Inversion method for hyperspectral characteristics and crop leaf nutrient element ecostoichiometry of unmanned aerial vehicle

Through drone hyperspectral imaging technology and machine learning algorithms, the inversion relationship of the molar ratio of carbon, nitrogen and phosphorus elements in crops is solved, and the problems of high cost and difficulty in real-time monitoring in the existing technology are solved, and the nutritional status of crops is achieved The precise monitoring and scientific management provide accurate crop growth management solutions.

CN118918985APending Publication Date: 2024-11-08DRYLAND AGRI INST GANSU ACADEMY OF AGRI SCI
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410972941.5
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-19
Publication Date
2024-11-08

AI Technical Summary

Technical Problem

The existing crop nutrient component analysis methods are costly and cannot meet the real-time monitoring needs. Most of them only focus on the impact of a single element, and cannot accurately measure the important role of the molar ratio of carbon, nitrogen and phosphorus in crop growth.

Method used

UAV hyperspectral imaging technology combined with machine learning algorithms, the inversion relationship between the molar ratio of carbon, nitrogen and phosphorus elements and spectral information is constructed by obtaining hyperspectral images of farmland crops, and the prediction is performed using a random forest algorithm, and user-friendly The interface provides scientific management advice.

Benefits of technology

It realizes accurate monitoring and management of crop nutritional status, provides a more scientific and accurate crop growth management plan, has predictive capabilities and decision-making support capabilities, and meets the real-time and accurate detection needs of large-area crops.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118918985A_ABST
    Figure CN118918985A_ABST
Patent Text Reader

Abstract

The invention discloses an unmanned aerial vehicle hyperspectral feature and crop leaf nutrient element ecostoichiometry inversion method, which comprises the following steps: acquiring a hyperspectral image of a crop in a farmland by adopting an airborne hyperspectral imager, and performing radiation correction in the acquisition process to remove the influence of the environment on the spectrum; the wavelengths related to carbon, nitrogen and phosphorus are reserved, other wavelengths are removed, the data volume is reduced, the model detection efficiency and performance are improved, and the molar ratio of samples is calculated; constructing an inversion relation between the molar ratio of carbon, nitrogen and phosphorus elements and the hyperspectral characteristics of the unmanned aerial vehicle through a random forest algorithm; the carbon, nitrogen and phosphorus element molar ratio is predicted, the random forest regression model votes a final prediction result through voting or majority, and the prediction result of each tree is integrated in an averaging or median taking mode; and scientific management. According to the method provided by the invention, the nutrient status of nitrogen and phosphorus elements of crops can be accurately mastered, so that a more scientific and accurate management scheme is provided.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to a crop nutrient element composition analysis technology, and in particular to an inversion method for unmanned aerial vehicle hyperspectral features and crop leaf nutrient element eco-stoichiometry based on machine learning. Background Art

[0002] At present, the molar ratio of carbon, nitrogen and phosphorus is an important factor affecting the growth, metabolism and overall health of plants in the elemental ecology of crop leaves. It plays an important role in the photosynthesis and respiration, nutritional balance, disease resistance, growth and development of crops. However, in the actual measurement process, most of them only focus on the impact of a single element on their crops. Therefore, whether the molar ratio of the elements can be accurately measured plays an important role in the growth process of crops.

[0003] There are some shortcomings in the current crop nutrient element analysis methods, such as X-ray fluorescence spectrometry (XRF) and other equipment costs are high, usually require professional operation, and will cause harm to crops. It is not possible to monitor crops in real time during their growth.

[0004] Based on the important role of the molar ratio of carbon, nitrogen and phosphorus in crop growth, it is urgent to find an advanced monitoring method to meet the needs of real-time and accurate monitoring of large-scale crops.

[0005] Airborne hyperspectral imaging technology is a new detection technology based on spectral technology and remote sensing technology. In the actual process, it collects spectral images of a large area to collect the element content in crops in the farmland. Combined with machine learning algorithms, it builds the relationship between the spectral information in the image and the molar ratio of carbon, nitrogen and phosphorus elements, and realizes rapid, real-time and accurate detection of the molar ratio of carbon, nitrogen and phosphorus elements in crops in the farmland.

[0006] Therefore, by establishing the inversion relationship between the spectral information in the image and the molar ratios of carbon, nitrogen and phosphorus elements, and applying it to the detection during the growth process of farmland crops, it is of great significance for scientifically guiding crop growth. Summary of the invention

[0007] The main purpose of the present invention is to provide an inversion method for the hyperspectral characteristics of unmanned aerial vehicles and the ecological stoichiometry of nutrient elements in crop leaves. By deeply analyzing the key elements and their proportions required for crop growth, accurate grasp of the nutritional status of crops can be achieved, thereby providing a more scientific and precise management plan.

[0008] The technical solution adopted by the present invention is: a method for ecological stoichiometry of UAV hyperspectral characteristics and nutrient elements in crop leaves, comprising:

[0009] S1, using an airborne hyperspectral imager to obtain hyperspectral images of crops in farmland, and performing radiation correction during the acquisition process to remove the impact of the environment on its spectrum;

[0010] S2, retains the wavelengths related to carbon, nitrogen and phosphorus, removes other wavelengths, reduces the amount of data, improves the model detection efficiency and performance, and calculates the molar ratio of the sample;

[0011] S3, constructing the inversion relationship between the molar ratio of carbon, nitrogen and phosphorus elements and the hyperspectral characteristics of drones through the random forest algorithm;

[0012] S4, prediction of carbon, nitrogen and phosphorus molar ratios, the random forest regression model predicts the final results by voting or majority voting, and integrates the prediction results of each tree by averaging or taking the median.

[0013] S5, scientific management.

[0014] Furthermore, in step S2,

[0015] When the wavelength of carbon is 820nm, the reflectivity is usually positively correlated with the total biomass of the plant, which is a key indicator for estimating carbon storage;

[0016] When the wavelength of nitrogen is 680nm, it is related to the chlorophyll content, which indirectly reflects the nitrogen nutritional status of the plant;

[0017] When the wavelength of nitrogen is 760nm, this band is related to the health status and nitrogen content of the plant;

[0018] When the wavelength of nitrogen is 850nm, the biomass and nitrogen content of plants can be estimated by combining it with the reflectance ratio or difference in the visible light region;

[0019] The wavelength of phosphorus is 600nm, which can indirectly reflect its phosphorus content. The molar ratio of carbon, nitrogen and phosphorus is calculated as follows:

[0020] Calculate the molar mass of an element:

[0021] The molar mass of carbon (C) is about 12 g / mol;

[0022] The molar mass of nitrogen (N) is about 14 g / mol;

[0023] The molar mass of phosphorus (P) is about 31 g / mol;

[0024] Determine the mass percentage of each element in the sample: Assume that the mass percentages of carbon, nitrogen, and phosphorus in the soil or plant sample are known, %C, %N, and %P respectively;

[0025] Calculate the molar ratio:

[0026] Convert the mass percentage of each element to mass. Assuming the total mass is 1 gram, the mass of carbon is %C×1 gram, and so on;

[0027] Divide the mass of each element by its molar mass to get the number of moles:

[0028]

[0029] Finally, calculate the C:N:P molar ratio

[0030] C:N:P ratio=moles ofC:moles ofN:moles of P.

[0031] Furthermore, the step S3 comprises:

[0032] Construction of decision trees: Each tree in a random forest is built by randomly selecting samples from the original dataset;

[0033] For each node, a portion of features are randomly selected from all features and the best split point is found. This randomness helps improve the generalization ability of the model. The construction of the tree usually involves the following calculations:

[0034] Information Gain:

[0035]

[0036] Where H(D) is the entropy of the dataset D, a is the feature, and D v It is a subset of feature a with value v; Gini Impurity:

[0037]

[0038] where p i is the relative frequency of the i-th category in the data set D;

[0039] Forest prediction: The prediction of a random forest is based on the prediction results of all its trees. For classification problems, each tree votes for one category, and the final result is the category with the most votes. For regression problems, the average of all tree predictions is usually taken, which can be expressed mathematically as follows:

[0040] Classification:

[0041] RF class (x)=mode{Tree1(x),Tree2(x),…,Tree n (x)}

[0042] return:

[0043]

[0044] Among them, Tree i (x) is the prediction result of the i-th tree for sample x, and n is the total number of trees;

[0045] Error estimation: Random forests use out-of-bag error to estimate the performance of the model; the out-of-bag error is the average of the prediction errors for the training samples not used by each tree; this is equivalent to a built-in cross-validation procedure;

[0046] Hyperspectral data preparation: Samples of different molar ratios in crops and determine the input feature values ​​and target variables;

[0047] Randomly select a data set; build a data set from classified samples, randomly select samples from them, and perform sampling with replacement to form a training subset for random selection;

[0048] Construct decision tree: Based on the feature threshold, recursively split the features of the training subset and construct a number of decision tree models;

[0049] Feature selection: During the segmentation process of each decision tree, only some features are randomly selected from the feature set.

[0050] Furthermore, the step S5 comprises:

[0051] Data collection: Using hyperspectral remote sensing technology, soil and plant tissue analysis and other methods, key data on crop growth are collected, including but not limited to the nutritional status of the soil, the carbon, nitrogen and phosphorus molar ratio in crop leaves, and the content of other trace elements;

[0052] Construction of experience knowledge base: Integrate agricultural prophet experience, historical crop growth data and the latest agricultural research results to build a comprehensive experience knowledge base; this knowledge base will serve as the basis for the system to provide decision support;

[0053] Establishment of decision support system: Based on data analysis results and experience knowledge base, the system can provide scientific decision support for agricultural producers, including but not limited to fertilizer use recommendations, irrigation strategy adjustments, and pest and disease control plans;

[0054] User Interface Management: Provides a user-friendly interface that allows agricultural producers to easily access system recommendations, view crop growth status and environmental data; users can also customize specific monitoring and management plans according to their own needs.

[0055] Advantages of the present invention:

[0056] The present invention is the first to use the molar ratio of element content as a detection standard for crops. It is a method for detecting the molar ratio of crop nutrient elements based on random forests and can make comprehensive scientific management decisions based on the molar ratio of nutrient elements.

[0057] Precision management: Through in-depth analysis of the key elements and their proportions required for crop growth, we can accurately grasp the nutritional status of crops and provide more scientific and precise management solutions;

[0058] Predictive capabilities: Combining historical data and advanced algorithms, the system has the ability to predict crop growth trends and potential problems, providing support for preventive management;

[0059] Decision support: By comprehensively utilizing the experience of the prophet and the results of data analysis, the system can provide comprehensive decision support for agricultural producers, helping them make more informed management decisions;

[0060] Ease of use: The user-friendly interactive interface ensures that all users can easily access and utilize system functions, regardless of their technical background.

[0061] In addition to the above-described purposes, features and advantages, the present invention has other purposes, features and advantages. The present invention will be further described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] The drawings constituting a part of this application are used to provide a further understanding of the present invention. The illustrative embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0063] Figure 1 It is a flow chart of the method for ecological stoichiometry of the hyperspectral characteristics of unmanned aerial vehicles and nutrient elements in crop leaves of the present invention;

[0064] Figure 2 It is a prediction diagram of the molar ratio of carbon, nitrogen and phosphorus elements of the present invention;

[0065] Figure 3 It is a schematic diagram of the operation of the crop nutrient element molar ratio detection method of the present invention. DETAILED DESCRIPTION

[0066] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0067] refer to Figures 1 to 3 , a method for inverting the hyperspectral characteristics of unmanned aerial vehicles and the ecological stoichiometry of nutrient elements in crop leaves, comprising the following steps:

[0068] Step 1: Use an airborne hyperspectral imager to obtain hyperspectral images of crops in the field, and perform radiation correction during the acquisition process to remove the impact of the environment on its spectrum.

[0069] Step 2: Keep the wavelengths related to carbon, nitrogen, and phosphorus, remove other wavelengths, reduce the amount of data, improve the model detection efficiency and performance, and calculate the molar ratio of the sample.

[0070] The wavelength of carbon is:

[0071] 820nm This reflectance is usually positively correlated with the total biomass of plants, which is a key indicator for estimating carbon storage.

[0072] Wavelength of Nitrogen:

[0073] 680nm is closely related to the chlorophyll content. Nitrogen is an important component of chlorophyll molecules, and its content indirectly reflects the nitrogen nutritional status of plants.

[0074] 760nm This band and slope are related to plant health and nitrogen content. A shift in the red edge position to longer wavelengths generally indicates higher nitrogen concentrations.

[0075] 850nm can be used in conjunction with the reflectance ratio or difference in the visible light region to estimate plant biomass and nitrogen content.

[0076] Phosphorus wavelength:

[0077] 600nm Phosphorus is a component of key molecules such as ATP and DNA. This band can indirectly reflect their phosphorus content.

[0078] The molar ratios of carbon, nitrogen and phosphorus are calculated as follows:

[0079] Calculate the molar mass of an element:

[0080] The molar mass of carbon (C) is about 12 g / mol.

[0081] The molar mass of nitrogen (N) is about 14 g / mol.

[0082] The molar mass of phosphorus (P) is about 31 g / mol.

[0083] Determine the mass percentage of each element in the sample: Assume that the mass percentage of carbon, nitrogen and phosphorus in the soil or plant sample is known, which are %C, %N and %P respectively.

[0084] Calculate the molar ratio:

[0085] First, convert the mass percentage of each element to mass (assuming the total mass is 1 gram, then the mass of carbon is %C x 1 gram, and so on).

[0086] Then, divide the mass of each element by its molar mass to get the number of moles:

[0087]

[0088] Finally, calculate the C:N:P molar ratio

[0089] C:N:P ratio=moles ofC:moles ofN:moles of P.

[0090] Step 3: Use the random forest algorithm to construct the inversion relationship between the molar ratio of carbon, nitrogen, and phosphorus elements and the hyperspectral characteristics of the drone.

[0091] Construction of decision trees: Each tree in a random forest is built by randomly sampling samples (with replacement) from the original data set. For each node, a portion of features is randomly selected from all features and the best split point is found. This randomness helps improve the generalization ability of the model. Tree construction usually involves the following calculations:

[0092] Information Gain:

[0093]

[0094] Where H(D) is the entropy of the dataset D, a is the feature, and D v It is a subset of feature a with value v. Gini Impurity:

[0095]

[0096] where p i is the relative frequency of the i-th class in the dataset D.

[0097] Forest prediction: The prediction of a random forest is based on the predictions of all its trees. For classification problems, each tree votes for one category, and the final result is the category with the most votes; for regression problems, the average of all tree predictions is usually taken. The mathematical expression is as follows:

[0098] Classification:

[0099] RF class (x)=mode{Tree1(x),Tree2(x),…,Tree n (x)}

[0100] return:

[0101]

[0102] Among them, Tree i(x) is the prediction result of the i-th tree for sample x, and n is the total number of trees.

[0103] Error estimation: Random Forest uses out-of-bag error (OOB) to estimate the performance of the model. The out-of-bag error is the average of the prediction errors for the training samples not used by each tree (i.e., out-of-bag samples). This is equivalent to a built-in cross-validation process.

[0104] Hyperspectral data preparation: Samples of different molar ratios in crops and determine the input feature values ​​and target variables;

[0105] Randomly select a data set; build a data set from classified samples, randomly select samples from them, and perform sampling with replacement to form a training subset for random selection;

[0106] Construct decision tree: Based on the feature threshold, recursively split the features of the training subset and construct a number of decision tree models;

[0107] Feature selection: During the segmentation process of each decision tree, only some features are randomly selected from the feature set.

[0108] Step 4: Prediction of carbon, nitrogen, and phosphorus molar ratios. The random forest regression model predicts the final results by voting or majority voting, and integrates the prediction results of each tree by averaging or taking the median. The results show that the R 2 The value is 0.6165 and the RMSE is 0.0779, which meets the needs of field detection.

[0109] Step 5: Scientific management. In order to improve crop production efficiency and quality in modern agricultural management, it is particularly important to develop a comprehensive scientific management decision-making system. The system will deeply explore the complex relationship between the molar ratio and content of key elements in the crop growth process, integrating traditional prophetic experience and advanced data analysis algorithms. Through careful observation and precise analysis of the nutritional status of crops, the system can accurately determine the growth status of crops and provide targeted management suggestions in a timely manner.

[0110] Data collection module: Use hyperspectral remote sensing technology, soil and plant tissue analysis and other methods to collect key data in the crop growth process, including but not limited to the nutritional status of the soil, the carbon, nitrogen and phosphorus molar ratio in crop leaves, and the content of other trace elements.

[0111] Experience knowledge base: Integrate agricultural prophet experience, historical crop growth data and the latest agricultural research results to build a comprehensive experience knowledge base. This knowledge base will serve as the basis for the system to provide decision support.

[0112] Decision support system: Based on data analysis results and experience knowledge base, the system can provide scientific decision support for agricultural producers, including but not limited to fertilizer use recommendations, irrigation strategy adjustments, pest and disease prevention and control plans, etc.

[0113] User Interface: Provides a user-friendly interface that allows agricultural producers to easily access system recommendations, view crop growth status and environmental data. Users can also customize specific monitoring and management solutions according to their needs.

[0114] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for inverting the hyperspectral characteristics of unmanned aerial vehicles and the ecological stoichiometry of nutrient elements in crop leaves, characterized in that: include: S1, using an airborne hyperspectral imager to obtain hyperspectral images of crops in farmland, and performing radiation correction during the acquisition process to remove the impact of the environment on its spectrum; S2, retains the wavelengths related to carbon, nitrogen and phosphorus, removes other wavelengths, reduces the amount of data, improves the model detection efficiency and performance, and calculates the molar ratio of the sample; S3, constructing the inversion relationship between the molar ratio of carbon, nitrogen and phosphorus elements and the hyperspectral characteristics of drones through the random forest algorithm; S4, prediction of carbon, nitrogen and phosphorus molar ratios, the random forest regression model predicts the final results by voting or majority voting, and integrates the prediction results of each tree by averaging or taking the median. S5, scientific management.

2. The inversion method of the UAV hyperspectral characteristics and the ecological stoichiometry of the nutrient elements in crop leaves according to claim 1 is characterized in that: In the step S2, When the wavelength of carbon is 820nm, the reflectivity is usually positively correlated with the total biomass of the plant, which is a key indicator for estimating carbon storage; When the wavelength of nitrogen is 680nm, it is related to the chlorophyll content, which indirectly reflects the nitrogen nutritional status of the plant; When the wavelength of nitrogen is 760nm, this band is related to the health status and nitrogen content of the plant; When the wavelength of nitrogen is 850nm, the biomass and nitrogen content of plants can be estimated by combining it with the reflectance ratio or difference in the visible light region; The wavelength of phosphorus is 600nm, which can indirectly reflect its phosphorus content. The molar ratio of carbon, nitrogen and phosphorus is calculated as follows: Calculate the molar mass of an element: The molar mass of carbon (C) is about 12 g / mol; The molar mass of nitrogen (N) is about 14 g / mol; The molar mass of phosphorus (P) is about 31 g / mol; Determine the mass percentage of each element in the sample: Assume that the mass percentages of carbon, nitrogen, and phosphorus in the soil or plant sample are known, %C, %N, and %P respectively; Calculate the molar ratio: Convert the mass percentage of each element to mass. Assuming the total mass is 1 gram, the mass of carbon is %C×1 gram, and so on; Divide the mass of each element by its molar mass to get the number of moles: Finally, calculate the C:N:P molar ratio C:N:P ratio=moles ofC:moles ofN:moles of P.

3. The inversion method of the UAV hyperspectral characteristics and the ecological stoichiometry of the nutrient elements in crop leaves according to claim 1 is characterized in that: The step S3 comprises: Construction of decision trees: Each tree in a random forest is built by randomly selecting samples from the original dataset; For each node, a portion of features are randomly selected from all features and the best split point is found. This randomness helps improve the generalization ability of the model. The construction of the tree usually involves the following calculations: Information Gain: Where H(D) is the entropy of the dataset D, a is the feature, and D v It is a subset of feature a with value v; Gini Impurity: where p i is the relative frequency of the i-th category in the data set D; Forest prediction: The prediction of a random forest is based on the prediction results of all its trees. For classification problems, each tree votes for one category, and the final result is the category with the most votes. For regression problems, the average of all tree predictions is usually taken, which can be expressed mathematically as follows: Classification: RF class (x)=mode{Tree1(x),Tree2(x),…,Tree n (x)} return: Among them, Tree i (x) is the prediction result of the i-th tree for sample x, and n is the total number of trees; Error estimation: Random forests use the out-of-bag error to estimate the performance of the model; the out-of-bag error is the average of the prediction errors for the training samples not used by each tree; this is equivalent to a built-in cross-validation procedure; Hyperspectral data preparation: Samples of different molar ratios in crops and determine the input feature values ​​and target variables; Randomly select a data set; build a data set from classified samples, randomly select samples from them, and perform sampling with replacement to form a training subset for random selection; Construct decision tree: Based on the feature threshold, recursively split the features of the training subset and construct a number of decision tree models; Feature selection: During the segmentation process of each decision tree, only some features are randomly selected from the feature set.

4. The inversion method of the UAV hyperspectral characteristics and the ecological stoichiometry of the nutrient elements in crop leaves according to claim 1 is characterized in that: The step S5 comprises: Data collection: Using hyperspectral remote sensing technology, soil and plant tissue analysis and other methods, key data on crop growth are collected, including but not limited to the nutritional status of the soil, the carbon, nitrogen and phosphorus molar ratio in crop leaves, and the content of other trace elements; Construction of experience knowledge base: Integrate agricultural prophet experience, historical crop growth data and the latest agricultural research results to build a comprehensive experience knowledge base; this knowledge base will serve as the basis for the system to provide decision support; Establishment of decision support system: Based on data analysis results and experience knowledge base, the system can provide scientific decision support for agricultural producers, including but not limited to fertilizer use recommendations, irrigation strategy adjustments, and pest and disease control plans; User Interface Management: Provides a user-friendly interface that allows agricultural producers to easily access system recommendations, view crop growth status and environmental data; users can also customize specific monitoring and management plans according to their own needs.