Corn canopy nitrogen content monitoring method and system, electronic equipment and storage medium

Abstract

The invention relates to the technical field of corn canopy nitrogen content monitoring, and discloses a corn canopy nitrogen content monitoring method and system, electronic equipment and a storage medium, and the method comprises the steps: obtaining multi-source remote sensing data of a target corn field area, the multi-source remote sensing data comprises a multi-spectral image and a hyperspectral reflectivity, the method comprises the following steps: constructing a multispectral vegetation index and a texture index based on a multispectral image, constructing a hyperspectral vegetation index, constructing a mixed feature data set by using the multispectral vegetation index, the hyperspectral vegetation index and the texture index, and carrying out feature screening on the mixed feature data set by adopting an LASSO regression algorithm to obtain an optimal feature combination; and training a random forest model through a training sample set formed by the optimal feature combination and a corresponding corn leaf nitrogen concentration measured value to form a corn canopy nitrogen content monitoring model, and obtaining a corn canopy leaf nitrogen content predicted value of the target corn field area through the corn canopy nitrogen content monitoring model. And the nitrogen condition of the corn plant in the target area can be effectively estimated.

Description

Technical Field

[0001] This invention relates to the field of corn canopy nitrogen content monitoring technology, and in particular to a method, system, electronic device and storage medium for monitoring corn canopy nitrogen content. Background Technology

[0002] Corn production accounts for more than one-third of my country's total grain output. Leaf nitrogen content affects corn canopy photosynthesis and productivity; nitrogen fertilizer application significantly increases corn yield. Nitrogen deficiency causes corn leaves to wither and turn yellow, while excessive nitrogen application increases planting costs and leads to environmental pollution. Therefore, real-time, rapid, and accurate monitoring of corn leaf nitrogen content can guide field decisions on precise fertilization, promoting corn growth, reducing planting costs, and protecting the environment.

[0003] Currently, leaf color diagnosis, laboratory chemical analysis, and satellite remote sensing monitoring are commonly used for plant nitrogen diagnosis, but all of them have some drawbacks: (1) Leaf color diagnosis: The most famous is the Japanese rice leaf color card. The color of rice leaves is compared with the standard color card and divided into different levels to determine the nitrogen abundance or deficiency and guide topdressing. Obviously, this is a semi-quantitative morphological diagnosis method that relies on human visual judgment and has a large error. (2) Laboratory chemical analysis requires the collection of plant tissues, which can damage crops and cannot continuously track and monitor the same plant. The cycle is long and cannot quickly guide the immediate management in the field. Moreover, the analysis results only represent the condition at the moment of sampling and the sampling point. They are greatly affected by individual plant differences and field variations and cannot fully reflect the nitrogen status of the entire field. (3) Satellite remote sensing monitoring: Many free satellite data (such as Sentinel-2, 10 meters / pixel) are still too "coarse" for the fine management required by precision agriculture. A single pixel may contain multiple information such as crop, soil, and shadow, resulting in nitrogen inversion results being an average of a mixture, which cannot be accurate to the variation area within a single row of crops or field. Although commercial high-resolution satellites can achieve sub-meter level accuracy, the cost is extremely high. Even satellites like Sentinel-2 have a revisit cycle of 5 days under ideal conditions, and this is only the "overhead" time. The most critical limitation is cloud cover. During the critical growth period of crops, if there are continuous rainy and cloudy days, it may be impossible to obtain effective clear sky data for several weeks, thus missing the best window for fertilization. "No data when needed" is the norm. (4) UAV remote sensing monitoring: Although more flexible than satellite remote sensing, its imaging quality is still affected by weather conditions (such as thick clouds and crosswinds) and the angle of illumination. At the same time, the vibration of the flight platform, changes in flight altitude and speed may introduce image noise, resulting in unstable spectral data and affecting the accuracy of diagnostic results.

[0004] Therefore, there is an urgent need to develop convenient new methods and means to meet the needs of accurate dynamic monitoring of crop nitrogen nutrition in a certain area. Summary of the Invention

[0005] To address the aforementioned technical problems, this invention provides a method, system, electronic device, and storage medium for monitoring nitrogen content in maize canopy, which can effectively estimate the nitrogen status of maize plants in a target area.

[0006] The first aspect of this invention provides a method for monitoring nitrogen content in the maize canopy, comprising the following steps: Multi-source remote sensing data of sample cornfields were acquired. The multi-source remote sensing data included multispectral images and hyperspectral reflectance. Multispectral vegetation index and texture index were constructed based on multispectral images, and hyperspectral vegetation index was constructed based on hyperspectral reflectance. A hybrid feature dataset was constructed by combining multispectral vegetation index, hyperspectral vegetation index, and texture index. The LASSO regression algorithm was used to select the optimal feature combination from the mixed feature dataset. A random forest model is trained using a training sample set consisting of the best feature combination and the corresponding measured values ​​of nitrogen concentration in maize leaves to form a maize canopy nitrogen content monitoring model. By inputting multi-source remote sensing data of the target maize field acquired during the collection period into the maize canopy nitrogen content monitoring model, the predicted value of maize canopy leaf nitrogen content in the target maize field is obtained.

[0007] Optionally, constructing multispectral vegetation indices based on multispectral image data includes: Multispectral images were combined to create an orthophoto of the experimental field, including RGB and single-band orthophotos. Single-band orthophotos are fused to create multispectral images. Regions of interest are defined, and the mean reflectance values ​​of five central bands (blue, green, red, red edge, and near-infrared) within each region of interest are extracted to construct a multispectral vegetation index.

[0008] Optionally, the texture index can be constructed by extracting and optimizing the texture index of the region of interest in the field area to be tested using a gray-level co-occurrence matrix.

[0009] Optionally, optimization includes the following steps: The normalized texture index, the difference texture index, and the ratio texture index are obtained through the texture index. Pearson correlation analysis was performed on the normalized texture index, the difference texture index, and the ratio texture index, respectively. The top 15 normalized texture indices, difference texture indices, and ratio texture indices with the highest correlation were selected as optimized texture indices and used to construct a hybrid feature dataset together with multispectral vegetation indices and hyperspectral vegetation indices.

[0010] Optional, normalized texture index We obtain it from the following formula: , In the formula, T1 and T2 both represent texture indices based on gray-level co-occurrence matrices from multispectral images; The Differential Texture Index (DTI) is obtained using the following formula: , The ratio texture index RTI is obtained by the following formula: .

[0011] Optionally, constructing a hyperspectral vegetation index based on hyperspectral reflectance includes: deriving the hyperspectral reflectance and then processing it with the second derivative to construct the hyperspectral vegetation index.

[0012] Optionally, the LASSO regression algorithm includes the following steps: Standardize the mixed feature dataset; Using five-fold cross-validation, the sample data is randomly divided into 5 parts, and 4 parts are used as the training set and 1 part as the validation set in turn. The prediction error is calculated. For each λ value, the average prediction error is calculated. The λ value that minimizes the cross-validation error is selected. For a series of λ values, the LASSO problem is solved using the coordinate descent method to obtain the coefficient estimate for each λ, and the coefficients are updated using the soft threshold operator. The LASSO regression algorithm is retrained using the selected λ value to obtain the optimal feature combination.

[0013] A second aspect of the present invention provides a maize canopy nitrogen content monitoring system, comprising: The data acquisition module is used to acquire the predicted values ​​of nitrogen content in maize canopy leaves obtained by the above methods; The decision-making module is used to generate fertilization recommendations for target cornfields based on predicted nitrogen content values.

[0014] A third aspect of the present invention provides an electronic device, including a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor executes the program to implement the method described above.

[0015] A fourth aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the above-described method.

[0016] The technical solution provided by the embodiments of the present invention has the following advantages compared with the prior art: This invention provides a method, system, electronic device, and storage medium for monitoring nitrogen content in maize canopy. First, it acquires the multispectral and hyperspectral reflectance of the maize canopy from sample points. Then, it constructs a multispectral vegetation index, a hyperspectral vegetation index, and a texture index. The multispectral vegetation index provides broad-band comprehensive information, the hyperspectral vegetation index provides narrow-band fine absorption characteristics, and the texture index introduces canopy heterogeneity and addresses light saturation. These three indices are combined to construct a hybrid dataset, enabling information complementarity and enhancement, providing more comprehensive and multi-dimensional plant phenotypic information. This provides data support for constructing a more robust and accurate maize canopy nitrogen content monitoring model. To address potential multicollinearity and overfitting in the hybrid dataset, the optimal feature combination is selected using the LASSO regression algorithm. Finally, a random forest algorithm is applied to construct the maize canopy nitrogen content monitoring model, achieving optimal estimation accuracy. Data shows that using multi-source data fusion significantly improves the estimation accuracy of maize nitrogen content and can effectively estimate the nitrogen status of maize plants in a region. Attached Figure Description

[0017] Figure 1 A flowchart illustrating a method for monitoring nitrogen content in the maize canopy, provided as an embodiment of the present invention; Figure 2 This is a schematic diagram of a maize canopy nitrogen content monitoring system provided in an embodiment of the present invention; Figure 3 shows the correlation analysis between the multispectral vegetation index at each growth stage and the nitrogen concentration of maize leaves at each layer, provided in the embodiments of the present invention. Figure 3a Pearson correlation analysis was conducted between VIs during the large trumpet stage and LNCs of different canopy layers in maize. Figure 3b Pearson correlation analysis was conducted to determine the relationship between VIs during the silking stage and LNCs in different canopy layers of maize. Figure 3c Pearson correlation analysis of VIs at the milk stage and LNCs of different canopies in maize; Figure 4 shows the correlation analysis between hyperspectral vegetation indices at each growth stage and nitrogen concentrations in maize leaves at each layer, provided in the embodiments of the present invention. Figure 4a Pearson correlation analysis was performed on HPs during the large trumpet stage and LNCs of various maize canopies. Figure 4b Pearson correlation analysis was performed on HPs during the silking stage and LNCs of different canopy layers in maize. Figure 4c Pearson correlation analysis of HPs and LNCs of maize at the large trumpet stage, silking stage, and milk stage; Figure 5 shows the correlation analysis between the texture index at each growth stage and the nitrogen concentration of maize leaves at each layer before texture index optimization, as provided in the embodiment of the present invention. Figure 5a Pearson correlation analysis of TIs during the large trumpet stage and LNCs of various maize canopies. Figure 5bPearson correlation analysis was performed between TIs during silking stage and LNCs of different canopy layers in maize. Figure 5c Pearson correlation analysis of TIs at milk stage and LNCs of different canopies in maize; Figure 6 shows the correlation analysis between the texture index at each milk-ripe stage and the nitrogen concentration of maize leaves at each layer after texture index optimization according to the embodiments of the present invention. Figure 6a Pearson correlation analysis was performed on the texture index (TIs) optimized at the milk stage and the LNC (lower canopy center) of the maize upper canopy. Figure 6b Pearson correlation analysis was performed on the texture index (TIs) optimized at the milk stage and the LNC (lower canopy center) of maize. Figure 6c Pearson correlation analysis between the texture index (TIs) optimized for milk-ripe maize and the lower canopy LNC; Figure 7 The following are fitting diagrams for predicting nitrogen concentration in leaves of maize at different canopy levels using a random forest algorithm with a single vegetation index, provided in this embodiment of the invention. Figure a shows the nitrogen concentration monitoring model for the upper leaves of maize plants at stage V12; Figure b shows the nitrogen concentration monitoring model for the upper leaves of maize plants at stage R1; Figure c shows the nitrogen concentration monitoring model for the upper leaves of maize plants at stage R3; Figure d shows the nitrogen concentration monitoring model for the middle leaves of maize plants at stage V12; Figure e shows the nitrogen concentration monitoring model for the middle leaves of maize plants at stage R1; Figure f shows the nitrogen concentration monitoring model for the middle leaves of maize plants at stage R3; Figure g shows the nitrogen concentration monitoring model for the lower leaves of maize plants at stage V12; Figure h shows the nitrogen concentration monitoring model for the lower leaves of maize plants at stage R1; and Figure i shows the nitrogen concentration monitoring model for the lower leaves of maize plants at stage R3. Figure 8 This invention provides a model for monitoring nitrogen content in maize leaves at various growth stages using the RF algorithm based on VIs+HPs+TIs data sources, and its validation. Figure a shows the nitrogen concentration monitoring model for the upper leaves of maize plants at stage V12; Figure b shows the nitrogen concentration monitoring model for the upper leaves of maize plants at stage R1; Figure c shows the nitrogen concentration monitoring model for the upper leaves of maize plants at stage R3; Figure d shows the nitrogen concentration monitoring model for the middle leaves of maize plants at stages V12, R1, and R3; Figure e shows the nitrogen concentration monitoring model for the middle leaves of maize plants at stage R1; Figure f shows the nitrogen concentration monitoring model for the middle leaves of maize plants at stage R3; Figure g shows the nitrogen concentration monitoring model for the lower leaves of maize plants at stage V12; Figure h shows the nitrogen concentration monitoring model for the lower leaves of maize plants at stage R1; and Figure i shows the nitrogen concentration monitoring model for the lower leaves of maize plants at stage R3. Detailed Implementation

[0018] The following detailed description of a specific embodiment of the present invention is provided in conjunction with the accompanying drawings. However, it should be understood that the scope of protection of the present invention is not limited to the specific embodiment.

[0019] In the description of this invention, it should be understood that the terms "center," "longitudinal," "lateral," "length," "width," "thickness," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," "outer," "axial," "radial," and "circumferential" indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing the technical solution of this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0020] The present invention will be described below through several specific embodiments. To keep the following description of the embodiments clear and concise, detailed descriptions of known functions and components may be omitted. When any component of an embodiment of the present invention appears in more than one drawing, the component may be represented by the same reference numerals in each drawing.

[0021] like Figure 1 As shown in the first embodiment of the present invention, a method for monitoring nitrogen content in the maize canopy is provided, comprising the following steps: Multi-source remote sensing data of cornfields were acquired, including multispectral images and hyperspectral reflectance. Multispectral vegetation index and texture index were constructed based on multispectral images, and hyperspectral vegetation index was constructed based on hyperspectral reflectance. A hybrid feature dataset was constructed by combining multispectral vegetation index, hyperspectral vegetation index, and texture index. The LASSO regression algorithm was used to select the optimal feature combination from the mixed feature dataset. A random forest model is trained using a training sample set consisting of the best feature combination and the corresponding measured values ​​of nitrogen concentration in maize leaves to form a maize canopy nitrogen content monitoring model. By inputting multi-source remote sensing data of the target maize field acquired during the collection period into the maize canopy nitrogen content monitoring model, the predicted value of maize canopy leaf nitrogen content in the target maize field is obtained.

[0022] This invention provides a method for monitoring nitrogen content in maize canopy. First, multispectral and hyperspectral reflectance of the maize canopy are obtained from samples in the maize field. Multispectral vegetation indices, hyperspectral vegetation indices, and texture indices are then constructed. The multispectral vegetation index provides comprehensive information across a wide band, the hyperspectral vegetation index provides fine absorption characteristics in a narrow band, and the texture index introduces canopy heterogeneity and addresses light saturation. These three indices are combined to construct a hybrid dataset, enabling information complementarity and enhancement, providing more comprehensive and multidimensional plant phenotypic information. This provides data support for building a more robust and accurate maize canopy nitrogen content monitoring model. To address potential multicollinearity and overfitting in the hybrid dataset, the optimal feature combination is selected using the LASSO regression algorithm. Finally, a random forest algorithm is applied to construct the maize canopy nitrogen content monitoring model, achieving optimal estimation accuracy. Data shows that using multi-source data fusion significantly improves the estimation accuracy of maize nitrogen content and can effectively estimate the nitrogen status of maize plants in a region.

[0023] Optionally, constructing multispectral vegetation indices based on multispectral image data includes: Multispectral images were combined to create an orthophoto of the experimental field, including RGB and single-band orthophotos. Single-band orthophotos are fused to create multispectral images. Regions of interest are defined, and the mean reflectance values ​​of five central bands (blue, green, red, red edge, and near-infrared) within each region of interest are extracted to construct a multispectral vegetation index.

[0024] Optionally, the texture index can be constructed by extracting and optimizing the texture index of the region of interest in the field area to be tested using a gray-level co-occurrence matrix.

[0025] Optionally, optimization includes the following steps: The normalized texture index, the difference texture index, and the ratio texture index are obtained through the texture index. Pearson correlation analysis was performed on the normalized texture index, the difference texture index, and the ratio texture index, respectively. The top 15 normalized texture indices, difference texture indices, and ratio texture indices with the highest correlation were selected as optimized texture indices and used to construct a hybrid feature dataset together with multispectral vegetation indices and hyperspectral vegetation indices.

[0026] Optional, normalized texture index We obtain it from the following formula: , In the formula, T1 and T2 both represent texture indices based on gray-level co-occurrence matrices from multispectral images; The Differential Texture Index (DTI) is obtained using the following formula: , The ratio texture index RTI is obtained by the following formula: .

[0027] Optionally, constructing a hyperspectral vegetation index based on hyperspectral reflectance includes: deriving the hyperspectral reflectance and then processing it with the second derivative to construct the hyperspectral vegetation index and improve its correlation.

[0028] (1) The experiment of this invention was carried out in 2024 at the International Agricultural High-tech Industrial Park of the Chinese Academy of Agricultural Sciences in Guangyang District, Langfang City, Hebei Province (116°36'58"E, 39°36'32"N). The average annual temperature is 13.5℃, the average annual sunshine duration is 2445.0h, and the rainfall is 640.3mm. It belongs to the warm temperate continental monsoon climate. There are four distinct seasons, with more rain in summer, mostly concentrated in June to August. The soil type of the experimental field is sandy loam. The basic chemical properties are as follows: soil organic matter content is 3.73g / kg, available phosphorus content is 2.17mg / kg, available potassium content is 85.1mg / kg, total nitrogen content is 0.376g / kg, ammonium nitrogen is 8.075mg / kg, and nitrate nitrogen is 26.55mg / kg.

[0029] (2) A field experiment was conducted to investigate the interaction of three maize varieties, three planting densities, and five nitrogen fertilizer application rates. The three varieties were Zhengdan 958, Xianyu 335, and Woyu 3, and the three planting densities were 60,000 plants per hectare. -1 75,000 ha -1 90,000 ha -1 The five nitrogen fertilizer application treatments were all 0 kg Nha. -1 100kgNha -1 200kgNha -1 300kgNha -1 400kgNha -1 The residential area is 79.2 square meters. 2 Planting pattern: (6.6m × 12m), with equal row spacing (60cm-60cm). Nitrogen fertilizer was applied at 1 / 3 and 2 / 3 of the total nitrogen fertilizer at the sowing and jointing stages, respectively. The application rate of phosphorus fertilizer (P2O5) and potassium fertilizer (K2O) was 100 kg / ha each. -1 All fertilizer should be applied as base fertilizer during the sowing period. Other cultivation and tillage practices should refer to local high-yield and high-efficiency maize cultivation techniques, and ensure the absence of biological stresses such as diseases, pests, and weeds.

[0030] (3) During key growth stages of maize, such as the large trumpet stage, silking stage, and milk stage, three representative maize plants with similar growth were selected from each experimental plot. The plants were divided into three layers (based on plant height) and only leaves were taken for nitrogen concentration determination using a Kjeldahl nitrogen analyzer. 70% of the plants were used for the model set, and 30% were used for model testing. At the same time as plant sampling, multispectral images were acquired using a drone (DJI Phantom 4RTK Multispectral Edition), and hyperspectral reflectance data (PSR-1100F ground object spectrometer) were collected simultaneously on the ground. The drone images were then synthesized into orthophotos of the experimental field using Pix4Dmapper software (drone photogrammetry and remote sensing data processing software), including RGB and single-band orthophotos. The single-band images were then imported into ENVI software (remote sensing image processing platform) for band fusion to create multispectral images. The Region of Interest (ROI) was divided, and the reflectance of five bands was extracted: the average reflectance of the five central bands (blue (B, 450nm), green (G, 560nm), red (R, 650nm), red edge (RE, 730nm), and near-infrared (NIR, 840nm). Multispectral vegetation indices were constructed according to the formulas in Table 1. Hyperspectral reflectance was exported from the built-in memory and processed using its second derivative to improve correlation; this was also used for hyperspectral vegetation index construction (see Table 2). Texture indices were extracted by importing the ROI images into MATLAB R2024a software and calculating them using the gray-level co-occurrence matrix algorithm (MATLAB is a commercial mathematical software developed by MathWorks, widely used in data analysis, wireless communication, deep learning, image processing and computer vision, signal processing, quantitative finance and risk management, robotics and control systems). Texture indices were constructed and optimized through Pearson correlation analysis (see Table 3). Since the nitrogen content in maize leaves is a complex physiological and biochemical parameter, a single type of data can only reflect one aspect (spectral response, spatial structure), resulting in limitations and redundancy in the information.

[0031] Table 1. Selection and Calculation Formula of Multispectral Vegetation Indices Note: B, G, R, RE, and NIR represent the reflectance of the five bands: blue, green, red, red edge, and near-infrared, respectively.

[0032] Table 2 Selection and Calculation Formula of Hyperspectral Vegetation Indices Note: R x Given the hyperspectral reflectance at point X, input the corresponding vegetation index using the vegetation index formula on the right, based on the reflectance at the corresponding band position.

[0033] Table 3. Selection and Calculation Formula of Texture Index Note: In the formula, represents the pixel brightness value in the i-th row and j-th column, and N represents the window size for texture analysis. P i,j These are the elements of the normalized gray-level co-occurrence matrix, representing the distance d and direction.

[0034] The probability that gray levels i and j co-occur at θ.

[0035] Therefore, in this embodiment of the invention, multispectral vegetation indices provide comprehensive information across a wide band, hyperspectral vegetation indices provide fine absorption features across a narrow band, and optimized texture indices introduce canopy heterogeneity and address light saturation by fusing them to form a hybrid dataset. The combination of these three indices achieves information complementarity and enhancement, providing more comprehensive and multidimensional plant phenotypic information, thereby constructing a more robust and accurate prediction model. Multispectral vegetation indices are common but have limited information, hyperspectral indices provide more detailed spectral information, and texture indices capture spatial structural changes. Combining these three indices allows them to complement each other; for example, hyperspectral indices compensate for the limitations of multispectral bands, and texture indices reflect spatial heterogeneity. The ultimate goal is to improve the final modeling accuracy and reduce prediction errors. In texture index optimization, normalized difference texture index (NDTI), difference texture index (DTI), and ratio texture index (RTI) are constructed to replace ordinary texture indices (see Table 3). These improve the correlation with canopy leaf nitrogen content, as shown in Figure 5. The calculation formulas for different combinations of texture indices are as follows:

[0036] , , , In this table, T1 and T2 both represent the texture indices based on the gray-level co-occurrence matrix (GLCM) from multispectral images, as shown in Table 3. Each multispectral band corresponds to 8 texture features, for a total of 40 texture features. These 40 texture features are then used for combination optimization.

[0037] The Pearson correlation calculation can be divided into four steps: 0) Data Preparation In R, data is typically organized in the form of a data.frame, with each row representing a sample. Column 1: Nitrogen content in canopy leaves (denoted as N). Multiple columns: Vegetation index (VI1, VI2, ..., VI) m ), The formal representation is as follows: {(VI1i VI 2i ,…,VI mi , Ni ), i=1,2,…,n}, Where n is the number of samples.

[0038] 2) Prerequisite check Remove missing values ​​to ensure that vegetation indices and canopy leaf nitrogen content come from the same temporal and spatial scales; if there are obvious outliers, they can be removed before analysis.

[0039] 3) Pearson correlation calculation The Pearson correlation coefficient between any vegetation index VI and canopy leaf nitrogen content (N) is defined as follows: , in, The sum of the mean and mean of the vegetation indices. This represents the average nitrogen content in the canopy leaves. ∈[−1,1].

[0040] 1) Significance test and result screening In hypothesis testing: Null hypothesis H0: r = 0 (no correlation), Alternative hypothesis H1: r ≠ 0, Pearson correlation coefficient test statistic. for: , This statistic follows a t-distribution with n−2 degrees of freedom. The p-value is calculated based on the t-value and used to determine whether the correlation is significant.

[0041] To address the potential multicollinearity of features in mixed datasets and avoid overfitting, a LASSO regression algorithm is introduced for feature selection. The parameter λ, controlling the regularization strength, is determined using five-fold cross-validation. In the five-fold cross-validation curve, a smaller Y-axis value indicates a better fit. λ is selected based on the minimum bias. At this value, the model achieves the best fit. A penalty term is established based on the regularization strength parameter λ, and the optimal feature combination is selected when the regression coefficients converge to 0. The regularization strength parameter λ controls the weight of the penalty term in the loss function.

[0042] Specifically, the LASSO regression algorithm includes the following steps: Standardize the mixed feature dataset; Five-fold cross-validation was used to randomly divide the sample data into five parts, and four parts were used as the training set and one part as the validation set in turn to calculate the prediction error. For each λ value, the average prediction error was calculated. The λ value that minimizes the cross-validation error was selected. The λ value that minimizes the error, i.e. the λ value that minimizes the difference between the predicted nitrogen content and the measured value, was selected. The ultimate goal is to reduce the error and make the predicted value close to the measured value of the nitrogen content in the canopy leaves. For a series of λ values, the LASSO problem is solved using the coordinate descent method to obtain the coefficient estimate for each λ, and the coefficients are updated using the soft threshold operator. The LASSO regression algorithm is retrained using the selected λ value to obtain the optimal feature combination.

[0043] The objective function expression for the Lasso regression model is: , Where N is the number of samples and p is the number of features. Let i be the response value of the i-th observation. For the j-th feature value of the i-th observation, For the intercept term, For the regression coefficient vector, Let be the regression coefficient of the j-th feature, and let λ be the regularization parameter ≥ 0.

[0044] (4) The steps for solving LASSO regression are as follows: 1) Data preprocessing Standardize continuous features (make the mean 0 and the variance 1) so that the regularization term can penalize the coefficients equally.

[0045] For categorical features, one-hot encoding or other encoding methods are required.

[0046] 2) Choose the regularization parameter λ λ is typically selected through cross-validation. In this embodiment of the invention, five-fold cross-validation is used. The data is randomly divided into five parts, and four parts are used alternately as the training set and one part as the validation set to calculate the prediction error. For each λ value, the average prediction error is calculated. The λ value that minimizes the cross-validation error is selected (choosing the value with the smallest error), or the λ value corresponding to the simplest model (i.e., the one with the fewest coefficients) whose cross-validation error is within one standard deviation is selected (referred to as the 1SE criterion).

[0047] 3) Solving the coefficient path For a series of λ values ​​(typically values ​​evenly spaced on a logarithmic scale), the LASSO problem is solved using coordinate descent to obtain coefficient paths (i.e., coefficient estimates for each λ). Coordinate descent is an iterative algorithm that updates one coefficient at a time while keeping the others fixed. Since the penalty term in LASSO is non-differentiable, a soft thresholding operator is used to update the coefficients.

[0048] 4) Model Evaluation The model is retrained using the selected λ value to obtain the final coefficients. The performance of the LASSO model is then evaluated on the test set.

[0049] (5) The best feature combination obtained is fed into the random forest algorithm for modeling. For model accuracy evaluation, the coefficient of determination (R²) is selected (both the modeling set and the test set will calculate and output the coefficient of determination and the root mean square (RMS)). 2 The following statement explains the error—used as an evaluation metric for training and validating models. The root mean square error (RMSE) is a fundamental metric for evaluating the performance of training and validating models. 2 A higher value indicates a lower RMSE value, which means the model performs better.

[0050] The main steps are as follows: I. Data Preparation: First, a dataset needs to be prepared for training and validating the model. The dataset should contain features and corresponding target variables, i.e., the measured canopy nitrogen content of each plot. Features are attributes or characteristics (vegetation indices) used to predict the target variable, while the target variable is the value to be predicted by regression. Typically, the dataset needs to be divided into a training set and a validation set (test set), where the training set is used to train the model, and the validation set is used to evaluate the performance of the nitrogen inversion model.

[0051] Building a Random Forest: In Python's Scikit-learn library, the RandomForestRegressor class can be used to build a random forest regression model. This embodiment of the invention uses the TreeBagger function in MATLAB to create a random forest model. Several parameters can be set to control the behavior of the random forest, such as the number of decision trees, the feature selection method, and the growth method of the decision trees. Parameters (sample size) can be adjusted according to the actual problem and requirements.

[0052] Training the model: The random forest regression model is trained using the training set. The model will construct multiple decision trees based on the samples in the training set and the measured canopy nitrogen content, and perform feature selection and splitting on each tree.

[0053] Prediction Results: The trained random forest regression model is used to predict the samples in the validation set. The model will average or weight the prediction results of each decision tree to obtain the final regression prediction result—the predicted value of nitrogen content in maize canopy leaves.

[0054] Model evaluation: The model's performance is evaluated by comparing it with the true target variable (measured canopy leaf nitrogen content). Various regression performance metrics, such as root mean squared error (RMSE), mean absolute error (MAE), and coefficient of determination (R-squared), can be used to assess the model's accuracy and generalization ability.

[0055] Model tuning: Based on the model evaluation results, the random forest regression model can be tuned. You can try adjusting the parameters of the random forest, such as increasing or decreasing the number of decision trees, adjusting the feature selection method, and adjusting the growth method of the decision trees, thereby improving the model's performance.

[0056] Model Application: After model evaluation and tuning, the trained random forest regression model can be used for actual predictions. New input samples can be fed into the model to obtain the corresponding regression prediction results.

[0057] The calculation formula is as follows: In the formula, As the coefficient of determination, The root mean square error, These are measured values. These are predicted values. The average of the measured values. For the sample size, =1,2,3..., ; The sample size is [number]. The complete workflow, including variable selection, modeling, cross-validation, and performance evaluation, is performed in MATLAB R2024a.

[0058] (6) This method can quantitatively monitor the nitrogen nutrition status of maize leaves in the field area, and is applicable to crops such as maize, rice, and wheat.

[0059] like Figure 2 As shown in the second embodiment of the present invention, a maize canopy nitrogen content monitoring system is provided, comprising: The data acquisition module is used to acquire the predicted values ​​of nitrogen content in maize canopy leaves obtained by the above methods; The decision-making module is used to generate fertilization recommendations for target cornfields based on predicted nitrogen content values.

[0060] The third embodiment of the present invention provides an electronic device, including a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor executes the program to implement the above-described method.

[0061] The fourth embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-described method.

[0062] Experimental Design: A field experiment of maize with the interaction of three cultivars, three planting densities, and five nitrogen fertilizer application rates. The three cultivars were Zhengdan 958, Xianyu 335, and Woyu 3. The three planting density treatments were 60,000, 75,000, and 90,000 plants / ha. The five nitrogen fertilizer application rates were 0, 100, 200, 300, and 400 kg N / ha. -1 The residential area is 79.2 square meters. 2 Planting pattern: 6.6 m × 12 m, with equal row spacing (60-60 cm). The drone used was a DJI Phantom 4 RTK Multispectral Edition; specific parameters are shown in Table 4. A PSR-1100F ground cover spectrometer was used with fiber optic cable, calibrated for radiometric intensity. Spectral range: 320-1100 nm. Sampling bandwidth: 1.5 nm, wavelength accuracy: 0.5 nm. When using the spectrometer, it should be pointed directly at the corn, maintaining an angle of ±10° with the normal to the horizontal plane, and kept at a certain distance. Nitrogen content in leaves of each canopy layer was measured at the corn's tasseling, silking, and milk stages.

[0063] Table 4. DJI Phantom 4 Multispectral Edition UAV Gimbal Camera Parameters The results are shown in Figures 3 to 4. Figure 8 In Figure 3, * indicates a significant difference (P < 0.05), and ** indicates a highly significant difference (P < 0.01). In Figure 4, * and ** indicate significant and highly significant differences (P < 0.01), respectively. In Figure 5, * and ** indicate significant and highly significant differences (P < 0.01), respectively. In Figure 6, * and ** indicate significant and highly significant differences (P < 0.01), respectively. Figure 7 In the diagram, the horizontal axis represents the measured LNC of maize leaves (determined by a Kjeldahl nitrogen analyzer); the vertical axis represents the predicted LNC of maize leaves, in units of %; V12, R1, and R3 represent the large trumpet stage, silking stage, and milk stage, respectively; R 2 The coefficient of determination is represented by RMSE, which is the root mean square error. To facilitate differentiation of leaf layers, the upper, middle, and lower sections are labeled in the figure. Figure 8 In the diagram, the horizontal axis represents the measured LNC of maize leaves (measured using a Kjeldahl nitrogen analyzer); the vertical axis represents the predicted LNC of maize leaves, in units of %; V12, R1, and R3 represent the large trumpet stage, silking stage, and milk stage, respectively; R 2 The coefficient of determination is represented by RMSE, which is the root mean square error. To facilitate the differentiation of leaf layers, the upper, middle, and lower parts have been marked on the figure.

[0064] Based on the combined fitting performance of the modeling and validation sets at various growth stages, the multi-source data (VIs+HPs+TIs) combined with the random forest (RF) model exhibits the best predictive ability, significantly improving the prediction accuracy of nitrogen concentration in maize leaves compared to traditional UAV remote sensing nitrogen diagnosis. It also achieves stratified prediction of nitrogen content in the maize canopy, with the R1 stage showing the highest accuracy for LNC-Upper nitrogen concentration. 2 The R² value was 0.6494, an improvement of 8.7% compared to the single vegetation index model, with a 7.4% decrease in RMSE. The LNC-Middle improved from 0.7355 to 0.7657, a 4.1% improvement in accuracy; the LNC-Lower improved from 0.6836 to 0.7519, a 10.0% improvement. RMSE decreased in both LNC-Upper and LNC-Middle. At R3, the R² value for LNC-Upper was 0.7834, a 7.6% improvement compared to the single VIs model. RMSE also decreased significantly. In LNC-Middle and LNC-Lower, R² improved from 0.7152 and 0.7366 to 0.7216 and 0.7369 respectively. Although the improvement was small, it still demonstrated the robust gain effect of multi-source fusion.

[0065] Therefore, this invention uses multi-source data combined with a maize canopy nitrogen content monitoring model, which significantly improves the accuracy and stability of maize canopy nitrogen content monitoring.

[0066] The above inventions are merely a few specific embodiments of the present invention. However, the embodiments of the present invention are not limited thereto, and any variations that can be conceived by those skilled in the art should fall within the protection scope of the present invention.

Claims

1. A method for monitoring nitrogen content in the maize canopy, characterized in that, The method comprises the following steps: Obtaining multi-source remote sensing data of a sample corn field area, wherein the multi-source remote sensing data comprises multi-spectral images and hyperspectral reflectivity, constructing multi-spectral vegetation indexes and texture indexes based on the multi-spectral images, and constructing hyperspectral vegetation indexes based on the hyperspectral reflectivity; Constructing a mixed feature dataset by using the multi-spectral vegetation indexes, the hyperspectral vegetation indexes and the texture indexes; Using a LASSO regression algorithm to perform feature screening on the mixed feature dataset to obtain an optimal feature combination; Training a random forest model by using a training sample set formed by the optimal feature combination and corresponding measured values of corn leaf nitrogen concentration to form a corn canopy nitrogen content monitoring model; Obtaining a corn canopy leaf nitrogen content prediction value of a target corn field area in a to-be-collected period by inputting multi-source remote sensing data of the target corn field area into the corn canopy nitrogen content monitoring model.

2. The method of claim 1, wherein the step of determining the nitrogen content of the corn canopy is performed by a method comprising: The method for constructing the multi-spectral vegetation indexes based on the multi-spectral image data comprises the following steps: Synthesizing the multi-spectral images into test field orthographic images, including RGB and single-band orthographic images; Performing band fusion on the single-band orthographic images, drawing multi-spectral images, dividing a region of interest, extracting mean values of reflectivity of five center bands including blue, green, red, red edge and near-infrared in the region of interest, and constructing multi-spectral vegetation indexes.

3. The method for monitoring nitrogen content in maize canopy as described in claim 2, characterized in that, The method for constructing the texture indexes comprises the following steps: extracting texture indexes of a region of interest in a to-be-measured field area by using a gray level co-occurrence matrix and performing optimization.

4. The method for monitoring nitrogen content in maize canopy as described in claim 3, characterized in that, The optimization comprises the following steps: Obtaining normalized texture indexes, difference texture indexes and ratio texture indexes by using the texture indexes; Performing Pearson correlation analysis on the normalized texture indexes, the difference texture indexes and the ratio texture indexes respectively; Selecting the first 15 normalized texture indexes, the first 15 difference texture indexes and the first 15 ratio texture indexes with the highest correlation as optimized texture indexes, and constructing a mixed feature dataset together with the multi-spectral vegetation indexes and the hyperspectral vegetation indexes.

5. The method for monitoring nitrogen content in maize canopy as described in claim 4, characterized in that, the normalized texture index is obtained by the formula: ， In the formula, T1 and T2 both represent texture indexes based on a gray level co-occurrence matrix from the multi-spectral images; The difference texture index DTI is obtained by the following formula: ， The ratio texture index RTI is obtained by the following formula: 。 6. The method of claim 1, wherein, The method for constructing the hyperspectral vegetation indexes based on the hyperspectral reflectivity comprises the following steps: deriving the hyperspectral reflectivity, performing second derivative processing, and constructing hyperspectral vegetation indexes.

7. The method for monitoring nitrogen content in maize canopy as described in claim 1, characterized in that, The LASSO regression algorithm comprises the following steps: Performing standardization processing on the mixed feature dataset; Using five-fold cross-validation, randomly dividing the sample data into five parts, using four parts as a training set and one part as a validation set in turn, calculating a prediction error, calculating an average prediction error for each λ value, and selecting a λ value that minimizes the cross-validation error; For a series of λ values, using a coordinate descent method to solve the LASSO problem to obtain a coefficient estimate corresponding to each λ, and using a soft threshold operator to update the coefficient; Using the selected λ value to retrain the LASSO regression algorithm to obtain an optimal feature combination.

8. A corn canopy nitrogen content monitoring system characterized by, The method comprises the following steps: A data acquisition module is configured to acquire a corn canopy leaf nitrogen content prediction value obtained by the method in any one of claims 1 to 7. a decision module configured to generate a fertilization recommendation for the target corn field area based on the nitrogen content prediction.

9. An electronic device comprising a memory, a processor, and a computer program stored on the memory and running on the processor, characterized in that, The processor implements the method of any one of claims 1 to 7 when executing the program.

10. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program implements the method of any one of claims 1 to 7 when executed by a processor.

Images