A method and system for monitoring grassland vegetation cover based on spectral inflection points

By using an adaptive interval empirical model and a multi-factor comparative prediction formula, combined with a physical constraint loss function and a graded penalty verification, the adaptability and stability issues of grassland vegetation coverage monitoring were resolved, and accurate monitoring of grassland vegetation coverage was achieved.

CN123066076APending Publication Date: 2026-11-10INNER MONGOLIA AUTONOMOUS REGION FORESTRY MONITORING & PLANNING INST
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611546935.9
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-10-08
Publication Date
2026-11-10

AI Technical Summary

Technical Problem

Existing vegetation cover monitoring technologies cannot adaptively match the canopy structure characteristics of different grassland types, and suffer from problems such as large spectral detection errors and insufficient detection stability, especially in arid and semi-arid clump grassland environments where adaptability is poor.

Method used

By constructing an adaptive interval empirical model, the coverage interval is divided by identifying spectral inflection points. Combining multi-factor comparison prediction formulas and physical constraint loss functions, hierarchical adaptive boundary violation penalties and physical verification are adopted to achieve accurate monitoring of grassland vegetation coverage.

Benefits of technology

It achieves adaptive adaptation for grassland vegetation coverage monitoring, reduces spectral detection errors, improves detection stability and continuity, and meets the monitoring needs of complex grassland environments.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123066076A_ABST
    Figure CN123066076A_ABST
Patent Text Reader

Abstract

This invention discloses a grassland vegetation cover monitoring method and system based on spectral inflection points. Using vegetation spectral reflectance and canopy morphology physical parameters as the core detection basis, the method adaptively divides the grassland cover into low, medium, and high coverage intervals by fitting the coverage spectral response curve and using the second derivative physical inflection point. A canopy structure-climate modulation type nonlinear empirical formula is constructed for each interval to quantitatively correct spectral detection errors caused by leaf overlap. A composite loss function with physical consistency constraints is used to correct the global prediction formula, and a hierarchical adaptive out-of-bounds penalty and a secondary verification using the canopy radiation law are combined to correct abnormal results. Finally, the detection results are output after Sigmoid weighted smoothing of the transition zone. This invention has strong interval adaptability, significantly improves the full gradient detection accuracy, effectively suppresses outliers caused by gravel interference and canopy saturation, eliminates numerical jumps at segment boundaries, and provides continuous and stable results. It is suitable for grassland ecological monitoring, degraded grassland assessment, and other scenarios.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of grassland vegetation monitoring technology, specifically to a method and system for monitoring grassland vegetation coverage based on spectral inflection points. Background Technology

[0002] Current vegetation cover monitoring primarily focuses on the spectral reflectance physical characteristics of the vegetation canopy, using linear fitting of vegetation indices to invert coverage. Some schemes introduce leaf area index as an auxiliary parameter to attempt to correct errors caused by canopy structure. However, existing technologies still have the following shortcomings: To adapt to the nonlinear response characteristics of different coverage, existing segmented detection methods generally use artificially fixed thresholds to divide intervals. However, this method does not take into account the inherent physical inflection point of the vegetation canopy optical response, and cannot adaptively match the canopy structure characteristics of different grassland types. It has poor adaptability to the arid and semi-arid tufted grass grasslands of Inner Mongolia.

[0003] Existing empirical formulas mostly adopt multiple linear regression structures, without taking into account the physical laws of canopy radiative transmission. They cannot specifically quantify and correct the underestimation error in spectral detection caused by leaf overlap. They have large fitting deviations in low-coverage bare soil interference and high-coverage spectral saturation scenarios, and are prone to abnormal results that do not conform to the physical logic of plant growth.

[0004] Existing technical solutions often use fixed thresholds to truncate detection results that exceed reasonable ranges, lack differentiated grading and penalty mechanisms, and do not combine canopy morphology and physical parameters for physical consistency verification. As a result, the detection stability is insufficient and it is difficult to adapt to the complex field detection environment of grasslands. Summary of the Invention

[0005] The purpose of this invention is to provide a method and system for monitoring grassland vegetation coverage based on spectral inflection points, so as to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: A method for monitoring grassland vegetation cover based on spectral inflection points includes: S1. Construction of an adaptive interval division empirical model Full-gradient quadrats with coverage from 0% to 100% were set up to determine the gold standard of true vegetation cover, extract the soil-adjusted vegetation index, leaf area index, and auxiliary climate parameters; effective sample response curves were obtained by fitting local weighted regression, the second derivative of the response curves was calculated, the two maximum points of the absolute value of the second derivative were identified, three coverage intervals were divided, and the empirical formula was fitted to the samples in the three intervals using the weighted nonlinear least squares method. S2. Construction of Multi-Factor Comparison Prediction Formula and Correction of Physically Constrained Loss Function For the same input variables collected from the detection sample plots, the adaptive gradient descent algorithm is used to optimize the solution and obtain the corrected global comparison prediction formula; a loss function consisting of three parts is constructed, namely, the accuracy term, the physical consistency constraint term, and the coefficient regularization term. S3, Hierarchical Adaptive Boundary Penalty and Physical Verification Correction The corrected output value is compared with the total effective range of the empirical model. Adaptive penalties are applied to the prediction results, which are divided into mild and severe levels according to the degree of exceeding the limit. The physical consistency of the results after penalty is verified by combining the Beer-Lambert canopy law. S4. Precise matching of experience intervals and smooth output of transition zones. The final coverage after penalty correction is compared with the adaptive boundary point to determine the main interval to which it belongs. The Sigmoid weighted fusion of adjacent interval model results is applied to the samples in the transition zone of the boundary point. S5. Result Output The system outputs two types of core detection results: the final vegetation coverage value; the empirical interval to which the coverage belongs and the corresponding grassland type classification.

[0007] Preferably, the gold standard for obtaining the true vegetation cover in step S1 is as follows: the true vegetation cover is determined by the quadrat cut-out method. The true value of vegetation cover is obtained by converting the vertical projection area of ​​vegetation, which serves as the gold standard for model fitting. Soil-adjusted vegetation index was determined by measuring the reflectance spectrum of the canopy in the 350–2500 nm range using a ground-based spectrometer. Calculate the modified soil-adjusted vegetation index (MSAVI); where, Near-infrared band surface reflectance; Red band surface reflectance; As a translation term, setting +1 raises the baseline, reducing the sensitivity of the vegetation index to changes in soil background under bare soil conditions; It is a square construct; The red-near-infrared reflectance difference term characterizes the difference in vegetation absorption or reflection; output value The range of values ​​is limited to .

[0008] Preferably, the three coverage intervals are interval 1: Interval 2: (Corresponds to low-coverage degraded grassland) Corresponding to typical grassland with medium coverage; Interval 3: This corresponds to a high-coverage meadow grassland.

[0009] Preferably, the three intervals share a unified empirical formula structure, with only the coefficients differing. This empirical formula is based on As dependent variables, spectral index x and leaf area index The precipitation anomaly z is the independent variable, where, Represents the optical contribution module. For structural correction module, For baseline correction terms; By performing a weighted nonlinear fit on the above formulas and applying physical constraints to the parameters, three sets of interval-specific optimal coefficients are obtained: Scope of application ; Scope of application ; Scope of application ; The coefficients are interval-specific fitting coefficients, which are obtained independently by weighted nonlinear least squares fitting with physical constraints in intervals. n=1, 2, and 3 correspond to the low, medium, and high coverage intervals, respectively.

[0010] Preferably, the corrected global contrast prediction formula in step S2 is as follows: Solve the loss function and obtain the final comparative prediction formula. , The initial prediction coverage is obtained through the following process: Using ordinary least squares with all valid samples, a global linear prediction formula is fitted and used as the initial value for loss function optimization. The initial formula is: ,in, , , , The initial global coefficients are the basic parameters obtained by fitting the entire calibrated sample set without dividing the coverage interval; Spectral index; Leaf area index, For precipitation anomaly; The adaptive learning rate gradient descent algorithm is used to find the minimum value of the loss function, obtaining the optimal coefficients of the comparison formula. Based on the optimal coefficients, the final comparison prediction formula is derived. ; The loss function is as follows: The fitting accuracy term of the composite loss function uses an interval-weighted setting, with lower coverage intervals having a higher weight than higher coverage intervals; the physical consistency constraint term is used to penalize physical anomalies where leaf area index increases but leaf cover decreases, as detailed below: The loss function consists of three parts: fitting accuracy term, physical consistency constraint term, and coefficient regularization term, as shown in the following formula: , This represents the physical constraint weighting coefficient, ranging from 0.1 to 0.3. The regularization coefficient is 0.01 to 0.05, determined by the prior physical boundary of the canopy. Fitting accuracy term The weighted mean square error is used to measure the deviation between the output of the comparison formula and the output of the corresponding interval empirical model. , where n is the total number of samples; The predicted coverage of the i-th sample is calculated using the comparison formula; The output coverage of the empirical model for the interval corresponding to the i-th sample; Interval weights are set to 1.2 for interval 1, 1.0 for interval 2, and 0.8 for interval 3 to enhance the fitting accuracy for low-coverage scenarios. Physical consistency constraints Penalize predictions that do not conform to the physical laws of plant canopy. Physically, the larger the leaf area index (LAI), the more monotonically the vegetation cover decreases. After sorting the samples in ascending order of LAI, penalize the samples that "increase LAI but decrease predicted coverage" to ensure that the model conforms to physical logic. ,in, This is an indicator function. It takes the value 1 when the condition in parentheses is true, and 0 otherwise. When the LAI of adjacent samples increases, but the predicted coverage decreases, a physical penalty is triggered. The larger the deviation, the stronger the penalty. Coefficient regularization term L2 regularization is used to constrain the magnitude of the coefficients, avoiding the excessive dominance of a single factor in the results. .

[0011] Preferably, the adaptive penalty in step three is divided into a mild linear penalty and a severe exponential shrinkage penalty. The lower bound penalty intensity is negatively correlated with the reliability of the spectral data, and the upper bound penalty intensity is positively correlated with the degree of leaf area index saturation, as detailed below: The corrected comparison output value Total effective range of the empirical model That is, comparing ([0, 1]), we divide the states into three categories and five levels: Valid range: No penalty is required; the process proceeds directly to interval matching. Mild lower bound crossing: Weak punishment; Severe lower bound crossing: Severe punishment; Mild upper boundary violation: Weak punishment; Severe upper boundary violation: Severe punishment; Lower bound out-of-bounds adaptive penalty correction: For out-of-bounds values ​​that are less than the lower bounds of all empirical formulas, different penalty strategies are adopted according to the out-of-bounds level. The penalty strength is negatively correlated with the reliability of the spectral data, and finally corrected to the effective range of interval 1. Calculate the spectral confidence penalty coefficient: ; in The confidence level of the spectral data of the target sample plot (value from 0 to 1). The larger the value, the worse the spectral quality and the higher the penalty intensity.

[0012] Mild lower bound out-of-bounds linear penalty: The result is corrected to the 0~0.1 range using linear contraction, as shown in the formula: ; This is the final coverage after penalty correction; Severe lower bound out-of-bounds exponential shrinkage penalty: The result is shrunk to the limit using an exponential decay function. The larger the deviation from the boundary, the stronger the contraction. The formula is: ; in This is the lower bound contraction coefficient, typically set to 0.8, to ensure that severely out-of-bounds results ultimately fall within a reasonable range of low coverage. Upper bound out-of-bounds adaptive penalty correction: For out-of-bounds values ​​exceeding the upper limits of all empirical formulas, different penalty strategies are applied according to the out-of-bounds level. The penalty strength is positively correlated with the LAI saturation degree, and is finally corrected to the effective range of interval 3. Calculate the LAI saturation penalty coefficient: ; in, The grassland canopy LAI saturation threshold, The larger the value, the more severe the canopy saturation and the higher the penalty intensity. Mild upper bound out-of-bounds linear penalty: The result is corrected to the 0.9~1 range using linear contraction, as shown in the formula: ; Severe upper bound out-of-bounds exponential shrinkage penalty: The result is shrunk to the limit using an exponential decay function. The larger the deviation from the boundary, the stronger the contraction. The formula is: ; in This is the upper bound contraction coefficient, typically set to 0.6, to ensure that severely out-of-bounds results ultimately fall within a reasonable range of the high coverage interval; Post-penalty physical consistency verification: The physical rationality of the results after penalty correction is verified to ensure that the correction value conforms to the structural and physical laws of the plant canopy. Based on Beer-Lambert's law, the theoretical coverage range is calculated using measured LAI: , where k is the canopy extinction coefficient, preferably taken as 0.7; If the correction and If the relative deviation is greater than 20%, then... A second correction is made based on the baseline, and the weighted average of the two is taken: This ensures that the final result conforms to the physical properties of the plant canopy and avoids errors introduced by purely mathematical corrections.

[0013] Preferably, the boundary transition zone is divided into a low-to-medium coverage transition zone, a medium-to-high coverage transition zone, and a non-transition zone. For the low-to-medium coverage transition zone and the medium-to-high coverage transition zone, the Sigmoid weighting function is used to smoothly fuse the empirical model results of the two adjacent intervals to eliminate the numerical jump at the boundary point. For the non-transition zone, the final coverage after penalty correction is directly retained.

[0014] A grassland vegetation cover monitoring system based on spectral inflection points, used to implement a grassland vegetation cover monitoring method based on spectral inflection points, includes: Data acquisition unit: includes a ground object spectrometer to collect canopy reflectance spectral data in the full band of 350~2500nm, used to extract the modified soil adjusted vegetation index MSAVI; and a canopy LAI analyzer to measure the leaf area index (LAI) in the sample plot based on the three-dimensional structural physical properties of the vegetation canopy, used to characterize the total amount and degree of overlap of canopy leaves. Meteorological auxiliary data acquisition unit: includes an in-situ rainfall sensor or regional meteorological station docking module, which collects the cumulative precipitation data of the corresponding time period of the sample plot, and is used to calculate the precipitation anomaly percentage as a climate modulation auxiliary factor. Data processing unit: Equipped with five processing modules corresponding to the five steps in the monitoring method, namely: sample preprocessing and adaptive interval division module, interval semi-physical empirical model fitting module, global comparison prediction formula correction module, hierarchical adaptive over-boundary penalty and physical verification module, and interval matching and transition zone smoothing module. These modules are used to process the data acquired by the data acquisition unit.

[0015] Preferably, the sample preprocessing and adaptive interval division module is used to complete the quality control and screening of calibrated samples. By fitting the coverage-spectral response curve and identifying physical inflection points, it automatically divides three types of coverage intervals, providing a reasonable interval benchmark for subsequent segmented model construction. The segmented semi-physical empirical model fitting module targets three types of coverage intervals. Combining sample confidence weights and physical constraint rules, it uses a nonlinear fitting algorithm to obtain three sets of exclusive semi-physical empirical formulas, which serve as the segmented benchmark model for coverage detection. The global comparison prediction formula correction module is used to construct a composite loss function that includes fitting accuracy, physical consistency, and regularization constraints. The corrected global prediction formula is obtained through iterative optimization, balancing the detection accuracy and physical rationality of the model. The hierarchical adaptive outbound penalty and physical verification module is used to perform differentiated adaptive penalty correction on the abnormal outbound prediction results according to the level, and then complete the physical consistency secondary verification based on the canopy radiative transfer law to suppress outlier errors and ensure that the results conform to the laws of plant physics. The interval matching and transition zone smoothing module is used to match the coverage interval and grassland type corresponding to the detection results, perform adjacent model weighted smoothing on the results of the boundary transition zone, eliminate the numerical jumps at the segment boundaries, and output the final continuous and stable detection results.

[0016] Compared with the prior art, the beneficial effects of the present invention are: 1. This invention automatically divides the coverage interval by fitting the coverage spectral response curve and identifying the physical inflection point of the second derivative, replacing the traditional method of manually fixing the threshold. The interval division conforms to the physical laws of the canopy optical response and can adaptively adapt to the canopy characteristics of different grassland types.

[0017] 2. This invention constructs a nonlinear empirical formula for canopy structure and climate modulation, with the optical contribution module and structure correction module corresponding to the optical and morphological physical properties of vegetation, respectively, to quantitatively correct the spectral detection error caused by leaf overlap; compared with the traditional linear empirical model, the error is significantly reduced.

[0018] 3. This invention uses a composite loss function with physical consistency constraints to correct the prediction formula. Combined with hierarchical adaptive out-of-bounds penalty and secondary verification of the canopy radiation law, it not only ensures that the model output conforms to the physical logic of plants, but also differentiates and corrects abnormal results caused by gravel interference and canopy saturation, with significant outlier suppression effect.

[0019] 4. This invention uses the Sigmoid weighting function to smooth the detection results of the interval boundary transition zone, eliminating the numerical jump problem at the boundary point of the traditional segmented model. The coverage detection results are more continuous and adaptable to the continuous monitoring needs of large gradient coverage in grassland areas. Attached Figure Description

[0020] Figure 1 This is a schematic diagram of the main process of a grassland vegetation coverage monitoring method based on spectral inflection points according to the present invention; Figure 2 This is a data processing diagram of a grassland vegetation coverage monitoring method based on spectral inflection points according to the present invention. Figure 3 This is a schematic diagram illustrating the data penalty and judgment of a grassland vegetation coverage monitoring method based on spectral inflection points according to the present invention. Figure 4 This is a schematic diagram of the module functions of a grassland vegetation coverage monitoring system based on spectral inflection points according to the present invention. Detailed Implementation

[0021] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention. Please refer to... Figure 1-3 This invention proposes a grassland vegetation cover monitoring method based on spectral inflection points, which is implemented through the grassland vegetation cover monitoring system constructed below: S1. Construction of an adaptive interval division empirical model 1.1 Graded Sample Collection and Multi-Source Parameter Quality Control (1) Gold standard data collection: In the monitoring area, no less than 120 1m×1m standard quadrats were set up in grasslands with different degrees of degradation and different dominant species (such as sheepgrass, needlegrass, and Artemisia frigida, native species of Inner Mongolia). The coverage of 0%~100% of the full gradient was uniformly distributed. The actual vegetation coverage was determined by quadrat cutting method for the standard quadrat frames. The true value of vegetation coverage is obtained by converting the vertical projection area of ​​vegetation, which serves as the gold standard for model fitting.

[0022] (2) Collection of core physical parameters: The reflectance spectrum of the canopy in the range of 350~2500nm was measured using a ground object spectrometer, and the modified soil adjusted vegetation index MSAVI (denoted as variable x) was extracted. Leaf area index was determined using a canopy analyzer. (denoted as variable y), three points are measured evenly in each quadrat and the average value is taken to characterize the total number of canopy leaves and the degree of overlap.

[0023] (3) Auxiliary climate parameter collection: Obtain the cumulative precipitation of each sample plot for the past 30 days, calculate the precipitation anomaly percentage (denoted as variable z), and eliminate the difference in precipitation baselines in different regions.

[0024] 1.2 Adaptive Partitioning of Coverage Intervals Based on Inflection Point Detection

[0025] Instead of the traditional manually set fixed thresholds of 30% and 60%, the system automatically divides the intervals based on the physical inflection points of the vegetation spectral response curve, making the segmented model conform to the canopy optical response law: (1) Locally weighted regression is used to fit all valid samples. The response curve yields a smooth, continuous response function. (2) Calculate the second derivative of the response curve and identify the two maximum points of the absolute value of the second derivative, which are the two physical boundary points of the coverage interval. Divide the area into three coverage zones: Interval 1 (low-coverage degraded grassland): ; Interval 2 (typical grassland with medium cover): ; Section 3 (high-coverage meadow steppe): .

[0026] 1.3 Fitting of Interval-Weighted Multiple Linear Empirical Model

[0027] For the samples in the three intervals, a weighted nonlinear least squares method with physical constraints is used to fit the specific empirical formula. The sample weights are positively correlated with the measurement reliability. (1) Sample weight calculation: the weight of each sample It is determined by both the confidence level of the gold standard measurement and the quality of the spectral data: (in The confidence level for artificial vegetation cover measurement ranges from 0 to 1. The confidence level for artificial vegetation cover measurement is obtained by combining the ratio of the standard deviation to the mean of multiple groups of artificial vegetation cover measurements and normalizing the field observation conditions to obtain a confidence value of 0 to 1. Specifically, the confidence level for artificial vegetation cover measurement is: 1 - the ratio of the standard deviation to the mean of multiple groups of artificial vegetation cover measurements / the maximum value of the ratio of the standard deviation to the mean. The confidence level for spectral data quality ranges from 0 to 1. It involves screening for noise and cloud shadow interference pixel by pixel in the spectrum, calculating indicators such as signal-to-noise ratio and spectral continuity, and then weighting and normalizing them to obtain a confidence value of 0 to 1 that represents the usability of the spectrum. Specifically, it is calculated as: Spectral Data Quality Confidence Level = Indicator Weight 1 × Signal-to-Noise Ratio + Indicator Weight 2 × Spectral Continuity Index + Indicator Weight 3 × Effective Ratio of Cloud Shadow Mask. The weights of the three indicators are calculated using the entropy weight method and satisfy the condition: Indicator Weight 1 + Indicator Weight 2 + Indicator Weight 3 = 1.

[0028] (2) Interval empirical formula fitting: with As the dependent variable, the spectral index x, With precipitation anomaly z as the independent variable, a nonlinear formula for canopy structure and climate modulation is constructed. The three intervals share a unified structure, differing only in coefficients. The empirical formulas for the three intervals correspond to the three coverage intervals respectively:

[0029] ( The interval-specific fitting coefficients are obtained by independently solving the weighted nonlinear least squares fitting with physical constraints in the intervals. n=1, 2, 3 correspond to the low, medium and high coverage intervals, respectively. Represents the optical contribution module. For structural correction module, (This is a baseline correction term).

[0030] (3) Constraint fitting solution: The Levenberg-Marquardt algorithm is used for weighted nonlinear fitting, and physical constraints (extinction coefficient) are applied. The solution yields three sets of interval-specific optimal coefficients: Scope of application , The proportion of bare soil is high, and spectral optical differences dominate the identification of coverage, with weak overlap of canopy leaves; Scope of application , Optical and canopy structure contribute similarly, while precipitation has the most significant effect on vegetation growth. Scope of application , As the canopy spectrum approaches saturation, the contribution of optical detection decreases, and the proportion of errors caused by leaf overlap increases.

[0031] S2. Construction of Multi-Factor Comparison Prediction Formula and Correction of Physically Constrained Loss Function

[0032] (1) Data acquisition of target quadrats: For the target detection quadrats, the same equipment, acquisition conditions and preprocessing methods as in step S1 are used to obtain the three input variables of the target quadrats: spectral index Leaf area index Precipitation anomaly This ensures the homology of variables and eliminates systematic errors.

[0033] (2) Initialization of the initial global contrastive prediction formula: Using the ordinary least squares method with all valid samples, a global linear prediction formula is fitted and used as the initial value for loss function optimization. The initial formula is: (in, , , , The initial global coefficients are the basic parameters obtained by fitting the entire calibrated sample set without dividing the coverage interval. The optimal coefficients are obtained by solving the loss function using an adaptive learning rate gradient descent algorithm to find the minimum value. , , , The final comparative prediction formula is obtained based on the optimal coefficients. ; (Initial prediction coverage).

[0034] (3) Loss function construction: The loss function consists of three parts: fitting accuracy term, physical consistency constraint term, and coefficient regularization term, as shown in the following formula: , This represents the physical constraint weighting coefficient, ranging from 0.1 to 0.3. The regularization coefficient is 0.01 to 0.05, determined by the prior physical boundary of the canopy.

[0035] Fitting accuracy term The weighted mean square error is used to measure the deviation between the output of the comparison formula and the output of the corresponding interval empirical model. (Where, n is the total number of samples;) The predicted coverage of the i-th sample calculated using the comparison formula; The output coverage of the empirical model for the corresponding interval of the i-th sample; Interval weights are set: 1.2 for interval 1, 1.0 for interval 2, and 0.8 for interval 3, to enhance the fitting accuracy in low-coverage scenarios.

[0036] Physical consistency constraints Penalize predictions that do not conform to the physical laws of plant canopy. Physically, the larger the leaf area index (LAI), the more monotonically the vegetation cover decreases. After sorting the samples in ascending order of LAI, penalize the samples that "increase LAI but decrease predicted coverage" to ensure that the model conforms to physical logic. (in, This represents the predicted coverage of the (i+1)th sample. Let be the leaf area index of the i-th sample; Let be the leaf area index of the (i+1)th sample; (This is an indicator function; it takes the value 1 if the condition in parentheses is true, and 0 otherwise.) When the LAI of adjacent samples increases but the predicted coverage decreases, a physical penalty is triggered, and the larger the deviation, the stronger the penalty.

[0037] Coefficient regularization term L2 regularization is used to constrain the magnitude of the coefficients, avoiding the excessive dominance of a single factor in the results. .

[0038] S3, Hierarchical Adaptive Boundary Penalty and Physical Verification Correction

[0039] (1) Determination of boundary crossing status

[0040] The corrected comparison output value Total effective range of the empirical model That is, comparing ([0, 1]), we divide the states into three categories and five levels: Valid range: No penalty is required; the process proceeds directly to interval matching. Mild lower bound crossing: Weak punishment; Severe lower bound crossing: Severe punishment; Mild upper boundary violation: Weak punishment; Severe upper boundary violation: Severe punishment.

[0041] (2) Adaptive penalty correction for lower bound out-of-bounds

[0042] For out-of-bounds values ​​that are less than the lower limits of all empirical formulas, different penalty strategies are applied according to the degree of out-of-bounds violation. The penalty strength is negatively correlated with the reliability of the spectral data (the worse the spectral quality, the stronger the penalty), and finally corrected to the effective range of interval 1. 1) Calculate the spectral confidence penalty coefficient: ; in The confidence level of the spectral data of the target sample plot (value from 0 to 1). The larger the value, the worse the spectral quality and the higher the penalty intensity.

[0043] 2) Mild lower bound out-of-bounds linear penalty: The result is corrected to the 0~0.1 range using linear contraction, as shown in the formula: ; 3) Severe lower bound out-of-bounds exponential contraction penalty: The result is contracted to the specified value using an exponential decay function. The larger the deviation from the boundary, the stronger the contraction. The formula is: ; in This is the lower bound contraction coefficient, typically set to 0.8, to ensure that severely out-of-bounds results ultimately fall within a reasonable range of low coverage.

[0044] (3) Adaptive penalty correction for upper bound overrun

[0045] For out-of-bounds values ​​exceeding the upper limits of all empirical formulas, different penalty strategies are applied according to the out-of-bounds level. The penalty strength is positively correlated with the degree of LAI saturation (the higher the LAI, the more severe the canopy saturation, and the stronger the penalty), ultimately correcting it to the effective range of interval 3. 1) Calculate the LAI saturation penalty coefficient: ; in, The grassland canopy LAI saturation threshold is set at 4.0 (preferably). The larger the value, the more severe the canopy saturation and the higher the penalty intensity.

[0046] 2) Mild upper bound out-of-bounds linear penalty: The result is corrected to the 0.9~1 range using linear contraction, as shown in the formula: .

[0047] 3) Severe upper bound out-of-bounds exponential contraction penalty: The result is contracted to the limit using an exponential decay function. The larger the deviation from the boundary, the stronger the contraction. The formula is: ; in This is the upper bound contraction coefficient, typically set to 0.6, ensuring that severely out-of-bounds results ultimately fall within a reasonable range of the high coverage interval.

[0048] (4) Physical consistency check after penalty

[0049] The physical rationality of the corrected results is verified to ensure that the corrected values ​​conform to the structural and physical laws of the plant canopy. Based on Beer-Lambert's law, the theoretical coverage range is calculated using measured LAI: , where k is the extinction coefficient of the canopy, preferably taken as 0.7.

[0050] If the correction and If the relative deviation is greater than 20%, then... A second correction is made based on the baseline, and the weighted average of the two is taken: This ensures that the final result conforms to the physical properties of the plant canopy and avoids errors introduced by purely mathematical corrections.

[0051] S4. Precise matching of experience intervals and smooth output of transition zones.

[0052] (1) Preliminary determination of the main interval

[0053] Final coverage after penalty correction With adaptive boundary point Based on the comparison, the main interval to which it belongs can be preliminarily determined: 1) If Preliminary determination indicates it falls within interval 1 (low-coverage degraded grassland), corresponding to the empirical formula. ; 2) If Preliminary determination indicates it falls within interval 2 (medium-coverage typical grassland), corresponding to the empirical formula. ; 3) If Preliminary determination indicates it falls within interval 3 (high-coverage meadow-steppe), corresponding to the empirical formula. .

[0054] (2) Smoothing correction of transition zone

[0055] like Falling into the transition zone specified by S1 ( The Sigmoid weighting function is used to smoothly fuse the empirical model results of two adjacent intervals, eliminating numerical jumps at the boundary points. 1) Low-to-medium coverage transition zone ( ); Calculate the smoothing weights: ; in The smoothing coefficient is preferably set to 50 to control the smoothness of the transition zone.

[0056] Final coverage after smoothing correction: .

[0057] 2) Medium-to-high coverage transition zone ( ); Similarly, the empirical model results of interval 2 and interval 3 are fused using Sigmoid weights to ensure that the coverage transitions smoothly and continuously at the boundary.

[0058] 3) Non-transition zone areas: directly preserved As a final result .

[0059] S5. Result Output

[0060] Output two types of core detection results: final vegetation coverage value The empirical range of grassland coverage and the corresponding grassland type classification.

[0061] A grassland vegetation cover monitoring system based on spectral inflection points, used to implement the aforementioned grassland vegetation cover monitoring method based on spectral inflection points, includes: Data acquisition unit: includes a ground object spectrometer to collect canopy reflectance spectral data in the full band of 350~2500nm, used to extract the modified soil adjusted vegetation index MSAVI; and a canopy LAI analyzer to measure the leaf area index (LAI) in the sample plot based on the three-dimensional structural physical properties of the vegetation canopy, used to characterize the total amount and degree of overlap of canopy leaves. Meteorological auxiliary data acquisition unit: includes an in-situ rainfall sensor or regional meteorological station docking module, which collects the cumulative precipitation data of the corresponding time period of the sample plot, and is used to calculate the precipitation anomaly percentage as a climate modulation auxiliary factor. Data processing unit: Equipped with five processing modules corresponding to the five steps in the monitoring method, namely: sample preprocessing and adaptive interval division module, interval semi-physical empirical model fitting module, global comparison prediction formula correction module, hierarchical adaptive over-boundary penalty and physical verification module, and interval matching and transition zone smoothing module. These modules are used to process the data acquired by the data acquisition unit.

[0062] The sample preprocessing and adaptive interval division module is used to complete the quality control and screening of calibrated samples. By fitting the coverage-spectral response curve and identifying physical inflection points, it automatically divides three types of coverage intervals, providing a reasonable interval benchmark for subsequent segmented model construction.

[0063] The segmented physical empirical model fitting module, for three types of coverage intervals, combines sample confidence weights and physical constraint rules, and uses a nonlinear fitting algorithm to obtain three sets of exclusive physical empirical formulas, which serve as the segmented benchmark model for coverage detection.

[0064] The global comparison prediction formula correction module is used to construct a composite loss function that includes fitting accuracy, physical consistency, and regularization constraints. The corrected global prediction formula is obtained through iterative optimization, balancing the detection accuracy and physical rationality of the model.

[0065] The hierarchical adaptive out-of-bounds penalty and physical verification module is used to perform differentiated adaptive penalty correction on the abnormal out-of-bounds prediction results according to the level, and then complete the physical consistency secondary verification based on the canopy radiative transfer law to suppress outlier errors and ensure that the results conform to the physical laws of plants.

[0066] The interval matching and transition zone smoothing module is used to match the coverage interval and grassland type corresponding to the detection results, perform adjacent model weighted smoothing on the results of the boundary transition zone, eliminate the numerical jumps at the segment boundaries, and output the final continuous and stable detection results.

[0067] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for monitoring grassland vegetation cover based on spectral inflection points, characterized in that, include: S1. Construction of an adaptive interval division empirical model Full-gradient quadrats with coverage from 0% to 100% were set up to determine the gold standard of true vegetation cover. The soil-adjusted vegetation index, leaf area index, and auxiliary climate parameters were extracted. The response curves of effective samples were obtained by fitting local weighted regression. The second derivative of the response curves was calculated, and the two maximum points of the absolute value of the second derivative were identified. Three coverage intervals were divided, and the empirical formula was fitted to the samples in the three intervals using the weighted nonlinear least squares method. S2. Construction of Multi-Factor Comparison Prediction Formula and Correction of Physically Constrained Loss Function For the same input variables collected from the detection sample plots, the adaptive gradient descent algorithm is used to optimize the solution and obtain the corrected global comparison prediction formula; a loss function consisting of three parts is constructed, namely, the accuracy term, the physical consistency constraint term, and the coefficient regularization term. S3, Hierarchical Adaptive Boundary Penalty and Physical Verification Correction The corrected output value is compared with the total effective range of the empirical model. Adaptive penalties are applied to the prediction results, which are divided into mild and severe levels according to the degree of exceeding the limit. The physical consistency of the results after penalty is verified by combining the Beer-Lambert canopy law. S4. Precise matching of experience intervals and smooth output of transition zones. The coverage after penalty correction is compared with the adaptive boundary point to determine the main interval to which it belongs. The Sigmoid weighted fusion of adjacent interval model results is applied to the samples in the transition zone of the boundary point. S5. Result Output The system outputs two types of core detection results: the final vegetation coverage value; the empirical interval to which the coverage belongs and the corresponding grassland type classification.

2. The grassland vegetation cover monitoring method based on spectral inflection points according to claim 1, characterized in that: The process for obtaining the gold standard for true vegetation cover in step S1 is as follows: the true vegetation cover is determined by the quadrat clipping method. The true value of vegetation coverage is obtained by converting the vertical projection area of ​​vegetation, which serves as the gold standard for model fitting.

3. The grassland vegetation cover monitoring method based on spectral inflection points according to claim 1, characterized in that: The three coverage intervals are as follows: Interval 1: Interval 2: (Corresponds to low-coverage degraded grassland) Corresponding to typical grassland with medium coverage; Interval 3: This corresponds to a high-coverage meadow grassland.

4. The grassland vegetation cover monitoring method based on spectral inflection points according to claim 1, characterized in that: The three intervals share a unified empirical formula structure, differing only in their coefficients. This empirical formula is based on As dependent variables, spectral index x and leaf area index The precipitation anomaly z is the independent variable, where, Represents the optical contribution module. For structural correction module, For baseline correction terms; By performing a weighted nonlinear fit on the above formulas and applying physical constraints to the parameters, three sets of interval-specific optimal coefficients are obtained: Scope of application ; Scope of application ; Scope of application ; The coefficients are interval-specific fitting coefficients, obtained independently through weighted nonlinear least squares fitting; n=1, 2, 3, corresponding to the low, medium, and high coverage intervals, respectively.

5. The grassland vegetation cover monitoring method based on spectral inflection points according to claim 1, characterized in that: The corrected global contrast prediction formula in step S2 is as follows: Solve the loss function and obtain the final comparative prediction formula. , The initial prediction coverage is obtained through the following process: Using ordinary least squares with all valid samples, a global linear prediction formula is fitted and used as the initial value for loss function optimization. The initial formula is: ,in, , , , The initial global coefficients are the basic parameters obtained by fitting the entire calibrated sample set without dividing the coverage interval; Spectral index; Leaf area index, For precipitation anomaly; The adaptive learning rate gradient descent algorithm is used to find the minimum value of the loss function, obtaining the optimal coefficients of the comparison formula. Based on the optimal coefficients, the final comparison prediction formula is derived. ; The loss function is as follows: The fitting accuracy term of the composite loss function uses an interval-weighted setting, with lower coverage intervals having a higher weight than higher coverage intervals; the physical consistency constraint term is used to penalize physical anomalies where leaf area index increases but leaf cover decreases, as detailed below: The loss function consists of three parts: fitting accuracy term, physical consistency constraint term, and coefficient regularization term, as shown in the following formula: , This represents the physical constraint weighting coefficient, ranging from 0.1 to 0.

3. The regularization coefficient is 0.01 to 0.05, determined by the prior physical boundary of the canopy. Fitting accuracy term The weighted mean square error is used to measure the deviation between the output of the comparison formula and the output of the corresponding interval empirical model. , where n is the total number of samples; The predicted coverage of the i-th sample is calculated using the comparison formula; The output coverage of the empirical model for the interval corresponding to the i-th sample; Interval weights are set to 1.2 for interval 1, 1.0 for interval 2, and 0.8 for interval 3 to enhance the fitting accuracy for low-coverage scenarios. Physical consistency constraints To penalize predictions that do not conform to the physical laws of plant canopy, samples are sorted in ascending order of LAI, and samples with "increased LAI but decreased predicted coverage" are penalized to ensure that the model conforms to physical logic. ,in, This represents the predicted coverage of the (i+1)th sample. Let be the leaf area index of the i-th sample; Let be the leaf area index of the (i+1)th sample; This is an indicator function. It takes the value 1 when the condition in parentheses is true, and 0 otherwise. When the LAI of adjacent samples increases, but the predicted coverage decreases, a physical penalty is triggered. The larger the deviation, the stronger the penalty. Coefficient regularization term L2 regularization is used to constrain the magnitude of the coefficients, avoiding the excessive dominance of a single factor in the results. .

6. The grassland vegetation cover monitoring method based on spectral inflection points according to claim 1, characterized in that: The adaptive penalty in step S3 is divided into a mild linear penalty and a severe exponential shrinkage penalty. The lower bound penalty intensity is negatively correlated with the reliability of the spectral data, and the upper bound penalty intensity is positively correlated with the degree of leaf area index saturation, as detailed below: The corrected comparison output value Total effective range of the empirical model In comparison, the states are divided into three categories and five levels: Valid range: No penalty is required; the process proceeds directly to interval matching. Mild lower bound crossing: Weak punishment; Severe lower bound crossing: Severe punishment; Mild upper boundary violation: Weak punishment; Severe upper boundary violation: Severe punishment; Lower bound out-of-bounds adaptive penalty correction: For out-of-bounds values ​​that are less than the lower bounds of all empirical formulas, different penalty strategies are adopted according to the out-of-bounds level. The penalty strength is negatively correlated with the reliability of the spectral data, and finally corrected to the effective range of interval 1. Calculate the spectral confidence penalty coefficient: ; in The confidence level of the spectral data of the target sample plot, with a value of 0 to 1. The larger the value, the worse the spectral quality and the higher the penalty intensity. Mild lower bound out-of-bounds linear penalty: The result is corrected to the 0~0.1 range using linear contraction, as shown in the formula: ; This is the final coverage after penalty correction; Severe lower bound out-of-bounds exponential shrinkage penalty: The result is shrunk to the limit using an exponential decay function. The larger the deviation from the boundary, the stronger the contraction. The formula is: ; in The lower bound contraction coefficient is set to 0.8 to ensure that the results of severe boundary violations eventually fall within a reasonable range of low coverage intervals. Upper bound out-of-bounds adaptive penalty correction: For out-of-bounds values ​​exceeding the upper limits of all empirical formulas, different penalty strategies are applied according to the out-of-bounds level. The penalty strength is positively correlated with the LAI saturation degree, and is finally corrected to the effective range of interval 3. Calculate the LAI saturation penalty coefficient: ; in, The grassland canopy LAI saturation threshold, The larger the value, the more severe the canopy saturation and the higher the penalty intensity. Mild upper bound out-of-bounds linear penalty: The result is corrected to the 0.9~1 range using linear contraction, as shown in the formula: ; Severe upper bound out-of-bounds exponential shrinkage penalty: The result is shrunk to the limit using an exponential decay function. The larger the deviation from the boundary, the stronger the contraction. The formula is: ; in This is the upper bound contraction coefficient, typically set to 0.6, to ensure that severely out-of-bounds results ultimately fall within a reasonable range of the high coverage interval; Post-penalty physical consistency verification: The physical rationality of the results after penalty correction is verified to ensure that the correction value conforms to the structural and physical laws of the plant canopy. Based on Beer-Lambert's law, the theoretical coverage range is calculated using measured LAI: , where k is the canopy extinction coefficient, preferably taken as 0.7; Leaf area index; If the correction and If the relative deviation is greater than 20%, then... A second correction is made based on the baseline, and the weighted average of the two is taken: .

7. The grassland vegetation cover monitoring method based on spectral inflection points according to claim 1, characterized in that: The boundary transition zone is divided into low-to-medium coverage transition zone, medium-to-high coverage transition zone, and non-transition zone region. For the low-to-medium coverage transition zone and the medium-to-high coverage transition zone, a weighted function is used to smoothly fuse the empirical model results of two adjacent intervals to eliminate the numerical jump at the boundary point. For the non-transition zone region, the final coverage after penalty correction is directly retained.

8. A grassland vegetation cover monitoring system based on spectral inflection points, used to implement the grassland vegetation cover monitoring method based on spectral inflection points as described in any one of claims 1-7, characterized in that... ,include: Data acquisition unit: includes a ground object spectrometer for collecting canopy reflectance spectral data and extracting the modified soil vegetation index (MSAVI); and a canopy LAI analyzer for measuring the leaf area index (LAI) within the quadrat based on the three-dimensional structural physical properties of the vegetation canopy, which characterizes the total amount and overlap of canopy leaves. Meteorological auxiliary data acquisition unit: includes in-situ rainfall sensor and regional meteorological station docking module, collects cumulative precipitation data of corresponding time period of sample plot, and uses it to calculate precipitation anomaly percentage as climate modulation auxiliary factor; Data processing unit: Equipped with processing modules corresponding to each step in the vegetation cover monitoring method, namely: sample preprocessing and adaptive interval division module, interval semi-physical empirical model fitting module, global comparison prediction formula correction module, hierarchical adaptive boundary penalty and physical verification module, and interval matching and transition zone smoothing module. These modules are used to process the data acquired by the data acquisition unit.

9. A grassland vegetation cover monitoring system based on spectral inflection points according to claim 8, characterized in that: The sample preprocessing and adaptive interval division module is used to complete the quality control and screening of calibrated samples. It automatically divides three types of coverage intervals by fitting coverage, spectral response curves, and identifying physical inflection points. The segmented physical empirical model fitting module, for three types of coverage intervals, combines sample confidence weights and physical constraint rules, and uses a nonlinear fitting algorithm to obtain three sets of exclusive physical empirical formulas, which serve as the segmented benchmark model for coverage detection. The global comparison prediction formula correction module is used to construct a composite loss function that includes fitting accuracy, physical consistency, and regularization constraints, and obtains the corrected global prediction formula through iterative optimization. The hierarchical adaptive out-of-bounds penalty and physical verification module is used to perform differentiated adaptive penalty correction on the prediction results of abnormal out-of-bounds according to the level, and then complete the physical consistency secondary verification based on the canopy radiative transfer law. The interval matching and transition zone smoothing module is used to match the coverage interval and grassland type corresponding to the detection results, perform adjacent model weighted smoothing on the results of the boundary transition zone, eliminate the numerical jumps at the segment boundaries, and output the final continuous and stable detection results.