A comprehensive evaluation method for drought resistance of maize varieties

Abstract

This invention discloses a comprehensive evaluation method for drought resistance of maize varieties, belonging to the field of maize variety evaluation technology. The method includes the following steps: acquiring multi-temporal, multi-band reflectance spectral data under continuous drought conditions, and combining this with the structural characteristics of the waxy coating on maize leaf epidermis to perform structural coupling modeling of the spectral data, constructing a time-series curve characterizing the dynamic changes in leaf wax distribution as a dynamic spectral monitoring baseline; performing differential analysis between consecutive time points based on the dynamic spectral monitoring baseline to identify spectral abrupt change points caused by leaf wax migration, and combining this with spatial consistency testing. This invention accurately identifies wax migration interference by constructing a dynamic spectral monitoring baseline and energy shift fingerprint, overcoming the assumption of spectral stability, and achieving high accuracy, adaptability, and consistency in evapotranspiration rate inversion under drought conditions through multi-level correction and closed-loop update strategies, thereby improving the scientific rigor and comparability of drought resistance evaluation.

Description

Technical Field

[0001] This invention relates to the field of maize variety evaluation technology, specifically to a comprehensive evaluation method for the drought resistance of maize varieties. Background Technology

[0002] Comprehensive drought resistance evaluation of maize varieties refers to a systematic and quantitative analysis of maize's growth and development, yield formation, physiological and biochemical responses, and water use efficiency under drought stress, conducted under unified experimental conditions and a scientific evaluation system. This analysis is achieved through field planting trials, controlled environmental trials, and relevant physicochemical index testing. The results are then combined with different indicator weights to form a comprehensive score or grade classification, objectively reflecting the drought adaptability of the varieties. The aim is to scientifically screen and identify superior varieties that can maintain high productivity and stability under drought conditions, providing a reliable basis for maize variety promotion, regional layout optimization, drought-resistant breeding and selection, and the development of dryland agriculture. This will reduce the risk of yield reduction caused by drought and enhance food security.

[0003] The existing technology has the following shortcomings:

[0004] In existing technologies, methods for retrieving evapotranspiration rates based on optical sensors typically assume that the distribution of leaf epidermal wax remains relatively stable before and after drought stress. However, under sustained drought conditions, the leaf epidermal wax in some maize varieties migrates from the interveinal to the vein margin or from the leaf tip to the leaf base, causing abrupt changes in leaf surface reflectivity within a short period. This change in optical properties directly interferes with the sensor's analysis of the leaf surface energy exchange process, leading to abnormal shifts in evapotranspiration rate retrieval parameters. This results in severely distorted water consumption estimates, ultimately affecting the accuracy and comparability of comprehensive drought resistance evaluations.

[0005] The information disclosed in the background section is only intended to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0006] The purpose of this invention is to provide a comprehensive evaluation method for drought resistance of maize varieties, so as to solve the problems in the background art mentioned above.

[0007] To achieve the above objectives, the present invention provides the following technical solution: a method for comprehensive evaluation of drought resistance in maize varieties, comprising the following steps:

[0008] Establish a dynamic spectral monitoring baseline: By acquiring multi-temporal and multi-band reflectance spectral data under continuous drought conditions in maize, and combining it with the structural characteristics of the epidermal wax of maize leaves, structural coupling modeling of the spectral data is performed to construct a time series curve that can characterize the dynamic changes in the distribution of leaf wax, which serves as the dynamic spectral monitoring baseline.

[0009] Extraction of wax migration interference signals: Based on the dynamic spectral monitoring baseline, differential analysis between continuous time points is performed to identify spectral abrupt change points caused by leaf wax migration. Combined with spatial consistency test, interference signals related to wax migration are extracted to form a wax migration interference signal set.

[0010] Constructing an energy exchange shift fingerprint: The set of wax migration interference signals is time-matched with key inversion parameters in the energy exchange process of maize leaves to construct an energy exchange shift fingerprint that reflects the impact of spectral abrupt changes on energy exchange interference.

[0011] Establish a multi-level correction function: Based on the energy exchange offset fingerprint, a multi-level correction function is constructed, and the spectral abrupt change parameter is coupled with the maize leaf stomatal conductance estimation model to correct the evapotranspiration rate inversion offset caused by wax migration in real time.

[0012] Introducing a residual backtracking mechanism and generating a dynamic weight distribution: Based on the multi-level correction function, the corrected evaporation rate inversion result is iteratively compared with the dynamic spectral monitoring baseline to calculate the residual across time periods and generate the corresponding dynamic weight distribution.

[0013] An adaptive closed-loop update strategy is formed: Based on the dynamic weight distribution, an adaptive closed-loop update strategy is constructed, and the update results are written back to the dynamic spectral monitoring baseline and the multi-level correction function, respectively, so as to maintain the accuracy and consistency of evapotranspiration rate inversion under long-term drought conditions.

[0014] Preferably, the steps for establishing a dynamic spectral monitoring baseline include:

[0015] Multi-time-point spectral acquisition experiments were conducted on maize under continuous drought stress to obtain leaf reflectance spectral data in the 400nm to 1000nm band with 5nm intervals. The data were then normalized by standard whiteboard reflectance calibration and environmental parameter recording.

[0016] Simultaneously with spectral acquisition, data on the structural characteristics of leaf epidermal wax were obtained, including observing the wax distribution and crystal morphology in different areas of the leaf using low-temperature scanning electron microscopy, analyzing wax composition using Fourier transform infrared spectroscopy, reflecting the degree of wax coverage by contact angle measurement, and performing time and space correspondence registration with spectral data.

[0017] Using multi-temporal spectral data and wax structure characteristics as input and output variables, a structural coupling model was established using partial least squares regression. A time delay factor and spatial region identification parameters were introduced for fitting, resulting in a set of spectral fitting functions based on the time dimension of drought stress.

[0018] A dynamic spectral monitoring baseline is constructed based on a set of fitting functions. The output curves include the principal component variation curve of reflectance, the wax coverage index and the reflectance curve of the characteristic band, the perturbation residual matrix, and the spectral evolution curve of the spatial sub-region. These curves provide a long-term reference standard for subsequent interference identification and evapotranspiration rate inversion.

[0019] Preferably, during the construction of the dynamic spectral monitoring baseline, the wax coverage index is jointly calculated with the ratio of 765nm band reflectance to 700nm band reflectance to generate a spectral response sensitivity curve. The perturbation residual matrix is ​​formed by performing difference processing on the spectra at each time point and the pre-drought control state, which is used as a preliminary reference for subsequent evapotranspiration rate inversion and migration identification.

[0020] Preferably, the step of extracting wax migration interference signals includes:

[0021] Based on the continuous multi-temporal leaf reflectance spectrum data obtained from the dynamic spectral monitoring baseline, differential operation is performed on adjacent time points to obtain differential spectral curves. The rate of change curve is identified by processing with the first derivative. Then, a threshold is set according to the standard deviation of the spectral change of the control group to screen out candidate segments of spectral abrupt change.

[0022] Spatial consistency is tested on the candidate segments of spectral abrupt change. The consistency index of pixel value changes in the local neighborhood is calculated. If the abrupt band shows synchronous change in most areas of the leaf, it is judged as having high spatial consistency; otherwise, it is removed as noise.

[0023] The abrupt bands selected through two rounds of screening will be integrated to generate a set of feature vectors containing location, amplitude of change, rate of change of derivative and spatial consistency score. These vectors will then be physically verified using wax structure observation data to identify high-confidence interference signals.

[0024] The identified interference signals are uniformly encoded according to the time dimension and spatial label to form a set of wax migration interference signals containing timestamp, abrupt band number, change direction, change amplitude, spatial consistency score and corresponding wax index state, which is used for the subsequent construction of energy exchange offset fingerprint.

[0025] Preferably, the spectral abrupt change candidate segments in the wax migration interference signal set are limited to the near-infrared 780nm to 880nm range and the red-edge 720nm to 740nm range. They are required to simultaneously satisfy the spatial variation coefficient being lower than a set threshold and the local information entropy being lower than a set threshold in the spatial consistency test in order to be confirmed as valid interference signals.

[0026] Preferably, the steps for constructing the energy exchange offset fingerprint include:

[0027] The set of wax migration interference signals was aligned and matched with the time series of blade energy exchange parameters, and spatial resampling and interpolation were used to ensure that each interference signal corresponds to complete energy parameters.

[0028] The matched interference signal and energy parameters are subjected to differential operation to calculate the change value of energy parameters, and compared with the parameter difference of the control area to obtain the normalized interference offset value.

[0029] The normalized interference offset values ​​are encoded using multidimensional features to form a set of feature vectors that include time labels, spatial attributes, band categories, abrupt change magnitudes, and energy impact weights.

[0030] Cluster analysis is performed on the feature vector set to extract typical features and group them to construct energy exchange offset fingerprint spectrum, generating a fingerprint set that can characterize the evaporation rate and stomatal conductance offset.

[0031] The energy exchange offset fingerprint is matched with the newly observed wax migration interference signal at time points. If the similarity exceeds a set threshold, the interference identification is completed; otherwise, the new features are saved to update the fingerprint set.

[0032] Preferably, the steps of establishing a multi-level correction function and coupling it with the porosity estimation model include:

[0033] Based on the interference behavior patterns characterized in the energy exchange migration fingerprint spectrum, feature parameters of different abrupt change bands are extracted, and first-order and second-order response functions are established respectively to form a multi-level set of correction functions.

[0034] The multi-level correction function is coupled with the stomatal conductance estimation model of maize leaves for joint modeling. The position, amplitude and spatial distribution parameters of the abrupt change band are used as covariates to input the stomatal conductance estimation model, and an expression that considers both spectral abrupt change and physiological response characteristics is formed through joint fitting.

[0035] During the evaporation rate inversion process, when a new wax migration interference signal is identified, the matching correction function is called and the spectral abrupt change parameter is input to generate a correction factor. The correction value is then fused with the coupled stomatal conductance estimation result. Through iterative comparison and feedback adjustment, the real-time adaptive correction of the evaporation rate inversion value is achieved.

[0036] Preferably, the step of iteratively comparing the corrected evaporation rate inversion result with the dynamic spectral monitoring baseline and generating a dynamic weight distribution includes:

[0037] The evaporation rate correction results were compared with the spectral response curves at the corresponding time points in the dynamic spectral monitoring baseline, and the residual values ​​on the near-infrared, red edge, and moisture absorption bands were calculated.

[0038] The residuals at each time point are arranged in sequence to form a time series residual chain, and cumulative drift, periodic oscillation or sudden jump characteristics are identified by smoothing and derivative calculation.

[0039] A dynamic weight distribution function is constructed based on the residual amplitude, rate of change, and spectral sensitivity to generate a weight vector corresponding to the confidence level at each time point.

[0040] Dynamic weight distribution is applied to the weighting of historical correction results and the prediction and adjustment of future correction values, and weights are updated when new data is added to ensure the long-term continuity and stability of the inverted sequence.

[0041] Preferably, the steps of constructing an adaptive closed-loop update strategy and writing the update results back to the dynamic spectral monitoring baseline and the multi-level correction function include:

[0042] An adaptive weight adjustment factor is constructed based on dynamic weight distribution, and a feedback control factor matrix is ​​generated. The feedback intensity characterizes the importance of each time point to the model update, and the temporal continuity is maintained through a time window smoothing mechanism.

[0043] The feedback factor matrix is ​​applied to the dynamic spectral monitoring baseline and the multi-level correction function. The baseline achieves iterative adjustment of the spectral response curve through weighted sliding update and trend reconstruction. The correction function redistributes the parameters of the first-order and higher-order correction functions and introduces new offset fingerprint data through the feedback factor.

[0044] The updated spectral monitoring baseline and multi-level correction function are reloaded into the evaporation rate inversion process. When new data is input, iterative feedback is triggered to form a closed-loop inversion structure with continuous self-correction capability.

[0045] The technical effects and advantages provided by the present invention in the above technical solution are as follows:

[0046] This invention achieves a breakthrough correction to the spectral input stability assumption of traditional evapotranspiration rate inversion models by constructing a dynamic spectral monitoring baseline that integrates dynamic spectral changes and physiological structural responses, and by combining the accurate identification of wax migration interference signals with the construction of energy exchange offset fingerprints. Compared to the limitations of existing technologies that lack a sensing and response mechanism for changes in leaf optical properties, this invention can immediately identify perturbation behavior and perform intervention-based modeling at the early stage of spectral abrupt changes caused by wax migration. This makes the inversion model more environmentally adaptable and interference-resistant, effectively avoiding evapotranspiration inversion distortion caused by wax migration.

[0047] This invention establishes a multi-level correction function and a dynamic weight feedback mechanism, forming a closed-loop update strategy with self-learning and adaptive characteristics. This allows the evapotranspiration rate inversion system to continuously optimize its structure and parameters based on error feedback during long-term operation, thereby maintaining the stability and comparability of the output results across different drought stages, environmental backgrounds, and crop varieties. This adaptive mechanism significantly enhances the model's long-term ability to cope with complex spectral perturbations, overcoming problems such as error accumulation, model rigidity, and parameter drift in traditional methods. In comprehensive drought resistance evaluation applications, it ensures that the inversion results maintain a high degree of consistency and interpretability throughout the entire drought cycle, thereby enhancing the evaluation results' value for generalization and guidance. Attached Figure Description

[0048] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0049] Figure 1 This is a flowchart of a method for comprehensive evaluation of drought resistance of maize varieties according to the present invention. Detailed Implementation

[0050] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, they are provided so that the description of this disclosure will be more complete and fully convey the concept of the exemplary embodiments to those skilled in the art.

[0051] This invention provides, for example Figure 1 The method for comprehensive evaluation of drought resistance in maize varieties, as shown, includes the following steps:

[0052] Establish a dynamic spectral monitoring baseline: By acquiring multi-temporal and multi-band reflectance spectral data under continuous drought conditions in maize, and combining it with the structural characteristics of the epidermal wax of maize leaves, structural coupling modeling of the spectral data is performed to construct a time series curve that can characterize the dynamic changes in the distribution of leaf wax, which serves as the dynamic spectral monitoring baseline.

[0053] To address the objective phenomenon of spatial migration and structural changes in the epidermal wax of maize leaves under persistent drought conditions, and to monitor and quantify the impact of these changes on leaf spectral reflectance characteristics in real time, thus providing stable and reliable time-series reference data for interference identification and shift correction in subsequent evapotranspiration rate inversion, a specific method for constructing a dynamic spectral monitoring baseline is as follows:

[0054] A multi-time-point spectral acquisition experiment was conducted on maize under continuous drought stress to obtain leaf reflectance spectral data with temporal resolution. Specifically, maize varieties with consistent growth periods, similar plant heights, and stable phenotypes were selected as monitoring subjects. These selected plants were planted in a controlled-water field experimental area, which was divided into a control area and a drought treatment area based on irrigation settings. The drought treatment area was designed to receive only one initial irrigation after sowing, followed by a halt to irrigation, creating a typical moderate to severe natural drought stress environment. At key time points—days 1, 3, 5, 7, 10, 14, and 21 after the onset of drought stress—near-ground imaging of maize functional leaves was performed using a portable hyperspectral imaging device. The acquisition height was controlled at 20 cm vertically above the leaf surface, with an imaging angle of 90° vertically to avoid sidelight interference. The acquisition wavelength range was between 400 nm and 1000 nm, with a band interval of 5 nm to ensure fine resolution of the spectral data. Meanwhile, to ensure the comparability and normalization of spectral data, the equipment was calibrated for reflectance using a standard white board before each imaging session, and the ambient light intensity, air humidity, temperature, and wind speed at that time point were recorded for subsequent modeling to facilitate the coordinated correction of environmental factors. During the data processing stage, images acquired at each time point were uniformly cropped into Regions of Interest (ROIs) containing only the leaf area, removing background soil, leaf sheaths, and other plant interference areas, resulting in a continuous, multi-temporal, structurally standardized set of leaf spectral reflectance data.

[0055] At each spectral acquisition time point, a synchronous experiment was conducted to acquire the waxy structure characteristics of maize leaves, which was used to assist in analyzing the underlying physical mechanisms behind the spectral changes. The specific process included: immediately after each spectral acquisition, the corresponding area of ​​the leaf to be tested was marked and cut from the same plant. A 2cm × 2cm area of ​​the leaf surface was cut, flash-frozen in liquid nitrogen, and then sent to a cryogenic scanning electron microscope for low-temperature scanning observation. The microscopic images were used to identify the distribution pattern, crystal morphology, and density changes of the wax on the leaf surface, with a focus on observing the microstructural changes in the interveinal region, vein margin region, leaf tip region, and leaf base region. In addition, at each sampling point, Fourier transform infrared spectroscopy (FTIR) was used to analyze the molecular structure and composition of the wax film, obtaining the relative abundance changes of long-chain fatty acids, alcohols, and esters in the wax; simultaneously, a contact angle meter was used to rapidly measure the contact angle of the leaf surface after water was dripped on it, to quantitatively reflect the changing trend of the wax coverage. All these structural and physical data are numbered according to time series and registered one-to-one with the sampling location and spectral image region to ensure that spectral changes can establish an accurate correspondence with specific wax structure changes.

[0056] After acquiring multi-temporal spectral data and wax structure data, structural coupling modeling was performed on the two types of data to construct a spectral response function model that can express the driving force of wax changes. The specific modeling process included: First, using the high-dimensional spectral data of the leaves at each time point (containing 121 bands) as the input variable vector, and using the wax structure indices (such as crystal density, average wax thickness, contact angle, and infrared peak intensity ratio) collected at the corresponding time point and spatial location as the output target variables. Partial least squares regression (PLSR) was used to construct a locally fitted model for each set of temporal data. The optimal number of principal components was determined through cross-validation to ensure the model's stability and generalization ability in the context of high-dimensional data. To model the lag in the spectral response to wax structure evolution over time, a time delay factor was introduced into the modeling process, interactively pairing the current spectral data with the wax indices from the previous day to enhance the model's ability to capture dynamic evolution processes. In the modeling process, the spectral curves of the interveinal and vein-marginal regions, as well as the leaf tip and leaf base regions, were modeled as sub-regions, and region identification parameters were set to participate in the fitting, forming a structure-spectral coupled response model that combines temporal continuity and spatial heterogeneity. Finally, the model outputs a set of spectral fitting functions based on the time dimension of drought stress, which can express the continuous process of spectral response induced by wax changes.

[0057] Finally, based on the set of fitting functions generated by the aforementioned structural coupling model, a dynamic spectral monitoring baseline with temporal and spatial hierarchical dimensions was constructed. This baseline includes the following: first, the leaf reflectance principal component variation curves arranged chronologically, showing the evolution trajectory of overall leaf reflectance with the drought process; second, the spectral response sensitivity curve jointly generated by the wax coverage index (WCI) calculated based on the fitting function and the reflectance of characteristic bands (e.g., 765nm / 700nm), reflecting the degree of spectral perturbation caused by wax migration; third, the perturbation residual matrix formed by differentiating the spectrum at each time point from the initial control state (pre-drought state), used as a preliminary reference for subsequent evapotranspiration inversion parameter shift identification; and fourth, the spectral evolution curve of each spatial sub-region, accompanied by significant change time points and corresponding wax change characteristic values. This dynamic monitoring baseline not only provides the entire trajectory of spectral changes in maize under drought conditions but also establishes a logical correlation channel between spectral perturbation and wax migration, providing a quantifiable and traceable analytical basis for subsequent differential identification, interference signal extraction, energy shift fingerprint construction, and dynamic correction. This baseline has the ability to be dynamically updated. In actual implementation, new temporal data can be continuously added to achieve adaptive spectral-structure linkage updates, providing a long-term, highly consistent reference standard for steady-state inversion of evapotranspiration rate.

[0058] This implementation method, through a dynamic monitoring baseline construction method with strong temporal continuity, clear structural response, and accurate data registration, is the first to highly integrate the microstructural change of maize wax migration with large-scale reflectance spectral monitoring. This frees traditional inversion models from dependence on wax invariance and fundamentally improves the stability and accuracy of evapotranspiration parameter inversion under drought conditions.

[0059] It should be noted that:

[0060] The Wax Coverage Index (WCI) is a characterizing indicator used to quantify the degree of wax coverage on the surface of plant leaves. It mainly reflects the distribution density, thickness, or optical influence of the epidermal wax layer. It is usually used as an indirect parameter for evaluating wax levels, calculated through optical or contact angle measurements, and has important applications in plant drought resistance, physiological and ecological adaptability, and remote sensing inversion models.

[0061] In existing technologies, methods for obtaining the wax coverage index mainly fall into two categories:

[0062] Non-contact estimation based on spectral reflectance characteristics: Since epidermal wax can alter leaf surface reflectance, especially showing significant spectral responses in the near-infrared (e.g., 750-900 nm) and short-wave infrared bands, researchers often establish empirical models using specific band ratios or vegetation indices (e.g., R765 / R700, WI, NDWI, etc.). After obtaining leaf spectral data through remote sensing equipment (e.g., hyperspectral imagers), the data is substituted into the model to calculate WCI. Some studies introduce multivariate methods such as principal component analysis and partial least squares regression for modeling to improve estimation accuracy.

[0063] Direct measurement based on contact angle and microstructure: By adding a small amount of water droplets to the blade surface and measuring the change in its static contact angle, the hydrophobicity and coverage of the wax can be deduced; a larger contact angle generally indicates a thicker wax layer. In addition, scanning electron microscopy (SEM) can directly observe the wax crystal density and calculate the coverage ratio of wax crystals per unit area using image processing techniques, further quantifying it as WCI.

[0064] Extraction of wax migration interference signals: Based on the dynamic spectral monitoring baseline, differential analysis between continuous time points is performed to identify spectral abrupt change points caused by leaf wax migration. Combined with spatial consistency test, interference signals related to wax migration are extracted to form a wax migration interference signal set.

[0065] To identify abrupt changes in spectral reflectance characteristics of maize caused by wax migration from leaf epidermis under sustained drought stress, and to ensure that interference sources in subsequent evapotranspiration rate inversion parameters can be accurately identified and effectively separated, this implementation proposes a wax migration interference signal extraction method based on time-series difference analysis combined with spatial consistency testing, building upon an established dynamic spectral monitoring baseline. This method includes the following steps:

[0066] In continuous multi-temporal leaf reflectance spectral data acquired based on dynamic spectral monitoring baselines, pairwise difference analysis was performed on each consecutive time point pair to identify abrupt changes in the spectral curves during drought stress evolution. Specifically, assuming that hyperspectral data of maize functional leaves were collected on day n and day n+1, the reflectance value of each spectral band on day n+1 was subtracted from the value of the corresponding band on day n to generate a complete difference spectral curve. This difference curve reflects the changing trend of leaf reflectance after a 24-hour drought evolution. To improve the sensitivity of abrupt change identification, the difference curve was processed with its first derivative to obtain the rate of change curve, and a threshold was further set to determine whether the change exceeded the normal fluctuation range of the system. The threshold was statistically derived based on the standard deviation of the spectral change range of the control group (maize leaves under fully irrigated conditions). That is, only band changes with significantly higher difference values ​​or rates of change than the normal fluctuation range of the control group were marked as candidate segments for spectral abrupt changes. In addition, based on the known pattern that wax has a significant impact on the reflectivity of specific wavelengths (such as the near-infrared 780-880nm range and the red edge 720-740nm range), we can prioritize the differential response of these sensitive wavelengths to improve the targeting and efficiency of mutation identification.

[0067] Secondly, after completing the temporal difference analysis and initially identifying candidate segments for spectral mutations, a spatial consistency test was performed on the spectral images to verify whether these mutations possessed consistent characteristics within the leaf region, thereby eliminating the influence of occasional disturbances such as random noise, shadows, and dust deposition. The specific test method was as follows: using the spectral mutation band in the hyperspectral image as input data, the pixel value distribution map of the entire leaf region under that band was extracted, and the consistency index of pixel value changes within the local neighborhood was calculated, such as the coefficient of variation (CV), local entropy (LIE), and mean shift (MSD). Spatially, if the mutation band showed a synchronous enhancement or weakening trend in most areas of the leaf, i.e., low spatial variability and low local entropy, then the spectral mutation could be determined to have high spatial consistency and reflect a structural response change. Conversely, if the mutation band showed obvious dispersion in the image or was limited to jumps in only a few pixels, it could be judged as random noise not caused by wax migration and should be eliminated. The implementation of spatial consistency testing enables the identification of spectral mutations to not only rely on the magnitude of changes in the spectral curve, but also to verify the rationality of spatial distribution, thereby improving the accuracy and robustness of wax migration interference signal identification.

[0068] After completing temporal difference analysis and spatial consistency verification, multidimensional feature integration was performed on all spectral abrupt change bands that passed two rounds of screening to construct a structural expression vector for the wax migration interference signal. This process first assembles vector elements from the location, amplitude, rate of change of derivative, and corresponding spatial consistency score of the abrupt change bands, forming a feature vector set containing each abrupt change point. Then, cluster analysis was performed on the distribution of all abrupt change bands in the time series to identify the temporal patterns of abrupt events during drought evolution, determining whether they exhibit typical "continuous accumulation abrupt changes" or "intermittent jumps." Continuous accumulation abrupt changes often better match the gradual change in spectral reflectance during slow wax migration and are considered high-confidence interference signals. Simultaneously, by further combining registered wax structure observation data, the presence of synchronous changes in wax distribution characteristics, such as increased wax thickness or shifts in distribution areas, was searched at corresponding time points to reverse-verify the physical causes of the spectral abrupt changes. By combining multidimensional fusion with physical response linkage, the extracted interference signals are ensured to not only have abrupt change characteristics at the spectral level, but also have occurrence evidence at the structural level, thus achieving an upgrade in identification from data-driven to mechanism-driven.

[0069] After extracting abrupt change bands, verifying spatial consistency, and confirming structural features, a set of wax migration interference signals is formed. These signals are then uniformly encoded according to time and spatial labels for use in the subsequent construction of energy exchange migration fingerprints and dynamic correction of inversion parameters. The interference signal set specifically includes: timestamp information (indicating the specific time phase of the abrupt change), abrupt change band number (indicating the perturbation location in the specific reflectance spectrum), direction of change (enhancement or weakening), amplitude of change, spatial consistency score, corresponding wax index status, and the functional attribute of the band in the inversion model (e.g., whether it is used for NDVI, PRI, or thermal infrared estimation). This set constitutes a systematic signal matrix, which can serve as a leading indicator affecting the accuracy of evapotranspiration calculations, enabling the identification of potential interference factors at the source. In practical applications, this set can be dynamically expanded according to the actual acquisition frequency, and the signal identification threshold can be continuously corrected through machine learning, giving the interference identification capability long-term adaptability and timely response capability.

[0070] Through the above steps, the spectral interference signals related to wax migration can be effectively identified and accurately extracted in both the time and spatial domains. This not only ensures the scientific validity and repeatability of the interference source identification, but also provides a solid data foundation and logical basis for subsequent energy exchange parameter modeling and multi-level offset correction.

[0071] Constructing an energy exchange shift fingerprint: The set of wax migration interference signals is time-matched with key inversion parameters in the energy exchange process of maize leaves to construct an energy exchange shift fingerprint that reflects the impact of spectral abrupt changes on energy exchange interference.

[0072] To further clarify the interference pathways and quantitative impacts of these spectral abrupt changes in energy exchange on maize leaves, this implementation proposes a method for constructing an energy exchange migration fingerprint, based on the established set of waxy migration interference signals. This method involves time-point matching, difference extraction, feature parsing, and joint modeling of the waxy migration interference signal set with multiple key energy exchange inversion parameters. The result is a structured migration feature fingerprint representing the interference behavior across multidimensional energy parameters, which is then used to accurately drive subsequent calibration models. The entire process includes the following sub-steps:

[0073] Based on the acquired set of waxy migration interference signals, a corresponding time series of energy exchange parameters was constructed, ensuring strict alignment between the two in terms of temporal accuracy. Specifically, using meteorological and infrared thermal imaging data acquired synchronously with the spectral data, key energy exchange parameters such as leaf surface temperature, estimated stomatal conductance, net radiation, initial inversion values ​​of evapotranspiration rate, and air-leaf temperature difference of maize functional leaves were extracted at each time point. These parameters were then indexed and paired according to the timestamps in the waxy migration signal set. To ensure data consistency, all energy exchange parameters underwent unified spatial resampling, retaining only image data that completely overlapped with the waxy migration abrupt change region. Simultaneously, bidirectional interpolation was used to complete intermittent data caused by equipment errors or environmental fluctuations, ensuring that each waxy interference signal could find a clearly corresponding response value in the energy parameter space. This step established a unified multidimensional time-matching framework for subsequent analysis, providing a computational basis for the causal coupling of spectral abrupt changes and energy variations.

[0074] Based on time alignment, the differences between the wax migration interference signal and its corresponding energy exchange parameters at each time point are calculated to uncover the energy shift caused by the mutation. The specific method includes: using the amplitude of the band change in the wax migration signal as the interference intensity input, and the difference in energy parameters between that time point and the previous time point as the output, calculating the change ΔE of each energy parameter before and after the mutation, and comparing it with the energy parameter change ΔE_ref of the control plant or control area during the same time period to form a normalized interference shift value δE = ΔE - ΔE_ref. To eliminate the common influence of natural environmental changes on energy parameters, a reference model is constructed by introducing the average values ​​of multiple control samples and spatial blocks to further improve the characterization ability of the interference shift value. Essentially, this step converts the wax signal perturbation into an energy response perturbation, clarifying which type of spectral mutation behavior may have a specific direction and degree of influence on which energy parameter.

[0075] Based on the normalized interference offset values ​​mentioned above, the response features of abrupt interference in the energy space are extracted and multidimensionally characterized and encoded to form offset feature units. This process includes: performing principal component analysis on the δE vectors at all time points to identify the contribution rate of each energy parameter to the offset behavior; then, performing grouped statistical analysis according to the band attributes of wax migration (e.g., whether it is located in the red-edge region or near-infrared region), abrupt amplitude level (low, medium, high), and abrupt direction (enhancement or weakening) to extract the interference probability and average offset intensity of specific energy parameters (e.g., evaporation rate, stomatal conductance). Furthermore, spatial consistency indicators (e.g., the area ratio of abrupt regions) are added to the encoding to form a multidimensional feature vector containing time labels, spatial attributes, band categories, abrupt amplitude, and energy influence weights, used to represent the interference behavior of a typical wax migration event in the energy parameter space. The output of this step is a set of multiple offset feature units, constituting an energy response feature library.

[0076] Based on the construction of multiple offset feature units, hierarchical clustering and fingerprint matching algorithms are used to classify and analyze the above features, constructing an identifiable and reusable energy exchange offset fingerprint map. Specifically, all feature units are measured by distance according to the similarity of mutation behavior, and a feature distance matrix is ​​constructed using Euclidean distance or cosine similarity. Hierarchical clustering is then used to divide the features into several classes, and the most representative typical features in each class are extracted as the central fingerprint. Based on this, typical offset features are grouped according to the affected objects, such as fingerprint sets for evapotranspiration rate offset, fingerprint sets for stomatal conductance error, etc., and the spectral behavior combination patterns causing interference are marked in each fingerprint class. Finally, these representative features are encoded in a map format, forming a multi-type, clearly structured, and traceable energy exchange offset fingerprint set. This map not only records the feature expressions of historical interference events but also has the ability to quickly compare, assess similarity, and predict responses to new mutation events, serving as an important basis for evapotranspiration rate correction and dynamic weight adjustment.

[0077] The constructed energy exchange migration fingerprint map is matched and identified against newly observed wax migration interference signals at specific time points, enabling real-time detection and response to online interference. The matching method includes: for wax migration signals identified at the current time point, their feature vectors are calculated and compared with various fingerprint types in the fingerprint map. If the similarity exceeds a set threshold, the mutation behavior is determined to belong to a known fingerprint class, and the migration pattern in that fingerprint class is used to predict and adjust the current energy parameters. For new types of mutation events that have not yet matched any existing fingerprints, their features are stored in a cache, awaiting future observations to form a complete pattern before being added to the fingerprint set. Through this strategy, this scheme achieves a rapid response mechanism for identifying the causes of changes, continuously improving the stability and predictability of evapotranspiration inversion parameters in arid environments.

[0078] Through the close integration and high-precision calculation of the above steps, this implementation method not only achieves quantifiable modeling of the shift from abrupt changes in the waxy migration spectrum to the shift in energy exchange parameters, but also forms a generalizable and scalable structured shift fingerprint spectrum. This provides full-process data-driven and model fusion support for the accurate control of evapotranspiration parameters in comprehensive drought resistance evaluation. Compared with traditional methods, this implementation method has stronger disturbance adaptability and parameter correction closed-loop performance, exhibiting better stability and recognition sensitivity under long-term drought dynamics.

[0079] Establish a multi-level correction function: Based on the energy exchange offset fingerprint, a multi-level correction function is constructed, and the spectral abrupt change parameter is coupled with the maize leaf stomatal conductance estimation model to correct the evapotranspiration rate inversion offset caused by wax migration in real time.

[0080] To further address the systematic shift in maize evapotranspiration rate inversion parameters caused by abrupt changes in spectral reflectance characteristics due to wax migration, this paper proposes a strategy based on the previously constructed energy exchange migration fingerprint. This strategy involves constructing a multi-level correction function and coupling it with the stomatal conductance estimation process. Through multi-level interference modeling, joint physiological parameter fitting, and real-time data fusion correction, dynamic correction of evapotranspiration rate inversion results under interference conditions is achieved. This method specifically includes the following sub-steps:

[0081] Based on the typical disturbance behavior patterns characterized by energy exchange migration fingerprints, a multi-level set of correction functions applicable to various spectral mutation paths is constructed. This process specifically includes: extracting all representative disturbance types from the migration fingerprint and classifying and summarizing their impact on energy exchange parameters, such as evapotranspiration rate migration, net radiation disturbance, and leaf surface temperature error. Subsequently, for each type of disturbance behavior, first-order and second-order response functions are established based on its mutation characteristic parameters (such as mutation amplitude, mutation rate, and spatial consistency score) in different bands. The first-order function directly corrects the deviation in inversion values ​​caused by mutations, while the second-order function corrects indirect errors caused by parameter coupling during the inversion process. For example, for the overestimation of evapotranspiration caused by wax enhancement in the near-infrared region, a three-variable polynomial function is established, including the intensity of the mutation band, the sensitivity weight of the evapotranspiration inversion formula, and the correction residual from the previous day, as the basic correction expression. All correction functions are fitted based on historical samples, and uncertainty boundary terms are introduced during construction, enabling automatic matching of the closest function model for new types of mutations in practical applications, ensuring the correction mechanism has diverse response capabilities and structural stability.

[0082] The constructed multi-level correction function is coupled with the stomatal conductance estimation model of maize leaves to form a dynamic correction mechanism that simultaneously considers the effects of spectral abrupt changes and physiological response characteristics. The key to this step lies in establishing a two-dimensional fusion path: first, the fusion of spectral abrupt change parameters with the input factors of the stomatal conductance physiological estimation model; and second, the fusion of the feedback mechanism between the correction function output and the evapotranspiration inversion value. In the implementation process, spectral abrupt change parameters corresponding to the wax migration signal at each time point are first collected, including the location of the abrupt change band, the amplitude of the abrupt change, and the spatial distribution area. These parameters are then used as covariates and introduced into the stomatal conductance estimation model. The original stomatal conductance model mainly relies on input factors such as evapotranspiration rate, leaf temperature, and air temperature and humidity. In the coupled model, spectral abrupt change characteristics are added as adjustment factors, and joint fitting is performed through partial least squares regression or a generalized additive model to form a more physiologically reasonable stomatal conductance estimation expression that is adaptable to optical interference. In this expression, the stomatal conductance value reflects both the true physiological state of plant water and the perturbation trend brought about by the current spectral anomaly to the parameter inversion, thereby improving the stability of the entire inversion model to anomalous optical inputs. Finally, the corrected value output by the correction function is fused with the coupled stomatal conductance estimation result, and the evapotranspiration rate inversion value is adjusted in real time through a feedback correction mechanism to keep it dynamically within the dual constraints of physiological boundaries and optical logic.

[0083] Subsequently, the coupled correction mechanism was deployed at each time point in the actual evapotranspiration rate inversion process, enabling dynamic identification, immediate response, and automatic correction of waxy abrupt changes under drought conditions. In actual operation, whenever a new waxy migration interference signal is identified, a matching multi-level correction function is automatically invoked, and the current spectral abrupt change parameters are input to generate a correction factor. Simultaneously, the initial evapotranspiration rate inversion value at that time point is compared with the expected conductance value calculated by the stomatal conductance coupling model. If the deviation exceeds a set threshold, the correction mechanism is triggered, and the inversion parameters are adjusted according to the correction function output value. Based on this, the adjusted inversion results are input back into the coupling model for iterative comparison and adjustment until the corrected evapotranspiration rate value and the estimated stomatal conductance value reach the statistically consistent judgment condition, at which point the correction process ends. The entire correction process is completed within seconds, achieving real-time adaptive correction of the evapotranspiration rate inversion results and possessing the ability to automatically adjust the response intensity at different drought stages, thereby ensuring that the final evaluation results have high stability, repeatability, and spectral-physiological consistency.

[0084] Through the refined modeling and joint coupling of the above sub-steps, dynamic multi-correction of evapotranspiration inversion values ​​under the background of waxy migration interference was realized. This not only ensured the continuity of parameters in the time dimension, but also ensured the rationality of the estimation in physiological logic, and significantly improved the reliability of the comprehensive evaluation method of drought resistance of maize varieties in real drought environment.

[0085] Introducing a residual backtracking mechanism and generating a dynamic weight distribution: Based on the multi-level correction function, the corrected evaporation rate inversion result is iteratively compared with the dynamic spectral monitoring baseline to calculate the residual across time periods and generate the corresponding dynamic weight distribution.

[0086] To ensure the sustained accuracy and dynamic consistency of evapotranspiration rate inversion results across multiple time scales during long-term drought stress, this implementation method introduces an iterative comparison of the output of a multi-level correction function with a dynamic spectral monitoring baseline to generate a dynamic weight distribution. This method effectively tracks and regulates the propagation behavior of the correction residuals without compromising temporal continuity and physiological consistency, providing structured parameter support for the subsequent closed-loop adaptive update mechanism. The method consists of the following steps:

[0087] Based on the evapotranspiration rate correction value obtained in the previous stage driven by the multi-level correction function, this correction result is compared with the original spectral response curve recorded at the same time point in the dynamic spectral monitoring baseline, and the response residuals in multiple spectrally sensitive regions are calculated. Specifically, the comparison method is as follows: using the evapotranspiration rate correction value at the current time point as a reference standard, the corresponding spectral characteristic segments in the dynamic spectral monitoring baseline at that time point are identified, with particular attention paid to the near-infrared band (760-900nm), the red-edge band (700-740nm), and the moisture absorption band (970nm), which are strongly correlated with evapotranspiration. The difference between the actual reflectance and the spectral state implied by the correction result is extracted from these spectral segments, and the spectral trend represented by the evapotranspiration correction value is mapped through inverse modeling to perform direct residual calculations with the actual spectral curve. The goal of this step is to use the current correction value as the core to verify whether it possesses an evolutionary trend consistent with the time baseline in the spectral characteristic space, thereby determining the spectral rationality and traceability of the current correction result.

[0088] After obtaining the temporal residuals in the spectral space, cross-time comparisons are performed on multiple consecutive time points to construct a temporal residual curve of the evapotranspiration correction value in the dynamic monitoring baseline. Specifically, all evapotranspiration correction results from the start of drought stress to the current time point are selected, along with the monitoring baseline spectral records. The residual values ​​are arranged chronologically to form a complete time-series residual chain. This series, with time → residual amplitude as its core axis, constructs the trajectory of the difference between the evapotranspiration correction results and the spectral response throughout the entire drought process. To enhance the ability to judge trends, moving average and exponential smoothing algorithms are introduced to smooth the residual chain and identify systematic drift trends, such as cumulative shifts, periodic oscillations, or sudden jumps. Furthermore, residual gradient and first derivative curves are introduced to measure the rate of change of the residual at each time point, identifying the key nodes with the strongest residual fluctuations. Through these calculations, the spectral consistency behavior of the evapotranspiration correction value can be dynamically quantified, providing data for subsequent weight distribution generation.

[0089] Based on the structured characteristics of the aforementioned time series residual chain, a dynamic weight distribution function is constructed to clarify the confidence level and adjustment priority of the evapotranspiration correction result at each time point in the long-term inversion process. The weight function is constructed using residual amplitude, residual change rate, and spectral sensitivity coupling degree as the main input factors, and is calculated using a multi-factor exponential decay model or logistic regression function. Specifically, if the residual amplitude at a certain time point is significantly lower than the long-term mean, the residual change rate remains stable, and the corresponding band has high spectral sensitivity (i.e., a large weight influencing evapotranspiration estimation), then that time point is given a higher weight, indicating high reliability of its correction result and its potential as a dominant reference for subsequent evolution trends. Conversely, if the residual amplitude is large and the change rate is drastic, it indicates instability and drift risk in the correction value, and a lower weight is given. The final output of the dynamic weight distribution is a weight vector of the same length as the time series, with each time point assigned a real-valued weight between 0 and 1, forming a time point-confidence level correspondence table, which can be considered as the dynamic adjustment index of the evapotranspiration correction value at that time point.

[0090] The aforementioned dynamic weight distribution results are applied to the entire inversion result sequence, providing guiding feedback for subsequent correction values. The specific application strategy includes two aspects: first, re-evaluating and reassigning historical correction results by weighting the original evaporation correction results according to the weight vector using a weighted moving average, improving the long-term stability of the sequence; second, providing prior guidance for the dynamic prediction of correction values ​​at future time points. Especially when new wax migration interference signals appear, if their time point is adjacent to a historical high residual low weight interval, an early warning mechanism is triggered before evaporation inversion, automatically lowering the initial confidence level at that point or using a higher-order correction function for compensation. Simultaneously, an adaptive update channel is set in the dynamic weight vector. Whenever a new correction result is added at a subsequent time point, the current result is coupled with the weight trend of the previous time point to form a progressively evolving weight update trajectory, ensuring that the weight distribution possesses temporal continuity, memory capacity, and responsive sensitivity, thereby laying a data foundation for the long-term closed-loop stable operation of the inversion system.

[0091] Through the continuous implementation of the above sub-steps, this implementation method not only establishes a complete computational path from spectral residual quantification to weight strategy generation, but also realizes the ability to dynamically identify and predict risks in the temporal reliability of evapotranspiration rate correction in actual operation, significantly enhancing the model's long-term adaptability and control over spectral perturbation behavior during drought processes. Compared with traditional inversion methods that mainly rely on single-time-point fitting, this method has obvious advantages in sequence control and evolutionary modeling capabilities, providing strong technical support for achieving long-term continuous evaluation of drought resistance monitoring of maize varieties on a large scale.

[0092] An adaptive closed-loop update strategy is formed: Based on the dynamic weight distribution, an adaptive closed-loop update strategy is constructed, and the update results are written back to the dynamic spectral monitoring baseline and the multi-level correction function, respectively, so as to maintain the accuracy and consistency of evapotranspiration rate inversion under long-term drought conditions.

[0093] To ensure that the evapotranspiration rate inversion results maintain high accuracy and consistency across time scales under long-term drought stress, after generating the aforementioned dynamic weight distribution, an adaptive closed-loop update strategy with evolvable capabilities is further constructed. The updated results are then written back to the dynamic spectral monitoring baseline and the multi-level correction function, thereby achieving dynamic correction and continuous self-optimization capabilities for the entire inversion system. This closed-loop update process specifically includes the following steps, each using dynamic weights as the core decision-making mechanism to ensure coordinated evolution between upstream and downstream models. The specific implementation steps are as follows:

[0094] Based on the generated dynamic weight distribution, an adaptive weight adjustment factor is constructed to drive model updates, and a feedback control factor matrix is ​​generated accordingly. Specifically, for each time point, the dynamic weight value of the evaporation correction result is converted into a feedback intensity coefficient. The feedback intensity measures the importance of that time point to the overall model structure update. When the weight of a certain time point is significantly higher than the average of its surrounding time periods, it indicates that the correction result at that point is representative and stable, and it should be given a higher model-driving capability; conversely, if the weight is low, it indicates that there is high uncertainty at that time point, and its feedback intensity should be reduced accordingly. Furthermore, to avoid drastic oscillations in the feedback process, a time window smoothing mechanism is introduced to smoothly adjust the feedback intensity of adjacent time points, ensuring temporal continuity of the feedback behavior. Finally, a set of feedback factor matrices is formed, serving as the core input for subsequent model structure evolution and parameter refitting.

[0095] The aforementioned feedback factor matrix is ​​applied to the dynamic spectral monitoring baseline and the multi-level correction function, respectively, to achieve real-time iterative updates of the two core structures. For updating the dynamic spectral monitoring baseline, a weighted sliding update method is used, with feedback intensity as the weight coefficient. The spectral response curves corresponding to the same time point in the baseline are dynamically adjusted to gradually approach the spectral state implied by the current high-weight correction value. During this process, if multiple consecutive time points show a stable trend of difference from the original baseline, the entire baseline is reconstructed to ensure that its evolutionary path during long-term drought is closer to the actual physiological and reflective response states. For updating the multi-level correction function, the parameter weights of each order correction function are dynamically redistributed according to the feedback factors. For example, the response weight of the first-order correction function is increased in high-confidence time periods, or offset fingerprint data from newly discovered wax migration patterns is introduced to adjust the perturbation sensitivity parameters of higher-order functions. Through this process, the original static function structure is transformed into a learnable function architecture, enhancing its generalization ability in future interference scenarios.

[0096] After updating the spectral monitoring baseline and correction function, the updated model structure and parameter results are reloaded into the evapotranspiration rate inversion process, constructing a complete closed-loop path. This forms an iterative feedback mechanism when new data is input, enabling continuous model self-correction. In practice, whenever new drought stage data enters the inversion process, a preliminary inversion is performed first. Then, the residual results are compared with the updated model based on the current weight distribution, triggering the current update strategy to run again. This process can be run multiple times throughout the entire drought season, forming an adaptive inversion closed-loop structure that gradually optimizes and enhances responsiveness over time. Ultimately, the goal is to achieve a high degree of consistency, stability, and physiological response consistency in the evapotranspiration rate inversion results under different drought intensities, plant conditions, and time periods.

[0097] Through the continuous implementation of the above sub-steps, this implementation method constructs a feedback closed-loop mechanism with dynamic adjustment capabilities, breaking through the technical bottleneck of static model parameters and rigid inversion process in traditional models. It realizes the collaborative evolution of the entire process from spectral signal perception to inversion result correction and model self-optimization, which greatly improves the stability and reliability of evapotranspiration rate estimation results in complex drought environments, and provides strong technical support and accuracy support for the comprehensive evaluation of maize drought resistance.

[0098] This invention achieves a breakthrough correction to the spectral input stability assumptions of traditional evapotranspiration rate inversion models by constructing a dynamic spectral monitoring baseline that integrates dynamic spectral changes and physiological structural responses, and by combining the accurate identification of wax migration interference signals with the construction of energy exchange shift fingerprints. Compared to the limitations of existing technologies that lack perception and response mechanisms to changes in leaf optical properties, this invention can immediately identify perturbation behavior and perform intervention-based modeling at the early stage of spectral abrupt changes caused by wax migration. This makes the inversion model more environmentally adaptable and robust to interference, effectively avoiding evapotranspiration inversion distortion caused by wax migration. Especially in complex drought scenarios, it achieves dynamic tracking and quantitative correction of optical abrupt changes, ensuring the accuracy of water consumption estimation from the source, and providing a highly consistent and reliable foundation for the scientific evaluation of drought resistance in maize varieties.

[0099] This invention further establishes a multi-level correction function and a dynamic weight feedback mechanism, forming a closed-loop update strategy with self-learning and adaptive characteristics. This allows the evapotranspiration rate inversion system to continuously optimize its structure and parameters based on error feedback during long-term operation, thereby maintaining the stability and comparability of the output results across different drought stages, environmental backgrounds, and varieties. This adaptive mechanism significantly enhances the model's long-term ability to cope with complex spectral perturbations, overcoming problems such as error accumulation, model rigidity, and parameter drift in traditional methods. In comprehensive drought resistance evaluation applications, it ensures that the inversion results maintain a high degree of consistency and interpretability throughout the entire drought season, thereby enhancing the promotion and guidance value of the evaluation results and providing strong technical support for regional suitability analysis, drought-resistant variety selection, and precision planting in arid areas.

[0100] The foregoing has only described certain exemplary embodiments of the present invention by way of illustration. Undoubtedly, those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the foregoing drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

Claims

1. A comprehensive evaluation method for drought resistance of maize varieties, characterized in that, Includes the following steps: By acquiring multi-temporal and multi-band reflectance spectral data of maize under continuous drought conditions, and combining the structural characteristics of the epidermal wax of maize leaves, structural coupling modeling of the spectral data was carried out to construct a time series curve that can characterize the dynamic changes in the distribution of leaf wax, which serves as a baseline for dynamic spectral monitoring. Based on the dynamic spectral monitoring baseline, differential analysis was performed between continuous time points to identify spectral abrupt change points caused by leaf wax migration. Combined with spatial consistency test, interference signals related to wax migration were extracted to form a set of wax migration interference signals. By temporally matching the set of wax migration interference signals with key inversion parameters in the energy exchange process of maize leaves, an energy exchange shift fingerprint reflecting the impact of spectral abrupt changes on energy exchange interference is constructed. Based on the energy exchange shift fingerprint, a multi-level correction function is constructed, and the spectral abrupt change parameter is coupled with the maize leaf stomatal conductance estimation model to correct the evapotranspiration rate inversion shift caused by wax migration in real time. Based on the multi-level correction function, the corrected evaporation rate inversion results are iteratively compared with the dynamic spectral monitoring baseline to calculate the residuals across time periods and generate the corresponding dynamic weight distribution. Based on the dynamic weight distribution, an adaptive closed-loop update strategy is constructed, and the update results are written back to the dynamic spectral monitoring baseline and the multi-level correction function to realize the evapotranspiration rate inversion.

2. The method for comprehensive evaluation of drought resistance of maize varieties according to claim 1, characterized in that, The steps for establishing a dynamic spectral monitoring baseline include: Multi-time-point spectral acquisition experiments were conducted on maize under continuous drought stress to obtain leaf reflectance spectral data in the 400nm to 1000nm band with 5nm intervals. The data were then normalized by standard whiteboard reflectance calibration and environmental parameter recording. Simultaneously with spectral acquisition, data on the structural characteristics of leaf epidermal wax were obtained, including observing the wax distribution and crystal morphology in different areas of the leaf using low-temperature scanning electron microscopy, analyzing wax composition using Fourier transform infrared spectroscopy, reflecting the degree of wax coverage by contact angle measurement, and performing time and space correspondence registration with spectral data. Using multi-temporal spectral data and wax structure characteristics as input and output variables, a structural coupling model was established using partial least squares regression. A time delay factor and spatial region identification parameters were introduced for fitting, resulting in a set of spectral fitting functions based on the time dimension of drought stress. A dynamic spectral monitoring baseline is constructed based on a set of fitting functions, and the output includes the principal component variation curve of reflectance, the wax coverage index and characteristic band reflectance curve, the perturbation residual matrix, and the spectral evolution curve of the spatial sub-region.

3. The method for comprehensive evaluation of drought resistance of maize varieties according to claim 2, characterized in that, In the process of constructing the dynamic spectral monitoring baseline, the wax coverage index is jointly calculated with the ratio of 765nm band reflectance to 700nm band reflectance to generate a spectral response sensitivity curve, and a perturbation residual matrix is ​​formed by performing difference processing on the spectra at each time point and the pre-drought control state.

4. The method for comprehensive evaluation of drought resistance of maize varieties according to claim 2, characterized in that, The steps for extracting wax migration interference signals include: Based on the continuous multi-temporal leaf reflectance spectrum data obtained from the dynamic spectral monitoring baseline, differential operation is performed on adjacent time points to obtain differential spectral curves. The rate of change curve is identified by processing with the first derivative. Then, a threshold is set according to the standard deviation of the spectral change of the control group to screen out candidate segments of spectral abrupt change. Spatial consistency is tested on the candidate segments of spectral abrupt change. The consistency index of pixel value changes in the local neighborhood is calculated. If the abrupt band shows synchronous change in most areas of the leaf, it is judged as having high spatial consistency; otherwise, it is removed as noise. The abrupt bands selected through two rounds of screening will be integrated to generate a set of feature vectors containing location, amplitude of change, rate of change of derivative and spatial consistency score. These vectors will then be physically verified using wax structure observation data to identify high-confidence interference signals. The identified interference signals are uniformly encoded according to the time dimension and spatial label to form a set of wax migration interference signals that includes timestamp, abrupt band number, change direction, change amplitude, spatial consistency score and corresponding wax index status.

5. The method for comprehensive evaluation of drought resistance of maize varieties according to claim 4, characterized in that, The spectral abrupt change candidate segments in the wax migration interference signal set are limited to the near-infrared 780nm to 880nm range and the red-edge 720nm to 740nm range. They are required to simultaneously meet the following conditions in the spatial consistency test: the spatial variation coefficient is lower than a set threshold and the local information entropy is lower than a set threshold, in order to be confirmed as a valid interference signal.

6. The method for comprehensive evaluation of drought resistance of maize varieties according to claim 1, characterized in that, The steps for constructing an energy exchange offset fingerprint include: The set of wax migration interference signals was aligned and matched with the time series of blade energy exchange parameters, and spatial resampling and interpolation were used to ensure that each interference signal corresponds to complete energy parameters. The matched interference signal and energy parameters are subjected to differential operation to calculate the change value of energy parameters, and compared with the parameter difference of the control area to obtain the normalized interference offset value. The normalized interference offset values ​​are encoded using multidimensional features to form a set of feature vectors that include time labels, spatial attributes, band categories, abrupt change magnitudes, and energy impact weights. Cluster analysis is performed on the feature vector set to extract typical features and group them to construct energy exchange offset fingerprint spectrum, generating a fingerprint set characterizing the offset of evapotranspiration rate and stomatal conductance. The energy exchange offset fingerprint is matched with the newly observed wax migration interference signal at time points. If the similarity exceeds a set threshold, the interference identification is completed; otherwise, the new features are saved to update the fingerprint set.

7. The method for comprehensive evaluation of drought resistance of maize varieties according to claim 6, characterized in that, The steps for establishing a multi-level correction function and coupling it with the porosity estimation model include: Based on the interference behavior patterns characterized in the energy exchange migration fingerprint spectrum, feature parameters of different abrupt change bands are extracted, and first-order and second-order response functions are established respectively to form a multi-level set of correction functions. The multi-level correction function is coupled with the stomatal conductance estimation model of maize leaves for joint modeling. The position, amplitude and spatial distribution parameters of the abrupt change band are used as covariates to input the stomatal conductance estimation model, and an expression that considers both spectral abrupt change and physiological response characteristics is formed through joint fitting. During the evaporation rate inversion process, when a new wax migration interference signal is identified, the matching correction function is called and the spectral abrupt change parameter is input to generate a correction factor. The correction value is then fused with the coupled stomatal conductance estimation result. Through iterative comparison and feedback adjustment, the evaporation rate inversion value is adaptively corrected in real time.

8. The method for comprehensive evaluation of drought resistance of maize varieties according to claim 7, characterized in that, The steps for iteratively comparing the corrected evaporation rate inversion results with the dynamic spectral monitoring baseline and generating a dynamic weight distribution include: The evaporation rate correction results were compared with the spectral response curves at the corresponding time points in the dynamic spectral monitoring baseline, and the residual values ​​on the near-infrared, red edge, and moisture absorption bands were calculated. The residuals at each time point are arranged in sequence to form a time series residual chain, and cumulative drift, periodic oscillation or sudden jump characteristics are identified by smoothing and derivative calculation. A dynamic weight distribution function is constructed based on the residual amplitude, rate of change, and spectral sensitivity to generate a weight vector corresponding to the confidence level at each time point. Dynamic weight distribution is applied to the weighting of historical correction results and the prediction and adjustment of future correction values, and the weights are updated when new data is added.

9. The method for comprehensive evaluation of drought resistance of maize varieties according to claim 1, characterized in that, The steps for constructing an adaptive closed-loop update strategy and writing the update results back to the dynamic spectral monitoring baseline and multi-level correction function include: An adaptive weight adjustment factor is constructed based on dynamic weight distribution, and a feedback control factor matrix is ​​generated. The feedback intensity characterizes the importance of each time point to the model update, and the temporal continuity is maintained through a time window smoothing mechanism. The feedback factor matrix is ​​applied to the dynamic spectral monitoring baseline and the multi-level correction function. The baseline achieves iterative adjustment of the spectral response curve through weighted sliding update and trend reconstruction. The correction function redistributes the parameters of the first-order and higher-order correction functions and introduces new offset fingerprint data through the feedback factor. The updated spectral monitoring baseline and multi-level correction function are reloaded into the evaporation rate inversion process, triggering iterative feedback operation when new data is input.

Images