A rice tillering stage fertilization decision method, system, computer device and medium
By constructing a rice nitrogen concentration dilution curve using UAV multi-source remote sensing technology and optimized algorithms, the problem of low dry matter measurement efficiency was solved, enabling precise fertilization during the rice tillering stage and improving the real-time nature of fertilization and fertilizer utilization.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- SHENYANG AGRI UNIV
- Filing Date
- 2026-04-27
- Publication Date
- 2026-07-21
Smart Images
Figure CN122434673A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of smart agriculture and precision agriculture technology, specifically relating to a method, system, computer equipment, and medium for making decisions on fertilization during the rice tillering stage. Background Technology
[0002] Nitrogen is a core nutrient element that determines the growth, development, and potential productivity of rice. Changes in its content play a crucial regulatory role in photosynthesis, protein synthesis, and carbon and nitrogen metabolism, significantly impacting the final yield of rice. In rice production, the rational application of nitrogen fertilizer, reducing nitrogen application rates, and improving nitrogen fertilizer utilization efficiency have become key objectives of precision agriculture. The tillering stage of rice is the process by which axillary buds on the basal nodes of the main stem germinate and grow into new stems. If these tillers grow and develop normally, they will each produce panicles, flowers, and grains, ultimately constituting the final effective panicle number, which determines the rice yield. Sufficient nitrogen supply promotes the germination of tillering buds and the rapid growth of tillers, thereby increasing the effective panicle number per unit area. Therefore, rapidly and accurately diagnosing the nitrogen requirements of rice fields during the tillering stage and developing precise fertilization strategies based on the diagnostic results has become an important way to achieve refined field management and ensure rice yield.
[0003] Current rice fertilization decision-making methods combine UAV hyperspectral remote sensing technology with critical nitrogen concentration theory for rice nitrogen nutrition diagnosis. Airborne multispectral technology is used to acquire canopy spectral data during key growth stages of rice. Combined with parameters such as leaf nitrogen content and dry matter weight, a cold-region rice nitrogen nutrition diagnostic model based on canopy NDVI is constructed. Furthermore, UAV remote sensing technology is used to acquire rice canopy spectral data, constructing multiple vegetation indices, and combining machine learning algorithms to inversely model rice aboveground biomass, nitrogen uptake, and nitrogen nutrition indices. However, existing critical nitrogen concentration dilution curves are usually based on plant or leaf dry matter accumulation. Dry matter measurement requires field sampling, which is complex and time-consuming, easily missing critical fertilization windows such as the tillering stage, resulting in poor real-time fertilization decisions. Summary of the Invention
[0004] To address the issues of low efficiency in dry matter measurement and poor real-time performance in fertilization decisions, this invention provides a method, system, computer equipment, and medium for fertilization decisions during the rice tillering stage.
[0005] A first aspect of this invention provides a fertilization decision-making method based on the tillering stage of rice, comprising the following steps: The UAV multi-source remote sensing data of rice in the tillering stage was acquired, including visible light imagery and hyperspectral imagery. The leaf area index of the rice canopy in the tillering stage was obtained by inversion based on the visible light imagery, and the plant nitrogen concentration of rice in the tillering stage was obtained by inversion based on the hyperspectral imagery. The relationship between aboveground dry matter and leaf area index (LAI) of rice during the tillering stage was determined using the RiceGrow rice growth model. Based on the critical nitrogen concentration dilution curve describing the relationship between LAI and critical nitrogen concentration in rice, and the inverted plant nitrogen concentration, the nitrogen deficit of rice was calculated according to the relationship between aboveground dry matter and LAI. The critical nitrogen concentration dilution curve describing the relationship between LAI and critical nitrogen concentration in rice was obtained in advance by fitting the leaf area index of experimental rice and the measured nitrogen concentration of rice in multiple growth cycles. Based on the nitrogen deficit, decision-making results are generated to guide variable fertilization during the tillering stage.
[0006] Furthermore, the fertilization decision-making method during the rice tillering stage also includes: Texture features are extracted from visible light images; the texture features are input into an optimized first machine learning model, and the leaf area index is output. The reflectance of the characteristic bands is obtained by selecting the spectral characteristic bands from the hyperspectral image. The reflectance of the characteristic band is input into the optimized second machine learning model, and the nitrogen concentration is output.
[0007] Furthermore, in the rice tillering stage fertilization decision-making method, the first machine learning model is a kernel extreme learning machine (KELM) model that optimizes the kernel parameters and regularization parameters using the Zebra Optimization Algorithm (ZOA); the second machine learning model is a kernel extreme learning machine (KELM) model that optimizes the kernel parameters and regularization parameters using the Dung Beetle Optimization Algorithm (DBO).
[0008] Furthermore, the fertilization decision-making method during the rice tillering stage also includes: Each spectral feature band in the hyperspectral image is projected onto other spectral feature bands, and the wavelength with the largest projection vector is selected as the candidate wavelength. The reflectance of the feature band is then determined based on the candidate wavelength.
[0009] Furthermore, the fertilization decision-making method during the rice tillering stage also includes: For experimental rice under each nitrogen fertilizer gradient condition in multiple growth cycles, the leaf area index of rice with the highest nitrogen fertilizer treatment in each growth cycle was used as a reference. The groups unrestricted by nitrogen nutrition and those restricted by nitrogen nutrition in that growth cycle were obtained through analysis of variance. For experimental rice in the nitrogen-limited group, the linear relationship between leaf area index and measured nitrogen concentration of rice was fitted; for experimental rice in the non-nitrogen-limited group, the mean leaf area index of each growth cycle was taken as the maximum leaf area index of that growth cycle; and the critical nitrogen concentration of each growth cycle was taken as the cutoff point of the linear relationship between leaf area index and measured nitrogen concentration of rice at the maximum leaf area index. Based on the fitting of leaf area index of experimental rice with critical nitrogen concentration at multiple growth stages, a critical nitrogen concentration dilution curve encompassing leaf area index and critical nitrogen concentration of rice at multiple growth stages was obtained.
[0010] Furthermore, in the rice tillering stage fertilization decision-making method, the critical nitrogen concentration dilution curve describing the relationship between leaf area index and critical nitrogen concentration in rice is a power function model with leaf area index as the independent variable and critical nitrogen concentration as the dependent variable. The relationship between aboveground dry matter and leaf area index of rice during the tillering stage is a power function model with aboveground dry matter as the independent variable and leaf area index as the dependent variable.
[0011] Furthermore, the fertilization decision-making methods during the rice tillering stage include: The critical nitrogen accumulation amount was calculated based on the critical nitrogen concentration dilution curve and the leaf area index obtained by inversion. The actual nitrogen accumulation is calculated based on the nitrogen concentration and leaf area index obtained from the inversion. The nitrogen deficit is calculated based on the difference between the critical nitrogen accumulation rate and the actual nitrogen accumulation rate.
[0012] A second aspect of this invention provides a fertilization decision-making system based on the rice tillering stage, comprising: The data inversion module is used to acquire multi-source remote sensing data of rice during the tillering stage from UAVs. The multi-source remote sensing data includes visible light images and hyperspectral images. The leaf area index of the rice canopy during the tillering stage is obtained by inversion based on the visible light images, and the plant nitrogen concentration of rice during the tillering stage is obtained by inversion based on the hyperspectral images. The curve construction module is used to determine the correspondence between aboveground dry matter and leaf area index of rice during the tillering stage using the RiceGrow rice growth model. Based on the critical nitrogen concentration dilution curve describing the relationship between leaf area index and critical nitrogen concentration in rice, and the inverted plant nitrogen concentration, the nitrogen deficit of rice is calculated according to the correspondence between aboveground dry matter and leaf area index. The critical nitrogen concentration dilution curve describing the relationship between leaf area index and critical nitrogen concentration in rice is obtained in advance by fitting the leaf area index of experimental rice and the measured nitrogen concentration of rice in multiple growth cycles. The decision-making module generates decision results based on the nitrogen deficit to guide variable fertilization during the tillering stage.
[0013] A third aspect of the present invention provides a computer device including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the above-described method.
[0014] A fourth aspect of the present invention provides a readable storage medium storing a computer program that, when executed by a processor, implements the steps of the above-described method.
[0015] The rice tillering stage fertilization decision-making method provided by this invention has the following beneficial effects: By constructing a critical nitrogen concentration dilution curve with leaf area index obtained from remote sensing inversion as the independent variable, the entire diagnostic process can be completely freed from destructive sampling, achieving a qualitative leap from theory to application and greatly improving the feasibility of promoting this technology in field practice. Based on this, using the curve and the inverted actual nitrogen concentration, the nitrogen deficit in rice can be quantitatively and accurately calculated, elevating traditional qualitative or semi-qualitative nutrient diagnosis to a quantitative decision-making level that can guide precision fertilization. This solves the key technical bottleneck that existing remote sensing technology is mostly used for growth monitoring and cannot directly support quantitative fertilization decisions. Finally, based on the precise nitrogen deficit, the diagnostic results are directly transformed into executable precision agronomic measures, guiding fertilizer machinery to perform variable-rate operations. While ensuring stable rice yield, this effectively reduces excessive nitrogen fertilizer application, thereby improving fertilizer utilization efficiency. Attached Figure Description
[0016] To more clearly illustrate the embodiments and design schemes of the present invention, the accompanying drawings required for this embodiment will be briefly described below. The drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0017] Figure 1 This is a flowchart illustrating the rice tillering stage fertilization decision-making method provided by the present invention according to an exemplary embodiment; Figure 2 This is a schematic diagram of a hyperspectral curve provided by the present invention according to an exemplary embodiment; Figure 3 The technical route provided by the present invention according to an exemplary embodiment is shown in the figure; Figure 4 This is a texture feature map of an experimental field provided by the present invention according to an exemplary embodiment; Figure 5 This is a schematic diagram of the LAI inversion model results provided by the present invention according to an exemplary embodiment; Figure 6This is a schematic diagram of the spectral characteristic bands provided by the present invention according to an exemplary embodiment; Figure 7 This is a schematic diagram of the selection results of the characteristic bands of the hyperspectral reflectance of rice canopy according to an exemplary embodiment of the present invention; Figure 8 This is a schematic diagram of the modeling results of the hyperspectral rice nitrogen concentration inversion model provided by the present invention according to an exemplary embodiment; Figure 9 This is a schematic diagram of the relationship between the critical nitrogen concentration and the corresponding LAI value according to an exemplary embodiment of the present invention; Figure 10 This is a schematic diagram of the curve equation of aboveground dry matter mass (LAI) of rice according to an exemplary embodiment of the present invention. Figure 11 This is a prescription diagram for drone-based plant protection operations provided by the present invention according to an exemplary embodiment; Figure 12 This is a schematic diagram of the fertilization situation and unit yield of an experimental field according to an exemplary embodiment of the present invention. Detailed Implementation
[0018] To enable those skilled in the art to better understand and implement the technical solutions of the present invention, the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. The following embodiments are only used to more clearly illustrate the technical solutions of the present invention and should not be construed as limiting the scope of protection of the present invention.
[0019] This invention combines the LAI (Laminated Area Intake) and critical nitrogen concentration dilution curve theory. It collects visible light and hyperspectral data of rice canopy using drones, uses a Kernel Extreme Learning Machine (KELM) model optimized by the Zebra Optimization Algorithm (ZOA) to invert the rice LAI, constructs a critical nitrogen concentration dilution curve based on LAI, and calculates the nitrogen deficit in rice by combining the nitrogen concentration inverted by the KELM model optimized by the Dung Beetle Optimizer (DBO) algorithm, thus formulating a precise fertilization plan for the rice tillering stage.
[0020] The technical solutions provided by the various embodiments of the present invention will be described in detail below with reference to the accompanying drawings.
[0021] This invention provides a method for making fertilization decisions during the tillering stage of rice, specifically as follows: Figure 1 As shown, it includes the following steps: S1. Acquire multi-source remote sensing data of the target rice field by UAV, wherein the multi-source remote sensing data includes at least visible light images and hyperspectral images.
[0022] S2. Obtain the leaf area index of the rice canopy based on the visible light image inversion, and obtain the plant nitrogen concentration based on the hyperspectral image inversion.
[0023] S3. Construct dilution curves for critical nitrogen concentration in rice with leaf area index as the independent variable.
[0024] S4. Based on the critical nitrogen concentration dilution curve and the plant nitrogen concentration obtained by inversion, calculate the nitrogen deficit of rice.
[0025] S5. Based on the nitrogen deficit, generate decision results to guide variable fertilization during the tillering stage.
[0026] Based on the above inventive concept, the present invention proposes an embodiment, including the following steps: S1, Data Acquisition.
[0027] S1.1 Acquisition of Visible Light Remote Sensing Imagery of Rice: Visible light remote sensing orthophoto data of the experimental area were acquired using a UAV. For flight parameter settings, the UAV flight altitude was set to 30m, the flight speed to 10m / s, and both the forward and lateral overlap rates were set to 80% to ensure image coverage and data quality. Flight operations were conducted between 10:00 and 14:00 when lighting conditions were favorable to ensure image clarity and consistency. After image acquisition, the acquired images were stitched together, and the output image coordinate system was set to WGS84 to ensure the georeferenced accuracy and global consistency of the data.
[0028] S1.2 Acquisition of Rice Spectral Parameters: The hyperspectral imaging system onboard the UAV covers a spectral range of 400-1000 nm, with a spectral resolution of 3.5 nm and 170 effective bands. Data acquisition was conducted between 10:00 and 14:00 under favorable lighting conditions, with the UAV flying at an altitude of 100 m. Before takeoff, dark current and whiteboard calibration were performed on the hyperspectral imager. Simultaneously, a 1.5 × 1.5 m diffuse reflector with 60% reflectivity was placed within the acquisition area for subsequent reflectivity data correction. The acquired hyperspectral data underwent lens correction, uniformity correction, and reflectivity preprocessing using SpectraVIEW software, ultimately generating a hyperspectral reflectivity image of the rice canopy. Subsequently, ENVI 5.3 + IDL software was used to extract hyperspectral data of the rice plot area from the hyperspectral image. Interference spectra were removed using spectral angle mapping, and the average spectrum of the region of interest was calculated. The average spectrum was first resampled using Matlab software, reducing the sampling frequency to 1 nm. Because hyperspectral data often contains high-frequency noise and minute fluctuations, which may affect the accuracy of the data, spectral smoothing is necessary. To address this, a Savitzky-Golay (SG) convolutional smoothing algorithm based on local polynomial fitting is employed. This method effectively suppresses noise while preserving key spectral features to the greatest extent possible. After SG smoothing, the final hyperspectral data of the experimental area is obtained, as shown in the example hyperspectral curves below. Figure 2 As shown.
[0029] S1.3 Field trials and acquisition of rice agronomic parameters: In this invention, the leaf area index (LAI) of rice canopy in the experimental area was measured using an LAI-2200C plant canopy analyzer. To improve the accuracy and representativeness of the measurement results, the average value of LAI data from three locations within the same experimental field was selected as the measured LAI for that experimental area.
[0030] To determine the nitrogen concentration in the samples, destructive sampling was first performed at each sampling point in the experimental field. To ensure leaf freshness, the rice plants, along with their roots and root soil, were dug up during sampling, placed in resealable bags labeled with the test number and date, and then placed in a low-temperature incubator before being quickly transported back to the laboratory. In the laboratory, the rice samples were cut at the root, and after removing any adhering soil and dust, they were placed in a drying oven at 105°C for 30 minutes to kill the greening, and then dried at 80°C to constant weight. Finally, the dried samples were thoroughly ground into a uniform powder, and the nitrogen concentration of the rice was determined using the Kjeldahl method.
[0031] S2, Inversion Model Construction.
[0032] S2.1 Kernel Extreme Learning Machine Optimization Method Based on Intelligent Optimization Algorithm: To reduce the uncertainty caused by manual parameter selection and improve model inversion accuracy during model parameter optimization, this invention introduces the Zebra Optimization Algorithm (ZOA) and the Dung Beetle Optimizer (DBO) to optimize key parameters of the Kernel Extreme Learning Machine (KELM). Both optimization algorithms use minimizing model prediction error as the objective function, independently searching and updating the regularization parameters and kernel function-related parameters in KELM, thereby constructing ZOA-KELM and DBO-KELM models respectively. DBO simulates various navigation strategies of dung beetles, such as rolling dung balls, using celestial cues like the sun, moon, and polarized light for location, and performing group cooperation. It treats the dung ball's position as a candidate solution, using the global optimum as a reference point for guidance. Combined with random perturbation and neighborhood rolling mechanisms, it enhances population diversity and search capabilities, thus exhibiting stronger performance in high-dimensional and multi-modal optimization. ZOA (Zebra Algorithm for Optimization) focuses on global exploration capabilities and convergence speed. It utilizes the collaborative and dynamic behaviors of zebras, simulating their foraging and defense strategies to achieve position updates and solve the optimization problem. In foraging, zebras update their positions based on their foraging behavior, with the zebra in the best position acting as a vanguard, leading other zebras to better locations. In defense, zebras adapt to different predators, such as using zigzag or random movement patterns to escape lions, or using grouping to confuse or intimidate other predators, demonstrating good adaptability and applicability to dynamic optimization scenarios. The optimization processes of ZOA and DBO are independent, and their results are not merged. Instead, the inversion accuracy of models under different optimization strategies is compared and analyzed under the same dataset and modeling conditions to evaluate the performance improvement of each optimization algorithm and select the better-performing model for subsequent research.
[0033] S2.2 Rice LAI Inversion Model Based on Visible Light Imagery: Preferably, considering that the texture features extracted from UAV visible light remote sensing images can effectively invert the leaf area index (LAI) of rice, the texture features in the red band can accurately delineate the boundaries of rice leaves, effectively distinguishing them from background areas such as water surfaces and soil. Therefore, rice and background can be accurately distinguished simply by the extracted texture features, without the need for further complex image processing steps. Furthermore, since LAI is defined as the total area of leaves per unit area, delineating the leaf outline allows for more accurate calculation of the leaf area. Based on this, this invention extracts texture features from UAV visible light remote sensing images as input data for the LAI inversion model, and uses LAI data acquired by the LAI-2200C plant canopy analyzer as model output data. ENVI 5.3 software was used to process the UAV visible light remote sensing images before tillering fertilizer decisions and before panicle fertilizer decisions, and the images of each experimental field were cropped into separate image files. Subsequently, the Co-occurrence Measures function in the ENVI Filter toolbox was used to extract the texture feature variance of the images. To ensure the extracted texture features have high spatial resolution and good representational ability, the pane size was set to 5. Subsequently, the mean of the variance of the red band texture features for each image was calculated. Finally, a kernel extreme learning machine (KEM) optimized using the dung beetle optimization algorithm and the zebra optimization algorithm was used to construct an LAI inversion model, and the results were compared with the LAI inversion model built using the KEM.
[0034] S2.3, Rice nitrogen concentration inversion model based on hyperspectral imaging: The collected nitrogen concentration data for rice are shown in Table 1, ranging from 1.554% to 4.972%. The data were randomly divided into two groups, with a training set to a test set ratio of 7:3. The average nitrogen concentration in the training set was 3.354%, with a standard deviation of 0.698%; the average nitrogen concentration in the test set was 3.398%, with a standard deviation of 0.734%. The overall nitrogen concentration of the samples showed high discrimination, and the randomized division into two datasets was reasonable and effectively met the requirements for establishing and validating the inversion model.
[0035] Table 1 Statistical results of nitrogen concentration in rice Compared to multispectral data, hyperspectral data has higher spectral resolution and can characterize more information about rice. However, hyperspectral data also contains a large amount of redundant information, which may lead to a decrease in modeling accuracy and efficiency. This invention uses the Successive Projections Algorithm (SPA) to reduce the dimensionality of hyperspectral information in the 400-1000nm range. SPA is a forward feature selection method, mainly used for centralized feature selection in high-dimensional datasets, improving computational efficiency and model performance by reducing the dimensionality of the dataset. The SPA algorithm projects each feature wavelength onto other wavelengths through vector projection analysis, compares the magnitudes of the projection vectors, and selects the wavelength with the largest projection vector as the candidate wavelength, thereby finding the feature combination with the least redundant information and the least collinearity. The reflectance of the dimensionality-reduced feature bands is used as the model input, and nitrogen concentration data is used as the model output. The kernel parameters and regularization parameters of the kernel extreme learning machine are optimized using the dung beetle optimization algorithm and the zebra optimization algorithm to construct a hyperspectral inversion model of nitrogen concentration, which is then compared with the nitrogen concentration inversion model established by the kernel extreme learning machine.
[0036] S2.4 Model Evaluation Methods: In the evaluation of inversion models, the coefficient of determination (R²) is used. 2 R and root mean square error (RMSE) are used as evaluation metrics. 2 The goodness of fit of a model is measured by a value closer to 1, indicating a better fit. The calculation method is shown in equation (1). RMSE is used to measure the error in model prediction; a value closer to 0 indicates higher prediction accuracy. The calculation method is shown in equation (2).
[0037] In the formula, For the sum of squared residuals, For the total sum of squares, This is the actual value. For predicted values, The average of the actual values. The number of samples.
[0038] S3. Construction of fertilization decision-making methods.
[0039] S3.1 RiceGrow crop growth model: A rice growth simulation model, or rice growth model, is a comprehensive numerical simulation system based on systems science. It combines the physiological processes of rice with various influencing factors such as climate, soil, variety characteristics, and field management practices to comprehensively simulate and predict the growth and development of rice. This model takes a holistic system perspective, aiming to accurately describe and dynamically predict the biological behavior of rice under specific environmental conditions. The rice growth model can simulate the growth and development of rice at a single-point scale through specific time steps, including changes in various biological parameters such as leaf area index, photosynthesis, nutrient absorption, and rice yield.
[0040] The RiceGrow model is a comprehensive rice growth simulation model based on ecophysiological processes, designed to quantify the physiological and ecological processes of rice and their responses to environmental factors, genotype parameters, and management measures. The model simulates the growth process, yield formation, and changes in yield structure of rice on a timescale of growth and development with a daily step size. The RiceGrow model can effectively predict the growth and development of rice under different light conditions, water supply, and nitrogen limitation.
[0041] Based on this, the present invention generates leaf area index (LAI) and above-ground dry matter (AGDM) data of rice through a locally calibrated RiceGrow model, further explores the relationship between the two, and constructs a quantitative relationship curve to provide a theoretical basis for subsequent fertilization decisions.
[0042] S3.2 Calculation method of nitrogen deficit based on critical nitrogen concentration dilution curve: Based on the measured nitrogen concentration and LAI of rice, a critical nitrogen concentration dilution curve for rice was constructed. First, based on the nitrogen concentration and LAI data measured under different nitrogen fertilizer gradients, the samples were divided into two groups using analysis of variance: a nitrogen-limited group and an unlimited group. For example, using the LAI of the highest nitrogen treatment as a reference, a significance test was used to separate data where growth was limited by nitrogen (data with significant differences) from data where nitrogen was sufficient (data without significant differences). The former was used as data for the nitrogen-limited group, and the latter as data for the unlimited group. Second, for the nitrogen-limited group data, a linear fit was performed between LAI and nitrogen concentration to analyze the trend of nitrogen concentration changing with LAI. For the unlimited group data, the average LAI of the samples within the same growth period was taken as the maximum LAI for that period. Finally, the intercept point of the fitted curve at the maximum LAI for each period was taken as the theoretical critical nitrogen concentration point for that period. Using the critical nitrogen concentration points for each period as data input, a power function was used for fitting to establish a complete critical nitrogen concentration dilution curve for rice. In the formula, Critical nitrogen concentration (mass ratio) for rice, g / kg; LAI is leaf area index. These are curve parameters.
[0043] Construct a curve equation for rice aboveground dry matter content (LAI): In the formula, Rice aboveground dry matter mass t / hm 2 , These are curve parameters.
[0044] The critical nitrogen concentration dilution curve model can effectively reflect the excess or deficiency of nitrogen during the vegetative growth stage of crops. Based on the definition and calculation method of critical nitrogen concentration, the critical nitrogen accumulation (N) can be derived. cna The equation was developed, and a nitrogen deficit equation was further established to determine the nitrogen deficit in rice.
[0045] Critical nitrogen accumulation capacity refers to the total amount of nitrogen that a rice plant can accumulate under critical nitrogen concentration conditions. This can be expressed by the equation:
[0046] In the formula, N cna Nitrogen accumulation under critical nitrogen concentration conditions, kg / hm² 2 .
[0047] Nitrogen deficit (N) and Nitrogen deficit (NBS) refers to the amount of nitrogen deficiency that occurs in rice plants when nitrogen supply fails to meet their growth requirements during actual growth. The equation is derived based on the difference between the nitrogen accumulation in the plant under the actual nitrogen application rate and the critical nitrogen accumulation rate. The nitrogen accumulation in rice plants under the actual nitrogen application rate can be accurately calculated using nitrogen concentration retrieved via hyperspectral technology. The specific equation for calculating nitrogen deficit is as follows:
[0048] In the formula, N and Nitrogen deficit, expressed in kg / hm² 2 N na Actual nitrogen accumulation in plants under different nitrogen application rates, expressed in kg / hm² 2 .
[0049] The nitrogen deficit of rice can be calculated using the above formula, thus providing a scientific basis for precision fertilization and ensuring that rice receives sufficient nitrogen supply at different growth stages. The technical route of this invention is as follows: Figure 3 As shown.
[0050] Experimental Analysis: The experiment was conducted over two years and employed a rice plot cultivation design with different nitrogen fertilizer gradients. Five basal fertilizer gradients were established: 360 kg / hm² for experimental plots 1 and 5. 2 The yields for experimental fields No. 2 and No. 6 were 420 kg / hm². 2 The concentration of experimental fields No. 3 and No. 7 was 300 kg / hm². 2 The yield for experimental fields No. 4 and No. 8 was 240 kg / hm². 2 Experimental fields 9, 10, and 11 had a yield of 0 kg / hm². 2 During the experiment, field sampling was conducted from the rice tillering stage to the heading stage. Simultaneously, drone-generated visible light images, canopy hyperspectral reflectance, and key data such as rice nitrogen concentration and LAI were collected. The standard nitrogen fertilizer basal fertilizer application rate during the rice tillering stage was 130 kg / hm². 2 .
[0051] LAI Inversion Results and Evaluation: This invention uses texture features extracted from visible light remote sensing images as input data for the LAI inversion model, and LAI data acquired by the LAI-2200C plant canopy analyzer as output data. Figure 4 Here is an example of the texture feature map of test field No. 1.
[0052] To construct the LAI inversion model, KELM, DBO-KELM, and ZOA-KELM were used for model construction and comparison. The datasets were divided in a 7:3 ratio, with the training set containing 68 sets of data and the test set containing 30 sets of data.
[0053] The LAI inversion model results are as follows: Figure 5 As shown, where, Figure 5 (a) shows the comparison between the predicted values and the measured values from the KELM model. Figure 5 (b) shows the comparison between the predicted values and the measured values of the DBO-KELM model. Figure 5 Figure (c) shows the comparison between the predicted and measured values of the ZOA-KELM model. As can be seen from the figure, the LAI inversion model based on ZOA-KELM performs consistently well on both the training and test sets, with the R² value on the training set being relatively high. 2 The R value for the test set is 0.75, and the RMSE is 0.64; 2 The R² value is 0.73, and the RMSE is 0.66. The difference between the training and test sets is small, and the R² values of the two sets are similar. 2 A value greater than 0.70 indicates that the model has a good fit and high prediction accuracy. Under the same dataset, the R-value of the DBO-KELM training set is... 2 The R value for the test set is 0.71, the RMSE is 0.69, and the R value for the test set is... 2 The R value is 0.68, the RMSE is 0.69, and the R value of the KELM training set is...2 The R value is 0.71, the RMSE is 0.65, and the R value on the test set is... 2 The LAI index was 0.65, and the RMSE was 0.82. In summary, the ZOA-KELM-based LAI retrieval model has certain advantages and high reliability in rice LAI retrieval, and can quickly obtain the leaf area index of large areas of rice canopy using UAV visible light imagery.
[0054] Results and evaluation of nitrogen concentration inversion.
[0055] Nitrogen-sensitive band selection based on canopy hyperspectral reflectance characteristics: This invention utilizes the Continuous Projection Algorithm (SPA) to select characteristic bands from rice canopy hyperspectral data, with the number of target bands ranging from 5 to 30. Internal cross-validation is performed on the selected bands using a test set, and the hyperspectral characteristic bands representing rice nitrogen concentration are determined based on the root mean square error (RMSE) of the validation results. The subset of bands selected when the RMSE reaches its minimum value is considered the optimal subset. Figure 6 As shown, when the minimum RMSE is 0.57, a total of 10 spectral feature bands are extracted.
[0056] The results of the selection of characteristic bands for hyperspectral reflectance of rice canopy are as follows: Figure 7 As shown, the corresponding wavelengths are 502, 692, 707, 731, 761, 779, 906, 964, 985, and 999 nm.
[0057] Nitrogen concentration inversion results: This invention uses 10 extracted hyperspectral characteristic variables of rice as model inputs and rice nitrogen concentration data measured by the Kjeldahl method as model outputs. UAV hyperspectral rice nitrogen concentration inversion models are established using KELM, DBO-KELM and ZOA-KELM respectively.
[0058] Modeling results are as follows Figure 8 As shown, where, Figure 8 (a) shows the modeling results of the KELM model. Figure 8 (b) shows the modeling results of the DBO-KELM model. Figure 8 (c) represents the modeling results of the ZOA-KELM model, and the R-values of the nitrogen concentration inversion model training set based on DBO-KELM. 2 The R value for the test set is 0.77, the RMSE is 0.38, and the R value for the test set is... 2 The R² value was 0.72, and the RMSE was 0.45. This indicates that the model fits well on both the training and test sets and can accurately predict nitrogen concentration in rice. Although the performance on the test set was slightly lower than that on the training set, overall, the model's fit and prediction accuracy are still quite ideal. Under the same dataset, the R² value of the ZOA-KELM training set is [not specified].2 The R value for the test set is 0.69, the RMSE is 0.46, and the R value for the test set is... 2 The R value is 0.68, the RMSE is 0.46, and the R value of the KELM training set is... 2 The R value for the test set is 0.68, the RMSE is 0.47, and the R value for the test set is... 2 The accuracy is 0.64 and the RMSE is 0.49. This indicates that the kernel extreme learning machine optimized using the dung beetle optimization algorithm has certain advantages in nitrogen concentration retrieval in rice. In conclusion, the nitrogen concentration retrieval model based on DBO-KELM has high reliability.
[0059] Construction of fertilization decision-making model.
[0060] Critical nitrogen concentration curve: Based on the critical nitrogen concentration dilution curve construction method, a relationship curve between the critical nitrogen concentration and the corresponding LAI value was established, such as... Figure 9 As shown in the figure, this curve characterizes the relationship between the critical nitrogen concentration and leaf area index at different growth stages of rice. Its specific form is as follows:
[0061] The R-value of the established critical nitrogen concentration dilution curve equation 2 A value of 0.87 indicates that the regression model has a high degree of fit to the data and can effectively reflect the relationship between nitrogen concentration and LAI. This critical nitrogen concentration curve can be used to determine the nitrogen nutrient abundance or deficiency status of rice at different growth stages, providing a scientific basis for fertilization decisions. Based on this, by comparing with the actual nitrogen concentration, the nitrogen deficit of rice can be further calculated, optimizing the application rate of nitrogen fertilizer and thus improving fertilization efficiency.
[0062] Nitrogen deficit calculation: This invention uses regression analysis to fit the relationship between daily aboveground dry matter and leaf area index (LAI) output by a locally calibrated RiceGrow model, constructing a curve equation for rice aboveground dry matter versus LAI, as shown below. Figure 10 As shown. This curve equation can be used to estimate the quantitative relationship between aboveground dry matter (LAI) and rice at different growth stages. Its specific form is as follows:
[0063] Through regression analysis, R is obtained. 2 =0.98, indicating that the model fits the data well. This result shows that the constructed curve equation can accurately reflect the relationship between aboveground dry matter and leaf area index at different growth stages of rice.
[0064] The critical nitrogen accumulation level can be calculated based on the above equation. Its expression is: By comparing the nitrogen accumulation of plants under actual nitrogen application conditions with the critical nitrogen accumulation level, the nitrogen deficit can be accurately quantified, thus providing a scientific basis for fertilization decisions.
[0065] Fertilization decisions based on critical nitrogen concentration curves: This invention uses the fertilizer application rate of experimental rice fields as a standard, including four rounds of fertilization: basal fertilizer, greening fertilizer, tillering fertilizer, and panicle fertilizer. The basal fertilizer, greening fertilizer, and panicle fertilizer are applied according to local standard fertilization plans. This invention mainly focuses on the precise application of tillering fertilizer. Based on the fertilization decision-making method, the nitrogen deficit of experimental fields 1 to 7 is calculated, thereby guiding precise topdressing during the tillering stage. Figure 11 This is a prescription map for drone-based plant protection operations; the topdressing prescription map is a variable prescription map. The fertilization details and yield per unit area for each experimental field are as follows: Figure 12 As shown, field 0 is the control group with standard fertilization.
[0066] In terms of yield, the yields of experimental fields 1 to 7 were similar to those of the standard experimental field 0, with minimal difference, indicating that the fertilizer application rate calculated based on the critical nitrogen concentration curve of leaf area index could adequately meet the growth needs of rice. Regarding fertilizer application rate, except for experimental field 2, the fertilizer application rates in the other experimental fields were reduced to varying degrees, possibly related to factors such as the lower soil nutrient content in experimental field 2. Overall, under the same yield conditions, the fertilization strategy based on the critical nitrogen concentration curve of leaf area index required less fertilizer, a reduction of 7.8% compared to the standard fertilization plan.
[0067] Compared with the prior art, the present invention has the following advantages: Visible light images of rice canopies were acquired using drones, and their texture features were extracted. Subsequently, a rice leaf area index (LAI) inversion model was constructed using a zebra-optimized kernel extreme learning machine algorithm and instrument-measured leaf area index. The training set for the LAI inversion model was R... 2 The R value is 0.75, the RMSE is 0.64, and the R value on the test set is... 2 The correlation coefficient was 0.73, and the RMSE was 0.66. The results indicate that there is a strong correlation between the texture features of visible light remote sensing images and the rice leaf area index, and rice canopy leaf area index data can be quickly obtained through inversion.
[0068] Hyperspectral images of rice canopies were acquired using drones. Cellular hyperspectral data extraction was performed on the acquired hyperspectral remote sensing images. The results after resampling and denoising were used as the hyperspectral information for each experimental area. A rice nitrogen concentration inversion model was constructed using a dung beetle-optimized kernel extreme learning machine algorithm and the measured nitrogen concentration. The R-value of the nitrogen concentration inversion model training set... 2 The R value for the test set is 0.77, the RMSE is 0.38, and the R value for the test set is...2 The reliability of the rice nitrogen concentration inversion model based on DBO-KELM is 0.72, and the RMSE is 0.45.
[0069] A critical nitrogen concentration dilution curve model for rice was constructed based on LAI and nitrogen concentration data. Using this model, the nitrogen deficit in rice was further calculated, and targeted fertilization strategies were proposed based on the deficit information. A precision fertilization plan was developed, and an operational prescription map was created using drones. Under the same yield conditions, the fertilization strategy based on the LAI-based critical nitrogen concentration dilution curve required less fertilizer, a reduction of 7.8% compared to the standard fertilization plan. Compared to traditional methods that only describe nitrogen nutrient status, this method can accurately quantify the nitrogen deficit by incorporating the actual nitrogen requirements of rice, thus providing more scientific and efficient guidance for precision fertilization in the field.
[0070] Based on the above inventive concept, the present invention also provides a rice tillering stage fertilization decision system, comprising: The data inversion module is used to acquire multi-source remote sensing data of rice during the tillering stage from UAVs. The multi-source remote sensing data includes visible light images and hyperspectral images. The leaf area index of the rice canopy during the tillering stage is obtained by inversion based on the visible light images, and the plant nitrogen concentration of rice during the tillering stage is obtained by inversion based on the hyperspectral images. The curve construction module is used to determine the correspondence between aboveground dry matter and leaf area index of rice during the tillering stage using the RiceGrow rice growth model. Based on the critical nitrogen concentration dilution curve describing the relationship between leaf area index and critical nitrogen concentration in rice, and the inverted plant nitrogen concentration, the nitrogen deficit of rice is calculated according to the correspondence between aboveground dry matter and leaf area index. The critical nitrogen concentration dilution curve describing the relationship between leaf area index and critical nitrogen concentration in rice is obtained in advance by fitting the leaf area index of experimental rice and the measured nitrogen concentration of rice in multiple growth cycles. The decision-making module generates decision results based on the nitrogen deficit to guide variable fertilization during the tillering stage.
[0071] The present invention also provides a computer-readable storage medium storing a computer program that can be used to execute the above-described... Figure 1 The steps of the method for making decisions on fertilization during the rice tillering stage are provided.
[0072] This invention also provides a computer device. At the hardware level, the computer device includes a processor, an internal bus, a network interface, memory, and non-volatile memory, and may also include other hardware required for various operations. The processor reads the corresponding computer program from the non-volatile memory into memory and then executes it to achieve the above-mentioned functions. Figure 1 The steps of the method for making decisions on fertilization during the rice tillering stage are provided.
[0073] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.
[0074] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, as well as combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.
[0075] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.
[0076] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.
[0077] It should be noted that the specific embodiments described above enable those skilled in the art to more fully understand the present invention, but do not limit the present invention in any way. Therefore, although the present invention has been described in detail in this specification, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the present invention; and all technical solutions and improvements that do not depart from the spirit and scope of the present invention are covered within the protection scope of the patent of the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.
Claims
1. A method for making fertilization decisions during the tillering stage of rice, characterized in that, Includes the following steps: Acquire multi-source remote sensing data of rice during the tillering stage using unmanned aerial vehicles (UAVs), wherein the multi-source remote sensing data includes visible light images and hyperspectral images; obtain the leaf area index of the rice canopy during the tillering stage based on the visible light images, and obtain the plant nitrogen concentration of rice during the tillering stage based on the hyperspectral images. The relationship between aboveground dry matter and leaf area index (LAI) of rice during the tillering stage was determined using the RiceGrow rice growth model. Based on the critical nitrogen concentration dilution curve describing the relationship between LAI and critical nitrogen concentration in rice, and the inverted plant nitrogen concentration, the nitrogen deficit of rice was calculated according to the relationship between aboveground dry matter and LAI. The critical nitrogen concentration dilution curve describing the relationship between LAI and critical nitrogen concentration in rice was obtained in advance by fitting the leaf area index of experimental rice and the measured nitrogen concentration of rice over multiple growth cycles. Based on the nitrogen deficit, decision-making results are generated to guide variable fertilization during the tillering stage.
2. The method according to claim 1, characterized in that, The leaf area index of rice canopy at the tillering stage was obtained based on the visible light image inversion, and the plant nitrogen concentration of rice at the tillering stage was obtained based on the hyperspectral image inversion, including: Texture features are extracted from the visible light image; the texture features are input into an optimized first machine learning model, and the leaf area index is output. The hyperspectral image is subjected to spectral feature band selection to obtain the reflectance of the feature band; The reflectance of the characteristic band is input into an optimized second machine learning model, and the nitrogen concentration is output.
3. The method according to claim 2, characterized in that, The first machine learning model is a Kernel Extreme Learning Machine (KELM) model that optimizes the kernel parameters and regularization parameters using the Zebra Optimization Algorithm (ZOA). The second machine learning model is a KELM model that optimizes the kernel parameters and regularization parameters using the Dung Beetle Optimization Algorithm (DBO).
4. The method according to claim 2, characterized in that, The hyperspectral image is subjected to spectral feature band selection to obtain the reflectance of the feature bands, including: Each spectral feature band in the hyperspectral image is projected onto other spectral feature bands, and the wavelength with the largest projection vector is selected as the candidate wavelength. The reflectance of the feature band is then determined based on the candidate wavelength.
5. The method according to claim 1, characterized in that, Before calculating the nitrogen deficit in rice, the following also includes: For the experimental rice under each nitrogen fertilizer gradient condition in multiple growth cycles, the leaf area index of the rice with the highest nitrogen fertilizer treatment in each growth cycle was used as a reference, and the nitrogen-free group and the nitrogen-restricted group for that growth cycle were obtained through analysis of variance. For the experimental rice in the nitrogen-limited group, a linear relationship between the leaf area index and the measured nitrogen concentration of rice was fitted; for the experimental rice in the non-nitrogen-limited group, the mean leaf area index of each growth cycle was taken as the maximum leaf area index of that growth cycle; and the critical nitrogen concentration of each growth cycle was taken as the cutoff point of the linear relationship between the leaf area index and the measured nitrogen concentration of rice at the maximum leaf area index. Based on the fitting of the leaf area index of rice with critical nitrogen concentrations at multiple growth stages in the experiment, a critical nitrogen concentration dilution curve encompassing the leaf area index and critical nitrogen concentration of rice at multiple growth stages was obtained.
6. The method according to claim 5, characterized in that, The critical nitrogen concentration dilution curve describing the relationship between leaf area index and critical nitrogen concentration in rice is a power function model with leaf area index as the independent variable and critical nitrogen concentration as the dependent variable. The relationship between aboveground dry matter and leaf area index of rice during the tillering stage is a power function model with aboveground dry matter as the independent variable and leaf area index as the dependent variable.
7. The method according to claim 6, characterized in that, The calculation of nitrogen deficit in rice based on the critical nitrogen concentration dilution curve and the inverted plant nitrogen concentration includes the following steps: The critical nitrogen accumulation amount is calculated based on the critical nitrogen concentration dilution curve and the leaf area index obtained by inversion. The actual nitrogen accumulation is calculated based on the nitrogen concentration and leaf area index obtained from the inversion. The nitrogen deficit is calculated based on the difference between the critical nitrogen accumulation level and the actual nitrogen accumulation level.
8. A rice tillering stage fertilization decision system, characterized in that, include: The data inversion module is used to acquire multi-source remote sensing data of rice during the tillering stage from unmanned aerial vehicles (UAVs), wherein the multi-source remote sensing data includes visible light images and hyperspectral images; the leaf area index of the rice canopy during the tillering stage is obtained by inversion based on the visible light images, and the plant nitrogen concentration of rice during the tillering stage is obtained by inversion based on the hyperspectral images. The curve construction module is used to determine the correspondence between aboveground dry matter and leaf area index of rice during the tillering stage using the RiceGrow rice growth model; based on the critical nitrogen concentration dilution curve describing the relationship between leaf area index and critical nitrogen concentration of rice and the inverted plant nitrogen concentration, the nitrogen deficit of rice is calculated according to the correspondence between aboveground dry matter and leaf area index. The critical nitrogen concentration dilution curve describing the relationship between leaf area index and critical nitrogen concentration of rice is obtained in advance by fitting the leaf area index of experimental rice and the measured nitrogen concentration of rice for multiple growth cycles. The decision module is used to generate decision results to guide variable fertilization during the tillering stage based on the nitrogen deficit.
9. A computer device, characterized in that, The method includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the method described in any one of claims 1 to 7.
10. A readable storage medium, characterized in that, The readable storage medium stores a computer program that, when executed by a processor, implements the method described in any one of claims 1 to 7.