Saline-alkali soil corn growth vigor and salt stress monitoring method based on unmanned aerial vehicle multispectrum

Abstract

The invention discloses a saline-alkali soil corn growth vigor and salt stress monitoring method based on unmanned aerial vehicle multispectrum, and the method comprises the steps: collecting canopy multispectral images and flight attitude data in different growth periods of corn through an unmanned aerial vehicle, and extracting spectral reflectivity, vegetation index and texture feature parameters after correction and registration through an unmanned aerial vehicle agricultural spectral data application platform. Constructing a salt-alkali gradient corn salt stress response model to determine a salt stress key spectrum sensitive wave band, constructing a time sequence data set by adopting a growth vigor dynamic spectrum sequence analysis algorithm, and inputting related data into a corn spectrum response random forest mechanism model to complete feature optimization and training; and finally outputting the corn growth grade, the salt stress degree and the spatial distribution characteristic parameters. According to the method, through multi-model cooperation and time sequence data processing, accurate and efficient monitoring of saline-alkali soil corn growth vigor and salt stress is achieved, and technical support is provided for saline-alkali soil corn planting management.

Description

A method for monitoring maize growth and salt stress in saline-alkali land using UAV multispectral data. Technical Field

[0001] This invention relates to the field of maize growth monitoring technology, and in particular to a method for monitoring maize growth and salt stress in saline-alkali land based on UAV multispectral imaging. Background Technology

[0002] Saline-alkali land, as an important untapped land resource in agricultural production, is prone to hindering maize growth and reducing yield due to its high soil salinity. Accurate monitoring of maize growth and salt stress is crucial for improving the utilization efficiency of saline-alkali land. Traditional monitoring methods rely heavily on manual field sampling and laboratory analysis, which are time-consuming, labor-intensive, have limited coverage, and poor timeliness, making them unsuitable for the dynamic monitoring needs of large-scale maize cultivation in saline-alkali land. With the rapid development of UAV technology and multispectral remote sensing technology, UAV multispectral monitoring, with its advantages of low-altitude flight, flexibility, and efficient data acquisition, can quickly acquire maize canopy spectral information. This provides technical support for large-scale, high-precision monitoring of maize growth and salt stress in saline-alkali land, and has become a research hotspot in the field of agricultural remote sensing monitoring.

[0003] Existing methods for monitoring maize in saline-alkali land based on spectral technology have two significant drawbacks: First, the spectral response signals of salt stress and maize growth are prone to mutual interference. Existing models often consider a single factor and do not fully couple the dynamic spectral characteristics of the saline-alkali gradient with the different growth stages of maize, resulting in insufficient accuracy in distinguishing between the degree of salt stress and the growth level. Second, data processing and model application lack systematic integration. Existing technologies often focus on the development of single algorithms or models and have not formed a complete technical system from multispectral data acquisition, feature extraction, model training to result output. Furthermore, the correction accuracy of UAV-collected data and the correlation processing of time-series data are not perfect, making it difficult to achieve accurate inversion of salt stress and growth parameters and effective characterization of spatial distribution characteristics. Summary of the Invention

[0004] In order to overcome the shortcomings and deficiencies of existing technologies, this invention provides a method for monitoring maize growth and salt stress in saline-alkali land based on UAV multispectral data.

[0005] The technical solution adopted in this invention is a method for monitoring the growth and salt stress of maize in saline-alkali land based on UAV multispectral imaging, comprising the following steps: S1, using a UAV equipped with a multispectral imaging module to fly at low altitude along a preset route at different growth stages of maize in saline-alkali land, simultaneously collecting multispectral image data of the maize canopy and flight attitude data; S2, using a UAV agricultural spectral data application platform to perform target correction and geometric registration on the collected multispectral image data, extracting maize canopy spectral reflectance, vegetation index, and texture feature parameters; S3, constructing a saline-alkali gradient maize salt stress response model, utilizing the spectral reflectance... S4. Coupled analysis of the growth rate and salinity gradient data to determine the spectral sensitive bands for salt stress calibration; S5. Using the growth dynamic spectral sequence analysis algorithm, the vegetation index and texture feature parameters at different growth stages are processed temporally to construct a maize growth dynamic spectral sequence dataset; S6. The calibrated spectral sensitive band data and the growth dynamic spectral sequence dataset are input into the maize spectral response random forest mechanism model for feature dimension optimization and model training; S7. The trained maize spectral response random forest mechanism model outputs maize growth level parameters, salt stress degree parameters, and spatial distribution feature parameters in saline-alkali land.

[0006] Furthermore, the expression for the salt-alkali gradient maize salt stress response model is as follows: ,in, This represents the salt stress response value. These are the model fitting coefficients. These are the spectral reflectances in the near-infrared, red, blue, and green bands, respectively. For parameters related to the salinity content of saline-alkali land, Normalized Difference Vegetation Index (NDVI) To calibrate the wavelength parameters of the spectral sensitive band, For parameters of the salinity gradient, The results are principal component analysis of reflectance in the blue and green bands.

[0007] Furthermore, the expression for the growth dynamic spectral sequence analysis algorithm is as follows: ,in, The value of the growth dynamic sequence at time t. For time series data length, Adjusting parameters for the algorithm, for Real-time vegetation index Let be the texture feature parameters at time t. As the baseline reproductive period, Let be the canopy texture feature parameters at time t.

[0008] Furthermore, the expression for the maize spectral response random forest mechanism model is as follows: ,in, To monitor the output composite value, For the number of decision trees, The number of feature dimensions. Let be the weight of the j-th feature in the k-th decision tree. For the bias term of the k-th decision tree, This represents the training output of the random forest model for the j-th feature. This is an index showing the correlation between plant growth and salt stress. For model gain parameters, Let j be the j-th input feature parameter.

[0009] Furthermore, the data processing model expression of the UAV agricultural spectral data application platform is as follows: ,in, For the processed standard data, This is the target temperature correction factor. For image geometric registration function, Let Variance be the spectral reflectance. This is the error adjustment factor. To acquire image quality parameters, The correlation coefficient between multispectral data and standard reflectance. This is flight attitude data.

[0010] Furthermore, the coupling model expression for maize growth and salt stress monitoring parameters in saline-alkali land is as follows: ,in, For parameter coupling values, The coupling coefficient is... For growth level parameters, This is a parameter representing the degree of salt stress. The standard deviation of the spatial distribution. To optimize parameters for features, For logistic regression function, This is the normalization function.

[0011] Further, step S3 includes the following sub-steps: S31 Select experimental plots with significant differences in salinity gradient, collect multispectral reflectance data of maize canopy and soil salinity content correlation parameters under different gradients, and establish a dataset corresponding to salinity gradient and spectral reflectance; S32 Screen the multispectral reflectance data using band sensitivity analysis, and retain candidate bands with a correlation to salt stress higher than a preset threshold; S33 Perform nonlinear fitting between the candidate band reflectance data and salinity gradient level parameters to determine the model coupling relationship and fitting coefficients; S34 Verify the accuracy of the constructed salinity gradient maize salt stress response model using cross-validation, remove abnormal data, and optimize model parameters.

[0012] Further, step S4 includes the following sub-steps: S41 performing temporal alignment on vegetation indices and texture feature parameters collected at different growth stages, and constructing an original spectral sequence dataset according to the growth stage order; S42 using the sliding window method to smooth the original spectral sequence dataset and eliminate the interference of random noise on the temporal features; S43 extracting the rate of change parameter and trend feature parameter of the spectral sequence through temporal difference operation, and constructing a dynamic feature subset; S44 fusing the original spectral sequence data with the dynamic feature subset to form a complete maize growth dynamic spectral sequence dataset.

[0013] Further, S5 includes the following sub-steps: S51 Perform feature standardization processing on the calibrated spectral sensitive band data and the growth dynamic spectral sequence dataset, and divide the dataset into training set and validation set; S52 Set the hyperparameters of the maize spectral response random forest mechanism model, including the number of decision trees, maximum depth, and feature sampling rate; S53 Input the training set into the model for iterative training, estimate feature importance through out-of-bag data, and remove redundant features; S54 Use the validation set to evaluate the performance of the trained model and adjust the hyperparameters to optimize the model's generalization ability.

[0014] A method for monitoring maize growth and salt stress in saline-alkali land based on UAV multispectral imaging is implemented through different units, including: a real-time multispectral data receiving and storage unit, used to receive multispectral image data and flight attitude data transmitted by UAVs, and to classify, store, and index the data; an image correction and feature extraction unit, connected to the real-time multispectral data receiving and storage unit, to perform target correction, geometric registration, and extraction of spectral reflectance, vegetation index, and texture feature parameters; a model parameter configuration and calculation unit, which establishes data interaction with the image correction and feature extraction unit, the saline-alkali gradient maize salt stress response model, the growth dynamic spectral sequence analysis algorithm, and the maize spectral response random forest mechanism model, respectively, to complete model parameter settings and data calculation; a time-series data processing and fusion unit, connected to the model parameter configuration and calculation unit, to perform time-series correlation and fusion processing on data from different growth stages; a monitoring result output and visualization unit, which receives the output data from the model parameter configuration and calculation unit and generates visualization results of growth level, salt stress degree, and spatial distribution characteristics; and a data interaction and update unit, connected to each unit, to perform external data interaction, online updating of model parameters, and historical tracing of monitoring data.

[0015] Beneficial Effects: This invention proposes a method for monitoring maize growth and salt stress in saline-alkali land based on UAV multispectral data. By collecting multispectral data of maize canopy at different growth stages through low-altitude UAV flights, and combining this with a dedicated platform for data correction and feature extraction, the limitations of traditional manual sampling are avoided. This enables large-scale, efficient dynamic monitoring. A salt-alkali gradient salt stress response model is constructed to determine key sensitive bands. A growth dynamic spectral sequence analysis algorithm is used for time-series correlation processing, followed by feature fusion and optimization training using a random forest mechanism model, achieving accurate differentiation and coupled analysis of the two. Furthermore, relying on a complete platform system including data reception, feature extraction, model computation, time-series fusion, result visualization, and data interaction, a complete technical chain from data acquisition to result output is formed. This improves the data correction and time-series processing stages, addressing the lack of systematic integration in existing technologies. Ultimately, it accurately outputs growth level, salt stress degree, and spatial distribution parameters, providing reliable technical support for the precision management of maize cultivation in saline-alkali land. Attached Figure Description

[0016] Figure 1 is a flowchart of the overall steps of the method of the present invention; Figure 2 is a flowchart of step S3 of the method of the present invention; Figure 3 is a flowchart of step S4 of the method of the present invention; Figure 4 is a flowchart of step S5 of the method of the present invention; Figure 5 is a diagram of the unit composition of the method implementation of the present invention. Detailed Implementation

[0017] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. The application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0018] As shown in Figure 1, the method for monitoring maize growth and salt stress in saline-alkali land based on UAV multispectral imaging includes the following steps: S1, using a UAV equipped with a multispectral imaging module to fly at low altitude along a preset route at different growth stages of maize in saline-alkali land, simultaneously collecting multispectral image data of the maize canopy and flight attitude data; specifically, in step S1, the data acquisition task is performed by the UAV equipped with a multispectral imaging module. The UAV is a quadcopter model, equipped with a multispectral camera including four core bands: blue, green, red, and near-infrared. The center wavelengths of each band are set to 450±10nm, 550±10nm, 650±10nm, and 850±10nm, respectively, with an imaging resolution of no less than 10 million pixels. Before flight, the route is planned according to the terrain characteristics of the saline-alkali land and the distribution range of maize planting, adopting a grid-like coverage pattern, with a route spacing of 50 meters, and the flight altitude controlled at 100-120 meters to ensure that the overlap rate of adjacent images reaches more than 60%. During flight, the UAV uses GPS and inertial navigation systems for real-time positioning with an accuracy better than 1 meter. It simultaneously records attitude data such as flight speed, pitch angle, roll angle, and yaw angle, with an attitude data sampling frequency of 10Hz. Data collection is conducted on clear, cloudless days from 10:00 AM to 2:00 PM to avoid backlighting and shadows. Data is collected at five key growth stages of maize: seedling stage, jointing stage, tasseling stage, grain-filling stage, and maturity stage. Each growth stage is collected at least three times, with a seven-day interval between each collection, ensuring the acquisition of complete spectral data sequences for the entire growth period, providing continuous and comprehensive basic data support for subsequent analysis.

[0019] S2, based on the UAV agricultural spectral data application platform, performs target correction and geometric registration on the collected multispectral image data, extracting maize canopy spectral reflectance, vegetation index, and texture feature parameters. Specifically, step S2 utilizes the UAV agricultural spectral data application platform for data preprocessing. The platform is equipped with a high-performance processor and dedicated spectral analysis software. First, target correction is performed using a diffuse reflectance standard white board with a reflectance set to 99%. By collecting spectral data of the white board under the same lighting conditions, the influence of light intensity variations on the image is eliminated. Geometric registration employs a ground control point-assisted method, with more than 20 ground control points evenly distributed in the saline-alkali land. The control points are positioned using GPS with a coordinate accuracy better than 0.5 meters. Spatial correction of the multispectral image is performed using the control points to ensure that the spatial resolution of the image remains between 0.1 and 0.2 meters. Subsequently, the platform's built-in algorithm was used to extract the spectral reflectance of the maize canopy, and statistical parameters such as the mean, standard deviation, maximum, and minimum reflectance values ​​for each band were calculated. At the same time, more than 10 core vegetation indices, such as the normalized vegetation index, enhanced vegetation index, and ratio vegetation index, were calculated. Texture feature parameters such as contrast, correlation, entropy, and evenness of the image were also extracted. All parameters were calculated pixel by pixel to ensure the precision of the data. The processed parameter data was stored in matrix form in the common TIFF format, which facilitates subsequent model calls and analysis, and provides standardized and highly reliable feature data for subsequent model construction.

[0020] S3 involves constructing a salt-alkali gradient maize salt stress response model. This involves coupling the spectral reflectance with the salt-alkali gradient data to determine the spectrally sensitive bands for salt stress calibration. Specifically, step S3 focuses on constructing the salt-alkali gradient maize salt stress response model. First, five salt-alkali gradient levels are defined, each based on soil salinity correlation parameters. The gradient interval is set to an equal interval, covering the range from mild to severe salt stress. The spectral reflectance of each band extracted in step S2 is coupled with the corresponding salt-alkali gradient level data. Correlation analysis and significance tests are used to screen spectral bands closely associated with salt stress. During the analysis, a significance level of 0.05 is set, retaining bands with an absolute correlation coefficient greater than 0.7 as candidate sensitive bands. Principal component analysis, factor analysis, and other multivariate statistical methods are used to reduce the dimensionality of the candidate bands, eliminating redundant bands, ultimately determining 3-5 key spectrally sensitive bands for salt stress. The model construction process employs a nonlinear fitting method combined with least squares for parameter estimation. The model fitting coefficients are determined through iterative calculations, with the number of iterations controlled to be no less than 100 to ensure a goodness-of-fit R² greater than 0.85. This step, by accurately locating key spectral sensitive bands of salt stress, effectively eliminates interference from non-salt stress factors on the spectral signal, providing core technical support for subsequent salt stress degree inversion and improving the targeting and accuracy of monitoring.

[0021] S4 employs a growth dynamic spectral sequence analysis algorithm to perform temporal correlation processing on vegetation indices and texture feature parameters at different growth stages, constructing a maize growth dynamic spectral sequence dataset. Specifically, step S4 uses the growth dynamic spectral sequence analysis algorithm to process the time-series data. First, the vegetation indices and texture feature parameters for the five growth stages of maize are temporally aligned. Using the number of days in the growth period as the time axis, the parameters are arranged in chronological order to construct the original spectral sequence dataset. The dataset spans the entire growth period of maize, with no fewer than 15 data sets. A sliding window method is used to smooth the original sequence, with the window size set to 3-5 time points. Random noise is eliminated by weighted averaging of the data within the window, and the weight coefficients are linearly distributed according to their distance from the center time point. The rate of change of parameters between adjacent growth stages is calculated through temporal difference operations, extracting trend feature parameters such as growth rate and decay rate to construct a dynamic feature subset. The dynamic feature subset includes indicators such as parameter change rate, cumulative change, and peak occurrence time. The original spectral sequence data and dynamic feature subsets are fused along the time dimension, and a complete dynamic spectral sequence dataset of maize growth is formed by data splicing. During the fusion process, the data dimensions are kept consistent, and the data are normalized to be of the same magnitude. This dataset can fully represent the dynamic changes of maize growth over time, providing comprehensive data support including time-series information for subsequent model training, and improving the dynamism and continuity of growth monitoring.

[0022] S5 involves inputting the calibrated spectral sensitive band data and the growth dynamic spectral sequence dataset into the maize spectral response random forest mechanism model for feature dimension optimization and model training. Specifically, step S5 trains and optimizes the maize spectral response random forest mechanism model. First, the key spectral sensitive band data determined in step S3 and the growth dynamic spectral sequence dataset constructed in step S4 are processed by feature standardization, transforming the data to the [0,1] interval. The training set and validation set are randomly divided in a 7:3 ratio, ensuring that the distribution ratio of each salinity gradient level data is consistent between the training and validation sets. Model hyperparameters are set, including 100-200 decision trees, a maximum decision tree depth of 10-15 layers, a feature sampling rate of 0.6-0.8, and a minimum number of leaf node samples of 5-10. The training set is input into the model for iterative training. During training, out-of-bag data is used to estimate the importance of each feature, and the bottom 20% of redundant features are removed based on importance to optimize feature dimensions and reduce model complexity. The trained model is evaluated using a validation set. The model performance is measured by metrics such as confusion matrix, accuracy, and recall. If the model accuracy is below 0.8, the hyperparameters are adjusted and the model is retrained. The optimization is iterated repeatedly until the model performance is stable. This step fully explores the intrinsic relationship between spectral data and maize growth and salt stress through feature dimension optimization and model training, improves the model's generalization ability and prediction accuracy, and provides reliable model support for the final monitoring results output.

[0023] S6 outputs parameters such as maize growth level, salt stress degree, and spatial distribution characteristics in saline-alkali land through the trained maize spectral response random forest mechanism model.

[0024] Specifically, step S6 outputs monitoring results through a trained maize spectral response random forest mechanism model. The model receives input features such as key spectral sensitive band data and growth dynamic spectral sequence data. After voting and ensemble calculation through an internal decision tree, it outputs maize growth level parameters, salt stress degree parameters, and spatial distribution characteristic parameters in saline-alkali land. The growth level parameter is divided into four levels: excellent, good, medium, and poor, based on a comprehensive judgment of vegetation index and biomass-related parameters. The level classification threshold is automatically optimized and determined during model training. The salt stress degree parameter is divided into four levels: no stress, mild stress, moderate stress, and severe stress, based on a comprehensive evaluation of key spectral sensitive band reflectance and salt stress response values. The spatial distribution characteristic parameters include spatial distribution maps of growth level, salt stress degree, and parameter variation coefficients. A continuous spatial distribution map is generated through a spatial interpolation algorithm, with the interpolation resolution consistent with the original image. The output results are presented in two forms: digital files and visual maps. The digital files include specific parameter values ​​for each monitoring unit, while the visual maps use a hierarchical coloring method to display spatial distribution characteristics, with a color gradient of 5-7 levels to ensure visual differentiation. This step enables quantitative and spatial monitoring of maize growth and salt stress in saline-alkali land. The output results can provide agricultural production managers with precise decision-making basis, guiding the implementation of precise water and fertilizer management and salt stress control measures for maize cultivation in saline-alkali land.

[0025] Preferably, the expression for the salt-alkali gradient maize salt stress response model is as follows: ,in, This represents the salt stress response value. These are the model fitting coefficients. These are the spectral reflectances in the near-infrared, red, blue, and green bands, respectively. For parameters related to the salinity content of saline-alkali land, Normalized Difference Vegetation Index (NDVI) To calibrate the wavelength parameters of the spectral sensitive band, For parameters of the salinity gradient, The results are principal component analysis of reflectance in the blue and green bands.

[0026] Specifically, the saline-alkali gradient maize salt stress response model is a quantitative model constructing the coupling relationship between salt stress and spectral signals in maize in saline-alkali land. It achieves accurate characterization of salt stress response values ​​through multi-dimensional parameter collaborative computation. The model fitting coefficients were determined through nonlinear regression iteration optimization using 30 sets of measured data covering mild to severe salt stress, with values ​​controlled within the ranges of 0.8-1.2, 0.3-0.7, 1.5-2.5, and 0.1-0.3 respectively, ensuring adaptability to different saline-alkali environments. The spectral reflectance of each band input to the model comes from UAV multispectral imagery after target correction. The salinity content correlation parameters of saline-alkali land are obtained through correlation modeling and transformation between soil sampling analysis data and spectral data. The key spectral sensitive band wavelength parameters focus on 3-5 core bands selected through sensitivity analysis within the 450-850nm range. The saline-alkali gradient level parameters are divided into 5 gradient intervals according to soil salinity content and quantified. Principal component analysis of reflectance in the blue and green bands reduces data dimensionality by extracting the first two principal components, eliminating redundant information between bands and reducing model computational complexity. This model, through multi-parameter coupling and nonlinear computation, effectively isolates the interference of non-salt stress factors such as light intensity and growth period on the spectral signal, accurately characterizing the intrinsic correlation between saline-alkali gradient and maize's salt stress response. This provides core technical support for the quantitative inversion of salt stress levels, significantly improving the targeting and accuracy of salt stress monitoring, and providing reliable quantitative data related to salt stress for subsequent model training.

[0027] Preferably, the expression for the growth dynamic spectral sequence analysis algorithm is: ,in, The value of the growth dynamic sequence at time t. For time series data length, Adjusting parameters for the algorithm, for Real-time vegetation index Let be the texture feature parameters at time t. As the baseline reproductive period, Let be the canopy texture feature parameters at time t.

[0028] Specifically, the growth dynamic spectral sequence analysis algorithm is used to uncover the temporal patterns of maize growth changes throughout its growth stages. It achieves precise characterization of growth dynamics by integrating spectral feature parameters from multiple growth stages. The length of the time-series data in the algorithm is determined based on the number of monitoring sessions throughout the entire maize growth period, set to 5-8 sessions, covering key nodes from seedling to maturity. Adjustment parameters are determined through time-series data fitting and optimization. One adjustment parameter is set between 0.1 and 0.3 based on the growth stage redistribution, another parameter ranges from 0.6 to 0.9, and a third parameter is controlled between 1.2 and 1.8 to ensure smooth transitions and highlight dynamic characteristics in the time-series data. Vegetation indices at different times include more than 10 core indices such as the Normalized Difference Vegetation Index (NDVI) and the Enhanced Vegetation Index (EVI). Texture feature parameters include four basic parameters such as contrast and correlation. The baseline growth stage is selected as the maize jointing stage to ensure the consistency of the time-series analysis benchmark. The algorithm constructs a complete growth dynamic spectral sequence dataset by weighting, temporal differencing, and exponential smoothing of parameters at each growth stage, and by fusing the original spectral data with dynamic change characteristics. This algorithm can effectively capture the gradual change pattern and abrupt change characteristics of corn growth over time, eliminate the random errors of data at a single time point, improve the dynamism and continuity of growth monitoring, provide comprehensive feature data including time-series dimensions for subsequent model input, and ensure the accuracy of growth level assessment.

[0029] Preferably, the expression for the maize spectral response random forest mechanism model is: ,in, To monitor the output composite value, For the number of decision trees, The number of feature dimensions. Let be the weight of the j-th feature in the k-th decision tree. For the bias term of the k-th decision tree, This represents the training output of the random forest model for the j-th feature. This is an index showing the correlation between plant growth and salt stress. For model gain parameters, Let j be the j-th input feature parameter.

[0030] Specifically, the maize spectral response random forest mechanism model is a monitoring model built on ensemble learning theory. It achieves accurate output of growth and salt stress parameters through multi-decision tree collaborative computation. The number of decision trees in the model is set to 100-200, and the number of feature dimensions is determined to be 15-25 based on the input parameter selection results, ensuring a balance between model computational efficiency and prediction accuracy. The weights of the features corresponding to each decision tree are determined through out-of-bag data estimation, ranging from 0 to 1. The bias term is optimized through fitting the training set data to a value between -0.5 and 0.5. The correlation index between growth and salt stress is determined to be between 0.6 and 0.9 through correlation analysis. The model gain parameter is set between 1.0 and 2.0 to strengthen the influence weight of key features. During model training, the input features are standardized and divided into training and validation sets in a 7:3 ratio. The splitting rules of each decision tree are optimized through iterative training. The importance of features is evaluated using out-of-bag data, and the bottom 20% of redundant features are removed. This model effectively reduces the risk of overfitting of a single model through multi-decision tree ensemble learning, fully explores the complex nonlinear relationship between spectral features and maize growth and salt stress, improves the model's generalization ability and prediction accuracy, and can simultaneously output quantitative parameters of growth level and salt stress degree, providing core algorithmic support for the quantification of monitoring results.

[0031] Preferably, the data processing model expression of the UAV agricultural spectral data application platform is as follows: ,in, For the processed standard data, This is the target temperature correction factor. For image geometric registration function, Let Variance be the spectral reflectance. This is the error adjustment factor. To acquire image quality parameters, The correlation coefficient between multispectral data and standard reflectance. This is flight attitude data.

[0032] Specifically, the data processing model of the UAV agricultural spectral data application platform is a standardized processing model built specifically for the characteristics of UAV multispectral data, realizing the transformation of raw data into effective feature parameters. The target temperature correction coefficient in the model is determined using measured data from a diffuse reflectance standard whiteboard, with a value range between 0.95 and 1.05, ensuring the correction effect for changes in light and temperature. The image geometric registration function is constructed using ground control point data, combined with UAV flight attitude data for spatial coordinate correction. The ground control point density is 4-6 points per hectare, with a positioning accuracy better than 0.5 meters. The spectral reflectance variance is obtained through statistical calculation of pixel reflectance in each band, with an error adjustment factor set between 0.1 and 0.3 to balance the weight of data noise and effective signal. Image quality parameters are quantitatively evaluated using indicators such as image sharpness and signal-to-noise ratio, with values ​​ranging from 0 to 1. The correlation coefficient between multispectral data and standard reflectance is obtained through comparison with a standard spectral library, ensuring data accuracy. This model effectively eliminates the interference of external factors such as flight attitude and lighting conditions on the original data through multi-step data correction and optimization. It transforms the multispectral image data collected by UAVs into standardized characteristic parameters such as spectral reflectance and vegetation index, providing high-quality data support for subsequent model construction and analysis, and ensuring the reliability of the entire monitoring process.

[0033] Preferably, the coupling model expression for maize growth and salt stress monitoring parameters in saline-alkali land is as follows: ,in, For parameter coupling values, The coupling coefficient is... For growth level parameters, This is a parameter representing the degree of salt stress. The standard deviation of the spatial distribution. To optimize parameters for features, For logistic regression function, This is the normalization function.

[0034] Specifically, the coupled model of maize growth and salt stress monitoring parameters in saline-alkali land is a fusion model used to integrate multi-dimensional monitoring parameters to achieve a comprehensive representation of growth and salt stress information. The coupling coefficients in the model are determined through fitting and optimization of multiple sets of measured data, with values ​​controlled within the ranges of 0.7-1.3, 0.4-0.8, 1.2-1.8, and 0.2-0.5 respectively, ensuring a reasonable weight allocation for each parameter. Growth level parameters are divided into four levels based on vegetation index and biomass-related indicators and quantified. Salt stress severity parameters are divided into four levels based on salt stress response values. The spatial distribution standard deviation is obtained through spatial statistical calculation of parameters within the monitoring area. Feature optimization parameters are determined to be between 0.6 and 0.9 through feature importance assessment. A normalization function transforms the input data to the 0-1 interval to ensure data consistency. The model strengthens the nonlinear correlation between growth and salt stress parameters through a logistic regression function, and achieves deep fusion of monitoring information through multi-parameter collaborative computation. This model can effectively integrate multi-dimensional parameters such as growth level, salt stress degree, and spatial distribution characteristics, eliminating the limitations of single parameter monitoring, realizing a comprehensive assessment of maize growth status in saline-alkali land, and providing quantitative basis for the comprehensive judgment of subsequent monitoring results with the output parameter coupling value, thereby improving the completeness and practicality of monitoring results and providing more comprehensive technical support for agricultural production management decisions.

[0035] Preferably, as shown in Figure 2, step S3 includes the following sub-steps: S31 Select experimental plots with significant differences in salinity gradient, collect multispectral reflectance data of maize canopy and soil salinity content correlation parameters under different gradients, and establish a dataset corresponding to salinity gradient and spectral reflectance; S32 Screen the multispectral reflectance data using band sensitivity analysis, and retain candidate bands whose correlation with salt stress is higher than a preset threshold; S33 Perform nonlinear fitting between the candidate band reflectance data and salinity gradient level parameters to determine the model coupling relationship and fitting coefficients; S34 Verify the accuracy of the constructed salinity gradient maize salt stress response model using cross-validation, remove abnormal data, and optimize model parameters.

[0036] Specifically, step S3 involves four sub-steps to fully construct and optimize the maize salt stress response model based on the salinity gradient. S31 selects five experimental plots with significant differences in salinity gradients, each plot measuring 100 square meters. These plots are divided into five gradient levels based on soil salinity content. Twenty sampling points are evenly distributed within each plot, simultaneously collecting multispectral reflectance data of the maize canopy and soil salinity correlation parameters, establishing a dataset containing 300 sets of data corresponding to the salinity gradient and spectral reflectance. S32 uses band sensitivity analysis to analyze all bands of the multispectral imagery, setting a correlation threshold of 0.7. Bands with a correlation to salt stress higher than this threshold are retained as candidate sensitive bands, typically selecting 8-12. S33 uses a nonlinear fitting method to couple the candidate band reflectance data with the salinity gradient level parameters. Through more than 100 iterations, the model coupling relationship and fitting coefficients are determined, ensuring a goodness of fit of no less than 0.85. S34 employs a 5-fold cross-validation method to verify the accuracy of the constructed model. The dataset is divided into a validation set at a ratio of 20%, and outliers with errors exceeding twice the mean are removed. The model parameters are then re-optimized, improving the model validation accuracy to above 0.9. This step, through systematic selection of sample plots, band screening, model fitting, and validation optimization, ensures that the constructed salt-alkali gradient maize salt stress response model can accurately capture the intrinsic correlation between salt stress and spectral signals, providing reliable technical support for the subsequent determination of key spectrally sensitive bands.

[0037] Preferably, as shown in Figure 3, step S4 includes the following sub-steps: S41, the vegetation index and texture feature parameters collected at different growth stages are time-series aligned to construct the original spectral sequence dataset in the order of growth stages; S42, the original spectral sequence dataset is smoothed using the sliding window method to eliminate the interference of random noise on the time-series features; S43, the rate of change parameters and trend feature parameters of the spectral sequence are extracted through time-series difference operations to construct a dynamic feature subset; S44, the original spectral sequence data and the dynamic feature subset are fused to form a complete dynamic spectral sequence dataset of maize growth.

[0038] Specifically, step S4 constructs the dynamic spectral sequence dataset of maize growth through four sub-steps. S41 first aligns the vegetation indices and texture parameters collected during the five key growth stages of maize—seedling, jointing, tasseling, grain-filling, and maturity—time-series. Using the number of days in the growth period as the time axis, the data for each parameter are organized chronologically to construct an original spectral sequence dataset containing 15 sets of data, ensuring uniform time intervals. S42 uses a sliding window method to smooth the original spectral sequence dataset, setting the window size to three time points. Random noise interference is eliminated through weighted averaging of the data within the window. The weight coefficients are set to 0.2, 0.6, and 0.2 based on their distance from the center time point, ensuring that the smoothed data retains the original trend characteristics. S43 calculates the rate of change of parameters between adjacent growth stages using temporal difference operations, extracting six trend characteristic parameters, including growth rate, decay rate, and peak occurrence time, to construct a dynamic feature subset. This subset maintains the same dimensionality as the original data. S44 uses a data stitching method to fuse the original spectral sequence data with a dynamic feature subset, aligning it in both the time and feature dimensions to form a complete dynamic spectral sequence dataset of maize growth including 30 features. This step, through time alignment, smoothing and denoising, dynamic feature extraction and data fusion, effectively captures the dynamic changes in maize growth with the growth stage, providing comprehensive feature data including time-series information for subsequent model training, and improving the continuity and accuracy of growth monitoring.

[0039] Preferably, as shown in Figure 4, step S5 includes the following sub-steps: S51 Perform feature standardization processing on the calibrated spectral sensitive band data and the growth dynamic spectral sequence dataset, and divide the dataset into training and validation sets; S52 Set the hyperparameters of the maize spectral response random forest mechanism model, including the number of decision trees, maximum depth, and feature sampling rate; S53 Input the training set into the model for iterative training, estimate feature importance through out-of-bag data, and remove redundant features; S54 Use the validation set to evaluate the performance of the trained model and adjust the hyperparameters to optimize the model's generalization ability.

[0040] Specifically, step S5 completes the feature optimization and training of the maize spectral response random forest mechanism model through four sub-steps. S51 first performs feature standardization on key spectral sensitive band data and the growth dynamic spectral sequence dataset. The max-min standardization method is used to transform all data into the 0-1 interval. The training and validation sets are randomly divided in a 7:3 ratio, ensuring that the salinity gradient levels and growth stage data are evenly distributed between the two sets to avoid data bias. S52 sets the model hyperparameters, including 150 decision trees, a maximum decision tree depth of 12 layers, a feature sampling rate of 0.7, and a minimum number of leaf nodes of 8, ensuring the model balances computational efficiency and prediction accuracy. S53 inputs the training set into the model for iterative training, setting the number of iterations to 200. Model performance is recorded every 50 iterations. The importance of each feature is estimated using out-of-bag data, and the bottom 20% of redundant features are removed based on importance, optimizing the feature dimension from 30 to 24 dimensions to reduce model complexity. S54 uses a validation set to evaluate the performance of the trained model, employing accuracy, recall, and F1 score as three metrics to measure model performance. If the accuracy is below 0.85, the hyperparameters are adjusted and the model is retrained. This iterative optimization continues until all three metrics are stable above 0.9. This step, through standardization, hyperparameter setting, iterative training, and performance evaluation, fully explores the complex correlation between spectral features and maize growth and salt stress, improving the model's generalization ability and prediction accuracy, and providing reliable model support for the final monitoring results output.

[0041] As shown in Figure 5, the method for monitoring maize growth and salt stress in saline-alkali land based on UAV multispectral imaging is characterized by being implemented through different units, including: a multispectral data real-time receiving and storage unit, used to receive multispectral image data and flight attitude data transmitted by the UAV, and to classify, store, and index the data; an image correction and feature extraction unit, connected to the multispectral data real-time receiving and storage unit, for target correction, geometric registration, and extraction of spectral reflectance, vegetation index, and texture feature parameters; and a model parameter configuration and calculation unit, connected to the image correction and feature extraction unit and the saline-alkali gradient maize salt stress unit. The system establishes data interaction among the response model, the growth dynamic spectral sequence analysis algorithm, and the maize spectral response random forest mechanism model to complete model parameter settings and data calculations. A time-series data processing and fusion unit, connected to the model parameter configuration and calculation unit, performs time-series correlation and fusion processing on data from different growth stages. A monitoring result output and visualization unit receives the output data from the model parameter configuration and calculation unit and generates visualization results of growth level, salt stress degree, and spatial distribution characteristics. A data interaction and update unit, connected to each unit, performs external data interaction, online updates of model parameters, and historical tracking of monitoring data.

[0042] A method for monitoring maize growth and salt stress in saline-alkali land based on UAV multispectral technology can rapidly cover large areas of saline-alkali land at different growth stages of maize, efficiently collecting canopy spectra and related data. Compared with traditional manual sampling methods, this method significantly improves monitoring efficiency and spatial coverage integrity. By constructing a dedicated salt-alkali gradient salt stress response model, a growth dynamic spectral sequence analysis algorithm, and a maize spectral response random forest mechanism model, deep coupling of spectral data with salt-alkali gradient and growth temporal characteristics is achieved, enhancing the correlation and accuracy of monitoring parameters. With the help of a fully functional UAV agricultural spectral data application platform, the entire process from data correction and feature extraction to model calculation and result output is standardized, ensuring the systematic nature and stability of the monitoring process.

[0043] This method uses a salt-alkali gradient salt stress response model to screen key sensitive bands, removes irrelevant spectral interference, and then uses a growth dynamic spectral sequence analysis algorithm to perform time-series correlation processing on data from different growth stages to clarify the dynamic change pattern of growth. Finally, it uses a random forest mechanism model to achieve accurate differentiation and fusion inversion of the characteristics of both. Addressing the lack of systematic integration in existing technologies, the UAV agricultural spectral data application platform it constructs integrates functional modules such as data reception and storage, correction and extraction, model calculation, time-series fusion, result visualization, and data interaction and updating, forming a complete technology chain. At the same time, through a multi-step data processing and model optimization process, it improves the data correction and time-series correlation links, realizing efficient collaboration throughout the entire process from data acquisition to monitoring result output, and significantly improving the accuracy and spatial distribution characterization ability of growth and salt stress monitoring.

[0044] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "set," "install," "connect," "link," and "fix" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal communication between two components. Those skilled in the art will understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0045] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various equivalent changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for monitoring maize growth and salt stress in saline-alkali land based on UAV multispectral imaging, characterized in that, Includes the following steps: S1. Using a drone equipped with a multispectral imaging module, corn in saline-alkali land undergoes low-altitude flight along a preset route at different growth stages, simultaneously collecting multispectral image data of the corn canopy and flight attitude data. S2. Based on a drone-based agricultural spectral data application platform, target correction and geometric registration are performed on the collected multispectral image data, extracting corn canopy spectral reflectance, vegetation index, and texture feature parameters. S3. A saline-alkali gradient corn salt stress response model is constructed, and the spectral reflectance and saline-alkali gradient data are coupled for analysis to determine the salt stress calibration spectral sensitive bands. S4. A growth dynamic spectral sequence analysis algorithm is used to perform temporal correlation processing on vegetation index and texture feature parameters at different growth stages, constructing a corn growth dynamic spectral sequence dataset. S5. The calibration spectral sensitive band data and the growth dynamic spectral sequence dataset are input into a corn spectral response random forest mechanism model for feature dimension optimization and model training. S6. The trained corn spectral response random forest mechanism model outputs saline-alkali land corn growth level parameters, salt stress degree parameters, and spatial distribution feature parameters.

2. The method for monitoring maize growth and salt stress in saline-alkali land based on UAV multispectral imaging according to claim 1, characterized in that, The expression for the maize salt stress response model based on the salinity gradient is as follows: ,in, This represents the salt stress response value. These are the model fitting coefficients. These are the spectral reflectances in the near-infrared, red, blue, and green bands, respectively. For parameters related to the salinity content of saline-alkali land, Normalized Difference Vegetation Index (NDVI) To calibrate the wavelength parameters of the spectral sensitive band, For parameters of the salinity gradient, The results are principal component analysis of reflectance in the blue and green bands.

3. The method for monitoring maize growth and salt stress in saline-alkali land based on UAV multispectral imaging according to claim 1, characterized in that, The algorithm expression for the dynamic spectral sequence analysis of the growth potential is as follows: ,in, The value of the growth dynamic sequence at time t. The length of the time series data. Adjusting parameters for the algorithm, for Real-time vegetation index Let be the texture feature parameters at time t. As the baseline reproductive period, Let be the canopy texture feature parameters at time t.

4. The method for monitoring maize growth and salt stress in saline-alkali land based on UAV multispectral imaging according to claim 1, characterized in that, The expression for the maize spectral response random forest mechanism model is as follows: ,in, To monitor the output composite value, For the number of decision trees, The number of feature dimensions. Let be the weight of the j-th feature in the k-th decision tree. For the bias term of the k-th decision tree, This represents the training output of the random forest model for the j-th feature. This is an index showing the correlation between plant growth and salt stress. For model gain parameters, Let j be the j-th input feature parameter.

5. The method for monitoring maize growth and salt stress in saline-alkali land based on UAV multispectral imaging according to claim 1, characterized in that, The data processing model expression of the UAV agricultural spectral data application platform is as follows: ,in, For the processed standard data, This is the target temperature correction factor. For image geometric registration function, Let Variance be the spectral reflectance. This is the error adjustment factor. To acquire image quality parameters, The correlation coefficient between multispectral data and standard reflectance. This is flight attitude data.

6. The method for monitoring maize growth and salt stress in saline-alkali land based on UAV multispectral imaging according to claim 1, characterized in that, The coupling model expression between maize growth and salt stress monitoring parameters in saline-alkali land is as follows: ,in, For parameter coupling values, The coupling coefficient is... For growth level parameters, This is a parameter representing the degree of salt stress. The standard deviation of the spatial distribution. To optimize parameters for features, For logistic regression function, This is the normalization function.

7. The method for monitoring maize growth and salt stress in saline-alkali land based on UAV multispectral imaging according to claim 1, characterized in that, S3 includes the following steps: S31, selecting experimental plots with significant differences in salinity gradient, collecting multispectral reflectance data of maize canopy and soil salinity content correlation parameters under different gradients, and establishing a dataset corresponding to salinity gradient and spectral reflectance; S32, screening multispectral reflectance data using band sensitivity analysis, retaining candidate bands with a correlation to salt stress higher than a preset threshold; S33, performing nonlinear fitting between candidate band reflectance data and salinity gradient level parameters to determine the model coupling relationship and fitting coefficients; S34, verifying the accuracy of the constructed salinity gradient maize salt stress response model using cross-validation, removing outlier data and optimizing model parameters.

8. The method for monitoring maize growth and salt stress in saline-alkali land based on UAV multispectral imaging according to claim 1, characterized in that, S4 includes the following sub-steps: S41, performing temporal alignment on vegetation indices and texture feature parameters collected at different growth stages, and constructing the original spectral sequence dataset in the order of growth stages; S42, using the sliding window method to smooth the original spectral sequence dataset and eliminate the interference of random noise on the temporal features. S43, extract the rate of change parameter and trend feature parameter of the spectral sequence through time-series difference operation, and construct a dynamic feature subset; S44, fuse the original spectral sequence data with the dynamic feature subset to form a complete maize growth dynamic spectral sequence dataset.

9. The method for monitoring maize growth and salt stress in saline-alkali land based on UAV multispectral imaging according to claim 1, characterized in that, S5 includes the following steps: S51, performing feature standardization on the calibrated spectral sensitive band data and the growth dynamic spectral sequence dataset, and dividing the dataset into training and validation sets; S52, setting the decision tree number, maximum depth, and feature sampling rate hyperparameters of the maize spectral response random forest mechanism model; S53, inputting the training set into the model for iterative training, estimating feature importance through out-of-bag data, and removing redundant features. S54 uses the validation set to evaluate the performance of the trained model and adjusts hyperparameters to optimize the model's generalization ability.

10. The method for monitoring maize growth and salt stress in saline-alkali land based on UAV multispectral imaging according to any one of claims 1-9, characterized in that, This method is implemented through different units, including: a real-time multispectral data receiving and storage unit, used to receive multispectral image data and flight attitude data transmitted by UAVs, and to classify, store, and index the data; an image correction and feature extraction unit, connected to the real-time multispectral data receiving and storage unit, to perform target correction, geometric registration, and extraction of spectral reflectance, vegetation index, and texture feature parameters; a model parameter configuration and computation unit, which establishes data interaction with the image correction and feature extraction unit, the saline-alkali gradient maize salt stress response model, the growth dynamic spectral sequence analysis algorithm, and the maize spectral response random forest mechanism model, respectively, to complete model parameter settings and data computation; a time-series data processing and fusion unit, connected to the model parameter configuration and computation unit, to perform time-series correlation and fusion processing on data from different growth stages; a monitoring result output and visualization unit, which receives the output data from the model parameter configuration and computation unit and generates visualization results of growth level, salt stress degree, and spatial distribution characteristics; and a data interaction and update unit, connected to each unit, to perform external data interaction, online updating of model parameters, and historical tracing of monitoring data.

Images