Vegetation leaf dry matter content remote sensing high-precision inversion method based on wavelet analysis

Abstract

The invention relates to a vegetation leaf dry matter content remote sensing high-precision inversion method based on wavelet analysis. The method comprises the following steps that S1, a leaf sample database is constructed; S11, a leaf actual measurement data set is constructed; s12, constructing a blade simulation data set; s13, data set division: dividing an actual measurement data set into an actual measurement training set and an actual measurement verification set according to wavelet features; s2, extracting dry matter feature information through wavelet transform; S2, performing multi-scale decomposition on the leaf spectral signal by using continuous wavelet transform; s22, adopting Morlet wavelet as a generating function to carry out continuous wavelet transform analysis on the original reflection spectrum of the blade; s23, carrying out correlation test on the obtained wavelet coefficient and an LMA value; s24, according to a set threshold value, wavelet coefficient characteristics sensitive to the dry matter content are screened out; and S3, constructing an LMA inversion model based on the wavelet coefficient characteristics and the spectral indexes. The method is high in inversion precision, strong in robustness and good in stability.

Description

Technical Field

[0001] This application relates to a high-precision remote sensing inversion method for the dry matter content of vegetation leaves based on wavelet analysis, which is mainly applicable to fields such as agricultural monitoring, forest ecosystem research, and ecological environment assessment. Background Technology

[0002] Leaf mass per area (LMA) is an important biochemical parameter characterizing the functional traits and physiological state of vegetation. This indicator not only reflects the vegetation's ability to regulate photosynthetic product allocation and resource utilization strategies, but is also closely related to plant adaptation to stresses such as drought and poor soil conditions. Therefore, obtaining high-precision information on leaf mass per area is of great significance for conducting research on vegetation physiological and ecological processes and assessing ecosystem function.

[0003] Currently, methods for obtaining leaf dry matter content mainly fall into two categories: laboratory chemical analysis and remote sensing inversion. The former, such as drying methods and chemical digestion methods, can provide relatively accurate results, but suffers from drawbacks such as being highly destructive, time-consuming, and costly, making it difficult to meet the needs of large-scale and continuous monitoring. In contrast, remote sensing inversion based on spectral information has advantages such as speed, non-destructive nature, and scalability, making it a research hotspot.

[0004] Existing remote sensing inversion methods can be broadly classified into three categories: (i) Empirical model methods: These methods involve constructing vegetation indices or using statistical regression analysis to determine the relationship between leaf spectra and dry matter content. While simple, they have poor universality and are significantly limited by sample and environmental conditions. (ii) Radiative transfer model methods: Represented by optical models such as PROSPECT, these methods can explain the relationship between spectra and biochemical parameters mechanistically. However, their inversion often relies on bidirectional inputs of reflectance and transmittance. In practical applications, transmittance data is difficult to obtain, leading to ill-posedness in model inversion. (iii) Machine learning methods: These methods utilize nonlinear algorithms to fit the complex relationship between spectra and biochemical parameters. While this can improve accuracy, the model results depend on the quality of the training dataset and are prone to "black box" problems.

[0005] It should be noted that LMA has weak spectral characteristics and is often masked by the spectral characteristics of components such as chlorophyll and water, which limits the accuracy and stability of traditional methods in dry matter retrieval. Summary of the Invention

[0006] To address the shortcomings of existing remote sensing methods for retrieving leaf dry matter content, such as insufficient accuracy, strong data dependence, and susceptibility to interference from moisture and chlorophyll, this application presents a high-precision remote sensing method for retrieving leaf dry matter content based on wavelet analysis. This method can effectively extract the most relevant sensitive bands in the leaf spectrum and construct a new dry matter index, achieving high-precision and stable retrieval of leaf dry matter content for different vegetation types.

[0007] The basic principle of this application is as follows: Wavelet analysis is a multi-scale spectral signal processing method that can decompose the continuous spectrum of vegetation into wavelet components of different frequencies and scales, thereby effectively extracting weak features masked by interference from other components. In vegetation remote sensing research, wavelet transform has been proven to have advantages in noise suppression and extraction of dry matter information in sensitive bands. By selecting bands highly correlated with LMA based on wavelet decomposition and then constructing a specialized dry matter index, the accuracy and robustness of the inversion results can be improved while maintaining the simplicity of the method.

[0008] The technical solution adopted in this application to solve the above-mentioned technical problems is: a high-precision remote sensing inversion method for the dry matter content of vegetation leaves based on wavelet analysis, which mainly includes the following steps: A high-precision remote sensing inversion method for vegetation leaf dry matter content based on wavelet analysis mainly includes the following steps: S1: Construction of the Leaf Sample Database S11: Construct a leaf measurement dataset, collect and organize the measurement data, collect the leaf reflectance spectrum and dry matter absorption coefficient, and ensure that the two have the same spectral resolution to ensure the accuracy of subsequent calculations; S12: Construct a simulated dataset of blades. Based on the PROSPECT radiative transfer model, construct a simulated dataset corresponding to the measured dataset of blades. At the same time, in order to improve the applicability of the model to actual observation conditions, two types of noise are superimposed on the simulated dataset: one is fixed noise (0.005), and the other is relative noise (0.003). S13: Dataset partitioning. Using a stratified sampling method, the measured dataset is divided into a measured training set and a measured validation set based on wavelet features, for example, using a 6:4 ratio. The measured training set is used to construct and calibrate the parameters of the leaf dry matter content inversion model, while the measured validation set is used to test the model's prediction performance under the same data source. S2: Wavelet transform extracts dry matter feature information S21: Multi-scale decomposition of the leaf spectral signal (reflectance data in the 400 nm–2500 nm range) is performed using continuous wavelet transform, expressed as follows:

[0009] In the formula, f(t) is the input spectral signal, W(a,b) are the corresponding wavelet coefficients; the scale parameter a controls the degree of scaling of the wavelet, with large scales corresponding to the overall features of low frequency and small scales corresponding to the detailed features of high frequency; the translation parameter b is used to locate the signal features at different wavelengths; ψ(t) is the mother wavelet function, which is essentially a local signal with zero mean and oscillatory characteristics. S22: Using Morlet wavelet as the mother function, a continuous wavelet decomposition of 128 scales is performed on each leaf spectral sample. The Morlet wavelet function contains a center frequency parameter ω0, which determines the oscillation frequency of the wavelet and thus affects the capture effect of spectral details at different scales. The expression is as follows:

[0010] In the formula, t is the wavelength of the reflection spectrum, and w o is the center frequency of the wave, used to control the oscillation frequency of the Morlet wavelet; i is the imaginary unit. S23: Perform a correlation test between the obtained wavelet coefficients and LMA values, using the Pearson correlation coefficient as the indicator. The calculation formula is as follows:

[0011] In the formula, Indicates the first i Wavelet coefficients; This represents the corresponding LMA value; and These are the means of the two sets of variables. The correlation coefficient ranges from [-1, 1]: when r is close to 1, it indicates a strong positive correlation; when it is close to -1, it indicates a strong negative correlation; when it is close to 0, it indicates that the two are almost uncorrelated. S24: Based on the set correlation threshold, select the wavelet coefficient feature intervals that are sensitive to dry matter content from all wavelet coefficients to provide input variables for subsequent inversion modeling; S3: LMA Inversion Modeling Based on Wavelet Coefficient Characteristics and Spectral Indices S31: Calculate the traditional dry matter spectral index using the original leaf reflectance data, and establish an LMA inversion model based on the spectral index. The formula is as follows: NDMI = (R 1649 -R 1722 ) / (R 1649 +R 1722 (1) ND LMA =(R 1368 -R 1722 ) / (R 1368+R 1722 (2) In the formula, R 1649 R 1722 R 1368 These represent the leaf reflectance at 1649 nm, 1722 nm, and 1368 nm, respectively. S32: The wavelet coefficient features and spectral index are fused into a model. The spectral bands from the sensitive wavelet coefficient feature interval obtained in step S24 are used as input variables. These spectral bands have a clear distribution in the scale-wavelength space and can effectively characterize the spectral response of the LMA value. Subsequently, for the two benchmark indices (Equation 1 and Equation 2), considering that the 1722nm band has a relatively stable response to leaf structure and dry matter, it is retained to reduce background interference while ensuring that the index calculation logic remains consistent with the existing normalized indices. Based on this, a new spectral index based on the sensitive wavelet coefficient feature interval data is constructed, with the following specific form: ND LN =(R bi -R 1722 ) / (R bi +R 1722 (i=1,2,3…) (3) Among them, ND LN For the newly constructed dry matter index, bi is the i-th highly correlated band (a special case is 21 bands from 1628nm to 1648nm), and R bi Let be the reflectance of the i-th highly correlated band.

[0012] S33: Based on Formula 3, an LMA inversion model was constructed, and its inversion effect was compared and verified with the LMA inversion models constructed based on Formula 1 and Formula 2. The optimal dry matter index (i.e., bi=1645nm was determined to be the optimal result of Formula 3) and the LMA inversion model based on the optimal dry matter index were obtained. This index better reflects the characteristic differences between the near-infrared and short-wave infrared bands and has a better correlation with the dry matter content of leaves.

[0013] The publicly available measured dataset of leaves mentioned in step S11 of this application includes the reflectance spectrum of fresh leaves and the dry matter content of fresh leaves, as detailed in Table 1.

[0014] In step S12 of this application, the radiative transfer model used is the PROSPECT-D model. This model includes seven input parameters: leaf structure parameter N (dimensionless), chlorophyll content Chl (μg / cm²), carotenoid content Car (μg / cm²), brown pigment content Cbp (dimensionless), dry matter content Cm (g / cm²), equivalent water thickness Cw (cm), and anthocyanin content Ant (μg / cm²). To ensure that the generated simulation dataset reflects the true characteristics of leaves under natural conditions, this application reviewed relevant published literature to statistically analyze the value ranges of biochemical indicators for normal leaves, and based on this, limited the value ranges of each input variable (see Table 2). Running the model under parameter constraints, a simulated sample dataset containing leaf reflectance spectra and corresponding dry matter content was finally obtained. The inversion effect comparison and verification in step S33 of this application includes the following aspects: 1. A comparative verification of the inversion results of the traditional vegetation index LMA universal inversion model based on raw reflectance and the inversion results of the new LMA universal inversion model constructed based on wavelet coefficients, with the evaluation based on the dry matter index ND. LN (Formula 3) The fitting accuracy and predictive ability of the LMA universal inversion model constructed from the relationship between LMA values; 2. Add absolute noise (0.005) and relative noise (0.003) to the data in the simulated dataset (this step can also be done in step S12), and extract wavelet coefficient features according to step S2, carry out correlation analysis, and evaluate the robustness of the newly constructed LMA universal inversion model under noise interference by comparing it with the inversion results under noise-free conditions. 3. The inversion results of several measured datasets are compared and verified to prove the applicability and reliability of the model established in this application under different vegetation types and data conditions.

[0015] This application uses normalized root mean square error (nRMSE) and coefficient of determination (R²). 2 These two evaluation metrics are used to verify the accuracy of the model, and the formulas are as follows:

[0016]

[0017] In the formula, n represents the total sample size. These are the measured observations of the sample. The predicted value output by the model. This is the sample mean. These correspond to the maximum and minimum values ​​in the actual measured values ​​of the sample, respectively.

[0018] Compared with existing technologies, this application effectively extracts weak spectral feature information of dry matter from vegetation leaves through continuous wavelet transform, and constructs new vegetation indices based on wavelet analysis, thereby enhancing the expressive power of spectral information and making the inversion accuracy significantly better than traditional dry matter spectral indices. Simultaneously, the wavelet coefficient features are highly sensitive to dry matter content, making the inversion results more accurate and reliable. Furthermore, continuous wavelet transform exhibits strong robustness to noise; even at high scales, the wavelet coefficient features of both measured and simulated datasets in the 1645nm band remain largely unaffected by noise. Attached Figure Description

[0019] Figure 1 This section presents correlation feature maps between wavelet coefficients and dry matter content on measured, simulated, and combined datasets, respectively. The study found that in the measured dataset (… Figure 1 (above) and on the PROSPECT simulation dataset ( Figure 1 In the study, wavelet coefficients related to leaf dry matter content were mainly distributed in the 1150-1300 nm and 1570-1700 nm bands. The wavelet coefficients in the 1150-1300 nm band were correlated with moisture content, stemming from the broad absorption state of the equivalent water thickness (EWT) in this band and its positive correlation with dry matter. The wavelet coefficients in the 1570-1700 nm band reflected the spectral characteristics of components such as starch and lignin in the dry matter. A comprehensive dataset was obtained by weighted averaging of the measured and simulated datasets, yielding 89 wavelet coefficient features that were significantly correlated in both datasets. Figure 1 (See the composite diagram below).

[0020] Figure 2 This section presents correlation feature maps between wavelet coefficients and dry matter content on the measured dataset, the noisy simulated dataset, and the combined dataset obtained by weighted averaging of the measured and noisy simulated datasets. To explore the robustness of continuous wavelet transform and verify the impact of noise on wavelet coefficient features, relative and absolute noise were added to the PROSPECT simulated dataset. Figure 2 As can be seen in the (middle) diagram, after adding noise to the simulated dataset, the Morlet wavelet, due to the low-scale wavelet coefficient characteristics, is quite sensitive to noise changes, and the signal-to-noise ratio is low in the 1150-1300nm band. Noise masks the weak absorption signal, and the relevant features disappear due to noise. However, in the comprehensive diagram (… Figure 2 As can be seen below, the wavelet coefficient features at high scales remain stable, and adding noise does not affect the expression of LMA spectral information of this part of the wavelet coefficients, indicating that this part of the wavelet coefficient features has strong robustness.

[0021] Figure 3This is a heatmap showing the correlation between the spectral bands corresponding to wavelet coefficient features and dry matter content. To further investigate the correlation of wavelet coefficient features on three measured datasets, their composite datasets, and simulated datasets, the correlation coefficients between wavelet coefficient features and dry matter content at different bands were calculated. Correlation analysis revealed that the corresponding spectral bands for wavelet coefficient features are concentrated in the 1628-1649 nm range. In the ANGERS dataset, the wavelet coefficients effectively capture the changes in LMA values, with the correlation slightly increasing from 1628 nm to 1640 nm, reaching a peak of 0.86 at 1639–1643 nm. Compared to ANGERS, the LOPEX dataset shows a slightly lower overall correlation, but it remains significant, and performs better in higher bands (after 1645 nm), indicating that this band better reflects leaf structure and dry matter signals. The LOTUS dataset shows a lower correlation, generally remaining between 0.26 and 0.31, although there is a slight increase in the 1633–1641 nm range, the fluctuation is small. On the simulated dataset PROSPECT, the correlation exhibits obvious fluctuations. Around 1628–1634 nm, the correlation is extremely low (as low as 0.07–0.15). Starting from 1636 nm, the correlation increases to around 0.77–0.80, which is close to the performance of the measured data.

[0022] Figure 4 This figure compares the accuracy of LMA estimation using a universal model and a general-purpose model (covering both noise-free and noisy training conditions) in the 1628nm–1649nm wavelength range on a composite measured dataset of ANGERS, LOPEX, and LOTUS, as well as a composite simulated dataset obtained using the PROSPECT radiative transfer model based on the composite measured dataset. The results show that the R̅² of the universal model gradually increases with wavelength, from 0.36 to approximately 0.48, while the nRMSE decreases from 7.3% to approximately 6.5%, indicating that the new index constructed in this range has strong robustness and applicability across different measured datasets. In contrast, under noise-free conditions, the R̅² of the general-purpose model in the 1636–1649nm range is approximately between 0.40 and 0.52, and the nRMSE remains stable at around 7–8%. After introducing noise, the RMSE decreases from 9.6% to approximately 6.7%, indicating that the constructed index is more sensitive to noise in the higher wavelength range. Therefore, considering both inversion accuracy and error, the best-performing band within this range is 1645nm. Thus, bi = 1645nm is determined in Formula 3, and the optimal spectral index model ND is constructed. LN High-precision inversion of leaf dry matter content.

[0023] Figure 5The traditional spectral index model (based on the 1368 nm and 1649 nm bands) and the newly constructed spectral index model in this application (based on ND in the 1645 nm band) are compared. LN A comparison of the LMA inversion performance of the models is shown in the figure. In the universal model, ND LN The model's inversion performance is significantly better than the benchmark spectral index model: in terms of accuracy, ND LN The model validation set R̅² reached 0.4905, and in terms of error metrics, its standardized root mean square error (nRMSE) was only 6.50%. Both of these factors indicate that ND... LN The model has stable and reliable prediction accuracy in LMA inversion of the same source measured samples.

[0024] In the general model, ND LN The model also exhibits superior performance: in a noise-free simulation environment, the R² of the index increases to 0.5204, and the nRMSE is 8.7%. Compared to the benchmark index, its fitting ability to LMA and error control are both more outstanding. In scenarios where noise is added to the simulation data, due to ND... LN The core feature bands of the model remained unchanged, and their R̅² remained stable without variation, while the nRMSE was further reduced by approximately 2 percentage points compared to the noise-free environment. This result not only reflects the ND LN The model's ability to stably preserve band characteristics further highlights its stronger anti-interference capability and inversion stability under complex interference conditions. Detailed Implementation

[0025] Key terms explained in this application: Continuous wavelet transform: Perform continuous wavelet transform analysis on the original reflectance spectrum of each leaf sample at 128 scales; Wavelet coefficient characteristics: The wavelet coefficients obtained by extracting continuous wavelet transform using Pearson coefficients have a high correlation with LMA. LMA universal model: The model is built using the tested training set and validated using the tested validation set.

[0026] LMA generalization model: The model is trained using a simulated dataset and validated using a real-world dataset.

[0027] A high-precision remote sensing inversion method for vegetation leaf dry matter content based on wavelet analysis mainly includes the following steps: S1: Construction of the Leaf Sample Database S11: Construct a leaf measurement dataset, collect and organize the measurement data, collect the leaf reflectance spectrum and dry matter absorption coefficient, and ensure that the two have the same spectral resolution to ensure the accuracy of subsequent calculations; S12: Construct a simulated leaf dataset. Based on the PROSPECT radiative transfer model, generate simulated data to model the spectral characteristics of the leaves and obtain the corresponding simulated sample data. Meanwhile, to improve the model's applicability to actual observation conditions, two types of noise are superimposed on the simulated data: one is fixed noise (0.005), and the other is relative noise (0.003). S13: Dataset partitioning. Using stratified sampling, the measured dataset is divided into a measured training set and a measured validation set based on wavelet features, for example, in a 6:4 ratio. The measured training set is used to construct and calibrate the parameters of the leaf dry matter content inversion model, while the measured validation set is used to test the model's prediction performance under the same data source.

[0028] S2: Wavelet transform extracts dry matter feature information S21: Multi-scale decomposition of the leaf spectral signal (reflectance data in the 400–2500 nm range) is performed using continuous wavelet transform (CWT). The mathematical expression is as follows:

[0029] In the formula, f(t) is the input spectral signal, W(a,b) are the corresponding wavelet coefficients; the scale parameter a controls the degree of scaling of the wavelet, with large scales corresponding to the overall features of low frequency and small scales corresponding to the detailed features of high frequency; the translation parameter b is used to locate the signal features at different wavelengths; ψ(t) is the mother wavelet function, which is essentially a local signal with zero mean and oscillatory characteristics.

[0030] S22: Using Morlet wavelet as the mother function, a continuous wavelet decomposition of 128 scales is performed on each leaf spectral sample. The Morlet wavelet function includes a center frequency parameter ω0, which determines the oscillation frequency of the wavelet, thus affecting the capture effect of spectral details at different scales. The expression is as follows:

[0031] In the formula, t is the wavelength of the reflection spectrum, and w o is the center frequency of the wave, used to control the oscillation frequency of the Morlet wavelet, and i is the imaginary unit.

[0032] S23: The obtained wavelet coefficients are correlated with LMA, using the Pearson correlation coefficient as the indicator. The calculation formula is as follows:

[0033] In the formula, Indicates the first i Wavelet coefficients; This represents the corresponding LMA value; and These are the means of the two sets of variables. The correlation coefficient ranges from [-1, 1]: when r is close to 1, it indicates a strong positive correlation; when it is close to -1, it indicates a strong negative correlation; and when it is close to 0, it indicates that the two are almost uncorrelated.

[0034] S24: Based on the set correlation threshold, select the wavelet coefficient feature intervals that are sensitive to dry matter content from all wavelet coefficients to provide input variables for subsequent inversion modeling.

[0035] S3: LMA Inversion Modeling Based on Wavelet Coefficient Characteristics and Spectral Indices S31: Calculate the traditional dry matter spectral index using raw leaf reflectance data, and establish an LMA estimation (inversion) model based on the spectral index. The formula is as follows: NDMI = (R 1649 -R 1722 ) / (R 1649 +R 1722 (1) ND LMA =(R 1368 -R 1722 ) / (R 1368 +R 1722 (2) In the formula, R 1649 R 1722 R 1368 These represent the leaf reflectance at 1649 nm, 1722 nm, and 1368 nm, respectively. This index reflects the characteristic differences between the near-infrared and short-wave infrared bands and has a certain correlation with the dry matter content of leaves.

[0036] S32: Wavelet coefficient features and spectral index fusion modeling, using the sensitive wavelet coefficient feature interval obtained by continuous wavelet transform as input variable to construct a new vegetation index (dry matter) model.

[0037] The specific steps include: using the spectral bands (1628~1648 nm) from the sensitive wavelet coefficient characteristic range obtained in step S24 as input variables. These spectral bands have a clear distribution in the scale-wavelength space and can effectively characterize the spectral response of the LMA value. Subsequently, for the two benchmark indices mentioned above, considering that the 1722 nm band has a relatively stable response to leaf structure and dry matter, it is retained to reduce background interference and ensure that the index calculation logic is consistent with the existing normalized index. Based on this, a new dry matter index based on the sensitive wavelet coefficient characteristic range data is constructed, and its specific form is as follows: ND LN =(R bi -R 1722 ) / (R bi+R 1722 (i=1,2,3…) (3) Among them, ND LN Let bi be the newly constructed dry matter index, and R be the i-th highly correlated band. bi Let be the reflectance of the i-th highly correlated band.

[0038] S33: Based on Formula 3, an LMA inversion model was constructed, and its inversion effect was compared and verified with the LMA inversion models constructed based on Formula 1 and Formula 2. The optimal dry matter index (i.e., bi=1645nm was determined to be the optimal result of Formula 3) and the LMA inversion model based on the optimal dry matter index were obtained. This index better reflects the characteristic differences between the near-infrared and short-wave infrared bands and has a better correlation with the dry matter content of leaves.

[0039] The publicly available measured dataset of leaves in step S11 of this application includes the reflectance spectrum of fresh leaves and the dry matter content of fresh leaves, as detailed in Table 1.

[0040] In step S12 of this application, the radiative transfer model used is the PROSPECT-D model. This model includes seven input parameters: leaf structure parameter N (dimensionless), chlorophyll content Chl (μg / cm²), carotenoid content Car (μg / cm²), brown pigment content Cbp (dimensionless), dry matter content Cm (g / cm²), equivalent water thickness Cw (cm), and anthocyanin content Ant (μg / cm²). To ensure that the generated simulation dataset reflects the true characteristics of leaves under natural conditions, this application reviewed relevant published literature to statistically analyze the value ranges of biochemical indicators for normal leaves, and based on this, limited the value intervals of each input variable (see Table 2). Running the model under parameter constraints, a simulated sample dataset containing leaf reflectance spectra and corresponding dry matter content was finally obtained.

[0041] Step S31 of this application includes the following steps: S311: A universal LMA inversion model based on raw reflectance to construct spectral indices, that is, using raw reflectance as input to the spectral index model, using the measured training set to participate in model construction, and using the measured validation set to validate the model; S312: A general LMA inversion model based on raw reflectance is constructed by using raw reflectance as input to the spectral index model, using simulation datasets to participate in the construction of the general model, and using measured datasets to validate the model.

[0042] The inversion effect comparison and verification in step S33 of this application includes the following aspects: 1. A comparative verification of the inversion results of the traditional vegetation index LMA universal inversion model based on raw reflectance and the inversion results of the new LMA universal inversion model constructed based on wavelet coefficients, with the evaluation based on the dry matter index ND. LN (Formula 3) The fitting accuracy and predictive ability of the LMA universal inversion model constructed from the relationship between LMA values; 2. Add absolute noise (0.005) and relative noise (0.003) to the data in the simulated dataset (this step can also be done in step S12), and extract wavelet coefficient features according to step S2. Perform correlation analysis, and evaluate the ND by comparing it with the inversion results under noise-free conditions. LN The robustness of the model under noise interference; 3. The inversion results of several measured datasets are compared and verified to prove the applicability and reliability of the model established in this application under different vegetation types and data conditions.

[0043] See Figure 5 The inversion results based on the traditional LMA spectral index inversion model in the 1368 nm band are shown in... Figure 5 The inversion results based on the LMA spectral index inversion model of this application in the 1645nm band are shown in the upper part. Figure 5 In the middle section, the inversion results based on the traditional LMA spectral index inversion model in the 1368nm band are shown. Figure 5 In the lower section, the traditional LMA spectral index inversion model constructed from the original reflectance performs worse than the LMA inversion model constructed in this application by combining wavelet coefficient features in all three types of inversion performance: universal model, noise-free universal model, and noise-added universal model. Overall, the inversion performance of the new LMA universal inversion model constructed based on wavelet coefficients is significantly better than the LMA universal inversion model based on the original spectral index, proving that the LMA spectral index universal inversion model constructed using the wavelet coefficient method is feasible.

[0044] This application uses normalized root mean square error (nRMSE) and coefficient of determination (R²). 2 These two evaluation metrics are used to verify the accuracy of the model, and the formulas are as follows:

[0045]

[0046] In the formula, n represents the total sample size. These are the measured observations of the sample. The predicted value output by the model. This is the sample mean. These correspond to the maximum and minimum values ​​in the actual measured values ​​of the sample, respectively.

[0047] This application proposes a high-precision remote sensing inversion method for dry matter content in vegetation leaves based on wavelet analysis. This method is based on a collaborative construction approach using multi-source data: on the one hand, it gathers large-scale measured spectral data of leaves acquired under different species, growth stages, and environmental conditions; on the other hand, it uses the PROSPECT radiative transfer model to generate a simulated sample set covering multiple scenarios. To enhance the stability and generalizability of the model under different application conditions, composite noise processing is introduced into the simulated data, combining absolute and relative noise, thereby making its distribution characteristics closer to actual observations.

[0048] The PROSPECT simulation dataset described in this application was generated using the PROSPECT-D radiative transfer model, which has seven input variables: leaf structure parameters (N, unitless), leaf chlorophyll content (μg / cm³), and leaf chlorophyll content (μg / cm³). 2 ), leaf carotenoid content (C ar μg / cm 2 ), brown pigment content (C bp (unitless), dry matter content (C) m g / cm 2 ), equivalent water thickness (C) w (cm) and anthocyanin content (Ant, μg / cm) 2 In order to make the simulation dataset representative, this application consulted publicly available literature to statistically analyze the extreme range of biochemical parameters of normal leaves, and based on this, constrained and set the input parameters. The generated simulation dataset sample includes leaf reflectance spectra and corresponding dry matter content.

[0049] This application uses normalized root mean square error (nRMSE) and coefficient of determination (…). This is used to evaluate the model's performance on the test set and to verify the model's accuracy.

[0050] Table 1. Description of the measured sample dataset

[0051] Table 2. Instructions for setting input parameters and value range for the PROSPECT-D model.

[0052] Table 3. Inversion results of the LMA inversion model constructed using Morlet continuous wavelet coefficient features in the relevant bands.

[0053] Table 3 illustrates the data using wavelet coefficient characteristics combined with the dry matter index (ND). LN The best inversion results were obtained when the wavelet coefficients were set to the 1645 nm band feature when constructing the LMA inversion model.

Claims

1. A high-precision remote sensing inversion method for dry matter content of vegetation leaves based on wavelet analysis, comprising the following steps: S1: Construction of the Leaf Sample Database S11: Construct a measured dataset of leaves, collect and organize leaf reflectance spectra and dry matter absorption coefficients, and ensure that the two have the same spectral resolution; S12: Construct a simulated dataset for the blade. Based on the PROSPECT radiative transfer model, construct a simulated dataset corresponding to the measured dataset of the blade. S13: Dataset partitioning. Using a stratified sampling method, the measured dataset is divided into a measured training set and a measured validation set based on wavelet features. The measured training set is used to construct and calibrate the parameters of the leaf dry matter content inversion model, while the measured validation set is used to verify the model's prediction performance under the same data source. S2: Wavelet transform extracts dry matter feature information S21: Multi-scale decomposition of the leaf spectral signal is performed using continuous wavelet transform, expressed as follows: In the formula, f(t) is the input spectral signal, W(a,b) are the corresponding wavelet coefficients; the scale parameter a controls the degree of scaling of the wavelet, with large scales corresponding to the overall features of low frequency and small scales corresponding to the detailed features of high frequency; the translation parameter b is used to locate the signal features at different wavelengths; ψ(t) is the mother wavelet function, which is essentially a local signal with zero mean and oscillatory characteristics. S22: Using Morlet wavelet as the mother function, a continuous wavelet decomposition of 128 scales is performed on each leaf spectral sample. The Morlet wavelet function contains a center frequency parameter ω0, which determines the oscillation frequency of the wavelet and thus affects the capture effect of spectral details at different scales. The expression is as follows: In the formula, t is the wavelength of the reflection spectrum, and w o is the center frequency of the wave, used to control the oscillation frequency of the Morlet wavelet; i is the imaginary unit. S23: The obtained wavelet coefficients are correlated with the LMA values ​​using the Pearson correlation coefficient as the indicator. The calculation formula is as follows: In the formula, Indicates the first i Wavelet coefficients; This represents the corresponding LMA value; and These are the means of the two sets of variables; the correlation coefficient ranges from [-1, 1]: when r is close to 1, it indicates a strong positive correlation; when it is close to -1, it indicates a strong negative correlation; when it is close to 0, it indicates that the two are almost uncorrelated. S24: Based on the set correlation threshold, select the wavelet coefficient feature intervals that are sensitive to dry matter content from all wavelet coefficients to provide input variables for subsequent inversion modeling; S3: Construct an LMA inversion model based on wavelet coefficient characteristics and spectral indices.

2. The high-precision remote sensing inversion method for determining the dry matter content of vegetation leaves according to claim 1, characterized in that: Step S3 includes: S31: Calculate the traditional dry matter spectral index using the original leaf reflectance data, and establish an LMA inversion model based on the spectral index. The formula is as follows: NDMI=(R 1649 -R 1722 ) / (R 1649 +R 1722 ) (1) ND LMA =(R 1368 -R 1722 ) / (R 1368 +R 1722 ) (2) In the formula, R 1649 R 1722 R 1368 These represent the leaf reflectance at 1649 nm, 1722 nm, and 1368 nm, respectively. S32: Wavelet coefficient features and spectral index fusion modeling, using the sensitive wavelet coefficient feature interval data obtained by continuous wavelet transform as input variables, constructs a new high-precision remote sensing inversion model of vegetation leaf dry matter content based on wavelet analysis. The inversion model satisfies formula (3): Fresh Dry Matter Index (ND) LN =(R bi -R 1722 ) / (R bi +R 1722 (i=1,2,3…) (3) Among them, ND LN Let bi be the newly constructed dry matter index, and R be the i-th highly correlated band. bi Let be the reflectance of the i-th highly correlated band; S33: Construct an LMA inversion model based on Formula 3, and compare and verify the inversion effect with the LMA inversion models constructed based on Formula 1 and Formula 2 to obtain the optimal dry matter index and the LMA inversion model based on the optimal dry matter index.

3. The high-precision remote sensing inversion method for determining the dry matter content of vegetation leaves according to claim 2, characterized in that: The optimal dry matter index ND LN =(R 1645 -R 1722 ) / (R 1645 +R 1722 ).

4. The high-precision remote sensing inversion method for vegetation leaf dry matter content according to claim 3, characterized in that: The inversion effect comparison and verification in step S33 includes the following aspects:

1. A comparative verification of the inversion results of the traditional vegetation index LMA universal inversion model based on raw reflectance and the inversion results of the new LMA universal inversion model constructed based on wavelet coefficients, with the evaluation based on the dry matter index ND. LN The fitting accuracy and predictive ability of the new LMA universal inversion model constructed based on the relationship with LMA values; 2. Superimpose absolute noise and relative noise on the data in the simulated dataset, extract wavelet coefficient features according to step S2, conduct correlation analysis, and evaluate the robustness of the new LMA universal inversion model under noise interference by comparing it with the inversion results under noise-free conditions.

3. Compare and verify the inversion results on several measured datasets to verify the applicability and reliability of the new LMA universal inversion model under different vegetation types and data conditions.

5. The high-precision remote sensing inversion method for determining the dry matter content of vegetation leaves according to claim 1, characterized in that: The PROSPECT radiative transfer model uses the PROSPECT-D radiative transfer model, and the simulation dataset includes the chlorophyll content, dry matter content, equivalent water thickness, carotenoid content, brown pigment content, structural parameters, and anthocyanin content of plant leaves.

6. The high-precision remote sensing inversion method for determining the dry matter content of vegetation leaves according to claim 4, characterized in that: The simulated dataset contains two types of noise superimposed: one is fixed noise, and the other is relative noise.

Images