Corn population nitrogen absorption amount monitoring method, system, device and storage medium

Abstract

The present application belongs to the field of crop growth monitoring and agricultural remote sensing, and discloses a corn population nitrogen uptake monitoring method, system, device and storage medium, the monitoring method comprising obtaining corn canopy hyperspectral image, pretreating the image and calculating vegetation index, using canopy nitrogen uptake inversion model to obtain observable canopy nitrogen uptake, calculating cumulative growing degree day according to meteorological data from sowing day to monitoring day, establishing a dynamic scale conversion model constrained by phenology, realizing the conversion of canopy scale nitrogen to population plant nitrogen uptake, and outputting field or plot scale monitoring results, the present application fully considers the canopy structure change and plant internal nitrogen redistribution law in the corn growth process, improves the accuracy, stability and cross-region applicability of population nitrogen uptake monitoring, and can be applied to corn nitrogen nutrition diagnosis, precision fertilization, growth monitoring and intelligent agricultural management.

Description

Technical Field

[0001] This invention belongs to the field of farmland crop condition monitoring and agricultural remote sensing technology, and particularly relates to a method, system, equipment and storage medium for monitoring nitrogen uptake of maize populations based on UAV hyperspectral and phenological constraints. Specifically, it relates to a method and system for high-precision and high-mobility monitoring of nitrogen uptake of maize plants at the population scale by inverting observable canopy nitrogen uptake from UAV hyperspectral images and combining the scale transformation relationship of phenological growth degree-day constraints. Background Technology

[0002] Plant nitrogen uptake Nitrogen uptake by maize plants is a crucial agronomic indicator reflecting the growth and development status, nitrogen nutrient levels, and yield potential of the maize population. It is also an important basis for precision fertilization, variable-rate fertilization, nutrient diagnosis, and nitrogen management in farmland. Accurately obtaining nitrogen uptake by maize plants is of great significance for improving nitrogen fertilizer utilization efficiency, reducing agricultural non-point source pollution, and achieving refined agricultural management. Traditional methods for obtaining nitrogen uptake mainly rely on manual field sampling, laboratory digestion analysis, and chemical determination, calculating nitrogen uptake by measuring plant dry matter mass and nitrogen content. While these methods provide relatively accurate results, they suffer from drawbacks such as being highly destructive, having long detection cycles, being labor-intensive, costly, and having limited spatial coverage. They can only obtain data from a small number of discrete sampling points, which is insufficient to meet the practical needs of modern agriculture for large-area, multi-temporal, rapid, and continuous acquisition of nitrogen information from maize populations. Therefore, utilizing remote sensing technology to achieve rapid and non-destructive monitoring of nitrogen uptake by maize plants has gradually become an important research direction in the field of agricultural remote sensing.

[0003] With the development of low-altitude remote sensing platforms and hyperspectral imaging technology, UAVs equipped with hyperspectral sensors can quickly acquire continuous narrow-band spectral information of crop canopies. They offer advantages such as high spatial resolution, strong maneuverability, minimal cloud cover influence, and strong repeatability, and have been widely applied in fields such as crop growth monitoring at the farmland scale, nitrogen nutrient diagnosis, yield prediction, and precision fertilization. Current technologies typically utilize UAVs to acquire hyperspectral reflectance data of maize canopies. By constructing vegetation indices, spectral characteristic parameters, or continuous spectral variables, and combining statistical analysis methods such as partial least squares regression, random forests, and support vector machines with machine learning algorithms, models for inverting plant nitrogen uptake or concentration are established to achieve rapid estimation of maize nitrogen information. Meanwhile, to further improve the inversion accuracy, existing studies have combined continuous wavelet transform, sensitive band screening, multi-scale spectral conversion, radiative transfer model, multi-source remote sensing data fusion, and deep learning to preprocess and extract features from hyperspectral data, thereby enhancing the responsiveness of spectral features to nitrogen information. Furthermore, the performance of the model has been improved through modeling different growth stages, canopy stratification analysis, and structural parameter fusion, thus continuously improving the application of UAV hyperspectral remote sensing in maize nitrogen monitoring.

[0004] In recent years, addressing the challenges of complex crop canopy structures and significant differences in spectral responses across different growth stages, existing technologies have further explored vertical canopy stratification monitoring, multi-scale information collaborative modeling, and the fusion of physical and data-driven models. For example, some schemes utilize the spectral response characteristics of different leaf layers in the canopy to establish stratified nitrogen monitoring models, improving the diagnostic capability for nitrogen levels in different leaf layers. Other schemes enhance model consistency across different spatial scales through information conversion between leaf, canopy, and regional scales. Still others employ radiative transfer models to simulate canopy spectral characteristics and combine them with machine learning algorithms to improve model generalization performance. Additionally, technologies integrate multi-source remote sensing information such as multispectral, hyperspectral, texture features, and lidar data to improve canopy structure representation and nitrogen retrieval accuracy. These technological solutions alleviate, to some extent, problems such as hyperspectral noise interference, limited representational power of single spectral features, and insufficient modeling accuracy in local areas, playing a positive role in improving nitrogen estimation capabilities in specific experimental areas or at specific growth stages.

[0005] However, existing technologies still primarily employ a modeling approach that directly correlates canopy spectral information acquired by drones with plant nitrogen uptake at the canopy scale. This involves establishing an empirical regression relationship between canopy spectral characteristics and measured overall plant nitrogen levels obtained through destructive field sampling. This approach still suffers from the following problems:

[0006] First, the optical sensors carried by UAVs are limited by the canopy structure and the visible angle. The reflected spectrum they acquire mainly comes from the top and upper middle leaves of the canopy. The lower leaves, stems, and ears of grain, which are covered by light, contribute little to the overall spectrum or are even difficult to observe effectively. Therefore, there is a natural scale difference between the actual objects observed by the sensors and the targets inverted by the model.

[0007] Second, as maize transitions from the vegetative growth stage to the reproductive growth stage, the plant canopy gradually closes, the degree of leaf shading increases, and the spectral saturation effect intensifies. Simultaneously, nitrogen within the plant is continuously transported from functional leaves to reproductive organs such as the ear. The proportion of nitrogen represented by the observable canopy changes continuously, resulting in a significant dynamic change in the correspondence between canopy spectral information and total nitrogen in the population. Existing technologies generally use fixed empirical relationships or static models for estimation, which makes it difficult to accurately reflect the changes in nitrogen distribution patterns within the plant at different growth stages. This leads to a decrease in the model's inversion accuracy in the later stages of growth and reduces the model's applicability in different years, different ecological regions, and different cultivation conditions.

[0008] Third: Agricultural remote sensing model training usually relies on destructive field sampling to obtain measured plant nitrogen data. Since the sampling process is time-consuming, labor-intensive, and costly, the number of labeled samples that can be obtained is usually limited. Existing inversion models based on traditional machine learning algorithms are easily limited by the sample size and are prone to problems such as overfitting, insufficient model stability, and decreased generalization ability under small sample conditions.

[0009] While existing technologies have improved the quality of input features through spectral denoising, sensitive feature screening, multi-source data fusion, and model optimization, these improvements mainly focus on the spectral information itself and do not address the inconsistency between the observation scale and the inversion target during model construction. They also lack constraint mechanisms that can describe crop phenological processes and the dynamic changes in nitrogen within plants. Therefore, the models remain susceptible to factors such as changes in growth stages, canopy structure differences, and ecological environment changes, making it difficult to balance monitoring accuracy, model stability, and cross-regional application capabilities. Consequently, they fail to meet the practical needs of modern precision agriculture for long-term, continuous, high-precision, and highly adaptable monitoring of nitrogen uptake in maize populations. Therefore, further research is necessary to address these existing problems. Summary of the Invention

[0010] This invention aims to address the technical problems of existing UAV remote sensing monitoring methods, which primarily rely on canopy-scale nitrogen information inversion, making it difficult to accurately characterize the nitrogen uptake of the entire maize plant population. Furthermore, fixed-scale conversion models cannot adapt to the dynamic changes in nitrogen redistribution within the plant at different growth stages, resulting in insufficient estimation accuracy and generalization ability. This application discloses a method, system, device, and storage medium for monitoring nitrogen uptake in maize populations. The monitoring method establishes a canopy nitrogen uptake inversion model and a phenologically constrained dynamic scale conversion model, achieving a continuous dynamic conversion from observable canopy nitrogen to population nitrogen, thus improving the accuracy, stability, and engineering application capability of population nitrogen uptake monitoring.

[0011] The technical solution of the present invention is as follows:

[0012] The first objective of this application is to disclose a method for monitoring nitrogen uptake in maize populations, including:

[0013] Based on the vertical structure of the corn canopy, the observable canopy leaf layers that can be directly observed by the UAV canopy optical sensor are determined, and the canopy nitrogen absorption corresponding to the observable canopy leaf layers is used as the canopy-scale nitrogen index.

[0014] The cumulative growth degree days are calculated based on the daily temperature data from the sowing date to the monitoring date in the target corn planting area.

[0015] Acquire UAV hyperspectral images of the target cornfield, preprocess the hyperspectral images and extract a set of vegetation index features;

[0016] The canopy nitrogen uptake inversion model, which takes the vegetation index feature set as input and the canopy-scale nitrogen index as output, is invoked to obtain the canopy-scale nitrogen index based on the vegetation index feature set.

[0017] The scale conversion relationship between the canopy-scale nitrogen index constrained by the cumulative growth days and the nitrogen uptake of the population plants is invoked, and the nitrogen uptake of the population plants in the target maize field is obtained based on the canopy-scale nitrogen index and the cumulative growth days.

[0018] Furthermore, the observable canopy leaf layer includes the entire canopy leaf layer during the vegetative growth stage, and includes the upper and middle leaf layers that can be directly observed by the UAV canopy optical sensor from the top of the canopy downwards during the reproductive growth stage, excluding the shaded lower leaf layer.

[0019] Furthermore, the canopy nitrogen uptake inversion model is constructed using a tabular prediction model based on the Transformer architecture. The tabular prediction model is the TabPFN model, which uses the set of vegetation index features extracted from UAV hyperspectral images as input features and the canopy nitrogen uptake corresponding to the observable canopy leaf layer as the target variable. The model is trained using historical field test samples.

[0020] Furthermore, the vegetation index feature set includes 22-dimensional vegetation index feature vectors of CIred-edge, DCNI, DVI, EVI, LICI, MCARI, mND705, MNLI, mRER, MTCI, NDRE, NDVI, OREA, OSAVI, REIP, SAVI, SIPI, SR, TCARI, TGI, TVI, and VOG2.

[0021] Furthermore, the formula for calculating the cumulative growth days is as follows:

[0022]

[0023] in, No. The highest temperature of the day For the first The lowest temperature of the day, The basic temperature for corn growth, The date sequence number between sowing date and monitoring date. This is the cumulative number of days from the sowing date to the monitoring date.

[0024] Furthermore, the scale transformation relationship satisfies:

[0025]

[0026] in,

[0027] Nitrogen uptake by the plant population;

[0028] The canopy-scale nitrogen index of the observable canopy leaf layer;

[0029] The slope of the scale transformation varies with the cumulative growth degree per day.

[0030] The scale transformation intercept varies with cumulative growth degree per day.

[0031] Furthermore, the scale transformation slope is represented by a linear function of the cumulative growth degree days, satisfying... ;

[0032] The scale transformation intercept is expressed as a power function of the cumulative growth degrees in days, satisfying: ;

[0033] in, , , , , All parameters were fixed and obtained by fitting historical field trial samples.

[0034] The first objective of this application is to disclose a system for monitoring nitrogen uptake in maize populations based on UAV hyperspectral and phenological constraints, including:

[0035] The hyperspectral image acquisition module is used to acquire UAV hyperspectral images of the target cornfield.

[0036] The preprocessing and feature calculation module is used to preprocess the hyperspectral image and extract a set of vegetation index features;

[0037] The canopy nitrogen inversion module is used to calculate the canopy nitrogen uptake corresponding to the observable canopy leaf layer based on the vegetation index feature set.

[0038] The phenological accumulated temperature calculation module is used to calculate the cumulative growth days based on daily temperature data;

[0039] The scale conversion module is used to establish the scale conversion relationship between the canopy-scale nitrogen index and the nitrogen uptake of the population plants based on the cumulative growth days, and to calculate the nitrogen uptake of the population plants.

[0040] The monitoring results output module is used to output the monitoring results of nitrogen uptake by plant populations at the field or plot scale.

[0041] The modules work together to execute the method for monitoring nitrogen uptake in maize populations as described above.

[0042] The purpose of this application is to disclose a device for monitoring nitrogen uptake in maize populations based on hyperspectral and phenological constraints from unmanned aerial vehicles (UAVs), comprising a processor, a memory, and a computer program stored in the memory. When the computer program is executed by the processor, it causes the processor to perform the nitrogen uptake monitoring method for maize populations as described above.

[0043] The fourth objective of this application is to disclose a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, causes the processor to perform the above-described method for monitoring nitrogen uptake in maize populations.

[0044] Compared with existing technologies, the method, system, equipment, and storage medium for monitoring nitrogen uptake in maize populations described in this invention have the following advantages:

[0045] This application proposes a method for monitoring nitrogen uptake in maize populations. Based on hyperspectral information acquired by unmanned aerial vehicles (UAVs), it integrates observable canopy nitrogen inversion results with phenological information characterized by cumulative growth days to establish a dynamic conversion mechanism from canopy scale to population scale. This enables continuous, non-destructive, and quantitative monitoring of nitrogen uptake by maize populations throughout their entire growth period. Compared to traditional methods that rely solely on vegetation indices to directly estimate population nitrogen, this application fully considers the impact of factors such as canopy structure changes, leaf shading, and nitrogen redistribution within the plant on remote sensing monitoring results during maize growth. This allows the scale conversion relationship to automatically adjust with phenological progress, thereby improving the consistency and stability of population nitrogen estimation results under different years, sowing dates, ecological regions, and planting management conditions. Furthermore, it constructs a unified monitoring process that balances monitoring accuracy, computational efficiency, and engineering application capabilities. This method can provide reliable data support for maize nitrogen nutrition diagnosis, precise variable fertilization, population growth monitoring, yield formation analysis, and smart agriculture management, and has high application value and promising prospects for industrial application. Attached Figure Description

[0046] Figure 1 This is a schematic diagram of the vertical layering structure of the maize canopy provided in an embodiment of the present invention;

[0047] Figure 2 This is a schematic diagram validating the maize canopy stratified nitrogen inversion model based on TabPFN in this application;

[0048] Figure 3 This invention provides observable canopy layers under different cumulative growth day stages according to embodiments of the invention. Linear regression diagram between canopy nitrogen uptake and total plant nitrogen uptake;

[0049] Figure 4 This is a comparison chart of the function fitting results of scale transformation slope and cumulative growth days provided in the embodiments of the present invention;

[0050] Figure 5 This is a comparison chart of the function fitting results of scale transformation intercept and cumulative growth days provided in the embodiments of the present invention;

[0051] Figure 6 This is a comparison chart of the estimation accuracy of nitrogen uptake by plant populations under different monitoring methods provided in the embodiments of the present invention;

[0052] Figure 7 This is a comparison chart of the transferability verification of the GDD-PNU model provided in this embodiment of the invention and the traditional VI-PNU empirical model under different conditions;

[0053] Figure 8This is a field application effect diagram based on the GDD-PNU model provided in an embodiment of the present invention. Detailed Implementation

[0054] The technical solutions of the embodiments of this application will be clearly described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application are within the scope of protection of this application.

[0055] It should be noted that all directional and positional terms used in this invention, such as "up," "down," "left," "right," "front," "back," "vertical," "horizontal," "inner," "outer," "top," "lower," "lateral," "longitudinal," and "center," are only used to explain the relative positional relationships and connections between components in a specific state (as shown in the accompanying drawings). They are merely for the convenience of describing the invention and do not require the invention to be constructed and operated in a specific orientation; therefore, they should not be construed as limitations on the invention. Furthermore, descriptions involving "first," "second," etc., are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated.

[0056] In the description of this invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" 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; 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 can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0057] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "illustrative embodiment," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0058] This application discloses a method for monitoring nitrogen uptake in maize populations, comprising the following steps:

[0059] Based on the vertical structure of the maize canopy, the observable canopy leaf layer directly observed by the UAV canopy optical sensor is determined, and the canopy nitrogen absorption corresponding to the observable canopy leaf layer is used as the canopy-scale nitrogen index. A canopy nitrogen absorption inversion model is established with the vegetation index feature set as input and the canopy-scale nitrogen index as output.

[0060] The cumulative growth days are calculated based on the daily temperature data from the sowing date to the monitoring date in the target maize planting area, and a scale conversion relationship is established between the canopy nitrogen uptake constrained by the cumulative growth days and the nitrogen uptake of the population plants.

[0061] A drone-generated hyperspectral image of the canopy of a target maize field is acquired. The hyperspectral image is preprocessed and a set of vegetation index features characterizing the nitrogen status and structure of the maize canopy is extracted. The set of vegetation index features is input into the canopy nitrogen uptake inversion model to obtain the canopy-scale nitrogen index. The cumulative growth rate of the corresponding monitoring day is then input into the scale conversion relationship to output the nitrogen uptake of the population of the target maize field.

[0062] The method for monitoring nitrogen uptake in maize canopies disclosed in this application does not follow the existing modeling approach of directly using UAV hyperspectral information to establish an empirical mapping relationship between canopy spectra and nitrogen uptake in canopy plants. Instead, it first redefines the monitoring object based on the actual observation capabilities of UAV canopy optical sensors. The traditional whole-plant nitrogen monitoring target is decomposed into two parts: a canopy-scale object that can be stably observed by the sensor and a canopy-scale object that cannot be directly observed. By analyzing the vertical structure of the maize canopy and the contribution of different leaf layers to the canopy reflectance spectrum, the leaf layers in the upper and middle sections that can stably acquire spectral response information are defined as observable canopy leaf layers. ), and based on the corresponding canopy nitrogen uptake ( As the first-stage inversion target, this ensures that the model output object is consistent with the actual object collected by the sensor. This reduces the problem of natural scale differences between the observed object and the predicted target in traditional methods from the modeling starting point. After completing the canopy-scale nitrogen prediction, instead of directly converting the overall plant nitrogen using a fixed ratio, it further incorporates the continuously changing phenological processes during maize growth, introducing cumulative growth days that continuously reflect the degree of maize development. As a dynamic constraint variable, the unit of cumulative growth degree-days is ℃·d. A scale conversion relationship is established between canopy-scale nitrogen and population nitrogen, dynamically changing with phenological stages. This allows the conversion process to automatically adjust the conversion ratio as the plant gradually transitions from vegetative to reproductive growth stages. This adapts to the characteristics of the continuously changing relationship between canopy spectral information and overall plant nitrogen, caused by the gradual closure of the canopy, increasing spectral saturation, and continuous nitrogen transport to organs such as the ear. In practical field applications, only UAV hyperspectral images need to be acquired and vegetation index feature sets extracted. Then, the final population nitrogen uptake can be obtained through two stages: canopy-scale nitrogen prediction and phenological constraint scale conversion. The entire monitoring process progressively completes the conversion between different scales of information according to three levels: observation capability matching, canopy nitrogen prediction, and dynamic scale conversion. Each stage establishes corresponding calculation logic for different interference factors, allowing various error sources to be separated and weakened layer by layer. The overall modeling approach is more consistent with the actual observation patterns of UAV remote sensing and the physiological changes during the formation of nitrogen in maize populations.

[0063] The method for monitoring nitrogen uptake in maize canopies disclosed in this application is based on the combination of the actual observation range of UAV canopy optical sensors and the dynamic distribution law of nitrogen in maize canopies. Firstly, the canopy leaf layers are divided according to the spatial distribution characteristics of different leaf layers and differences in light conditions. Then, based on the vertical observation method of the UAV, observable canopy leaf layers capable of stably receiving spectral reflectance information are determined, ensuring that the inversion model establishes a correspondence only for leaf layers for which the UAV can truly acquire information. During the model building process, multiple vegetation index features reflecting leaf nitrogen status, chlorophyll content, and canopy structure changes are calculated using hyperspectral images collected by the UAV. These vegetation index features are then used to establish a canopy-scale nitrogen uptake inversion model, realizing the conversion from hyperspectral information to canopy nitrogen uptake. Since the prediction target at this stage only corresponds to the actual observable leaf layers, the impact of canopy shading and spectral saturation on the prediction process can be effectively reduced. This method completes the canopy-scale nitrogen uptake inversion model. After forecasting, the cumulative growth days are calculated based on the daily maximum and minimum temperatures in the target area. The cumulative growth days continuously reflect the developmental process of maize throughout its entire growth cycle. Using the cumulative growth days as a control variable, a dynamic scale conversion relationship is established between canopy-scale nitrogen and plant-wide nitrogen. This allows the dynamic scale conversion model to automatically adjust the conversion ratio of canopy nitrogen to overall plant nitrogen according to different phenological stages. This simultaneously describes the continuous change in nitrogen distribution ratios among different organs such as canopy leaves, lower leaves, stems, and ears as the growth stage progresses. During real-time field monitoring, the UAV only needs to complete one hyperspectral data acquisition to quickly obtain canopy-scale nitrogen indicators. Combined with meteorological data for the corresponding period, the cumulative growth days are calculated. The dynamic scale conversion model then restores the nitrogen uptake of the entire maize plant population, creating a continuous connection between remote sensing observation information and the internal physiological changes of the crop. This achieves a layer-by-layer derivation and dynamic correction from canopy spectroscopy to plant-wide nitrogen uptake.

[0064] Compared with existing technologies, this application redefines the monitoring object of UAV hyperspectral remote sensing and introduces a phenologically constrained dynamic scale conversion mechanism, enabling the entire monitoring system to more accurately reflect the actual observation capabilities of UAV sensors and the nitrogen allocation changes during maize growth. During the monitoring process, the inversion target is first limited to the observable canopy and leaf layer, ensuring consistency between the model's predicted object and the UAV's collected object. This effectively reduces systematic errors caused by canopy shading, spectral saturation, and the involvement of unobservable organs in modeling, making the canopy-scale nitrogen prediction results more stable and reliable. Simultaneously, by accumulating growth days to continuously characterize the developmental process of maize at different growth stages, the conversion relationship between canopy-scale nitrogen and plant population nitrogen can be automatically adjusted according to different phenological stages, no longer relying on fixed empirical coefficients for scale conversion. This effectively alleviates the continuous shift of nitrogen from leaves to ears in the mid-to-late stages. The problem of nitrogen transport leading to decreased model accuracy has been addressed by this method, which allows monitoring results to more accurately reflect the overall nitrogen accumulation of plants at different growth stages. Furthermore, the entire monitoring process relies solely on hyperspectral imagery acquired by UAVs and conventional meteorological temperature data, eliminating the need for additional destructive sampling. This enables rapid acquisition of nitrogen uptake in large-scale farmland populations, reducing manual sampling costs and improving monitoring efficiency. Moreover, the two-tiered computational framework of canopy inversion and phenological scale conversion, with each stage addressing different error sources, ensures the model's adaptability and transferability across different years, ecological regions, varieties, and planting densities. This further enhances the stability, consistency, and engineering application value of the monitoring results, providing more reliable data support for precision fertilization, variable fertilization, growth diagnosis, and yield prediction in maize.

[0065] In some examples of this application, the observable canopy leaf layer includes the entire canopy leaf layer during the vegetative growth stage, and includes the upper and middle leaf layers that can be directly observed by the UAV canopy optical sensor from the top of the canopy downwards during the reproductive growth stage, excluding the completely obscured lower leaf layer.

[0066] This application further dynamically defines the observable canopy leaf layer, no longer using a fixed leaf layer division method, but adaptively determining the observable canopy range based on changes in canopy structure at different growth stages of maize and the actual observation capabilities of the UAV canopy optical sensor. Figure 1 As shown, the maize canopy is divided into: the upper leaf layer ( ), middle leaf layer ( ) and lower leaf layer ( The observable canopy leaf layer is defined as the cumulative leaf layer composed of the upper and middle leaf layers, i.e. The canopy is divided into three layers: the upper leaf layer, consisting of leaves at the top of the canopy that receive better light, and the assimilated products from photosynthesis primarily supply the development of the tassels and ears; the middle leaf layer, consisting of the ear-position leaf, one leaf above it, and one leaf below it, which has a larger area and is the most important source of photosynthetic assimilation during the kernel formation stage, as well as the functional leaf layer most sensitive to changes in nitrogen nutrition; and the lower leaf layer, consisting of basal leaves, which receive relatively less light and gradually age in the later stages of plant growth, with their photosynthetic assimilation products mainly used for early stem growth. When dividing the leaf layers, the middle leaf layer corresponding to the ear-position leaf is first identified, then the leaves above it are classified as the upper leaf layer, and the leaves below it as the lower leaf layer, to ensure consistency in leaf layer classification standards across different growth stages. Because the canopy structure of maize varies significantly across different growth stages, the observable canopy leaf layers are defined dynamically. During the vegetative growth stage, the plant height is relatively low, and the canopy has not yet formed significant shading. The lower leaves can receive sufficient sunlight, and the canopy reflectance spectrum can cover the entire leaf layer. Therefore, the canopy leaf layer that can be observed during this stage corresponds to the entire canopy leaf layer. After entering the reproductive growth stage, as the plant continues to grow, the canopy gradually closes, and the upper leaves continuously shade the lower leaves. The lower leaves are in a low-light or shaded environment for a long time, and their reflectance spectrum is difficult to be effectively transmitted to the UAV canopy optical sensor. Therefore, the canopy leaf layer that can be observed during this stage only includes the upper and middle leaves that can be directly observed by the UAV canopy optical sensor from the top of the canopy downwards, and does not include the completely shaded lower leaves. Under typical maize population conditions, this observable canopy layer usually corresponds to about eight leaves in the upper middle part of the canopy that can stably produce an effective spectral response.

[0067] This application uses the observable canopy leaf layer as the object of expression for canopy-scale nitrogen information, based primarily on two aspects: model inversion capability and canopy nitrogen distribution patterns. Firstly, as... Figure 2 As shown, TabPFN is used to construct vegetation index features and , , as well as , , The inversion model between them was verified, whereby, Corresponding to the upper leaf layer, Corresponding to the middle leaf layer, Corresponding to the lower leaf layer, For the reason An observable assemblage of canopy and foliage that exists independently. For the reason and The observable canopy-leaf layer combination formed by accumulation (i.e. the observable canopy ultimately selected in this application). For the reason , and The cumulative leaf layer assemblage of the entire canopy shows that, with When the corresponding canopy nitrogen uptake is used as the inversion target, higher prediction accuracy can be obtained compared to using only the upper leaf layer to build the model. However, after including the lower leaf layer as the inversion target, the model accuracy did not improve further, indicating that the effective spectral information obtained by the UAV canopy optical sensor mainly comes from the upper and middle leaf layers, while the lower leaf layer contributes little to the canopy spectrum. Secondly, the relationship between canopy nitrogen uptake corresponding to different canopy ranges and the nitrogen uptake of the entire plant population was analyzed. The results show that, as Figure 4 As shown, The canopy nitrogen uptake and the total nitrogen uptake of the plant population maintain a stable linear relationship. However, the canopy nitrogen uptake including the lower leaves shows a significant nonlinear relationship with the total nitrogen uptake of the plant population. This is because the lower leaves gradually age and fall off during the reproductive growth stage, and nitrogen is continuously transported to organs such as the spike, causing the nitrogen distribution relationship to change continuously with the phenological process. Therefore, it is not suitable as a target for direct inversion of the canopy spectrum by UAV.

[0068] In addition, Identifying the leaf layer as directly observed by the UAV's canopy optical sensor can also reduce the impact of spectral saturation on model building from the perspective of canopy radiative transfer mechanisms. When the number of canopy leaves does not exceed... Within the corresponding range, all leaf layers of the canopy can be effectively observed by the sensor, and the canopy spectrum can truly reflect the nitrogen status of the canopy leaf layers. When the canopy continues to grow and form more lower leaf layers, because the newly added leaf layers are located in the shaded area, their spectral information is difficult to participate in the canopy reflection signal acquired by the UAV. At this time, the canopy spectrum still mainly reflects The nitrogen information of the leaf layer within the range. Therefore, this application directly adopts this information during the model building stage. Using canopy nitrogen uptake as the inversion target ensures that the model's predicted object is consistent with the actual object observed by the UAV. This avoids the problem of inconsistency between the observation scale and the inversion target caused by canopy shading, spectral saturation, and the unobservable lower leaf layer. It also provides a stable and reliable canopy-scale nitrogen input for establishing a dynamic scale conversion relationship between canopy-scale nitrogen uptake and population nitrogen uptake.

[0069] In some examples of this application, the canopy nitrogen uptake inversion model is constructed using a tabular-based prediction model based on the Transformer architecture. The tabular-based prediction model is the TabPFN model, which uses a set of vegetation index features extracted from UAV hyperspectral images as input features and canopy nitrogen uptake corresponding to the observable canopy leaf layer as the target variable. The model is adapted through historical field trial samples. Each sample is based on the same experimental plot, the same sampling date, and the same leaf layer observation record as the basic unit, and includes a set of vegetation index features, variety information, planting density, nitrogen fertilizer treatment, sampling date or cumulative growth degree days information, and an observable canopy leaf layer nitrogen uptake label obtained through stratified destructive sampling. The TabPFN model uses a context learning mechanism based on the Transformer architecture to establish the correspondence between vegetation index features and observable canopy leaf layer canopy nitrogen uptake, and outputs the corresponding canopy nitrogen uptake based on the vegetation index features of the sample to be monitored, providing canopy-scale nitrogen input for the subsequent phenologically constrained dynamic scale conversion model.

[0070] This application constructs a canopy nitrogen uptake inversion model using a TabPFN table-based prediction model based on the Transformer architecture. It leverages the inherent tabular, multi-dimensional, and small-sample characteristics of UAV hyperspectral vegetation index data to build a specialized canopy nitrogen uptake inversion model, making the model structure more suitable for the data organization format of agricultural remote sensing monitoring. Compared to traditional random forests, partial least squares regression, or neural networks that rely on repeated training with a large number of samples, this application utilizes a contextual learning mechanism to directly establish the correlation between historical samples. This allows a limited number of field trial samples to fully realize their learning value, improving the model's adaptability in small-sample scenarios of agricultural remote sensing. Furthermore, the training samples are further limited to include not only the vegetation index feature set extracted from UAV hyperspectral data but also background information such as variety information, planting density, nitrogen fertilizer treatment, sampling date, and cumulative growth days. This ensures that each sample not only reflects the canopy spectral response but also the characteristics of plant nitrogen changes under different planting conditions and growth stages, thereby enhancing the model's adaptability to complex field environments. The model boasts superior generalization capabilities. Furthermore, the model labels directly adopt the canopy nitrogen uptake corresponding to the observable canopy leaf layer within the actual observation range of the UAV, ensuring that the model's learning target always corresponds to a stable information source that the sensor can acquire. This avoids the inconsistency between the observed object and the predicted target caused by traditional models directly fitting the nitrogen uptake of the entire plant population. It further reduces data interference caused by canopy shading, spectral saturation, and lower leaf senescence, ensuring that model training always revolves around the canopy spectral information that the UAV can realistically acquire. This guarantees a clear physical correspondence in the model building process. Simultaneously, the tabular-based prediction model can directly read vegetation index feature data without complex data format conversion, improving model deployment efficiency and further reducing the complexity of model training and practical applications. This forms a unified modeling system for the entire canopy nitrogen uptake retrieval process, supported by historical sample knowledge transfer, directly driven by spectral information, and constrained by real observation objects. This provides a stable, consistent, and reliable canopy nitrogen input for subsequent scale conversion relationships, improving the data continuity, model stability, and cross-regional application capability of the entire population nitrogen uptake monitoring method. Once the model is trained, its parameters are fixed. In practical applications, there is no need to retrain it. Only the corresponding vegetation index features need to be input to output the canopy nitrogen uptake.

[0071] In some examples of this application, the input features of the canopy nitrogen uptake inversion model include a set of vegetation index features of the canopy chlorophyll status, nitrogen nutrient status and canopy structure features. This set of vegetation index features includes 22-dimensional vegetation index feature vectors of CIred-edge, DCNI, DVI, EVI, LICI, MCARI, mND705, MNLI, mRER, MTCI, NDRE, NDVI, OREA, OSAVI, REIP, SAVI, SIPI, SR, TCARI, TGI, TVI and VOG2.

[0072] In this application, to further improve the ability of the canopy nitrogen uptake inversion model to express the true nitrogen status of the maize canopy and enhance the applicability of the model under different years, varieties, planting densities, and phenological stages, this application comprehensively selects vegetation index features that can reflect various information such as chlorophyll changes, nitrogen accumulation, canopy structure, leaf area changes, red edge position changes, and pigment composition changes to jointly constitute the model input features. This enables the vegetation index features to form a complementary relationship, improves the comprehensive expression ability of the input features for the true growth status of the canopy, reduces the impact of fluctuations in single spectral features on the inversion results, and improves the stability and reliability of the canopy nitrogen uptake prediction results.

[0073] Preferably, the canopy nitrogen uptake inversion model uses 22 vegetation index features with a fixed composition as input features. These 22 vegetation index features include CIred-edge, DCNI, DVI, EVI, LICI, MCARI, mND705, MNLI, mRER, MTCI, NDRE, NDVI, OREA, OSAVI, REIP, SAVI, SIPI, SR, TCARI, TGI, TVI, and VOG2. These 22 vegetation index features cover multiple key spectral regions, including the visible light band, red-edge band, and near-infrared band. Among these, CIred-edge, MCARI, MTCI, RE... IP, VOG2, mRER, and LICI primarily reflect changes in chlorophyll content and red-edge spectral characteristics; DCNI, NDRE, OREA, and mND705 primarily reflect changes in canopy nitrogen nutrient levels; NDVI, DVI, SR, OSAVI, SAVI, MNLI, EVI, TVI, and TGI primarily reflect plant growth vigor, leaf area, and canopy coverage; and SIPI primarily reflects changes in canopy pigment composition. Each vegetation index reflects the growth status of the maize canopy from different spectral perspectives, collectively forming a comprehensive feature set of information on canopy nitrogen status and canopy structure. All vegetation index features are calculated based on spectral reflectance obtained from UAV hyperspectral imagery. Indicates the corresponding wavelength The spectral reflectance at the location, and the above vegetation index features, can all be directly calculated using hyperspectral data acquired by UAVs. There is no need to add new sensing equipment or new spectral acquisition processes to complete the batch calculation of all input features. This ensures that a consistent data acquisition method can be used to form a standardized model input for different monitoring periods and different monitoring areas. At the same time, it enables the model to use multiple information dimensions to identify changes in the canopy spectrum, and improves the model's ability to express the differences in maize canopy spectrum under different fertilization levels, different growth conditions and different phenological stages.

[0074] The twenty-two vegetation index features are used as input to the TabPFN table prediction model. They are calculated based on the corresponding band reflectance according to the following preset calculation formula. For example, R840 represents the spectral reflectance at 840nm.

[0075]

[0076] After the UAV completes the hyperspectral image acquisition of the target field, the original images are first preprocessed by radiometric calibration, atmospheric correction, geometric correction and canopy region extraction to obtain stable canopy spectral reflectance data. Then, according to the calculation formulas corresponding to the above 22 vegetation index features, each vegetation index feature is calculated and arranged into a fixed-dimensional 22-dimensional vegetation index feature vector in a preset order, which serves as the unified input feature of the canopy nitrogen uptake inversion model. The arrangement order of each vegetation index feature is consistent in the model training stage and the actual prediction stage, thereby ensuring that the historical training samples and real-time monitoring samples adopt the same data organization form and improving the data consistency when the model calls the historical learning results for prediction.

[0077] Furthermore, the 22D vegetation index feature vector serves as the standard input data for the TabPFN table prediction model. The table-based prediction model employs a pre-trained table-based model for small-sample regression adaptation, with an input dimension of 22 and an output dimension of 1. Ten-fold cross-validation is used to evaluate model accuracy, and the random seed is set to 42. When using pre-trained weights for contextual learning regression, the learning rate, number of iterations, or tree depth are not used as necessary training hyperparameters. The table prediction model automatically learns the correspondence between different vegetation index features and observable canopy nitrogen uptake based on historical training samples. Simultaneously, it learns the synergistic and complementary relationships between different vegetation index features. During the prediction process, it comprehensively utilizes the information provided by each vegetation index feature to complete the regression prediction of observable canopy nitrogen uptake. Since all samples to be monitored use a unified dimension, unified calculation method, and unified arrangement order to form input features, it ensures that the model has a consistent data input format across different years, regions, maize varieties, and phenological stages, improving the model's data compatibility and generalization ability when applied across regions, years, and varieties.

[0078] In some examples of this application, the formula for calculating the cumulative growth days is:

[0079]

[0080] in, No. The highest temperature of the day For the first The lowest temperature of the day, The optimal temperature for corn growth is 10℃. The date sequence number between sowing date and monitoring date. This is the cumulative number of days from the sowing date to the monitoring date. As a continuous phenological quantitative index, it is input into the dynamic scale transformation model to characterize the growth and development process of maize and constrain the dynamic scale transformation relationship between canopy nitrogen uptake and population nitrogen uptake.

[0081] This application uses cumulative growth days as a unified phenological constraint parameter in a dynamic scale conversion model. By employing a fixed baseline temperature for maize growth and a standardized daily cumulative calculation method, the phenological constraint parameter has a unified data source, unified calculation rules, and a unified expression form. It no longer relies solely on the number of sowing days, growth stage divisions, or manual experience to determine the maize growth stage. Instead, it uses continuously accumulated effective heat to objectively reflect the entire growth and development process of maize, transforming phenological information from a staged description into a continuously changing numerical variable. This improves the consistency of phenological characterization across different years, regions, and cultivation conditions. Since the cumulative growth days are directly derived from daily meteorological observation data, it avoids subjective differences caused by human judgment of growth stages, ensuring that different monitoring areas can obtain consistent phenological parameters according to a unified standard. Furthermore, it uses 10℃ as the baseline temperature for maize growth. At that time, on that day A value of 0 is used to retain only the effective heat that can promote the normal growth and development of maize in the cumulative calculation, reducing the interference of low temperature environment on phenological parameters and making the obtained cumulative growth days more consistent with the actual growth pattern of maize. The phenological quantitative index formed on this basis can stably reflect the growth process of different developmental stages of the plant, providing unified and continuous phenological constraint parameters for the subsequent dynamic scale conversion model. This allows the canopy-scale nitrogen information to automatically match the corresponding scale conversion relationship according to different developmental stages, further reducing the impact of differences in growth process caused by different sowing dates, different years and different climatic conditions on the population nitrogen estimation results, improving the data comparability between monitoring results in different regions. At the same time, the unified calculation method also makes the entire monitoring process more standardized, which is conducive to long-term continuous monitoring, multi-regional promotion and application, and unified management between different data sources. This improves the data consistency, model stability and engineering application efficiency of the entire monitoring method, provides stable and reliable phenological basic parameters for subsequent dynamic scale conversion, and improves the continuity, accuracy and adaptability of the entire population nitrogen uptake monitoring process. The dynamic scaling transformation model of this application is applicable to spring maize and summer maize, and is preferably applicable to the cumulative growth temperature range of 400~1800℃·d. In other accumulated temperature ranges, the scaling transformation function can also be recalibrated according to the same method. Only the basic parameters need to be recalibrated to extend it to other regions.

[0082] In some examples of this application, the scale transformation relationship satisfies:

[0083]

[0084] in,

[0085] Nitrogen uptake by the plant population;

[0086] The canopy-scale nitrogen index of the observable canopy leaf layer;

[0087] To accumulate growth over time;

[0088] The slope of the scale transformation varies with the cumulative growth degree per day.

[0089] The scale transformation intercept varies with cumulative growth degree per day.

[0090] This application further dynamically defines the scale conversion relationship between canopy-scale nitrogen indices and canopy nitrogen uptake. A unified dynamic conversion model is established, eliminating the use of fixed conversion parameters and dynamically adjusting the scale conversion relationship based on phenological changes at different growth stages of maize. This allows canopy-scale nitrogen information to more accurately reflect the changing patterns of nitrogen uptake in the canopy. By using cumulative growth days as a unified phenological indicator reflecting maize development, the scale conversion relationship changes synchronously with plant growth and development. This makes the correspondence between canopy nitrogen and canopy nitrogen more consistent with the nitrogen accumulation and nutrient translocation patterns throughout the entire growth period of maize, further avoiding systematic biases caused by using a unified conversion relationship across different phenological stages. Simultaneously, the dynamic scale conversion relationship can take into account the rapid leaf growth during the vegetative growth stage, the continuous nitrogen translocation to the ear during the reproductive growth stage, and the gradual senescence of lower leaves. This dynamic scaling method ensures that canopy-scale nitrogen indices maintain a stable correlation with population-scale nitrogen uptake across different physiological processes, improving the consistency and continuity of scaling results across different growth stages. Furthermore, this method automatically adjusts the scaling relationship across different phenological stages using only real-time accumulated growth days, eliminating the need to rebuild new scaling models or perform extensive field sampling and calibration. This reduces repetitive model building and improves the consistency and efficiency of the monitoring process. Canopy-scale nitrogen data obtained from different years, regions, sowing times, and temperature conditions can all be converted to population-scale based on unified phenological constraints, further enhancing the model's adaptability to different environmental and cultivation management conditions. This improves the stability, reliability, and generalization ability of population nitrogen uptake estimation results, providing more stable and reliable data support for subsequent dynamic monitoring of maize population nitrogen, precision fertilization decisions, and cross-regional application. When there are significant changes in varieties, the scaling function can be refitted based on new samples.

[0091] In a preferred embodiment of this application, a scale-transformed sample set is first constructed based on synchronous observation data obtained from historical field trials. Each sample includes the observable canopy nitrogen uptake, canopy nitrogen uptake, and cumulative growth degree days corresponding to the same sampling time. All samples are grouped according to cumulative growth degree days, with each group corresponding to a similar phenological development stage. A linear relationship between the observable canopy nitrogen uptake and canopy nitrogen uptake is established within each cumulative growth degree day stage, yielding the scale-transformed slope coefficient and scale-transformed intercept coefficient for the corresponding stage. The relationship can be expressed as:

[0092]

[0093] in, This represents the s-th cumulative growth day. stage, This indicates the amount of nitrogen absorbed by the canopy and leaves at the corresponding stage. This indicates the amount of nitrogen absorbed by the plant population at the corresponding stage. and Let represent the scale transformation slope coefficient and scale transformation intercept coefficient corresponding to each stage, respectively. From the fitting results of each stage, a set of scale transformation parameters corresponding to different cumulative growth days can be obtained. Furthermore, the slope coefficient and intercept coefficient obtained at each stage are used to establish a functional relationship with the corresponding cumulative growth days, so as to realize the continuous change of scale transformation parameters with the phenological development process. Both the slope coefficient and intercept coefficient can be expressed as a function of the cumulative growth days, i.e., satisfying:

[0094]

[0095] To determine the most suitable functional form for describing the variation of scale transformation parameters, this application compared and fitted linear, power, and quadratic functions. Based on the goodness of fit, prediction error, and model stability of each function, a comprehensive evaluation was conducted, and the functional form with the best fit was selected to establish the final dynamic scale transformation relationship. This ensures that the scale transformation parameters corresponding to different cumulative growth degrees can be continuously determined, ultimately establishing the scale transformation relationship from canopy-scale nitrogen indices to nitrogen uptake by the plant population.

[0096]

[0097] This scale transformation relationship is the observable canopy leaf layer constrained by the cumulative growth degree per day ( Canopy-scale nitrogen index Scale conversion relationship between nitrogen uptake by the population and the amount of nitrogen absorbed by the plants.

[0098] Combination Figure 3The results shown further illustrate the basis for establishing the scale conversion relationship in this application. The figure divides all samples into multiple phenological stages according to cumulative growth days. Different colored scatter points represent sample data within different cumulative growth day intervals. Within each cumulative growth day interval, the observable canopy nitrogen uptake and canopy nitrogen uptake show a good linear correspondence. The fitted lines at each stage have high consistency, indicating a stable linear conversion law between canopy-scale nitrogen and canopy-scale nitrogen within the same phenological stage. Furthermore, as the cumulative growth days increase, the fitted lines corresponding to different stages undergo continuous changes, indicating that the ratio between canopy nitrogen and canopy nitrogen is constantly adjusted during maize growth and development. This change mainly originates from canopy structure evolution, nitrogen redistribution within the plant, and reproductive growth. Physiological processes such as the continuous transport of nutrients to the ear during the growth period make it difficult to simultaneously consider the scale conversion needs of different phenological stages if a unified conversion model with a fixed slope and fixed intercept is established. This can easily lead to systematic estimation errors at different growth stages. However, this application establishes a dynamic scale conversion relationship under the constraint of cumulative growth days, which allows the scale conversion parameters to be continuously adjusted according to the phenological development process. This enables the canopy-scale nitrogen index to be accurately converted into the corresponding population nitrogen uptake throughout the entire growth cycle of maize. This further improves the consistency, stability, and generalization ability of the monitoring results of population nitrogen uptake under different years, regions, sowing times, and cultivation management conditions. At the same time, it provides a unified, stable, and phenologically adaptable dynamic scale conversion basis for subsequent continuous monitoring of maize population nitrogen, precision fertilization decisions, and cross-regional promotion and application.

[0099] In some examples of this application, the scale transformation slope is represented by a linear function of the cumulative growth degree days, satisfying... ;

[0100] The scale transformation intercept is expressed as a power function of the cumulative growth degrees in days, satisfying: ;

[0101] in, , , , , All parameters are fixed parameters obtained by fitting historical field test samples. The linear function and the power function together constitute the dynamic scale conversion parameters in the dynamic scale conversion model that change continuously with the cumulative growth degree days. They are used to determine the scale conversion slope and scale conversion intercept corresponding to the conversion of canopy-scale nitrogen index to population plant nitrogen uptake based on the cumulative growth degree days at the time of monitoring.

[0102] This application further defines the specific expression of the scale transformation slope and scale transformation intercept in the dynamic scale transformation model. This ensures that the scale transformation relationship not only reflects the correspondence between canopy-scale nitrogen indices and nitrogen uptake by the plant population, but also achieves continuous changes in scale transformation parameters across different phenological stages through explicit mathematical functions. Compared to methods using fixed transformation parameters or only empirical piecewise correction, this application uses cumulative growth days as a unified driving variable, establishing the scale transformation parameters as functions that continuously change with phenology. This gives the entire scale transformation process a unified mathematical expression and continuous variation characteristics, avoiding parameter jumps caused by artificial division of growth stages. Furthermore, it allows maize from different monitoring dates, different climatic conditions, and different regions to complete scale transformations using the same computational framework, improving the consistency of model calls across different application scenarios. Additionally, all function parameters are derived from historical data. The model is obtained by fitting field trial samples. After the model is established, there is no need to re-establish the scale transformation relationship for new monitoring areas. The corresponding scale transformation parameters can be automatically obtained based on the cumulative growth days calculated in real time, realizing the automated calculation of the scale transformation process, improving the efficiency of model deployment and practical application. At the same time, the function expression is simple and clear, which can be directly embedded into UAV remote sensing monitoring platforms, agricultural information systems and field smart terminals, realizing the standardized calculation of the scale transformation process, improving the consistency of algorithm calls between different systems, and further ensuring that canopy-scale nitrogen data can be stably converted into population nitrogen uptake. This improves the consistency, stability and reliability of population nitrogen estimation results under different years, different sowing times, different planting densities and different climatic conditions, and provides a unified, standardized and easy-to-promote dynamic scale transformation parameter expression method for the continuous monitoring of nitrogen uptake in the entire maize population.

[0103] In some examples of this application, the canopy nitrogen uptake inversion model and the phenologically constrained dynamic scale conversion model are executed sequentially according to a fixed data processing flow. First, hyperspectral image data of the target maize planting area is acquired using a UAV, and hyperspectral image preprocessing and vegetation index feature calculation are performed. The extracted vegetation index features are then input into the trained canopy nitrogen uptake inversion model to obtain the observable canopy nitrogen uptake of the target field or plot. Subsequently, daily meteorological data from the sowing date to the monitoring date in the target area are acquired, the cumulative growth days for the corresponding monitoring date are calculated, and the scale conversion slope and scale conversion intercept corresponding to the current cumulative growth days are determined based on the established phenologically constrained dynamic scale conversion model. Finally, the observable... The nitrogen uptake of the canopy and leaves, along with the corresponding scale transformation parameters, are substituted into the scale transformation relationship to obtain the nitrogen uptake of the population at the target field or plot scale. This completes the dynamic conversion from UAV hyperspectral remote sensing information to population nitrogen uptake, achieving a continuous mapping between canopy-scale and population-scale information. The final output is the monitoring results of population nitrogen uptake in the target area. These monitoring results can be further used for maize nitrogen nutrition status diagnosis, variable topdressing decision-making, fertilization effect evaluation, dynamic monitoring of nutrients during the growth period, yield formation process analysis, and precision agricultural management. This provides continuous, stable, and phenologically constrained data support for precision nitrogen management in the field, and also provides a reliable monitoring basis for nutrient regulation and production decisions throughout the entire maize growth period.

[0104] In some examples of this application, the method is also applicable to gramineous crops with a layered canopy structure, which, in addition to maize, can also include wheat or rice, wherein the maize canopy is divided along the plant height direction into , and The layer With layers Constituting the observable canopy leaf layer ( ), An unobservable canopy is formed, while the observable canopy leaf layer is used to establish a canopy nitrogen uptake inversion model. Combined with the cumulative growth days, the dynamic scale conversion of nitrogen uptake by the population plants is completed. For other gramineous crops with stratified canopy structures, only the growth base temperature, canopy stratification standard, and scale conversion parameters of the corresponding crop need to be calibrated, and the same monitoring process can be used to complete the monitoring of population nitrogen uptake.

[0105] This application further specifies that the monitoring method is applicable to gramineous crops with a layered canopy structure, with maize as the preferred application. Maize's canopy naturally forms three relatively stable spatial layers: upper leaf layer, middle leaf layer, and lower leaf layer. When a drone equipped with a hyperspectral sensor acquires canopy reflectance information, the upper and middle leaf layers can directly receive solar radiation and provide the sensor with the main spectral information; therefore, they are collectively defined as the observable canopy leaf layer. The lower leaf layer, due to long-term shading by the upper or middle leaf layers, has less effective spectral information that can be directly acquired by the drone, and therefore does not directly participate in canopy-scale nitrogen inversion. Instead, it undergoes overall compensation in conjunction with phenological information during subsequent dynamic scale conversion, thereby forming the canopy inversion. The monitoring mechanism combined with scale transformation allows for the determination of corresponding stratification boundaries, basic growth parameters, and dynamic scale transformation parameters for gramineous crops such as wheat and rice, which also have stratified canopy structures. This ensures consistency across all processing steps, including UAV image acquisition, vegetation index feature extraction, canopy nitrogen inversion, phenological constraints, and population nitrogen calculation. The entire technical framework revolves around the stratified canopy structure for information acquisition, feature extraction, and scale transformation, achieving a data processing flow and monitoring method that is essentially consistent across different crops. This enables canopy spatial structure, growth and development processes, and nitrogen distribution patterns to jointly participate in the estimation of population nitrogen uptake, improving the continuity, consistency, and scalability of the entire monitoring process.

[0106] This application also discloses a monitoring device for nitrogen uptake in maize populations based on UAV hyperspectral and phenological constraints, used to perform the nitrogen uptake monitoring method for maize populations described in the aforementioned embodiments. The device includes a hyperspectral image acquisition module, a preprocessing and feature calculation module, a canopy nitrogen inversion module, a phenological accumulated temperature calculation module, a scale conversion module, and a monitoring result output module. The modules work together to form a monitoring system for nitrogen uptake in maize populations based on UAV hyperspectral, phenological constraints, and dynamic scale conversion.

[0107] The hyperspectral image acquisition module is used to acquire UAV hyperspectral image data of the target corn planting area. Specifically, it controls the UAV to complete field flight and acquisition according to the preset route, and acquires hyperspectral images covering the visible light and near-infrared bands. At the same time, it acquires the corresponding flight altitude, attitude information, time information and geographical location information to provide raw data for subsequent canopy nitrogen inversion.

[0108] The preprocessing and feature calculation module is used to complete hyperspectral image preprocessing and vegetation feature extraction. Specifically, it performs radiometric calibration, atmospheric correction, reflectance correction, geometric correction, image stitching and orthorectification on the acquired hyperspectral images in sequence, and extracts the target area based on the boundaries of the experimental field or the target plot. Furthermore, it calculates a preset vegetation index feature set based on the processed spectral reflectance. The vegetation index feature set includes twenty-two-dimensional vegetation indices, which are used to characterize the spectral response characteristics of the maize canopy and provide input features for the inversion of canopy nitrogen uptake.

[0109] The canopy nitrogen inversion module is used to call the tabular-based prediction model described in any of the aforementioned embodiments to invert canopy nitrogen uptake based on the vegetation index features output by the preprocessing and feature calculation module. Specifically, the trained TabPFN prediction model is used to predict the observable canopy leaf nitrogen uptake and output the canopy nitrogen uptake corresponding to the observable canopy leaf layer. The canopy nitrogen inversion module can also call the corresponding model parameters according to different monitoring areas to achieve consistent prediction of canopy nitrogen uptake in different monitoring areas.

[0110] The phenological accumulated temperature calculation module is used to calculate the cumulative growth degree days based on the daily meteorological data of the target monitoring area. Specifically, it obtains the daily maximum and minimum temperatures between the sowing date and the monitoring date, calculates the effective accumulated temperature of each day based on the preset base temperature, and completes the daily accumulation to obtain the cumulative growth degree days for the corresponding monitoring date. This is used to quantify the current phenological development status of maize and provide phenological constraint parameters for dynamic scale conversion.

[0111] The scale conversion module is used to perform dynamic conversion from canopy scale to population scale based on cumulative growth days. Specifically, it calculates the corresponding scale conversion slope and scale conversion intercept based on the cumulative growth days, and then combines the observable canopy nitrogen absorption corresponding to the canopy leaf layer output by the canopy nitrogen inversion module to calculate the population plant nitrogen absorption according to the scale conversion relationship, thereby realizing the dynamic mapping of canopy scale nitrogen information to population scale nitrogen information.

[0112] The monitoring result output module is used to output the monitoring results of nitrogen uptake by maize population plants. Specifically, it is used to generate spatial distribution maps of nitrogen uptake by the population at the field or plot scale, statistical analysis results, monitoring reports and variable fertilization decision data, and can output to agricultural management platforms, agricultural information systems, smart agriculture cloud platforms or mobile terminals in a preset format to realize the visualization, storage management and sharing application of maize nitrogen nutrition status monitoring results.

[0113] The hyperspectral image acquisition module, preprocessing and feature calculation module, canopy nitrogen inversion module, phenological accumulated temperature calculation module, scale conversion module, and monitoring result output module are uniformly scheduled and coordinated through the control unit in the processor, digital signal processor, edge computing terminal, embedded controller, cloud computing server, or agricultural remote sensing data processing platform. Each module is executed cyclically in the order of hyperspectral data acquisition, data preprocessing, canopy nitrogen inversion, phenological accumulated temperature calculation, dynamic scale conversion, and monitoring result output, thereby forming a complete closed-loop monitoring system covering UAV data acquisition, canopy nitrogen inversion, phenological dynamic constraints, and population nitrogen uptake calculation, realizing rapid, non-destructive, and dynamic monitoring of nitrogen uptake in maize populations.

[0114] This application also discloses a monitoring device for nitrogen uptake in maize populations based on UAV hyperspectral and phenological constraints, including at least one processor, a memory, and a data acquisition interface, a meteorological data interface, a network communication interface, and a monitoring result output interface that are communicatively connected to the processor;

[0115] The memory stores a computer program, which, when executed by the processor, causes the processor to perform the method for monitoring nitrogen uptake in maize populations as described in any of the foregoing embodiments, including steps such as UAV hyperspectral image acquisition, hyperspectral image preprocessing, vegetation index feature calculation, TabPFN canopy nitrogen uptake inversion, cumulative growth degree-day calculation, dynamic scale conversion, and output of nitrogen uptake in the population plants.

[0116] The monitoring equipment can be deployed on agricultural remote sensing monitoring platforms, smart agricultural management platforms, agricultural digital management systems, agricultural Internet of Things platforms, edge computing terminals, UAV ground stations, agricultural big data platforms, or cloud-based remote sensing analysis platforms to achieve online monitoring, real-time calculation, result analysis, and intelligent management of nitrogen uptake in maize populations.

[0117] This application also discloses a computer-readable storage medium storing a processor-executable computer program, which, when executed by a processor, performs the steps of the method for monitoring nitrogen uptake in maize populations as described in any of the foregoing embodiments.

[0118] The computer-readable storage medium includes, but is not limited to, read-only memory (ROM), random access memory (RAM), flash memory (Flash), solid-state drive (SSD), embedded multimedia memory (eMMC), storage server, network storage device, distributed storage system, or other storage medium capable of storing program code and allowing a processor to read and execute it.

[0119] By storing the corresponding program code in the computer-readable storage medium, the processor is able to perform functions such as UAV hyperspectral image acquisition, hyperspectral data preprocessing, vegetation index extraction, canopy nitrogen uptake inversion, cumulative growth degree-day calculation, phenological dynamic constraints, scale conversion parameter calculation, population plant nitrogen uptake estimation, and monitoring result output, thereby realizing fully automated monitoring of maize population nitrogen uptake based on UAV hyperspectral and phenological constraints.

[0120] This application constructs a hyperspectral image acquisition module, a preprocessing and feature calculation module, a canopy nitrogen inversion module, a phenological accumulated temperature calculation module, and a scale conversion module. It organically combines UAV hyperspectral remote sensing information, maize phenological development information, and the dynamic distribution patterns of nitrogen in the canopy. This enables continuous dynamic conversion of canopy-scale nitrogen information to canopy-scale nitrogen information, overcoming the problem that traditional static scale conversion models cannot adapt to changes in canopy structure and nitrogen transport patterns within plants at different growth stages. Simultaneously, it fully leverages the advantages of UAV hyperspectral monitoring—high efficiency, wide coverage, non-destructive testing, and strong repeatability—improving the accuracy, stability, and cross-year adaptability of canopy nitrogen uptake monitoring results. This provides reliable data support for precision fertilization of maize, nitrogen fertilizer use efficiency evaluation, farmland nutrient diagnosis, and smart agriculture precision management. Furthermore, by adjusting crop parameters, it can be further extended to monitoring scenarios of other gramineous crops with layered canopy structures, demonstrating good engineering application value and promising prospects for wider application.

[0121] Example 1

[0122] This embodiment discloses a specific monitoring scheme for nitrogen uptake in maize populations.

[0123] Experimental data source:

[0124] The experiment was conducted in Henan Province and Inner Mongolia Autonomous Region from 2020 to 2025. The tested maize varieties included JNK728, DK159, ZD958, R1852, and T397. Four planting density levels were set up: 52.5 × 10⁶. 3 60×10 3 75×10 3 and 90×10 3 Plants per hectare.

[0125] During the key growth stages of maize (jointing stage V6, large trumpet stage V12, tasseling stage VT, grain filling stage R3, and maturity stage R6), UAV hyperspectral flight and destructive sampling were carried out respectively. The UAV platform was equipped with a pushbroom hyperspectral imager (spectral range 400-1000nm), and the flight was carried out in clear, cloudless weather at an altitude of about 100m.

[0126] For destructive sampling, three representative plants were selected from each plot, harvested by leaf layer, and the stems and panicles were separated. All samples were blanched at 105℃ and dried at 85℃ to constant weight, and the dry weight and total nitrogen content were determined (Kjeldahl method).

[0127] Step S1: Constructing an inversion model for canopy nitrogen uptake

[0128] (1.1) Perform radiometric correction, geometric correction and plot extraction on the UAV hyperspectral imagery to obtain the average canopy reflectance spectrum of each plot;

[0129] (1.2) Calculate the 22 vegetation index characteristics (see parameter and formula description) to form a feature matrix.

[0130] (1.3) The model was adapted using the canopy nitrogen uptake as the target variable and historical field test samples. The measured value of each canopy nitrogen uptake was the target variable, and 22 vegetation index features were the input features. The model was trained using the TabPFN table and the accuracy was evaluated by 10-fold cross-validation.

[0131] (1.4) Training Dataset Construction Strategy: Based on multi-year, multi-location maize field trial data from 2020 to 2025, the average reflectance of plots, vegetation index, variety, density, nitrogen fertilizer treatment, sampling date, and cumulative growth degree daily information extracted from UAV hyperspectral images were compared with those obtained from destructive sampling on the same day. , , and , , Leaf nitrogen uptake was sample-level matched; among which... Canopy nitrogen uptake is used as a model label. Data used for model fitting may include samples from 2020, 2023, and 2024, while data used for independent validation may include samples from 2021, 2022, and 2025.

[0132] (1.5) Model input content: Each row of the input feature matrix X is a field or plot sample, and each column is a vegetation index feature, specifically including CIred-edge, DCNI, DVI, EVI, LICI, MCARI, mND705, MNLI, mRER, MTCI, NDRE, NDVI, OREA, OSAVI, REIP, SAVI, SIPI, SR, TCARI, TGI, TVI and VOG2, for a total of 22 dimensions; when sample screening or validation grouping is required, year, location, variety, density, nitrogen fertilizer treatment and cumulative growth days are used as sample metadata to participate in data partitioning and result interpretation.

[0133] (1.6) Model output: The output vector Y represents the observable canopy nitrogen uptake in the leaf layer, in kg / ha. In the field application stage, the model performs regression prediction on the 22-dimensional vegetation index vector of the field or plot to be monitored, and obtains the canopy nitrogen uptake at the corresponding field or plot scale. Estimated value.

[0134] (1.7) Model training process and hyperparameter settings: The training set feature matrix and labels are input into TabPFNRegressor for small-sample regression adaptation, and the validation set feature matrix is ​​input into the adapted model to obtain the predicted value of canopy nitrogen uptake; the model input dimension is set to 22, the output dimension is set to 1, the random seed is set to 42, the cross-validation fold number is set to 10, and the evaluation metric is R. 2 RMSE and rRMSE (relative root mean square error). The TabPFN model uses pre-trained tabular base model weights and completes regression prediction based on a context learning mechanism. In this embodiment, gradient update training for this task is not performed. Therefore, the learning rate, number of iterations, tree depth, or number of trees are not necessary training hyperparameters.

[0135] (1.8) Data processing: Before model training, dark current correction, radiometric calibration, geometric correction, stitching and cell area extraction are performed on the UAV hyperspectral images, and the average reflectance of each cell is calculated; samples with missing ground labels, obvious abnormal reflectance or that cannot be matched with the sampling records are removed; a 22-dimensional feature matrix is ​​constructed according to the vegetation index formula, and the canopy nitrogen uptake label is organized into a one-dimensional regression target.

[0136] (1.9) Scene integration and technical effect: The above model training process is designed for agricultural remote sensing scenes that mainly represent the upper and middle leaf layers in UAV vertical observation, so that the output target of the model is consistent with the actual observation object of the sensor; by first predicting the nitrogen uptake of the canopy and then combining it with the cumulative growth degree days to perform scale conversion of the nitrogen uptake of the population plants, the phenological deviation and spectral saturation error caused by directly fitting the nitrogen uptake of the population plants by the canopy spectrum are avoided.

[0137] Step S2: Construct the GDD-PNU scaling model

[0138] (2.1) Base temperature for maize growth Set the temperature to 10°C, and calculate using the formula for cumulative growth days:

[0139] Calculate the cumulative growth days for each sampling day, where and For respectively the first The highest and lowest temperatures of the day, This represents the cumulative number of days from the sowing date to the sampling date.

[0140] (2.2) The sampling data were grouped according to the cumulative growth period days, and the canopy nitrogen uptake and the population nitrogen uptake were established within each group. Linear regression was performed to obtain the scale transformation slope and scale transformation intercept.

[0141] To establish a dynamic scale conversion relationship between canopy-scale nitrogen uptake and canopy-scale nitrogen uptake, measured data of canopy-scale nitrogen uptake and canopy-scale nitrogen uptake obtained synchronously from multi-year, multi-location field trials were first used. Data were grouped according to the cumulative growth days corresponding to each sampling date. A linear regression model between canopy-scale nitrogen uptake and canopy-scale nitrogen uptake was established for each cumulative growth day period to obtain the scale conversion slope and intercept for that period. Then, the slope and intercept parameters obtained from different cumulative growth day periods were correlated with the corresponding cumulative growth days, and linear, power, and quadratic functions were used for fitting analysis. By comparing the goodness of fit and parameter stability of different function models, the optimal functional relationship between the scale conversion parameters and the cumulative growth days was determined.

[0142] This embodiment performed function fitting analysis on the nitrogen index of the upper canopy, the observable nitrogen index of the canopy leaves, and the nitrogen index of the entire canopy leaves. The fitting results are as follows: Figure 4 and Figure 5 As shown, where Figure 4 This represents the fitting result of the scale transformation slope varying with the daily cumulative growth degree. Figure 5 The fitting results represent the daily variation of the scale transformation intercept with cumulative growth degree. Each fitting group is compared simultaneously with three models: linear function, power function, and quadratic function, with the coefficient of determination R0. 2 As a model evaluation metric, different function forms are optimized.

[0143] The fitting results show that, for the observable canopy leaf nitrogen index used in this application, its scale transformation slope exhibits a stable linear variation with the cumulative growth degree per day. The linear function achieves high fitting accuracy and also possesses advantages such as simple expression, stable parameters, and high computational efficiency. Therefore, a linear function is chosen to describe the relationship between the scale transformation slope and the cumulative growth degree per day, and its expression is:

[0144]

[0145] Among them, the following was determined by fitting field trial data: , ;

[0146] This corresponds to the scale transformation slope function: ;

[0147] Furthermore, for the scale transformation intercept parameter, the fitting results show that the power function can more accurately describe its nonlinear change trend with the increase of cumulative growth degree days. Therefore, the power function is preferred to establish the relationship between the scale transformation intercept and the cumulative growth degree days, and its expression is: ;

[0148] Among them, the following was obtained through fitting: , , ;

[0149] Therefore, the scale transformation intercept function is formed accordingly: ;

[0150] Substituting the above scale transformation slope function and scale transformation intercept function into the dynamic scale transformation model, the scale transformation relationship used in the preferred embodiment of this application can be obtained:

[0151]

[0152] Right now:

[0153]

[0154] in,

[0155] Nitrogen uptake by the plant population;

[0156] The canopy-scale nitrogen index of the observable canopy leaf layer;

[0157] To accumulate growth over time;

[0158] The above parameters are obtained by least squares regression fitting of synchronous field sampling data of maize from multiple years, locations and varieties. Those skilled in the art can refit the corresponding parameters based on experimental data from different regions. Therefore, the above parameters are only preferred embodiments and do not constitute a limitation on the scope of protection of this application.

[0159] (2.3) The relationship between the scale transformation slope and the cumulative growth degree per day was fitted with a linear function, and the relationship between the scale transformation intercept and the cumulative growth degree per day was fitted with a power function. The results were then analyzed using R... 2 Determine the optimal function form.

[0160] (2.4) As a control, the VI-PNU empirical model directly fits the nitrogen uptake of the population using the same 22 vegetation index features (skipping the calculation of canopy nitrogen uptake and cumulative growth degree diurnal relationship), and is trained using the TabPFN table-based model. The accuracy is evaluated by 10-fold cross-validation, and R is obtained. 2 =0.25, rRMSE=36.12%.

[0161] To verify the effectiveness of the proposed method for monitoring nitrogen uptake in maize populations, the GDD-PNU dynamic scale conversion model established in this application was compared with the traditional VI-PNU static model based on empirical relationships of fixed vegetation indices. During the verification process, the measured nitrogen uptake per plant in the field was used as a reference value. Both the proposed model and the traditional empirical model were used to estimate the nitrogen uptake per plant in the population, and the coefficient of determination R0 was used. 2 The relative root mean square error (rRMSE) is used as a model evaluation index, where the coefficient of determination R... 2 The coefficient of determination R is used to evaluate the agreement between model predictions and measured values. 2 A larger value indicates a stronger model fit. The relative root mean square error (rRMSE) is used to evaluate the overall prediction error of the model. A smaller rRMSE indicates higher model prediction accuracy. The model accuracy verification results are as follows:

[0162]

[0163] like Figure 6 As shown, the prediction results of the GDD-PNU dynamic scaling transformation model established in this application have a high consistency with the measured values, with a coefficient of determination of 0.63 and a relative root mean square error of 25.23%. The predicted scatter points are mainly distributed near the 1:1 reference line, and the model maintains good prediction stability under different nitrogen levels. In contrast, the coefficient of determination R of the traditional VI-PNU empirical model is much lower. 2 The relative root mean square error (rRMSE) was only 0.25, while the relative root mean square error (rRMSE) reached 36.12%, indicating a significant increase in the dispersion of the predicted scatter points. Large systematic biases appeared in the medium-to-high nitrogen level region, suggesting that using fixed empirical relationships for population nitrogen estimation is difficult to adapt to the scale changes caused by the canopy structure and nitrogen redistribution within the plant at different phenological stages of maize.

[0164] Furthermore, to verify the stability of the model under different application conditions, this application conducted transfer validation for different cumulative growth degree daily intervals, different planting densities, different nitrogen application levels, different maize varieties, different experimental years, and different experimental locations, and statistically analyzed the model prediction errors under each condition. Figure 7As shown, under various validation conditions, the prediction error of the model in this application is significantly lower than that of the traditional empirical model. Specifically, under different cumulative growth degree day intervals, the relative root mean square error of the model in this application remains between 22.3% and 25.9%, around 23% under different planting densities, and around 20% to 30% under different nitrogen application treatments. The prediction accuracy remains relatively stable across different varieties, years, and locations, with a small overall error fluctuation. In contrast, the traditional empirical model exhibits larger error fluctuations under the same conditions, especially with more significant increases in error under different years, planting locations, and high accumulated temperature stages. This indicates that the fixed empirical model is highly sensitive to environmental and phenological changes.

[0165] The above verification results show that by introducing the cumulative growth days into the dynamic conversion process from the canopy scale to the population scale, this application enables the scale conversion parameters to be continuously adjusted with the phenological development of maize. This effectively compensates for the scale conversion errors caused by increased canopy closure, intensified spectral saturation, and the continuous translocation of nitrogen from the plant interior to the stems and ears. This allows observable nitrogen information from the canopy to be mapped more accurately to the nitrogen uptake of the population. Therefore, compared to traditional population nitrogen estimation models established using fixed empirical relationships, this application's model not only has higher prediction accuracy but also exhibits good stability and generalization ability under different accumulated temperature conditions, planting densities, nitrogen application levels, maize varieties, experimental years, and ecological regions. It can meet the practical needs of continuous monitoring of nitrogen uptake by the population throughout the entire maize growth period and its cross-regional application, providing more reliable technical support for precision fertilization, variable fertilization decision-making, nitrogen fertilizer use efficiency evaluation, and monitoring of maize yield formation.

[0166] By adopting the above-mentioned method for monitoring nitrogen uptake in maize populations, the following technical effects are achieved:

[0167] (1) Significantly improved inversion accuracy. The inversion target was limited to the observable canopy and leaf layer, matching the actual observation capability of the sensor; the TabPFN table-based model was introduced, which showed better generalization ability than traditional machine learning models under small sample conditions, with a determination coefficient R. 2 The value increased from 0.25 to 0.63, and the relative root mean square error (rRMSE) decreased from 36.12% to 25.23%.

[0168] (2) Significantly improved mobility. The scale transformation model constrained by GDD dynamically described the changes in nitrogen allocation from the canopy to the canopy. The rRMSE remained stable at 22.3%-25.9% at different phenological stages, approximately 24% at different densities, and stable at 23.9%-25.1% across years and locations.

[0169] (3) Strong applicability. It is especially suitable for scenarios where the canopy is closed in the middle and late stages, the spectrum is saturated and nitrogen is translocated to the ear, providing a more robust data foundation for precise nitrogen application decisions.

[0170] Example 2

[0171] To further verify the applicability of the proposed maize population nitrogen uptake monitoring method under actual production conditions, this embodiment selects different maize experimental fields for field application verification. During the monitoring process, firstly, during the critical growth period of maize, a drone equipped with a hyperspectral imaging device was used to complete the flight operation of the target field according to a preset route, acquiring hyperspectral image data covering the entire experimental area, and simultaneously recording the corresponding monitoring date and field information. Subsequently, the acquired hyperspectral images were preprocessed sequentially, including radiometric calibration, atmospheric correction, reflectance correction, image stitching, geometric correction, and field area extraction. Based on the aforementioned embodiment, the 22D vegetation index features corresponding to the target field were calculated, and the vegetation index features were input into the trained TabPFN canopy nitrogen uptake inversion model to obtain the observable canopy nitrogen uptake of the target field. At the same time, the cumulative growth degree days corresponding to the monitoring date were calculated based on the daily maximum and minimum temperatures between the sowing date and the monitoring date in the target area, and the corresponding scale conversion parameters were calculated based on the dynamic scale conversion model. and The obtained canopy nitrogen uptake is then substituted into the scale transformation relationship of population plant nitrogen uptake to output the population plant nitrogen uptake corresponding to the target field or target experimental plot. This allows for remote sensing monitoring of nitrogen uptake in maize populations.

[0172] like Figure 8 As shown, the left side displays the monitoring results for the early stage of maize growth, with a cumulative growth degree-day (°C·d) of approximately 923°C·d corresponding to the monitoring date. At this stage, the overall nitrogen uptake of the plant population is at a low level, and spatial differences already exist between different areas within the field. The right side displays the monitoring results for the middle and late stages of maize growth, with a cumulative growth degree-day (°C·d) of approximately 1492°C·d corresponding to the monitoring date. As the plants continue to grow and nitrogen accumulates, the overall nitrogen uptake of the plant population significantly increases, and the spatial distribution characteristics of nitrogen in different areas become clearer, intuitively reflecting the differences in nutrient accumulation and growth consistency within the field. The monitoring results are visualized using a spatial distribution map, with different colors corresponding to different nitrogen uptake levels in different plant populations. This allows for quick identification of areas with high and low nitrogen uptake, providing field managers with intuitive spatial distribution information.

[0173] This embodiment demonstrates that the monitoring method proposed in this application can monitor nitrogen uptake by a large-area field population in a single UAV flight, and dynamically correct the canopy-scale information by combining cumulative growth days, thereby achieving continuous estimation of nitrogen uptake by the population at different phenological stages. The monitoring results can not only accurately reflect the nitrogen accumulation level of maize at different growth stages, but also effectively reveal the spatial distribution differences of nitrogen within the field. It can be further used for maize nitrogen nutrition status diagnosis, variable fertilization zoning, topdressing timing determination, fertilization effect evaluation, population growth analysis, yield formation process monitoring, and precision agricultural management, providing reliable data support for precise nutrient regulation throughout the entire growth period of maize. At the same time, it verifies that the method of this application has good engineering application capabilities, stability, and promotion and application value in actual farmland monitoring scenarios.

[0174] The embodiments of this application have been described above with reference to the accompanying drawings. Unless otherwise specified, the embodiments and features in the embodiments of this application can be combined with each other. This application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of this application without departing from the spirit and scope of the claims, and all of these forms are within the protection scope of this application.

Claims

1. A method for monitoring nitrogen uptake in maize populations, characterized in that, include: Based on the vertical structure of the corn canopy, the observable canopy leaf layers that can be directly observed by the UAV canopy optical sensor are determined, and the canopy nitrogen absorption corresponding to the observable canopy leaf layers is used as the canopy-scale nitrogen index. The cumulative growth degree days are calculated based on the daily temperature data from the sowing date to the monitoring date in the target corn planting area. The formula for calculating the cumulative growth days is as follows: ； in, No. The highest temperature of the day For the first The lowest temperature of the day, The basic temperature for corn growth, The date sequence number between sowing date and monitoring date. This is the cumulative number of days from the sowing date to the monitoring date; Acquire UAV hyperspectral images of the target cornfield, preprocess the hyperspectral images and extract a set of vegetation index features; The canopy nitrogen uptake inversion model, which takes the vegetation index feature set as input and the canopy-scale nitrogen index as output, is invoked to obtain the canopy-scale nitrogen index based on the vegetation index feature set. The scale conversion relationship between the canopy-scale nitrogen index constrained by the cumulative growth days and the nitrogen uptake of the population plants is invoked, and the nitrogen uptake of the population plants of the target maize field is obtained based on the canopy-scale nitrogen index and the cumulative growth days. The scale transformation relationship satisfies: ； in, Nitrogen uptake by the plant population; The canopy-scale nitrogen index of the observable canopy leaf layer; The scale transformation slope varies with cumulative growth degrees per day, and is expressed as a linear function of cumulative growth degrees per day, satisfying the following conditions: ; The scale transformation intercept varies with cumulative growth degrees in days, and is expressed as a power function of the cumulative growth degrees in days, satisfying: ; in, , , , , All parameters were fixed and obtained by fitting historical field trial samples.

2. The method for monitoring nitrogen uptake in maize populations according to claim 1, characterized in that, The observable canopy leaf layer includes all the canopy leaf layers during the vegetative growth stage, and includes the upper and middle leaf layers that can be directly observed by the UAV canopy optical sensor from the top of the canopy downwards during the reproductive growth stage, excluding the shaded lower leaf layer.

3. The method for monitoring nitrogen uptake in maize populations according to claim 1, characterized in that, The canopy nitrogen uptake inversion model is constructed using a tabular prediction model based on the Transformer architecture. The tabular prediction model is the TabPFN model, which uses the set of vegetation index features extracted from UAV hyperspectral images as input features and the canopy nitrogen uptake corresponding to the observable canopy leaf layer as the target variable. The model is trained using historical field test samples.

4. The method for monitoring nitrogen uptake in maize populations according to claim 1, characterized in that, The vegetation index feature set includes 22-dimensional vegetation index feature vectors of CIred-edge, DCNI, DVI, EVI, LICI, MCARI, mND705, MNLI, mRER, MTCI, NDRE, NDVI, OREA, OSAVI, REIP, SAVI, SIPI, SR, TCARI, TGI, TVI, and VOG2.

5. A monitoring system for nitrogen uptake in maize populations based on UAV hyperspectral and phenological constraints, characterized in that, include: The hyperspectral image acquisition module is used to acquire UAV hyperspectral images of the target cornfield. The preprocessing and feature calculation module is used to preprocess the hyperspectral image and extract a set of vegetation index features; The canopy nitrogen inversion module is used to calculate the canopy nitrogen uptake corresponding to the observable canopy leaf layer based on the vegetation index feature set. The phenological accumulated temperature calculation module is used to calculate the cumulative growth days based on daily temperature data; The scale conversion module is used to establish the scale conversion relationship between the canopy-scale nitrogen index and the nitrogen uptake of the population plants based on the cumulative growth days, and to calculate the nitrogen uptake of the population plants. The monitoring results output module is used to output the monitoring results of nitrogen uptake by plant populations at the field or plot scale. Each module works in concert to execute the method for monitoring nitrogen uptake in maize populations as described in any one of claims 1 to 4.

6. A device for monitoring nitrogen uptake in maize populations based on UAV hyperspectral and phenological constraints, characterized in that, The system includes a processor, a memory, and a computer program stored in the memory, wherein when executed by the processor, the computer program causes the processor to perform the method for monitoring nitrogen uptake in maize populations as described in any one of claims 1 to 4.

7. A computer-readable storage medium, characterized in that, It stores a computer program, which, when executed by a processor, causes the processor to perform the method for monitoring nitrogen uptake in maize populations as described in any one of claims 1 to 4.

Images