Method and system for collaborative inversion of rice chlorophyll content and nitrogen content based on unmanned aerial vehicle remote sensing

Abstract

The application discloses a rice chlorophyll content and nitrogen content collaborative inversion method and system based on unmanned aerial vehicle remote sensing, and belongs to the technical field of crop growth remote sensing monitoring. The method first collects ground measured data of rice chlorophyll and nitrogen content and farmland unmanned aerial vehicle remote sensing images, and pre-processes the measured data and the images; then, based on the color space features of the visible light images, the OTSU threshold segmentation algorithm is combined to generate a rice canopy mask, and the pure canopy segmentation of the multispectral images is realized; the vegetation index is calculated, the sensitive spectral features are screened through correlation analysis, the binary linear regression collaborative inversion model is constructed by taking the sensitive spectral features as input, and the least square method is combined with Bootstrap resampling to optimize and solve the model parameters. The application is oriented to the natural rice field scene in the field, realizes synchronous quantitative inversion of the chlorophyll and nitrogen content, makes up for the low inversion accuracy and poor adaptability of the traditional single-parameter inversion, and has the advantages of high inversion accuracy, efficient operation and strong field practicability.

Description

Technical Field

[0001] This invention relates to the field of remote sensing monitoring technology for crop growth parameters, specifically a method for the synergistic inversion of chlorophyll content and nitrogen content in rice based on UAV remote sensing. Background Technology

[0002] Rice, as one of my country's major food crops, directly impacts national food security and agricultural economic development through its yield and quality. Chlorophyll and nitrogen are core indicators reflecting rice's photosynthetic capacity and nutritional status; accurately obtaining their contents is crucial for guiding field water and fertilizer management and predicting yield and quality. While traditional ground-based measurement methods offer high accuracy, they suffer from drawbacks such as cumbersome operation, high destructiveness, and poor spatiotemporal representativeness, making them unsuitable for large-scale real-time field monitoring. Existing remote sensing monitoring methods often employ single-parameter independent inversion, neglecting the close physiological coupling between chlorophyll and nitrogen, thus limiting further improvements in monitoring accuracy.

[0003] In view of this, the present invention aims to solve the problems of low efficiency and insufficient accuracy of manual monitoring of chlorophyll and nitrogen content in rice under natural conditions, and proposes a collaborative inversion method for chlorophyll and nitrogen content in rice based on UAV remote sensing, so as to achieve rapid, accurate and non-destructive synchronous monitoring. Summary of the Invention

[0004] To address the existing problems, the technical problem this invention aims to solve is to provide a method for the coordinated inversion of chlorophyll and nitrogen content in rice based on UAV remote sensing. This method first preprocesses multispectral images and constructs a canopy mask using visible light images to achieve accurate pure canopy extraction from the multispectral images, thereby obtaining spectral features that are sensitive to both chlorophyll and nitrogen content in rice. Second, a binary linear coordinated inversion model is constructed, and linear regression fitting is performed using the least squares method. Bootstrap resampling technology is introduced to generate multiple datasets for model training. Finally, the optimal inversion parameters are determined by calculating the statistical mean of the model parameters. This application fully utilizes the spectral coupling relationship between chlorophyll and nitrogen content, overcoming the limitations of single-parameter independent inversion and significantly improving the stability and reliability of dual-parameter estimation.

[0005] This application achieves the above effects through the following technical solution:

[0006] A method for synergistic inversion of chlorophyll and nitrogen content in rice based on UAV remote sensing, the method comprising the following steps:

[0007] S1 collects multispectral remote sensing images and visible light remote sensing images of rice under natural field scenes using drones. The multispectral remote sensing images include at least four bands: red light, green light, red edge, and near-infrared. Simultaneously, ground-based measured data of chlorophyll content and nitrogen content of rice in the same farmland are collected. The measured data should cover different phenological stages and growth states of rice to ensure the representativeness and integrity of the samples.

[0008] S2 performs black level correction, Gaussian filtering for noise reduction, and distortion correction on the multispectral remote sensing images of rice fields collected by UAVs. Then, it uses a spectral reflectance correction board to perform radiometric calibration on the preprocessed multispectral images to eliminate the influence of interference factors such as illumination and sensor differences, and finally obtains a rice canopy multispectral image with accurate spectral reflectance.

[0009] S3 first performs spatial registration between visible light and multispectral remote sensing images; then, it uses the visible light images to perform color space transformation, and combines the OTSU threshold segmentation algorithm to construct a rice canopy mask to remove non-vegetation background interference such as soil; finally, it applies the mask to the registered multispectral remote sensing images to complete the extraction of pure rice canopy pixels.

[0010] S4 uses multispectral remote sensing images of rice canopy extracted in S3 to calculate various vegetation indices. Correlation analysis is used to screen out feature indices that are significantly correlated with chlorophyll and nitrogen content. Subsequently, the screened features are matched with measured data according to the experimental area and collection time to construct a complete dataset, which is divided into training and validation sets. Finally, a binary linear co-inversion model is constructed based on the training set. The least squares method combined with Bootstrap resampling is used to solve for the optimal parameters of the model. The validation set is used for accuracy evaluation and model optimization to finally determine the optimal co-inversion model.

[0011] S5 inputs the multispectral image processed in step S2 into the trained collaborative inversion model to automatically extract sensitive spectral features; the collaborative inversion model performs calculations based on the input features to invert the distribution results of chlorophyll content and nitrogen content in field rice.

[0012] Furthermore, S2 specifically includes:

[0013] S21: Based on the camera's relative optical center parameters, the acquired spectral remote sensing images are cropped or filled to correct the optical center position and ensure the uniformity of the image optical reference.

[0014] S22: Perform black level correction on the remote sensing image processed in step S21. By subtracting the dark current signal of the sensor itself, eliminate the sensor noise interference and ensure that the image gray value can truly reflect the spectral information of the rice canopy.

[0015] S23: Use the Gaussian filtering algorithm to perform noise reduction and smoothing on the remote sensing image processed in S22, filter out random noise in the image, improve image clarity, and provide a high-quality image basis for subsequent correction steps.

[0016] S24: Based on the distortion coefficients in the metadata of the spectral remote sensing image, the corresponding distortion correction algorithm is used to remove the physical distortion caused by the lens optical characteristics, and the processed image is obtained to ensure the geometric accuracy of the image.

[0017] S25: Extract vignetting data parameters based on the metadata information inherent in the spectral remote sensing image, calculate the degree of illumination attenuation in the image edge area through the vignetting data parameters, and perform targeted brightness compensation to eliminate the influence of image vignetting on spectral reflectance.

[0018] S26: Using the sensor gain, exposure time, and parameters in the metadata, the raw digital quantization values ​​acquired by the camera are converted into initial physical reflectance values. The calculation expression is as follows:

[0019] ;

[0020] S27: Further calibrate the initial physical reflectance value calculated in step S26 using a spectral reflectance calibration plate. Using the standard reflectance of the spectral reflectance calibration plate as the dependent variable and the actual reflectance of the spectral reflectance calibration plate image as the independent variable, construct a linear regression equation. Solve the coefficients of this equation using the linear regression method. The formula for calculating the linear regression equation is:

[0021] ;

[0022] in, Given the known standard reflectance of the spectral reflectance correction plate, The actual reflectance of the spectral reflectance correction plate image calculated in step S26, where k is the regression coefficient and d is a constant term;

[0023] S28: Substitute the initial physical reflectance values ​​of each pixel in the multispectral image calculated in step S26 into the linear regression equation constructed in step S27 to complete the reflectance correction of the entire image and finally obtain accurate rice canopy spectral reflectance data.

[0024] Furthermore, S3 specifically refers to:

[0025] To address the issue of inconsistent spatial resolution between visible light and multispectral images, S31 utilizes ENVI software to assign geographic coordinates to visible light images, selects uniformly distributed corresponding control points, and achieves accurate geographic registration between visible light and multispectral images through downsampling and resampling.

[0026] S32 performs color space transformation on visible light images, and constructs a rice canopy mask using the OTSU threshold segmentation algorithm to remove background interference such as soil and weeds. The mask is then used to segment the preprocessed and registered multispectral images to extract pure rice canopy pixel multispectral remote sensing images.

[0027] Furthermore, S4 specifically includes:

[0028] S41 uses multispectral remote sensing images of pure rice canopy pixels, combined with the sensitivity characteristics of the red edge band, to calculate the vegetation indices NDVI, NDRE, GNDVI, OSAVI, VSI, CARI, DVI, GDVI, GOSAVI, GSAVI, MERIS, NGI, RERVI, CIG, GRVI, MCARIL, RVI, CIRE, and SR. The calculation formulas are as follows:

[0029]

[0030]

[0031]

[0032]

[0033]

[0034]

[0035]

[0036]

[0037]

[0038]

[0039]

[0040]

[0041]

[0042]

[0043]

[0044]

[0045]

[0046]

[0047]

[0048] Where red is the red band spectral reflectance, green is the green band spectral reflectance, nir is the near-infrared band spectral reflectance, and re is the red band spectral reflectance.

[0049] S42 performs correlation analysis between each vegetation index and the measured values ​​of chlorophyll and nitrogen, sorts them by the absolute value of the correlation coefficient, and selects vegetation indices that are significantly correlated with both as input features of the model.

[0050] S43 Based on the screening results of step S42, select the set vegetation index and match it with the measured chlorophyll and nitrogen data one by one according to region and time to construct a complete sample set; divide it into training set and validation set according to the set ratio.

[0051] Based on the training set partitioned in step S43, S44 constructs a binary linear co-inversion model with vegetation index as the independent variable and chlorophyll content and nitrogen content as the dependent variables. The expression for the co-inversion model is as follows:

[0052]

[0053] Wherein, VIs is a feature matrix composed of defined vegetation index features. This represents the measured nitrogen content of rice. This represents the measured value of chlorophyll content in rice. These are the regression coefficients of the model. For model constants;

[0054] S44 uses the least squares method combined with Bootstrap resampling to solve for the optimal parameters of the model; it achieves parameter estimation by minimizing the sum of squared residuals between the predicted and measured values, and improves the stability and reliability of the parameters by averaging the results of multiple resampling.

[0055] S46 uses a validation set to validate the model accuracy. Evaluation metrics include the coefficient of determination R², root mean square error RMSE, Pearson correlation coefficient, and Spearman correlation coefficient. If the model accuracy does not meet the preset standard, the process returns to the feature selection or dataset partitioning steps for re-optimization, ultimately determining the optimal collaborative inversion model.

[0056] Furthermore, the method also includes S6:

[0057] S6 analyzes the spatial distribution results of chlorophyll and nitrogen content in rice, and combines them with the grading standards for chlorophyll and nitrogen content at different phenological stages established based on field measurement data to comprehensively determine the current growth phenological stage of rice in the field.

[0058] This application also provides a system for the joint inversion of chlorophyll and nitrogen content in rice based on UAV remote sensing, the system comprising:

[0059] The image acquisition module is used to collect multispectral remote sensing image data of rice under natural field scenes by drones. The spectral remote sensing image includes at least four bands: red light, green light, red edge and near-infrared.

[0060] The image preprocessing module, the start calculation module is used to perform black level correction, Gaussian filtering noise reduction and distortion correction on the acquired multispectral remote sensing images of paddy rice from UAVs in sequence. Then, the preprocessed multispectral images are radiometrically calibrated using a spectral reflectance correction plate to eliminate the influence of interference factors such as illumination and sensor differences, and finally obtain a rice canopy multispectral image with accurate spectral reflectance.

[0061] The collaborative inversion model module is used to input the preprocessed multispectral image into the trained collaborative inversion model and automatically extract sensitive spectral features; the collaborative inversion model performs calculations based on the input features to invert the distribution results of chlorophyll content and nitrogen content in field rice.

[0062] The information display module is used to print the log information generated during the calculation process and the final calculation results. This module will provide the average chlorophyll content and nitrogen content of the farmland, as well as the reference phenological period information corresponding to the image.

[0063] The beneficial effects of this invention are as follows:

[0064] 1. To achieve simultaneous inversion of chlorophyll content and nitrogen content in rice under natural scenes.

[0065] 2. A binary linear regression model for the synergistic inversion of chlorophyll and nitrogen content in rice under natural scenarios was constructed.

[0066] 3. The least squares method and bootstrap sampling combination method are used to solve the model parameters, thereby improving the stability and reliability of parameter estimation. Attached Figure Description

[0067] Figure 1 This is a flowchart illustrating the overall method of the present invention;

[0068] Figure 2 This is a flowchart illustrating the construction process of the collaborative inversion model of the present invention.

[0069] Figure 3 This is a diagram showing the spatial distribution of rice chlorophyll content calculated by the system of the present invention.

[0070] Figure 4This is a spatial distribution diagram of the nitrogen content in rice calculated by the system of the present invention. Detailed Implementation

[0071] This application discloses a method for the coordinated inversion of chlorophyll and nitrogen content in rice based on multispectral remote sensing images. The method utilizes the least squares approach to construct a coordinated inversion model for rice chlorophyll and nitrogen content. This invention extends the traditional single chlorophyll or nitrogen content inversion model for rice through a coordinated approach, resulting in significant differences in technical path and implementation compared to traditional solutions. The specific details are as follows:

[0072] To clearly illustrate the core technical support of this invention, the OTSU threshold segmentation, least squares method, and Bootstrap resampling method used will first be described in detail:

[0073] OTSU thresholding, proposed by Otsu in 1979, is an adaptive image thresholding method based on grayscale clustering, widely used in machine vision, remote sensing image processing, and target extraction. This method eliminates the need for manual threshold setting, automatically calculating the optimal segmentation threshold based on the image's grayscale histogram to divide the image into foreground and background regions. It features high computational efficiency, strong adaptability, no manual intervention, and robustness to changes in natural lighting. In this invention, OTSU thresholding is primarily used for the automatic extraction of rice canopy layers from visible light images. Using a pre-processed color image as input, it performs binarization of the foreground and background and outputs a rice canopy mask, removing background interference such as soil and weeds, providing mask support for accurate segmentation of subsequent multispectral images.

[0074] The Least Squares Method (LSM) is a classic mathematical optimization and parameter estimation algorithm. It achieves optimal function matching and parameter solving by minimizing the sum of squared residuals between observed and predicted values. It is widely used in regression analysis, data fitting, and machine learning model training. This method is clear in principle, computationally efficient, and produces stable and unbiased fitting results, making it a mainstream method for linear model parameter estimation. In this invention, the LSM is used to solve the parameters of a binary linear co-inversion model of rice chlorophyll and nitrogen content. Using vegetation index characteristics as input and measured physiological parameters as output, it minimizes the sum of squared errors between predicted and measured values ​​to achieve optimal estimation of regression coefficients and constant terms, providing reliable model parameters for co-inversion. The calculation process is as follows:

[0075] First, the core principle of the least squares method is to obtain a set of parameters through optimization that minimizes the sum of squared residuals between the model-fitted values ​​and the measured values. The objective function expression is:

[0076]

[0077] During the iteration process, the model parameters are adjusted using gradient descent. Parameter updates are achieved by calculating the partial derivatives of the loss function with respect to each parameter, gradually minimizing the sum of squared residuals between the model's predicted and measured values ​​to obtain the optimal model parameters. The formulas for calculating the partial derivatives of each parameter are as follows:

[0078]

[0079]

[0080]

[0081] After obtaining the partial derivatives of the model parameters, iteratively update them with a learning rate η. By continuously iterating until convergence, the optimal regression coefficients can be obtained. The calculation formula is as follows:

[0082]

[0083]

[0084]

[0085] Bootstrap resampling is a robust estimation method based on statistical sampling. It generates multiple independent subsets by repeatedly sampling the original dataset with replacement, models each subset separately, and finally outputs the statistical mean. This effectively improves the stability and generalization ability of the model parameters. This method does not require prior assumptions about the data distribution, reducing the impact of sample fluctuations and outliers, making it particularly suitable for parameter estimation scenarios with small sample sizes and high robustness requirements. In this invention, Bootstrap resampling is combined with least squares to generate multiple sampled datasets from the training set and solve for the parameters separately. The average of the results from multiple sets is taken to obtain the final optimal parameters, further improving the accuracy, stability, and reliability of the chlorophyll and nitrogen content co-inversion model.

[0086] Example 1

[0087] This application provides a method for the coordinated inversion of chlorophyll and nitrogen content in rice leaves based on multispectral remote sensing images. The flowchart of the method is shown below. Figure 1 As shown, the overall implementation path includes the following steps:

[0088] S1 collects multispectral remote sensing images and visible light remote sensing images of rice under natural field scenes using drones. The multispectral remote sensing images include at least four bands: red light, green light, red edge, and near-infrared. Simultaneously, ground-based measured data of chlorophyll content and nitrogen content of rice in the same farmland are collected. The measured data should cover different phenological stages and growth states of rice to ensure the representativeness and integrity of the samples.

[0089] S2 performs preprocessing on the multispectral remote sensing images of rice fields collected by UAVs, including black level correction, Gaussian filtering for noise reduction, and distortion correction. Then, it uses a spectral reflectance correction board to perform radiometric calibration on the preprocessed multispectral images to eliminate the influence of interference factors such as illumination and sensor differences, and finally obtains a rice canopy multispectral image with accurate spectral reflectance.

[0090] S3 first performs spatial registration between visible light and multispectral remote sensing images; then, it uses the visible light images to perform color space transformation, and combines the OTSU threshold segmentation algorithm to construct a rice canopy mask to remove non-vegetation background interference such as soil; finally, it applies the mask to the registered multispectral remote sensing images to complete the extraction of pure rice canopy pixels.

[0091] S4 uses multispectral remote sensing images of rice canopy extracted in S3 to calculate various vegetation indices. Correlation analysis is used to screen out feature indices that are significantly correlated with chlorophyll and nitrogen content. Subsequently, the screened features are matched with measured data according to the experimental area and collection time to construct a complete dataset, which is divided into training and validation sets. Finally, a binary linear co-inversion model is constructed based on the training set. The least squares method combined with Bootstrap resampling is used to solve for the optimal parameters of the model. The validation set is used for accuracy evaluation and model optimization to finally determine the optimal co-inversion model.

[0092] S5 inputs the multispectral image processed in step S2 into the trained collaborative inversion model to automatically extract sensitive spectral features; the collaborative inversion model performs calculations based on the input features to invert the distribution results of chlorophyll content and nitrogen content in field rice.

[0093] S6 analyzes the spatial distribution results of chlorophyll and nitrogen content in rice, and combines them with the grading standards for chlorophyll and nitrogen content at different phenological stages established based on field measurement data to comprehensively determine the current growth phenological stage of rice in the field.

[0094] Firstly, step S2 of the above scheme provides a complete UAV multispectral image preprocessing and reflectance calibration process. By sequentially performing preprocessing such as black level correction, Gaussian filtering for noise reduction, and distortion correction, image geometric errors, optical noise, and sensor dark current interference are systematically eliminated. Then, a standard reflectance calibration board is used to complete accurate radiometric calibration, effectively reducing the influence of external factors such as light changes and atmospheric attenuation on spectral information. Finally, high-precision and high-fidelity rice canopy spectral reflectance data are obtained, laying a reliable data foundation for subsequent vegetation index calculation, spectral feature extraction, and parameter inversion.

[0095] S2 includes the following:

[0096] S21: Based on the camera's relative optical center parameters, the acquired spectral remote sensing images are cropped or filled to correct the optical center position and ensure the consistency of the image optical reference.

[0097] S22: Perform black level correction on the remote sensing image processed in step S21. By subtracting the dark current signal of the sensor itself, eliminate the sensor noise interference and ensure that the image gray value can truly reflect the spectral information of the rice canopy.

[0098] S23: Use the Gaussian filtering algorithm to perform noise reduction and smoothing on the remote sensing image processed in S22, filter out random noise in the image, improve image clarity, and provide a high-quality image basis for subsequent correction steps.

[0099] S24: Based on the distortion coefficients in the metadata of the spectral remote sensing image, the corresponding distortion correction algorithm is used to remove the physical distortion caused by the lens optical characteristics, and the processed image is obtained to ensure the geometric accuracy of the image.

[0100] S25: Extract vignetting data parameters based on the metadata information of the image, calculate the degree of illumination attenuation in the image edge area using these parameters, and perform targeted brightness compensation to eliminate the impact of image vignetting on spectral reflectance.

[0101] S26: Using the sensor gain, exposure time, and ambient light intensity parameters in the metadata, the raw DN (Digital Number) value acquired by the camera is converted into an initial physical reflectance value. The calculation formula is as follows:

[0102]

[0103] S27: Further calibrate the initial physical reflectance value calculated in step S26 using a standard reflectance calibration plate. Construct a linear regression equation with the standard reflectance of the calibration plate as the dependent variable and the actual reflectance of the calibration plate image as the independent variable. Solve the coefficients of this equation using the linear regression method. The formula for calculating the linear regression equation is as follows:

[0104]

[0105] in, The known standard reflectance of the standard reflectance calibration plate. d represents the actual reflectance of the calibration plate image calculated in step S26, k is the regression coefficient, and d is a constant term.

[0106] S28: Substitute the initial physical reflectance values ​​of each pixel in the multispectral image calculated in step S26 into the linear regression equation constructed in step S27 to complete the reflectance correction of the entire image and finally obtain accurate rice canopy spectral reflectance data.

[0107] Secondly, step S3 of the above scheme provides a rice canopy extraction scheme based on visible light images. This part is based on synchronously acquired visible light remote sensing images. First, the visible light images and multispectral images are georegistered to achieve spatial alignment between the two. Then, the background and canopy are distinguished by color space transformation on the registered visible light images, and a rice canopy mask is constructed by combining the OTSU threshold segmentation algorithm. Finally, the multispectral images are segmented using this mask to complete the canopy extraction, and finally a clean rice canopy multispectral remote sensing image is obtained.

[0108] S3 includes the following:

[0109] To address the issue of inconsistent spatial resolution between visible light and multispectral images, S31 utilizes ENVI software to assign geographic coordinates to visible light images, selects uniformly distributed corresponding control points, and achieves accurate geographic registration between visible light and multispectral images through downsampling and resampling.

[0110] S32 performs color space transformation on visible light images, and constructs a rice canopy mask using the OTSU threshold segmentation algorithm to remove background interference such as soil and weeds. The mask is then used to segment the preprocessed and registered multispectral images to extract pure rice canopy pixel multispectral remote sensing images.

[0111] Thirdly, step S4 of the above scheme provides a complete, closed-loop, and engineerable scheme for constructing and inferring a synergistic inversion model of rice chlorophyll and nitrogen content. This part is based on multispectral remote sensing images that have undergone a series of processing steps to calculate vegetation indices and screen correlations, thereby constructing a high-quality model sample set; the process of constructing a binary linear synergistic inversion model is as follows: Figure 2 As shown, the optimal estimation of model parameters is achieved by combining the least squares method and Bootstrap resampling, and multiple indicators are used to complete the model accuracy verification and iterative optimization. Finally, the spatial distribution results of chlorophyll content and nitrogen content of paddy rice in the field are obtained directly from the corrected multispectral images, which ensures the automation of the inversion process, the accuracy of the inversion results and the robustness of the model.

[0112] S4 includes the following:

[0113] Based on accurate canopy spectral reflectance data and combined with the sensitivity characteristics of the red edge band, S41 calculates vegetation indices such as NDVI, NDRE, GNDVI, OSAVI, VSI, CARI, DVI, GDVI, GOSAVI, GSAVI, MERIS, NGI, RERVI, CIG, GRVI, MCARIL, RVI, CIRE, and SR. The calculation formulas are as follows:

[0114]

[0115]

[0116]

[0117]

[0118]

[0119]

[0120]

[0121]

[0122]

[0123]

[0124]

[0125]

[0126]

[0127]

[0128]

[0129]

[0130]

[0131]

[0132]

[0133] Where red is the red band spectral reflectance, green is the green band spectral reflectance, nir is the near-infrared band spectral reflectance, and re is the red band spectral reflectance.

[0134] S42 performs correlation analysis between each vegetation index and the measured values ​​of chlorophyll and nitrogen, sorts them by the absolute value of the correlation coefficient, and selects vegetation indices that are significantly correlated with both as input features of the model.

[0135] S43 To ensure model accuracy and stability, based on the screening results of step S42, 12 preferred vegetation indices are selected and matched with the measured chlorophyll and nitrogen data by region and time to construct a complete sample set; the sample set is divided into training set and validation set in a 7:3 ratio to ensure the continuity of data time sequence and avoid training bias.

[0136] Based on the dataset partitioned in step S43, S44 constructs a binary linear joint inversion model with vegetation index as the independent variable and chlorophyll content and nitrogen content as the dependent variables. The model expression is as follows:

[0137]

[0138] Among them, VIs is a feature matrix composed of 12 preferred vegetation index features. This represents the measured nitrogen content of rice. This represents the measured value of chlorophyll content in rice. These are the regression coefficients of the model. This is the model constant term.

[0139] S45 uses the least squares method combined with Bootstrap resampling to solve for the optimal parameters of the model; it achieves parameter estimation by minimizing the sum of squared residuals between the predicted and measured values, and improves the stability and reliability of the parameters by averaging the results of multiple resampling.

[0140] S46 uses a validation set to validate the model accuracy. Evaluation metrics include the coefficient of determination R², root mean square error RMSE, Pearson correlation coefficient, and Spearman correlation coefficient. If the model accuracy does not meet the preset standards (R² ≥ 0.85, RMSE ≤ 5%), the feature selection or dataset partitioning steps are returned for re-optimization to finally determine the optimal collaborative inversion model.

[0141] Example 2

[0142] This application also provides a collaborative inversion system for chlorophyll and nitrogen content in rice based on multispectral remote sensing images, as shown in the system interface diagram below. Figure 3 Figure 4 As shown, the system includes:

[0143] The image acquisition module is used to collect multispectral remote sensing image data of rice under natural field scenes by drones. The spectral remote sensing image includes at least four bands: red light, green light, red edge and near-infrared.

[0144] The image preprocessing module, the start calculation module is used to perform black level correction, Gaussian filtering noise reduction and distortion correction on the acquired multispectral remote sensing images of paddy rice from UAVs in sequence. Then, the preprocessed multispectral images are radiometrically calibrated using a spectral reflectance correction plate to eliminate the influence of interference factors such as illumination and sensor differences, and finally obtain a rice canopy multispectral image with accurate spectral reflectance.

[0145] The collaborative inversion model module is used to input the preprocessed multispectral image into the trained collaborative inversion model and automatically extract sensitive spectral features. The collaborative inversion model performs calculations based on the input features to invert the distribution results of chlorophyll content and nitrogen content in paddy rice.

[0146] The information display module is used to print the log information generated during the calculation process and the final calculation results. This module will provide the average chlorophyll content and nitrogen content of the farmland, as well as the reference phenological period information corresponding to the image.

[0147] This application discloses a method for the coordinated inversion of chlorophyll and nitrogen content in rice based on multispectral remote sensing images. The method first preprocesses the multispectral image and constructs a canopy mask using visible light imagery to achieve accurate pure canopy extraction, thereby obtaining spectral features that are sensitive to both chlorophyll and nitrogen content in rice. Second, a binary linear coordinated inversion model is constructed, and linear regression fitting is performed using the least squares method. Bootstrap resampling technology is introduced to generate multiple datasets for model training. Finally, the optimal inversion parameters are determined by calculating the statistical mean of the model parameters. This application fully utilizes the spectral coupling relationship between chlorophyll and nitrogen content, overcoming the limitations of single-parameter independent inversion and significantly improving the stability and reliability of dual-parameter estimation.

Claims

1. A method for the synergistic inversion of chlorophyll content and nitrogen content in rice based on UAV remote sensing, characterized in that, The method includes the following steps: S1 collects multispectral remote sensing images and visible light remote sensing images of rice under natural field scenes using drones. The multispectral remote sensing images include at least four bands: red light, green light, red edge, and near-infrared. Simultaneously, ground-based measured data of chlorophyll content and nitrogen content of rice in the same farmland are collected. The measured data should cover different phenological stages and growth states of rice to ensure the representativeness and integrity of the samples. S2 performs black level correction, Gaussian filtering for noise reduction, and distortion correction on the multispectral remote sensing images of rice fields collected by UAVs. Then, it uses a spectral reflectance correction board to perform radiometric calibration on the preprocessed multispectral images to eliminate the influence of interference factors such as illumination and sensor differences, and finally obtains a rice canopy multispectral image with accurate spectral reflectance. S3 first performs spatial registration between visible light and multispectral remote sensing images; then, it uses the visible light images to perform color space transformation, and combines the OTSU threshold segmentation algorithm to construct a rice canopy mask to remove non-vegetation background interference such as soil; finally, it applies the mask to the registered multispectral remote sensing images to complete the extraction of pure rice canopy pixels. S4 uses multispectral remote sensing images of rice canopy extracted in S3 to calculate various vegetation indices. Correlation analysis is used to screen out feature indices that are significantly correlated with chlorophyll and nitrogen content. Subsequently, the screened features are matched with measured data according to the experimental area and collection time to construct a complete dataset, which is divided into training and validation sets. Finally, a binary linear co-inversion model is constructed based on the training set. The least squares method combined with Bootstrap resampling is used to solve for the optimal parameters of the model. The validation set is used for accuracy evaluation and model optimization to finally determine the optimal co-inversion model. S5 inputs the multispectral image processed in step S2 into the trained collaborative inversion model to automatically extract sensitive spectral features; the collaborative inversion model performs calculations based on the input features to invert the distribution results of chlorophyll content and nitrogen content in field rice.

2. The method for synergistic inversion of chlorophyll content and nitrogen content in rice based on UAV remote sensing according to claim 1, characterized in that, Specifically, S2 is: S21: Based on the camera's relative optical center parameters, the acquired spectral remote sensing images are cropped or filled to correct the optical center position and ensure the uniformity of the image optical reference. S22: Perform black level correction on the remote sensing image processed in step S21. By subtracting the dark current signal of the sensor itself, eliminate the sensor noise interference and ensure that the image gray value can truly reflect the spectral information of the rice canopy. S23: Use the Gaussian filtering algorithm to perform noise reduction and smoothing on the remote sensing image processed in S22, filter out random noise in the image, improve image clarity, and provide a high-quality image basis for subsequent correction steps. S24: Based on the distortion coefficients in the metadata of the spectral remote sensing image, the corresponding distortion correction algorithm is used to remove the physical distortion caused by the lens optical characteristics, and the processed image is obtained to ensure the geometric accuracy of the image. S25: Extract vignetting data parameters based on the metadata information inherent in the spectral remote sensing image, calculate the degree of illumination attenuation in the image edge area through the vignetting data parameters, and perform targeted brightness compensation to eliminate the influence of image vignetting on spectral reflectance. S26: Using the sensor gain, exposure time, and parameters in the metadata, the raw digital quantization values ​​acquired by the camera are converted into initial physical reflectance values. The calculation expression is as follows: ； S27: Further calibrate the initial physical reflectance value calculated in step S26 using a spectral reflectance calibration plate. Using the standard reflectance of the spectral reflectance calibration plate as the dependent variable and the actual reflectance of the spectral reflectance calibration plate image as the independent variable, construct a linear regression equation. Solve the coefficients of this equation using the linear regression method. The formula for calculating the linear regression equation is: ； in, Given the known standard reflectance of the spectral reflectance correction plate, The actual reflectance of the spectral reflectance correction plate image calculated in step S26, where k is the regression coefficient and d is a constant term; S28: Substitute the initial physical reflectance values ​​of each pixel in the multispectral image calculated in step S26 into the linear regression equation constructed in step S27 to complete the reflectance correction of the entire image and finally obtain accurate rice canopy spectral reflectance data.

3. The method for synergistic inversion of chlorophyll content and nitrogen content in rice based on UAV remote sensing according to claim 1, characterized in that, Specifically, S3 is: To address the issue of inconsistent spatial resolution between visible light and multispectral images, S31 utilizes ENVI software to assign geographic coordinates to visible light images, selects uniformly distributed corresponding control points, and achieves accurate geographic registration between visible light and multispectral images through downsampling and resampling. S32 performs color space transformation on visible light images and constructs a rice canopy mask using the OTSU threshold segmentation algorithm to remove background interference such as soil and weeds. The mask is then used to segment the preprocessed and registered multispectral images to extract pure rice canopy pixel multispectral remote sensing images.

4. The method for synergistic inversion of chlorophyll content and nitrogen content in rice based on UAV remote sensing according to claim 1, characterized in that, Specifically, S4 is: S41 uses multispectral remote sensing images of pure rice canopy pixels, combined with the sensitivity characteristics of the red edge band, to calculate the vegetation indices NDVI, NDRE, GNDVI, OSAVI, VSI, CARI, DVI, GDVI, GOSAVI, GSAVI, MERIS, NGI, RERVI, CIG, GRVI, MCARIL, RVI, CIRE, and SR. The calculation formulas are as follows: ； ； ； ； ； ； ； ； ； ； ； ； ； ；； ； ； ； ； ； Where red is the red band spectral reflectance, green is the green band spectral reflectance, nir is the near-infrared band spectral reflectance, and re is the red band spectral reflectance. S42 performs correlation analysis between each vegetation index and the measured values ​​of chlorophyll and nitrogen, sorts them by the absolute value of the correlation coefficient, and selects vegetation indices that are significantly correlated with both as input features of the model. S43 Based on the screening results of step S42, select the set vegetation index and match it with the measured chlorophyll and nitrogen data one by one according to region and time to construct a complete sample set; divide it into training set and validation set according to the set ratio. Based on the training set partitioned in step S43, S44 constructs a binary linear co-inversion model with vegetation index as the independent variable and chlorophyll content and nitrogen content as the dependent variables. The expression for the co-inversion model is as follows: ； Wherein, VIs is a feature matrix composed of defined vegetation index features. This represents the measured nitrogen content of rice. This represents the measured value of chlorophyll content in rice. These are the regression coefficients of the model. For model constants; S44 uses the least squares method combined with Bootstrap resampling to solve for the optimal parameters of the model; it achieves parameter estimation by minimizing the sum of squared residuals between the predicted and measured values, and improves the stability and reliability of the parameters by averaging the results of multiple resampling. S45 uses a validation set to validate the model accuracy. Evaluation metrics include the coefficient of determination R², root mean square error RMSE, Pearson correlation coefficient, and Spearman correlation coefficient. If the model accuracy does not meet the preset standard, the process returns to the feature selection or dataset partitioning steps for re-optimization, ultimately determining the optimal collaborative inversion model.

5. The method for synergistic inversion of chlorophyll content and nitrogen content in rice based on UAV remote sensing according to claim 1, characterized in that, The method further includes S6: S6 analyzes the spatial distribution results of chlorophyll and nitrogen content in rice, and combines them with the grading standards for chlorophyll and nitrogen content at different phenological stages established based on field measurement data to comprehensively determine the current growth phenological stage of rice in the field.

6. A system for the coordinated inversion of chlorophyll and nitrogen content in rice based on UAV remote sensing, characterized in that, The system includes: The image acquisition module is used to collect multispectral remote sensing image data of rice under natural field scenes by drones. The spectral remote sensing image includes at least four bands: red light, green light, red edge and near-infrared. The image preprocessing module, the start calculation module is used to perform black level correction, Gaussian filtering noise reduction and distortion correction on the acquired multispectral remote sensing images of paddy rice from UAVs in sequence. Then, the preprocessed multispectral images are radiometrically calibrated using a spectral reflectance correction plate to eliminate the influence of interference factors such as illumination and sensor differences, and finally obtain a rice canopy multispectral image with accurate spectral reflectance. The collaborative inversion model module is used to input the preprocessed multispectral image into the trained collaborative inversion model and automatically extract sensitive spectral features; the collaborative inversion model performs calculations based on the input features to invert the distribution results of chlorophyll content and nitrogen content in field rice. The information display module is used to print the log information generated during the calculation process and the final calculation results. This module will provide the average chlorophyll content and nitrogen content of the farmland, as well as the reference phenological period information corresponding to the image.

Images