In some embodiments, one or more (or all) of the second machine learning models used in the second layer can be nonlinear multivariate regression models, such as but not limited to support vector regression (SVR), random forest (RF), or extreme gradient boosting (XGB).

[0096] As described above, the first machine learning model in the first layer can be trained using a dataset containing leaf spectral data and a matching set of laboratory measured petiole nutrient values. This allows the first layer machine learning model to establish a relationship between the leaf spectral data values ​​and the individual petiole nutrient values.

[0097] Typically, due to the intricate relationships between nutrient concentrations in plants, there are overlapping effects on reflectivity at different wavelengths. A plant species has a certain tendency to absorb multiple nutrients (i.e., a species has specific nutrient absorption characteristics). The absorption of certain nutrients will be affected by the absorption of other nutrients. For example, for a plant species, when the plant absorbs a certain nutrient "x", it will also tend to absorb nutrient "y" (i.e., there is a positive correlation between nutrient "x" and nutrient "y"). But when the plant absorbs nutrient "x", it may also tend to reduce the absorption of nutrient "z" (i.e., there is a negative correlation between nutrient "x" and nutrient "z"). For example, it is known that potassium (K) and magnesium (Mg) are competing cations for plant absorption due to their antagonism.

[0098] Therefore, refining preliminary petiole nutrient values ​​determined from leaf spectral data based on interconnections between nutrients for a given plant species may provide statistically more accurate results.

[0099] The nonlinearity of the second machine learning model in the second layer allows the model to consider these nutrient absorption characteristics and potentially related results simultaneously. The second machine learning model in the second layer can be trained using a dataset containing a large number of different petiole nutrient values. The second machine learning model in the second layer can process this data to establish absorption relationships between different nutrients for a given plant species. Importantly, nutrient absorption characteristics will vary between different plant species.

[0100] Figure 6 A table 400 is shown which illustrates example nutrient absorption characteristics of a potato plant. Figure 6 As can be seen in the graph, for example, the uptake of nitrogen (N) by the potato plant tends to promote the uptake of phosphorus (P) (shown as a positive value of 0.83). However, when the potato plant absorbs nitrogen, it also tends to reduce the uptake of boron (B) (shown as a negative value of -0.66). In other words, there is a positive correlation between the uptake of nitrogen and phosphorus, but a negative correlation between the uptake of nitrogen and boron. Figure 6 As shown, correlations between each expected petiole nutrient can be determined. Taking into account the correlations between different nutrient absorptions allows the refined petiole nutrient values ​​to be statistically more accurate than the preliminary petiole nutrient values ​​output by the first machine learning model of the first layer input from the leaf spectral data.

[0101] There are many reasons for positive and negative correlations in the absorption of different nutrients. Some examples that may apply to one or more potato plants are illustrated below. The positive relationship between N and P may be because N fixation may depend on the availability of P in the soil for plant absorption. The offsetting effect between calcium (Ca) and P may be caused by the precipitation of insoluble calcium phosphate near the nutrient absorption roots. K and Mg may be competitive cations for plant absorption. Ca and Mg may have a synergistic effect on maintaining a stable pH value in the soil. N may be negatively interconnected with Mg because the rate of increase of N can be reduced by the rate of addition of Mg. B has a negative correlation with P, but the availability of B is positively affected by the presence of Ca. Plant absorption of K can be attributed to the role of N in leaf area development and photosynthesis. Applying K in the form of sulfate may result in the assimilation distribution of sulfur (S). For example, some potash fertilizers include potassium as K2SO4, so that the application of potassium is accompanied by sulfur. An increase in K supply (e.g., supplied to the ground by fertilizer) generally leads to an increase in plant absorption of K, which can reduce the concentration of manganese (Mn) in the leaves. Increased S supply (e.g., supplied to the ground via fertilizers) generally results in increased S uptake by the plant, which can improve the uptake of K and P. High doses of K can reduce the uptake of iron (Fe) due to increased tolerance to Fe toxicity. There is an inverse effect between copper (Cu) and Ca, as the transport of Ca from roots to leaves can be disrupted in the presence of Cu.

[0102] Reference now Figure 7 , which shows a flowchart illustrating an example method 500 for determining a refined petiole nutrient value based on leaf spectral data from a plant leaf. The method 500 can be performed, for example, using the system 100 ( Figure 2 ) or device 200( Figure 4 ) to execute, and also refer to Figure 2 and Figure 4 .

[0103] The method 500 may be performed while carrying a portable spectrophotometer in a crop field. For example, the portable spectrophotometer 104 may be carried by a person or vehicle in the crop field 34. The method 500 may also be performed using any other suitable spectrophotometer, including non-portable spectrophotometers.

[0104] Method 500 may optionally include step 504. At 504, spectral measurements of leaves of a plant may be taken using a spectrophotometer (eg, portable spectrophotometer 104) to generate leaf spectral data. For example, spectral measurements of leaves of a potato plant may be taken.

[0105] At 506, the leaf spectral data may be stored in a memory. In some embodiments, the generated leaf spectral data may be sent to a remote server via a network and stored in a memory of the remote server. For example, the portable spectrophotometer 104 may send the generated leaf spectral data to the remote server 108 via the network 112, and the remote server 108 may store the leaf spectral data in the memory 120.

[0106] In other embodiments, the generated leaf spectral data may be sent to a mobile device, and the mobile device may send the leaf spectral data to a remote server via a network. The leaf spectral data may be stored in a memory of the remote server. For example, the portable spectrophotometer 104 may send the generated leaf spectral data to the mobile device 42, and the mobile device 42 may send the leaf spectral data to the remote server 108 via the network 112. The remote server 108 may store the leaf spectral data in the memory 120.

[0107] At 508, the processor may convert the leaf spectral data into preliminary petiole nutrient values. The processor may convert the leaf spectral data by inputting the leaf spectral data into one or more linear machine learning models that are trained to provide petiole nutrient value outputs based on the leaf spectral data input. In response, the linear machine learning model may output the preliminary petiole nutrient value to the processor. For example, the processor 116 (or processor 228) may input the leaf spectral data into one or more trained linear machine learning models stored in the memory 120 (or memory 208). One or more (or all) of the linear machine learning models may be lasso regression models.

[0108] At 510, the processor may refine the preliminary petiole nutrient value by inputting the preliminary petiole nutrient value into one or more nonlinear machine learning models trained to provide a refined petiole nutrient value output based on the preliminary petiole nutrient value input. In response, the nonlinear machine learning model may output the refined petiole nutrient value to the processor. For example, the processor 116 (or processor 228) may input the preliminary petiole nutrient value into a trained nonlinear machine learning model stored in the memory 120 (or memory 208). One or more (or all) of the nonlinear machine learning models may be a multiple regression model.

[0109] At 512, a plant nutrient report indicating the refined petiole nutrient value can be output. The plant nutrient report can have any suitable format, including but not limited to a list, table, chart, graphical representation, any type of visualization (e.g., infographic, map, etc.), or animation of the refined petiole nutrient value. In some embodiments, the plant nutrient report can be output by sending the refined petiole nutrient value to a mobile device. For example, the remote server 108 or device 200 can provide near real-time results (e.g., within less than 1 minute from the completion of step 506) by sending the refined petiole nutrient value to the mobile device 42. The mobile device 42 can display the petiole nutrient value graphically.

[0110] In some embodiments, if there are additional leaves for measurement, method 500 may proceed to step 504 or 506. For example, method 500 may cycle through steps 504 and / or 506 to 512 for at least five other plant leaves in a crop field within approximately one hour. This may enable results to be obtained that indicate refined petiole nutrient values ​​corresponding to spectral measurements from multiple leaves over a short period of time. This in turn would allow the farm to react simultaneously to the returned refined petiole nutrient values ​​(e.g., by adjusting the composition and / or amount of fertilizer) before changes occur in the measured nutrient composition of the plant. This may increase the effectiveness of remedial measures, thereby increasing the yield and / or amount of the crop.

[0111] At 514, treatments may optionally be applied to the crop (e.g., including the measured plants and / or in the vicinity thereof). Treatments may include applying (e.g., spraying or spreading) liquid and / or solid (e.g., granular) fertilizers to the crop. Treatments may be determined based at least in part on the refined nutrient values ​​of step 512 (e.g., by processor 116 or 228). Treatments may be applied under the control of a human operator and / or as directed by electronic automation (e.g., by processor 116 or 228).

[0112] The two-layer approach described in this paper includes both linear and nonlinear machine learning models, which can greatly improve the accuracy of nutrient predictions. In particular, estimates for Zn, Fe, and Al are substantially more accurate than when only linear machine learning models are used in the single-layer approach.

[0113] Although the above description provides examples of embodiments, it should be understood that some features and / or functions of the described embodiments are susceptible to modification without departing from the spirit and principles of operation of the described embodiments. Therefore, the above description is intended to illustrate the present invention and is non-limiting, and those skilled in the art will understand that other variations and modifications may be made without departing from the scope of the present invention as defined by the appended claims. The scope of the claims should not be limited by the preferred embodiments and examples, but should be given the broadest interpretation consistent with the entire specification.

[0114] project

[0115] Item 1: A method for non-destructively determining petiole nutrient values ​​from plant leaves in a crop field in near real time using a portable spectrophotometer, the method comprising: (a) taking non-destructive spectral measurements of leaves of plants in a crop field using a portable spectrophotometer to generate leaf spectral data, (b) storing the leaf spectral data in a memory; (c) calculating, by a processor, the petiole nutrient value based on the stored leaf spectral data; and (d) providing near real-time results indicating the petiole nutrient value.

[0116] Item 2: The method of any preceding item, wherein the providing comprises sending the petiole nutrient value to a mobile device.

[0117] Item 3: The method of any preceding item, wherein the providing further comprises graphically displaying the petiole nutrient value on a mobile device.

[0118] Item 4: The method of any preceding item, wherein the storing comprises sending the leaf spectral data to a remote server via a network, wherein the remote server comprises a memory and a processor.

[0119] Item 5: The method of any preceding item, wherein the storing comprises a mobile device receiving the leaf spectral data, and the mobile device sending the leaf spectral data to a remote server via a network, wherein the remote server comprises a memory and a processor.

[0120] Item 6: The method of any preceding item, wherein the providing comprises sending the petiole nutrient value from the remote server to the mobile device and graphically displaying the petiole nutrient value on the mobile device.

[0121] Item 7: The method of any preceding item, further comprising repeating steps (a) to (d) for at least five other plant leaves in the crop field within ten minutes.

[0122] Item 8: The method as in any preceding item, wherein the plant is a potato plant.

[0123] Item 9: A method as in any of the preceding items, wherein the calculation comprises inputting leaf spectral data into one or more machine learning models trained to provide petiole nutrient value outputs based on the leaf spectral data input, the one or more machine learning models outputting petiole nutrient values.

[0124] Item 10: A method as in any of the preceding items, wherein each of the one or more machine learning models is a lasso regression model.

[0125] Item 11: A device for non-destructively determining petiole nutrient values ​​from plant leaves in a crop field in near real time, the device comprising: a memory storing one or more machine learning models trained to provide petiole nutrient value outputs based on leaf spectral data input; and at least one processor, the at least one processor being communicatively coupled to the memory, the at least one processor being configured to jointly: receive leaf spectral data generated by a portable spectrophotometer based on non-destructive spectral measurements of leaves of plants in a crop field; store the leaf spectral data in the memory; input the leaf spectral data into one or more machine learning models; receive outputs including petiole nutrient values ​​determined by one or more machine learning models based on the leaf spectral data; and provide near real-time results to a mobile device, the near real-time results indicating the petiole nutrient values.

[0126] Item 12: The device of any preceding item, wherein the petiole nutrient value is displayed graphically on the mobile device.

[0127] Item 13: The apparatus of any preceding item, wherein the leaf spectral data generated by the portable spectrophotometer is received from a mobile device via a network.

[0128] Item 14: The device as in any preceding item, wherein the plant is a potato plant.

[0129] Item 15: An apparatus as in any of the preceding items, wherein each of the one or more machine learning models is a lasso regression model.

[0130] Item 16: A device as in any preceding item, wherein the memory comprises one or more of a random access memory (RAM), a read only memory (ROM), a hard drive, and a flash memory.

[0131] Item 17: A non-transitory computer-readable medium comprising instructions executable by a processor, wherein when the instructions are executed, the processor is configured to: receive leaf spectral data generated by a portable spectrophotometer based on non-destructive spectral measurements of leaves of plants in a crop field; store the leaf spectral data in a memory; input the leaf spectral data into one or more machine learning models trained to provide a petiole nutrient value output based on the leaf spectral data input; receive an output comprising a petiole nutrient value determined by one or more machine learning models based on the leaf spectral data; and provide a near real-time result indicative of the petiole nutrient value.

[0132] Item 18: The non-transitory computer readable medium of any preceding item, wherein the leaf spectral data generated by the portable spectrophotometer is received from a mobile device over a network.

[0133] Item 19: The non-transitory computer readable medium as in any preceding item, wherein the plant is a potato plant.

[0134] Item 20: A non-transitory computer-readable medium as in any preceding item, wherein each of the one or more machine learning models is a lasso regression model.

[0135] Item 21: A non-transitory computer-readable medium as in any preceding item, wherein the memory comprises one or more of a random access memory (RAM), a read only memory (ROM), a hard drive, and a flash memory.

[0136] Item 22: A system for non-destructively determining petiole nutrient values ​​from plant leaves in a crop field in near real time, the system comprising: a portable spectrophotometer configured to: generate leaf spectral data based on non-destructive spectral measurements of leaves of plants in the crop field; and transmit the leaf spectral data via a network to be received by a remote server; and a remote server comprising a processor and a memory, the memory storing one or more machine learning models trained to provide petiole nutrient value outputs based on leaf spectral data input, the remote server configured to: receive the leaf spectral data; store the leaf spectral data in the memory; input the leaf spectral data into one or more machine learning models; receive the petiole nutrient value determined by the one or more machine learning models based on the leaf spectral data; and provide near real-time results indicating the petiole nutrient value to a mobile device.

[0137] Item 23: The system of any preceding item, wherein the petiole nutrient value is graphically displayed on a mobile device.

[0138] Item 24: The system as in any preceding item, wherein the portable spectrophotometer is configured to transmit the leaf spectral data to a mobile device, and the mobile device transmits the leaf spectral data to a remote server. Item 25: The system as in any preceding item, wherein the plant is a potato plant.

[0139] Item 26: A system as in any of the preceding items, wherein each of the one or more machine learning models is a lasso regression model.

[0140] Item 27: A system as in any preceding item, wherein the memory comprises one or more of a random access memory (RAM), a read only memory (ROM), a hard drive, and a flash memory.

[0141] Item 28: A method for determining a refined petiole nutrient value based on leaf spectral data from a plant leaf, the method comprising: (a) storing the leaf spectral data in a memory; (b) using a processor to convert the leaf spectral data into a preliminary petiole nutrient value by inputting the leaf spectral data into one or more linear machine learning models trained to provide a petiole nutrient value output based on the leaf spectral data input, the one or more linear machine learning models outputting the preliminary petiole nutrient value; (c) using a processor to refine the preliminary petiole nutrient value by inputting the preliminary petiole nutrient value into one or more nonlinear machine learning models trained to provide a refined petiole nutrient value output based on the preliminary petiole nutrient value input, the one or more nonlinear machine learning models outputting a refined petiole nutrient value; and (d) outputting a plant nutrient report indicating the refined petiole nutrient value.

[0142] Item 29: A method as in any preceding item, wherein one or more nonlinear machine learning models are trained to model the interrelationships of nutrient absorption characteristics of different phytonutrients.

[0143] Item 30: The method of any preceding item, further comprising, prior to said storing, taking spectral measurements of leaves of the plant using a spectrophotometer to generate leaf spectral data.

[0144] Item 31: A method as in any of the preceding items, wherein each of the one or more linear machine learning models is a lasso regression model.

[0145] Item 32: A method as in any of the preceding items, wherein each of the one or more nonlinear machine learning models is a multivariate regression model.

[0146] Item 33: A method as in any of the preceding items, wherein each of the one or more nonlinear machine learning models is one of the following: a support vector regression (SVR) model, a random forest (RF) model, and an extreme gradient boosting (XGB) model.

[0147] Item 34: The method of any preceding item, wherein the outputting comprises sending the refined petiole nutrient value to a mobile device.

[0148] Item 35: The method of any preceding item, wherein the outputting further comprises graphically displaying the refined petiole nutrient value on a mobile device.

[0149] Item 36: The method of any preceding item, wherein the storing comprises sending the leaf spectral data to a remote server via a network, wherein the remote server comprises a memory and a processor.

[0150] Item 37: The method of any preceding item, wherein the storing comprises the mobile device receiving the leaf spectral data, and the mobile device sending the leaf spectral data to a remote server via a network, wherein the remote server comprises a memory and a processor.

[0151] Item 38: The method of any preceding item, wherein the outputting comprises sending the refined petiole nutrient value from the remote server to the mobile device and graphically displaying the refined petiole nutrient value on the mobile device.

[0152] Item 39: The method as in any preceding item, wherein the plant leaves are potato plant leaves.

[0153] Item 40: The method of any preceding item, further comprising applying a fertilizer treatment to the plant based at least in part on the refined petiole nutrient value.

[0154] Item 41: A non-transitory computer-readable medium comprising instructions executable by a processor, wherein when the instructions are executed, the processor is configured to: store leaf spectral data from plant leaves in a memory; use the processor to convert the leaf spectral data into a preliminary petiole nutrient value by inputting the leaf spectral data into one or more linear machine learning models trained to provide a petiole nutrient value output based on the leaf spectral data input; receive an output including the preliminary petiole nutrient value from the one or more linear machine learning models; use the processor to refine the preliminary petiole nutrient value by inputting the preliminary petiole nutrient value into one or more nonlinear machine learning models trained to provide a refined petiole nutrient value output based on the preliminary petiole nutrient value input; receive an output including the refined petiole nutrient value from the one or more nonlinear machine learning models; and output a plant nutrient report indicating the refined petiole nutrient value.

[0155] Item 42: The non-transitory computer readable medium of any preceding item, wherein the one or more non-linear machine learning models are trained to model interrelationships of nutrient absorption characteristics of different phytonutrients.

[0156] Item 43: The non-transitory computer readable medium of any preceding item, wherein when the instructions are executed, the processor is configured to receive spectral data of a leaf of a plant taken using a spectrophotometer.

[0157] Item 44: A non-transitory computer-readable medium as in any preceding item, wherein each of the one or more linear machine learning models is a lasso regression model.

[0158] Item 45: A non-transitory computer-readable medium as in any preceding item, wherein each of the one or more nonlinear machine learning models is a multivariate regression model.

[0159] Item 46: A non-transitory computer-readable medium as in any of the preceding items, wherein each of the one or more non-linear machine learning models is one of: a support vector regression (SVR) model, a random forest (RF) model, and an extreme gradient boosting (XGB) model. Item 47: A non-transitory computer-readable medium as in any of the preceding items, wherein the output includes sending a refined petiole nutrient value to a mobile device.

[0160] Item 48: The non-transitory computer readable medium of any preceding item, wherein the instructions, when executed, configure the processor to apply a fertilizer treatment to the plant based at least in part on the refined petiole nutrient value.

[0161] Item 49: A system for determining a refined petiole nutrient value based on leaf spectral data from plant leaves, the system comprising: a memory configured to store leaf spectral data from plant leaves; and a processor configured to: convert the leaf spectral data into a preliminary petiole nutrient value by inputting the leaf spectral data into one or more linear machine learning models trained to provide a petiole nutrient value output based on the leaf spectral data input; receive an output containing the preliminary petiole nutrient value from the one or more linear machine learning models; refine the preliminary petiole nutrient value by inputting the preliminary petiole nutrient value into one or more nonlinear machine learning models trained to provide a refined petiole nutrient value output based on the preliminary petiole nutrient value input; receive an output containing the refined petiole nutrient value from the one or more nonlinear machine learning models; and output a plant nutrient report indicating the refined petiole nutrient value.

[0162] Item 50: The system of any preceding item, wherein the one or more nonlinear machine learning models are trained to model the interrelationships of nutrient absorption characteristics of different plant nutrients. Item 51: The system of any preceding item, wherein the processor is further configured to receive spectral data of a leaf of the plant taken using a spectrophotometer.

[0163] Item 52: A system as in any of the preceding items, wherein each of the one or more linear machine learning models is a lasso regression model.

[0164] Item 53: A system as in any of the preceding items, wherein each of the one or more nonlinear machine learning models is a multivariate regression model.

[0165] Item 54: A system as in any of the preceding items, wherein each of the one or more nonlinear machine learning models is one of: a support vector regression (SVR) model, a random forest (RF) model, and an extreme gradient boosting (XGB) model.

[0166] Item 55: The system of any preceding item, wherein the outputting comprises sending the refined petiole nutrient value to a mobile device.

[0167] Item 56: The system of any preceding item, wherein the processor is further configured to apply a fertilizer treatment to the plant based at least in part on the refined petiole nutrient value.

Claims

1. A method for determining a refined petiole nutrient value based on leaf spectral data from a plant leaf, the method comprising: (a) storing the leaf spectrum data in a memory; (b) using a processor, converting the leaf spectral data into a preliminary petiole nutrient value by inputting the leaf spectral data into one or more linear machine learning models trained to provide a petiole nutrient value output based on the leaf spectral data input, the one or more linear machine learning models outputting the preliminary petiole nutrient value; (c) using the processor, refining the preliminary petiole nutrient value by inputting the preliminary petiole nutrient value into one or more non-linear machine learning models trained to provide a refined petiole nutrient value output based on the preliminary petiole nutrient value input, the one or more non-linear machine learning models outputting a refined petiole nutrient value; as well as (d) outputting a plant nutrient report indicating the refined petiole nutrient value.

2. The method of claim 1, wherein the one or more nonlinear machine learning models are trained to model the interrelationships of nutrient absorption characteristics of different phytonutrients.

3. The method of claim 1, further comprising, prior to said storing, taking spectral measurements of leaves of the plant using a spectrophotometer to generate leaf spectral data.

4. The method of claim 1, wherein each of the one or more nonlinear machine learning models is a multivariate regression model.

5. The method of claim 1, wherein each of the one or more nonlinear machine learning models is one of: a support vector regression (SVR) model, a random forest (RF) model, and an extreme gradient boosting (XGB) model.

6. The method of claim 1, wherein the plant leaves are potato plant leaves.

7. The method of claim 1, further comprising applying a fertilizer treatment to the plant based at least in part on the refined petiole nutrient value.

8. A non-transitory computer-readable medium comprising instructions executable by a processor, wherein when the instructions are executed, the processor is configured to: storing leaf spectral data from leaves of the plant in a memory; using the processor, converting the leaf spectral data into preliminary petiole nutrient values ​​by inputting the leaf spectral data into one or more linear machine learning models trained to provide petiole nutrient value outputs based on the leaf spectral data inputs; receiving outputs from the one or more linear machine learning models including the preliminary petiole nutrient value; using the processor, refining the preliminary petiole nutrient value by inputting the preliminary petiole nutrient value into one or more non-linear machine learning models trained to provide a refined petiole nutrient value output based on the preliminary petiole nutrient value input; receiving output from the one or more nonlinear machine learning models including the refined petiole nutrient value; and A plant nutrient report is output indicating the refined petiole nutrient value.

9. The non-transitory computer readable medium of claim 8, wherein the one or more non-linear machine learning models are trained to model interrelationships of nutrient absorption characteristics of different phytonutrients.

10. The non-transitory computer readable medium of claim 8, wherein when the instructions are executed, the processor is configured to receive spectral data of a leaf of a plant taken using a spectrophotometer.

11. The non-transitory computer-readable medium of claim 8, wherein each of the one or more linear machine learning models is a lasso regression model.

12. The non-transitory computer-readable medium of claim 8, wherein each of the one or more non-linear machine learning models is a multivariate regression model.

13. The non-transitory computer-readable medium of claim 8, wherein each of the one or more nonlinear machine learning models is one of: a support vector regression (SVR) model, a random forest (RF) model, and an extreme gradient boosting (XGB) model.

14. The non-transitory computer readable medium of claim 8, wherein the instructions, when executed, cause the processor to be configured to apply a fertilizer treatment to a plant based at least in part on the refined petiole nutrient value.

15. A system for determining a refined petiole nutrient value based on leaf spectral data from a plant leaf, the system comprising: a memory configured to store leaf spectral data from leaves of a plant; as well as A processor, the processor being configured to: converting the leaf spectral data into preliminary petiole nutrient values ​​by inputting the leaf spectral data into one or more linear machine learning models trained to provide petiole nutrient value outputs based on the leaf spectral data inputs; receiving an output from the one or more linear machine learning models comprising the preliminary petiole nutrient value; refining the preliminary petiole nutrient value by inputting the preliminary petiole nutrient value into one or more nonlinear machine learning models trained to provide a refined petiole nutrient value output based on the preliminary petiole nutrient value input; receiving an output from the one or more nonlinear machine learning models comprising the refined petiole nutrient value; and A plant nutrient report is output indicating the refined petiole nutrient value.

16. The system of claim 15, wherein the one or more nonlinear machine learning models are trained to model interrelationships of nutrient absorption characteristics of different phytonutrients.

17. The system of claim 15, wherein the processor is further configured to receive spectral data of a leaf of a plant taken using a spectrophotometer.

18. The system of claim 15, wherein each of the one or more nonlinear machine learning models is a multivariate regression model.

19. The system of claim 15, wherein each of the one or more nonlinear machine learning models is one of: a support vector regression (SVR) model, a random forest (RF) model, and an extreme gradient boosting (XGB) model.

20. The system of claim 15, wherein the processor is further configured to apply a fertilizer treatment to the plant based at least in part on the refined petiole nutrient value.