Method and system for inverting total leaf area index of rape

Abstract

The application belongs to the technical field of agricultural remote sensing, and particularly relates to a method and system for inverting total photosynthetic area index of rape, comprising: constructing MLMSTM of rape based on vector radiation transfer theory to obtain HH, VV, VH and HV total backscattering coefficients and plant layer backscattering coefficient; determining the corrected MLMSTM according to Sobol parameter sensitivity analysis; determining the microwave scattering ratio of each pixel in the Radarsat-2 image region based on spatial interpolation method according to the corrected MLMSTM; and inverting the total photosynthetic area index TPAI of rape by using the trained machine learning model based on the microwave scattering ratio of each pixel. The method and system focus on the vertical layering scattering characteristics of rape at different phenological stages, realize high-precision inversion of rape canopy parameters, and provide technical support for rape growth monitoring and yield estimation.

Description

Technical Field

[0001] This invention belongs to the field of agricultural remote sensing technology, specifically, it relates to a method and system for retrieving the total photosynthetic area index (TPAI) of rapeseed. Background Technology

[0002] Crop canopy parameters reflect crop growth status and trends, and are among the most effective indicators for monitoring crop growth. Rapeseed plants possess unique morphology and photosynthetic organ succession characteristics; their primary photosynthetic organ gradually transitions from leaves to siliques as the rapeseed grows and develops. Specifically, from seedling to flowering, leaves are the main photosynthetic organ; after flowering, siliques develop, and the siliques begin photosynthesis, with both leaves and siliques serving as the main photosynthetic organs; at maturity, the siliques reach their final shape, leaves wither, and the rapeseed relies primarily on silique photosynthesis to fill the seeds. The maximum area of ​​the rapeseed silique is 1.54 times the maximum leaf area, and the net photosynthetic rate, transpiration rate, and light radiation intensity of the silique layer are all higher than those of the leaf layer. Therefore, rapeseed growth monitoring requires comprehensive consideration of both leaf area and silique area, i.e., the Total Photosynthetic Area Index (TPAI). Here, the Total Photosynthetic Area Index refers to the proportion of the sum of leaf area and silique area per unit land area, and it is a key canopy parameter for monitoring rapeseed growth.

[0003] However, TPAI (Total Potential Aspect Ratio) inversion for rapeseed based on optical remote sensing faces certain difficulties because optical satellites are easily affected by weather conditions, making it difficult to obtain stable, effective, and continuous data. Considering the regional characteristics of major rapeseed producing areas and the plant characteristics at different growth stages, Synthetic Aperture Radar (SAR) data is more suitable for remote sensing inversion of regional rapeseed TPAI. SAR data not only has all-weather, all-time observation capabilities unaffected by weather conditions, but its side-looking slant-range projection imaging capability also shows strong sensitivity to the three-dimensional structure of the spiral arrangement of rapeseed siliques. In particular, C-band radar microwaves can penetrate the rapeseed canopy, scattering multiple times between rapeseed flowers, siliques, and leaves, containing information from each layer of the rapeseed plant. Compared to optical remote sensing data, SAR data plays an irreplaceable and crucial role.

[0004] Currently, models for SAR-based crop canopy parameter retrieval mainly include empirical models, semi-empirical models, and mechanistic models. Empirical models are simple in principle and can well reflect the relationship between microwave characteristics and crop plant or canopy parameters. However, they require a large amount of measured data for training in the early stages, are greatly affected by data, and have poor portability, with significant differences in the accuracy of crop parameter retrieval across different regions. Semi-empirical models introduce basic theoretical content on the basis of empirical models, simplifying model parameters and theoretical logic. However, the solution of their empirical coefficients still relies on sample data, and the applicability of the model's fitting parameters is limited by factors such as crop type and planting environment. Mechanistic models combine crop microwave scattering characteristics with mathematical and physical models to reveal the scattering characteristics of various crop components and their influencing mechanisms, quantifying and numericalizing them. Only by deeply understanding the scattering mechanism of crops and quantitatively linking it with crop canopy parameters can we provide technical support for subsequent crop growth monitoring, crop yield estimation, and SAR agricultural applications. However, detailed theoretical research leads to complex analysis and detailed parameters, making the research difficult and computationally costly.

[0005] Secondly, crop canopies possess complex multi-layered structures, necessitating the development of microwave scattering models capable of comprehensively characterizing their vertical structure. Compared to multi-layered microwave scattering models for rice and wheat, rapeseed exhibits more complex dynamic evolutionary characteristics in its plant structure. Throughout its growth cycle, rapeseed undergoes significant changes in plant height and canopy vertical components, particularly in leaf development, from germination and growth to maximum size and eventual senescence and shedding. These dynamic changes in structure and composition lead to alterations in the composition and microwave scattering response of each vertical layer at different phenological stages, posing a significant challenge to the inversion of crop canopy parameters.

[0006] Furthermore, existing studies primarily focus on Gramineae crops and have not yet considered the characteristics of photosynthetic organ succession. Unlike Gramineae crops, whose canopy components and photosynthetic organs are mainly leaves throughout the phenological period, and whose leaf area index (LAI) is an important canopy parameter, rapeseed (Brassica oleracea) is a cruciferous crop. In the mid-to-late stages of canopy development, the canopy components and photosynthetic organs transition from leaves to siliques. Therefore, canopy parameters must consider the contribution of siliques in addition to LAI. It is evident that early SAR-based inversion models had limitations in retrieving the rapeseed canopy parameter TPAI. Summary of the Invention

[0007] To address the problems in the background technology, this invention proposes a method and system for retrieving rapeseed TPAI by combining the advantages of collaborative theory and machine learning.

[0008] In a first aspect, embodiments of the present invention provide a method for retrieving the total photosynthetic area index (TPAI) of rapeseed, comprising: constructing a multi-layer microwave scattering theoretical model (MLMSTM) of rapeseed based on vector radiative transfer theory to obtain the total backscattering coefficients HH, VV, VH, and HV and the plant layer backscattering coefficient, wherein the plant layer backscattering coefficient includes the stem-leaf layer backscattering coefficient and the silique layer backscattering coefficient; verifying the effectiveness of the MLMSTM based on data from the Laboratory of Target Microwave Properties (LAMP), and determining a corrected MLMSTM based on Sobol parameter sensitivity analysis; determining the pixel-by-pixel microwave scattering ratio of the Radarsat-2 image region based on the corrected MLMSTM using a spatial interpolation method; and retrieving the total photosynthetic area index (TPAI) of rapeseed using a trained machine learning model based on the pixel-by-pixel microwave scattering ratio, wherein the machine learning model is a random forest (RF) model.

[0009] Secondly, embodiments of this application provide a system for inverting the total photosynthetic area index of rapeseed, including a computer-executable program, which, when run, implements the steps of the method described above.

[0010] The beneficial effects of this invention are: focusing on the vertical stratified scattering characteristics of rapeseed at different phenological stages, constructing a multi-phenological stage multi-layer microwave scattering model that can characterize the dynamic changes of rapeseed canopy components (such as leaves, stems, and siliques) and photosynthetic organs, realizing high-precision inversion of rapeseed canopy parameters, and providing technical support for rapeseed growth monitoring and yield estimation. Attached Figure Description

[0011] To facilitate understanding of the invention, it will be described in more detail with reference to the specific embodiments shown in the accompanying drawings. These drawings depict only typical embodiments of the invention and should not be considered as limiting the scope of protection of the invention.

[0012] Figure 1 This is a flowchart illustrating the method for inverting rapeseed TPAI using the synergistic theory method and machine learning advantages according to an embodiment of the present invention.

[0013] Figure 2 This is a technical roadmap for inverting rapeseed TPAI using the synergistic theory method and the advantages of machine learning according to embodiments of the present invention.

[0014] Figure 3 This is a schematic diagram of the leaf-soil stratification and fruit-soil stratification scattering models according to an embodiment of the present invention.

[0015] Figure 4 This is a schematic diagram of the flower-leaf-soil stratification, flower / fruit-leaf-soil stratification, and fruit-leaf-soil stratification scattering models according to embodiments of the present invention.

[0016] Figure 5 This is a comparison chart of simulated values ​​and actual backscattering coefficient values ​​of MOh and Oh models based on LAMP data according to an embodiment of the present invention (40° incident angle).

[0017] Figure 6 The simulation results show the temporal differences in the backscattering coefficients of soils with different polarization modes (Moh and Oh) at 5.405 GHz according to embodiments of the present invention.

[0018] Figure 7 This is a comparison chart of the predicted and actual values ​​of the MLMSTM backscattering coefficient based on LAMP data according to an embodiment of the present invention.

[0019] Figure 8 The results of Sobol sensitivity analysis of MLMSTM parameters for each phenological stage according to embodiments of the present invention are shown.

[0020] Figure 9 The prediction accuracy of the RF training set for various phenological stages of rapeseed according to an embodiment of the present invention is shown.

[0021] Figure 10 The importance scores of radar features at various phenological stages of rapeseed according to embodiments of the present invention are displayed.

[0022] Figure 11 The results of TPAI inversion for rapeseed in the Radarsat-2 region according to an embodiment of the present invention are shown.

[0023] Figure 12 The accuracy of TPAI inversion verification for various phenological stages of rapeseed according to embodiments of the present invention is shown.

[0024] Figure 13 The accuracy of TPAI inversion for the entire growth period of rapeseed according to an embodiment of the present invention is shown.

[0025] Figure 14 The paper presents a comparison of the accuracy of TPAI (Taiwan Backscattering Coefficient Inversion) derived directly from SAR images and TPAI derived from MLMSTM (Metal Mover System Matrix) extracted from plant layer scattering information, according to an embodiment of the present invention. Detailed Implementation

[0026] The embodiments of the present invention are described below with reference to the accompanying drawings to enable those skilled in the art to better understand and implement the present invention. However, the listed embodiments are not intended to limit the present invention. In the absence of conflict, the following embodiments and the technical features in the embodiments can be combined with each other, wherein the same components are indicated by the same reference numerals.

[0027] Figure 1 A flowchart illustrating the method for TPAI (Total Potential Artificial Intelligence) in rapeseed using the synergistic theory method and machine learning advantage inversion according to an embodiment of the present invention is shown.

[0028] The method for retrieving the TPAI of rapeseed according to an embodiment of this application includes: S1, constructing an MLMSTM of rapeseed based on vector radiative transfer theory to obtain the total backscattering coefficients HH, VV, VH, and HV and the plant layer backscattering coefficient, wherein the plant layer backscattering coefficient includes the stem and leaf layer backscattering coefficient and the silique layer backscattering coefficient; S2, determining a corrected MLMSTM based on the validation of the effectiveness of the MLMSTM using LAMP data and the Sobol parameter sensitivity analysis; S3, determining the microwave scattering ratio per pixel of the Radarsat-2 image region based on the corrected MLMSTM using a spatial interpolation method; S4, retrieving the TPAI of rapeseed using a trained machine learning model (i.e., RF model) based on the microwave scattering ratio per pixel.

[0029] The following will refer to Figure 2 This paper details the technical route for TPAI (Total Potential Artificial Intelligence) in rapeseed based on the collaborative theory method and the advantages of machine learning according to embodiments of the present invention.

[0030] The construction of MLMSTM for rapeseed based on vector radiative transfer theory includes the following three steps: S1.1, S1.2, and S1.3.

[0031] S1.1, Based on the plant characteristics of rapeseed at different key phenological stages, the phase matrix and extinction matrix of rapeseed in different media are determined based on the microwave scattering amplitude function of single scattering particles, where different media refer to leaves, flowers and / or siliques.

[0032] That is, firstly, plant characteristic data of rapeseed at different key phenological stages are collected, and then the amplitude function of a single scattering particle is solved by the generalized Rayleigh-Gans approximation, thereby obtaining the phase matrix and extinction matrix of different media.

[0033] S1.2, Based on LAMP data, the coefficients of the Oh model are calibrated to obtain an improved Oh model, and the Muller matrix of the soil layer is determined using the improved Oh model.

[0034] Specifically, firstly, soil moisture content, root mean square height, and other data were collected at different phenological stages. kThe values ​​(wavenumbers) and the true values ​​of the bare soil backscattering coefficients corresponding to 20°, 30°, and 40° were used as the fitting dataset, which was then used as LAMP data. Then, for each frequency in the 4-8 GHz range with a step size of 0.005, the Oh model was calibrated based on the fitting dataset to obtain an improved Oh model. Next, based on the improved Oh model, the Muller matrix of the soil layer was determined according to the root mean square height of the soil and the soil moisture content.

[0035] Since the construction of vegetation scattering models is inseparable from the analysis of soil scattering, microwaves penetrate vegetation to reach the soil, and then interact with plant components, thereby affecting the total backscattering of the plant. Currently, the most commonly used model for soil scattering modeling is the Oh model. Equations (1) to (3) are the final expressions of the Oh model.

[0036] (1), (2), (3), In the formula, , , The backscattering coefficients are those under VH, VV, and HH polarizations. q For cross-polarization ratio, p For the same polarization ratio, k For wave number, , λ is the wavelength. s The root mean square height of the soil. m is the radar incident angle. s This refers to soil moisture content.

[0037] The soil samples used to construct the Oh model primarily consisted of silt, loam, and sandy loam. However, the soil texture in the study area is mainly clay, with finer particles. Therefore, when applying the Oh model to the study area, the model's empirical coefficients need to be recalibrated. The calibration coefficients corresponding to the original formulas are as follows: (4), (5), (6), In equation (4), It remains unchanged because, for typical agricultural soils, Fresnel reflectance has an approximately linear relationship with soil moisture content, remaining constant at a soil moisture level of 0.03%. <m s Valid within the range <0.35. a, b, c, d, e, and f are coefficients to be calibrated. Based on LAMP experimental data, coefficients are solved using multiple incident angles and frequencies to construct an improved Oh (Modified Oh, MOh) model based on clay.

[0038] A nonlinear least squares fitting method was used to calibrate a set of universal parameters for the MOh model at each frequency within the 4GHz to 8GHz band. Input data included soil moisture content, root-mean-square height of soil, etc., for five phenological stages. k The values ​​and the true values ​​of the backscattering coefficients of bare soil at 20°, 30°, and 40° are obtained. Data processing adopts a frequency-based strategy, that is, for each frequency within 4-8GHz with a step size of 0.005, the data of 20°, 30°, and 40° corresponding to the frequency under five phenological periods are combined to form a fitting dataset containing 15 samples. The independent variables of this dataset are the surface parameters and the incident angle, and the dependent variable is the observed value of the backscattering coefficient. Based on this, by minimizing the residual between the model prediction value and the true value, the parameters a, b, c, d, e, and f in the model defined by equations (4), (5), and (6) are solved iteratively. Finally, the accuracy of the calibrated model is quantitatively evaluated by calculating the root mean square error (RMSE) and mean absolute error (MAE) between the model fitting value and the true value throughout the entire frequency band.

[0039] In rapeseed mechanistic models, the Muller matrix of the soil layer is needed as soil scattering information. The differential Muller matrix of the soil layer is calculated based on the Oh model. , , The correlation (α) and co-polarization phase difference (ξ) are composed of: correlation degree (α) and co-polarization phase difference (ξ). Correlation degree refers to the degree of similarity or correlation between two signals from different radar channels, at different times, or at different frequencies. Co-polarization phase difference refers to the phase difference between co-polarized scattering channels during electromagnetic wave scattering. Both are also related to soil roughness and soil moisture content, and can be calculated using the following formulas: (7), (8), k For wave number, s The root mean square height of the soil. l For the relevant length, m s Soil moisture content, Let be the radar incident angle. In equation (7-8) These are the coefficients to be solved. The empirical parameters used in the calculation are applicable within a range of 0.03. <m s <0.3, 0.1< ks <7, 1.8< kl <22, the soil conditions are met. The coefficient corresponding to equation (7) The coefficients are 0.17, 0.01, 0.5, 1.1, and 0.4, respectively. The coefficients corresponding to equation (8) are... The values ​​are 0.44, 0.95, and 1.0, respectively.

[0040] According to the reciprocity relation, the cross-polarization scattering amplitude is the same in the backscattering direction, i.e., S vh =S hv . S S is a scattering matrix with a value of 0. vh This indicates that the incident wave is vertically polarized. v The scattered wave is horizontally polarized. h The complex scattering amplitude of ). S hv This indicates that the incident wave is horizontally polarized. h The scattered wave is vertically polarized. v The complex scattering amplitude of ). Therefore, the differential Muller matrix can be approximated as follows: (9), In the formula, and It can be derived from α and ξ: (10) (11), in, Re , Im Represent the real and imaginary parts respectively, * indicates conjugate, S vv This indicates that both the incident wave and the scattered wave are vertically polarized. v The complex scattering amplitude of ). S hh This indicates that both the incident wave and the scattered wave are horizontally polarized. h The complex scattering amplitude of ).

[0041] According to equations (10) and (11) and , , The elements of the differential Muller matrix are calculated as follows: (12) (13) (14) (15) (16) (17) Substituting equations (12) to (17) into equation (9) yields the Muller matrix of the soil layer: (18) S1.3, Based on the phase matrix and extinction matrix of different media and the Muller matrix of the soil layer, the total Muller matrix is ​​determined, thereby constructing the MLMSTM of rapeseed.

[0042] In one embodiment, the total Muller matrix is ​​calculated by weighted averaging of the extinction matrix, phase matrix, soil layer Muller matrix, and sensor parameters.

[0043] Generally, based on the crop coverage and physical and geometric characteristics of rapeseed during its growth stages, the rapeseed growth period can be divided into the following five phenological stages: six-leaf stage, 30% flowering stage, 90% flowering stage, initial maturity stage, and full maturity stage. Microwave scattering models are then established for each of these five phenological stages.

[0044] Specifically, in the first stage (six-leaf stage), the rapeseed plant consists only of stems and leaves with indistinct layering, and the stems are short and do not completely cover the ground. In the second stage (30% flowering stage), the rapeseed stems grow and the rapeseed flowers bloom; at this time, the plant includes not only stems and leaves but also rapeseed flowers. In the third stage (90% flowering stage), the rapeseed flowers gradually wither, and the siliques begin to grow; the plant includes stems, leaves, siliques, and rapeseed flowers. In the fourth stage (early maturity stage), all rapeseed flowers wither, and the siliques continue to grow; the plant includes stems, leaves, and siliques. In the second, third, and fourth stages, the rapeseed flowers and siliques do not completely cover the rapeseed stems and leaves. In the fifth stage (maturity stage), the siliques develop to full maturity, the leaves wither, and the plant mainly consists of siliques and stems. Therefore, the MLMSTM for rapeseed includes: a leaf-soil stratified scattering model at the six-leaf stage; a flower-leaf-soil stratified scattering model at 30% flowering stage; a flower / fruit-leaf-soil stratified scattering model at 90% flowering stage; a fruit-leaf-soil stratified scattering model at the early maturity stage; and a fruit-soil stratified scattering model at the maturity stage.

[0045] The derivation of the above scattering mechanism model is as follows.

[0046] (1) Leaf-soil and fruit-soil stratified scattering model: At the six-leaf stage of rapeseed, the plant is mainly composed of leaves, with relatively short stems. The stratification of stems and leaves is not obvious and does not completely cover the ground surface. Therefore, a leaf-soil stratified scattering model is proposed. The elements of the stem and leaf layer are assumed to be ellipsoids and uniformly distributed in the soil. h (Plant height) within. Based on the theory of vector radiative transfer, a set of vector radiative transfer equations for the leaf-soil stratification model is constructed: (19) (20) in, It is the zenith corner. It is the azimuth angle, and z represents the spatial position. and These are the Stokes vector intensities for the upward and downward directions, respectively. , The extinction matrix of the plant layer. and Let be the scattering source function, and be the Stokes vector intensity in each direction after scattering to . and Energy of direction.

[0047] Using an iterative method, the zeroth-order solution to the vector radiative transfer equations is ground scattering, such as... Figure 3 In A1; the first-order solution includes scattering from the vegetation layer and interaction scattering between the ground and the vegetation layer, such as Figure 3 A2, A3, A4, A5. In Figure 3 In the diagram, A1 represents direct scattering information from the soil layer, and A2 represents direct scattering information from the plant layer. A3, A4, and A5 represent multiple scattering information between the plant and soil layers. The multiple scattering component in case A5 is relatively small and can be ignored. For vegetation, the scattering satisfies... K e For the plant layer in and Extinction matrix in the direction.

[0048] Solving the vector radiative transfer equation, its zeroth-order solution corresponds to scattering from the ground, and the backscattering coefficient of the soil layer is: (twenty one), The soil scattering coefficient, , S Represents the scattering matrix. p , q Indicates the polarization direction; and for p and q Attenuation factor in the incident direction of polarized wave Indicates the scattering (emission) angle. The incident angle is calculated from the extinction matrix. (twenty two), (twenty three), K e For extinction matrix, h This refers to the height of the plant.

[0049] The stem-leaf layer backscattering coefficient is obtained from the first-order solution of the vector radiative transfer equation: (twenty four), in, It is direct scattering from the stem and leaf layer. and denoted as the scattering coefficient of the interaction between the stem and leaf layer and the soil layer.

[0050] (25), P pq It is the phase matrix. and Here are the extinction matrices in the incident and exit directions. h This refers to the height of the plant.

[0051] and The solution is quite complex. Current research mainly considers coherent scattering from the soil layer, while scattering in other directions is weak and can be ignored.

[0052] (26) The coherent scattering coefficient of the soil layer is obtained from the Muller matrix of the soil layer. h This represents the plant height. According to the reciprocity theorem... .

[0053] At this stage, the rapeseed leaves at the six-leaf stage do not cover the ground surface, therefore, a rapeseed planting ratio coefficient c needs to be introduced into the microwave scattering model. p A leaf-soil stratified discrete scattering model is constructed. Therefore, based on the zeroth and first-order solutions of the vector radiative transfer equation, the backscattering coefficients of the leaf-soil stratified discrete scattering model are obtained as follows: (27) c p This represents the rapeseed planting ratio coefficient within a pixel.

[0054] When rapeseed reaches maturity, its leaves wither, and the plant mainly consists of siliques and stems. The siliques grow extensively, essentially covering the ground surface, but the scattering effect from the soil layer still exists. Therefore, when constructing a fruit-soil layered scattering model for mature rapeseed, the rapeseed planting ratio coefficient c is not considered when calculating the total backscattering coefficient. p The parameter settings and calculation steps are consistent with the leaf-soil layered discrete scattering model: (28) (2) Layered scattering models of flower-leaf-soil, flower / fruit-leaf-soil and fruit-leaf-soil As rapeseed flowers bloom and pods develop, the vertical structure of rapeseed becomes increasingly complex. There are distinct stratifications between rapeseed flowers, pods, and stems / leaves, requiring a separate microwave scattering model to establish each stratified layer. The approach to solving the flower-leaf-soil, flower / fruit-leaf-soil, and fruit-leaf-soil stratified scattering models involves solving them as two separate factor stratified models. First, the leaf-soil stratified microwave scattering is calculated to obtain the overall upward Muller matrix. This matrix is ​​then used as the underlying surface of the flower or pod layer to construct the flower / fruit-underlying surface stratified model. The scattering processes of flower-leaf-soil, flower / fruit-leaf-soil, and fruit-leaf-soil are as follows: Figure 4 As shown, This refers to the microwave scattering component that directly penetrates below the rapeseed flower layer or silique layer, and After the component is incident on the rapeseed flower layer or silique layer, it is decomposed into components A1 to A5. The meanings of these five components are basically the same as those of components A1 to A5 in the leaf-soil layered scattering model. This represents the direct scattering component of the rapeseed flower layer or silique layer; , as well as The meanings of the components are basically the same as those of the A3, A4, and A5 components in the leaf-soil stratified scattering model, but the underlying surface is calculated using the leaf-soil stratified scattering model. Specifically, This represents information scattered from the stem and leaf layer. This refers to information scattered from the flower or silique layer. , , A1 represents the multiple scattering information between the flower or silique layer and the stem and leaf layer; A2, A3, A4, and A5 represent the scattering of the incident wave after it passes through the top layer and reaches the stem and leaf layer. h 1 The height of the flower or silique layer. h 2 This represents the height of the stem-leaf layer. Additionally, the flower or silique layer does not completely cover the stem-leaf layer; therefore, similar to the leaf-soil layered scattering model, the flower / fruit-leaf-soil layered scattering model requires inputting the proportion coefficient of the flower or silique. The formula for its total backscattering coefficient is as follows: (29) (30) (31), c is the total backscattering coefficient. f This represents the proportion of flowers and siliques within a pixel. , , , for Figure 4 The microwave scattering components in each direction are shown. is the backscattering coefficient of the stem and leaf layer. and for p and q Attenuation factor in the incident direction of polarized wave Indicates the scattering (emission) angle. Indicates the angle of incidence. P pq It is the phase matrix. and Here are the extinction matrices in the incident and exit directions. h This refers to the height of the plant. is the coherent scattering coefficient of the soil layer.

[0055] In summary, using Radarsat-2 fully polarimetric data as the data source, an MLMSTM was constructed based on vector radiative transfer theory to acquire scattering information from different layers of rapeseed. Focusing on scattering from the rapeseed plant layer and eliminating the influence of the underlying surface, this provides a theoretical basis for the rapeseed TPAI inversion model. Based on the plant characteristics at different key phenological stages of rapeseed, the extinction matrix and phase matrix of different rapeseed media (leaves, flowers, and siliques) required for establishing the equations for the microwave scattering amplitude function of single-scattering particles were calculated. The Muller matrix for soil scattering was calculated using an improved Oh model. Then, the extinction matrix, phase matrix, soil layer Muller matrix, and sensor parameters were weighted and averaged to obtain the overall Muller matrix solution. The backscattering coefficients of different vertical layers of rapeseed were then obtained by solving these solutions sequentially.

[0056] See Figure 2 After constructing the MLMSTM, the TPAI of rapeseed in the region will be inverted and validated. Specifically, it includes two parts: First, based on LAMP, the layered microwave scattering characteristics of rapeseed at five phenological stages are measured in a controlled environment (interference-free microwave anechoic chamber). Microwave scattering information of rapeseed at different phenological stages, different layers, and different polarization modes at an incident angle of 40° is obtained. After validating the model effectiveness based on LAMP data, the sensitive parameters are determined using Sobol parameter sensitivity to simplify the model complexity and obtain the parameter-calibrated MLMSTM. Second, taking the Radarsat-2 image area as the study area and the rapeseed microwave scattering ratio as the linking point, the RF model is used to realize the inversion of the regional rapeseed TPAI, further validating the model's feasibility.

[0057] Part One Rapeseed crop parameters and soil parameters vary at different locations. Numerous and unstable parameters can introduce errors into model predictions, and multi-parameter inputs are easily limited by sample data when applied regionally. Furthermore, the main canopy components of rapeseed plants change at different phenological stages, and the sensitive parameters also differ at these stages. To accurately retrieve the regional rapeseed TPAI, this embodiment employs the Sobol global sensitivity analysis method to identify the main sensitive crop and soil parameters at each phenological stage of rapeseed, simplifying model complexity.

[0058] The correction of the MLMSTM was determined based on Sobol parameter sensitivity analysis, including: identifying m parameters and n sample points from ground-collected data, where n is greater than 65 times m; determining the variance generated by individual parameters, the variance generated by the interaction of multiple parameters, and the variance generated by the combined effect of all parameters; determining the sensitivity of each parameter based on its influence on the backscattering coefficients of individual parameters and the interaction of multiple parameters; and extracting the main crop and soil parameters based on the sensitivity to obtain the correction of the MLMSTM. The MLMSTM includes several crop and soil parameters, such as rapeseed leaf length, thickness, and width; rapeseed flower diameter and height; and soil roughness and moisture content.

[0059] Specifically, the Sobol global sensitivity analysis method assesses parameter sensitivity by calculating the contribution of individual parameters and multiple parameters to the output variance. It assumes a nonlinear model using... express, y This refers to the objective function value output by the model; in this embodiment, it refers to the output backscattering coefficient value. This is the parameter set of the model, corresponding to the crop and soil parameters measured in this embodiment. y The total variance can be decomposed into several sub-variances: (32), D i It is x i The resulting variance, D ij It is x i and x j The variance generated by the interaction It is the variance generated by the combined effect of all parameters. Normalizing equation (32) yields the parameters and the sensitive parameters of their interactions: (33), S i It is parameter x i Sensitivity, S ij It is parameter x i and x j Interaction sensitivity.

[0060] The first-order Sobol sensitivity index represents the effect of a single input parameter on the output: (34), The overall Sobol sensitivity index reflects the impact of parameter interactions on the output parameters: (35), D To output the total variance of the data, To divide the parameter x i The variance is generated by the combined effect of other parameters. Calculating the first-order Sobol sensitivity index and the total Sobol sensitivity index requires running... n * m Second-rate, n For the sample size, m Set the number of input parameters for the model. The number of sample points should be more than 65 times the number of parameters. The maximum number of parameters is 14, so the number of samples should be set to 1000.

[0061] Part Two The determination of the per-pixel microwave scattering ratio of the Radarsat-2 image region based on the corrected MLMSTM and the parameters in the ground acquisition data includes: determining the backscattering coefficient of the vegetation layer based on the corrected MLMSTM and the parameters in the ground acquisition data; extracting the SAR image backscattering coefficient of a first number of sample points in the Radarsat-2 image region; determining the microwave scattering ratio of each sample point based on the backscattering coefficient of the vegetation layer and the backscattering coefficient of the SAR image; and determining the per-pixel microwave scattering ratio of the Radarsat-2 image region based on the microwave scattering ratio of each sample point using the spatial interpolation method.

[0062] Furthermore, based on the pixel-by-pixel microwave scattering ratio, the total photosynthetic area index (TPAI) of rapeseed is retrieved using a trained machine learning (RF algorithm) model, including: obtaining the leaf area index and silique pericarp area index from ground-collected data; training the RF model using the leaf area index and silique pericarp area index of a second number of samples, determining the optimal parameters of the RF model, wherein the second number is less than the first number; and retrieving the rapeseed TPAI value pixel-by-pixel based on the optimal parameters of the RF model.

[0063] Furthermore, the RF model was validated using the leaf area index and angular pericarp area index of the third number of samples, where the sum of the second and third numbers equaled the first number.

[0064] Specifically, based on the calibrated MLMSSM and combined with parameters collected from sampling points at various phenological stages of rapeseed, the scattering coefficients of different media layers (stem-leaf layer and pod layer) at each phenological stage can be obtained. The scattering coefficients of different layers are then mapped to LAI and pod area index (P). od_AI), and then TPAI is inverted using RF. That is: at the six-leaf stage, rapeseed plants are mainly composed of leaves, and the stem-leaf layer scattering coefficient is calculated using MLMSTM to invert LAI, at which point TPAI=LAI; at the 30% flowering stage, rapeseed plants are mainly composed of leaves and flowers, but flowers do not participate in photosynthesis, so LAI is inverted using the stem-leaf layer scattering coefficient, at which point TPAI=LAI; at the 90% flowering and early maturity stage, siliques are growing, and leaves have not yet completely fallen off, the photosynthetic organs are leaves and silique pericarps, at which point TPAI=LAI+P od _AI, then the stem-leaf layer scattering coefficient is inverted to retrieve LAI, and the silique layer scattering coefficient is inverted to retrieve P. od _AI; At maturity, the leaves completely wither, and the photosynthetic organ becomes the silique pericarp. At this time, TPAI=P od _AI, then the stratum corneum scattering coefficient is inverted to P od _AI. Furthermore, the radar backscattering characteristics of the input RF are: VV, HH, HV, HH / VV, and HH / HV. Based on the reciprocity theorem and LAMP data visualization analysis, the backscattering intensities of VH and HV are similar; to reduce information redundancy, only one is selected. Since the same polarization ratio HH / VV and cross-polarization ratio HH / HV also show good potential for inverting crop parameters, they are considered as radar characteristics in this embodiment to further improve the TPAI inversion accuracy.

[0065] The specific steps for TPAI inversion of Radarsat-2 image regions are as follows: (1) For the corresponding phenological stage of rapeseed, select the corresponding calibrated MLMSTM sub-model, and substitute the collected crop and soil parameters to obtain the scattering information of the stem and leaf layer and silique layer of rapeseed plants.

[0066] (2) Extract the SAR backscattering coefficients of each sample point in the corresponding Radarsat-2 image according to the sample point location, and calculate the rapeseed microwave scattering ratio: (36) R The microwave scattering ratios for different phenological stages of rapeseed are given. The backscattering coefficient of the main component layer of rapeseed plant. This represents the backscattering coefficient of the SAR image.

[0067] (3) Based on the microwave scattering ratio of the sample points, spatial interpolation is used to extend to regional pixels to obtain the microwave scattering ratio per pixel of the region. It has been found that Kriging interpolation shows better performance in the comparison of spatial interpolation methods for various factors such as soil nutrients, temperature, and precipitation. Therefore, the spatial interpolation method adopted in this embodiment is Kriging interpolation, and the interpolation model is an exponential model. The backscattering coefficient of the main component layer of rapeseed plants per pixel is calculated based on the microwave scattering ratio of rapeseed per pixel and the backscattering coefficient of SAR image.

[0068] (4) For example, the Radarsat-2 image covers a total of 30 sample points. The measured sample point data is divided into training set and validation set in a 2:1 ratio. The RF model is trained using the training set, and the optimal parameters of the RF model are determined. Then, the rapeseed TPAI value is retrieved pixel by pixel.

[0069] The final rapeseed TPAI inversion result was obtained by masking the rapeseed TPAI inversion result using the Radarsat-2 image area spatial distribution map, and the accuracy was evaluated.

[0070] To obtain better RF model parameters, this embodiment makes full use of limited samples, for example, by performing 5000 independent repetitions of machine learning. In each machine learning iteration, the sample points of the Radarsat-2 image region are randomly divided into training and validation sets at a 2:1 ratio to ensure the robustness of the results. Hyperparameter optimization is performed using a grid search method, with the following parameter settings: ① Number of decision trees: 50, 100, 200; ② Maximum depth: 3, 5, 7, 10; ③ Minimum number of split samples: 2, 5, 10; ④ Minimum number of leaf node samples: 1, 2, 4. Model training employs 5-fold cross-validation, using the minimum MAE as the evaluation criterion to select the optimal parameter combination and the corresponding RF model. After determining the optimal model, predictions are made on the training set to obtain the prediction accuracy. Simultaneously, the importance of VV, HH, HV, HH / VV, and HH / HV features is calculated to analyze the response of different radar features to TPAI. Feature importance is calculated by averaging the contribution of each feature across all trees after calculating the sum of impurity reduction during node splitting. The higher the feature importance value, the greater the influence of the feature on the model's prediction.

[0071] In this embodiment, the accuracy evaluation index can be the coefficient of determination (R²). 2 ), RMSE, and MAE. 2 R is an important indicator of the model's fit. 2 The closer the value is to 1, the closer the model's predicted value is to the actual value. RMSE is the square root of the mean square error, which is the square root of the average of the squares of the differences between the measured and predicted values, reflecting the dispersion of the model's prediction error. MAE is the average of the absolute errors between the measured and predicted values, reflecting the average level of the prediction error. The smaller the RMSE and MAE, the smaller the deviation between the model's predicted value and the actual value.

[0072] (37) (38), (39) N The total number of samples, iFor the first i One sample, y i These are the measured values ​​of TPAI at the sample points. These are the predicted values ​​from the TPAI model. This represents the average value of the measured TPAI values.

[0073] Therefore, the beneficial effects of the embodiments of the present invention are as follows: (1) Based on the unique characteristics of canopy composition changes and photosynthetic organ succession in rapeseed, a TPAI inversion model suitable for key phenological stages of rapeseed was constructed for the first time. Existing microwave scattering models are insufficient to describe the microwave scattering characteristics of key phenological stages of rapeseed, and have not yet considered the dominant role of P in the later stages of rapeseed growth. od This invention, in conjunction with the microwave scattering mechanism, constructs a stratified microwave scattering mechanism model for rapeseed at different phenological stages, links the main photosynthetic organs (leaves, siliques) of rapeseed with the SAR backscattering coefficient, and uses the mainstream satellite Radarsat-2 as the data source to achieve high-precision TPAI inversion for rapeseed, providing technical support for rapeseed growth monitoring and yield prediction.

[0074] (2) Focusing on the stratified microwave scattering characteristics of rapeseed at different phenological stages, an interpretable model of the rapeseed microwave scattering mechanism was developed. Considering the significant changes in rapeseed canopy composition, vertical structure, and physical geometric parameters with the growth stage, an MLMSTM was constructed based on fully polarized Radarsat-2 data and combined with vector radiative transfer theory. This model quantifies the contribution of rapeseed canopy composition (leaves, flowers, siliques) and vertical stratification structure to microwave scattering, eliminates the influence of underlying surface scattering, extracts scattering information from major photosynthetic organs, and inverts LAI and P by layer. od _AI, and then TPAI, provides theoretical support for the rapeseed canopy parameter inversion model.

[0075] Another embodiment of this application proposes a system for inverting the total photosynthetic area index of rapeseed, including a computer-executable program that implements the above-described method when the program is run.

[0076] The method of this invention was verified in both an indoor laboratory and a rapeseed growing area. First, based on LAMP, layered microwave scattering characteristics of rapeseed at the six-leaf stage, 30% flowering stage, 90% flowering stage, initial maturity stage, and maturity stage were measured in a controlled environment (interference-free microwave anechoic chamber). Microwave scattering information for different phenological stages, different layers, and different polarization modes of rapeseed at an incident angle of 40° was obtained. The simulation accuracy of the soil layer MOh model and the effectiveness of the MLMSSM model were verified based on the LAMP data. Second, four Radarsat-2 remote sensing images were selected in Hengyang City, a major rapeseed producing area. The model was verified by combining it with sample data collected in Hengyang City, and the results were compared with those obtained by directly using the backscattering coefficients of SAR remote sensing images through RF inversion, verifying the effectiveness of the MLMSTM proposed in this patent.

[0077] (1) Improve the validation of the Oh model Comparison of predicted and actual values ​​of soil backscattering coefficients from the Moh model after calibration for each phenological stage: Figure 5 First, overall, the MOh model achieved high simulation accuracy across various phenological stages and polarization modes. The curve trends from 4 GHz to 8 GHz are largely consistent with the actual values. Meanwhile, from... Figure 5 The results also show that the original Oh model performed poorly in the study area, exhibiting a smooth upward curve. The sample data used in the parameter correction of the original Oh model also included different frequencies (1.25 GHz, 5.3 GHz), but the range was large, leading to a decrease in prediction performance when applied to smaller frequency differences. Secondly, from... Figure 6 The temporal changes in soil backscattering were observed. The actual values ​​show that backscattering gradually decreases with phenological changes, which may be related to the decrease in soil roughness in the later stages. The MOh model's variation curve is basically consistent with the actual values, and the MOh prediction is closer to the actual values. The original Oh model's predictions also show a similar trend, but the Oh model's predictions deviate significantly from the actual values. Finally, based on the accuracy results in Table 1, the MOh model significantly reduces the deviation from the actual values ​​compared to the Oh model, and is closer to the actual values; the prediction accuracy for different polarization modes is also improved. In summary, the MOh model calibrated based on LAMP experimental data is more suitable than the original Oh model for simulating soil scattering information in the study area. The specific parameters of the MOh model at 5.405 GHz (C-band center frequency) after calibration are as follows: (40) (41), (42), in, ρ is the backscattering coefficient, q is the cross-polarization ratio, p is the same polarization ratio, and k is the wavenumber. , λ is the wavelength. s is the root mean square height of the soil. m is the radar incident angle. s This refers to soil moisture content.

[0078] Table 1. Accuracy verification of simulated and actual backscattering coefficient values ​​based on LAMP data using the MOh and Oh models.

[0079] (2) Verification of MLMSTM accuracy based on LAMP data Based on the MLMSTM constructed in sections 2.1 and 2.2, the backscattering coefficients of different layers at different phenological stages of rapeseed can be calculated. In this section, based on the crop and soil parameters measured at each phenological stage in the LAMP experiment, we input the corresponding MLMSTM sub-models for each phenological stage to predict the microwave scattering coefficients of each layer under VV, HH, VH, and HV polarization modes within the C-band from 4 GHz to 8 GHz of rapeseed. Using the layered backscattering coefficients of rapeseed in the LAMP experiment as the true values, we verify the effectiveness of MLMSTM by calculating the RMSE and MAE between the true values ​​and the MLMSTM simulated values.

[0080] Figure 7 The table shows the actual and predicted values ​​of the backscattering coefficients of rapeseed at each layer from 4 GHz to 8 GHz. Overall, the predicted values ​​are basically consistent with the actual values. From 5.405 GHz, the predicted backscattering coefficients for each phenological stage are close to the actual values. Table 2 shows that RMSE and MAE are both relatively small, with an average RMSE of 1.30 and an average MAE of 1.05. The RMSE and MAE of the stem-leaf layer at the six-leaf stage under VV polarization are relatively large, possibly related to the fact that rapeseed plants at this stage are dominated by large leaves and have less pronounced vertical structure. The backscattering coefficient values ​​at the 90% flowering stage and the initial ripening stage under VH and HV polarization differ significantly, possibly due to the influence of the spiral arrangement of the siliques on their three-dimensional structure during this period.

[0081] In summary, the prediction of rapeseed backscattering coefficient based on LAMP data and MLMSTM has high accuracy. MLMSTM can simulate the scattering information of each layer of the rapeseed plant well, and the trend of variation is basically consistent with the actual value. Moreover, at 5.405 GHz, the difference between the predicted value and the actual value is small, indicating that MLMSTM can be used in Radarsat-2 data and can be further applied in regional applications.

[0082] Table 2. Accuracy verification of MLMSTM backscattering coefficients based on LAMP data

[0083] (3) Results of sensitivity analysis of model parameters The results of MLMSTM sensitivity analysis of rapeseed at different phenological stages are as follows: Figure 8 As shown.

[0084] according to Figure 8 Sensitivity analysis of parameters at various phenological stages showed that, from the perspective of polarization mode, the main sensitive parameters for different polarization modes at each phenological stage were largely consistent. At the six-leaf stage of rapeseed, the mean square root height in the soil (… RMSH s Leaf water content ( M leaf It is more sensitive to backscattering results. During the flowering period, when the rapeseed flowers are in full bloom, the water content of the rapeseed flower layer ( M f ) and rapeseed flower density ( D f The results are more sensitive to backscattering, followed by leaf-related parameters. During the initial ripening stage, the rapeseed siliques grow, at which point the water content of the silique layer ( M p ), silique radius ( T p ), length of silique ( L p The geometric parameters of the siliques are sensitive parameters. At maturity, when all leaves have fallen and the siliques remain the focus of the plant, the geometric parameters of the siliques are more sensitive to backscattering results, including the silique radius (…). T p ), length of silique ( L p ) and silique density ( D p Meanwhile, the order of the first-order Sobol coefficient and the total Sobol coefficient at each growth stage of rapeseed is basically the same. At the six-leaf stage, the rapeseed plant is dominated by large leaves, but has not yet completely covered the ground surface; at this time, the rapeseed leaf and soil parameters are relatively sensitive. As the rapeseed flowers bloom, the canopy is mainly composed of rapeseed flowers; radar incident waves first interact with the rapeseed flowers at the top of the plant, at which point the backscattering coefficient is significantly affected by the rapeseed flower parameters. After the rapeseed flowers wither, the siliques begin to grow. The siliques have a spirally arranged three-dimensional structure, which easily interacts strongly with radar incident waves; at this time, the geometric parameters of the siliques are relatively sensitive. In summary, the results of the parameter sensitivity analysis are basically consistent with the changes in the main components of rapeseed at each growth stage, indicating the rationality of the parameters considered in MLMSTM and providing a good foundation for subsequent regional applications.

[0085] Based on the parameter sensitivity analysis results, a parameter calibration scheme for MLMSTM was determined to simplify model complexity and improve its applicability in regional applications. For crop and soil parameters sensitive to various phenological stages, such as soil root mean square height (…),… RMSH s Leaf water content ( M leafFor crop and soil parameters that are not sensitive to phenological stages, such as soil-related length (…), input the values ​​measured from field samples during each phenological stage. CL s ), leaf thickness ( T l ) and plant height ( H Parameters such as are expressed as the average values ​​of ground samples measured during each phenological period. Table 3 shows the reference range of MLMSTM input parameter values ​​for each phenological period after parameter calibration.

[0086] Table 3 Reference values ​​of MLMSTM parameters after parameter calibration

[0087] (4) Results and validation of TPAI inversion for rapeseed in Radarsat-2 region The optimal hyperparameters of the RF model corresponding to each phenological stage of rapeseed are shown in Table 4. The optimal parameters of the RF differ for different phenological stages, indicating that the model complexity and dominant factor parameters vary at different phenological stages. The RF model was trained on the training set based on the optimal parameter settings. Due to limitations in satellite image transit time, this embodiment of the invention selected four Radarsat-2 images for retrieving the TPAI of the rapeseed region, corresponding to the six-leaf stage, flowering stage (30% flowering stage), early maturity stage, and mature stage, respectively. Following the steps in Section 2.4, the RF model was trained sequentially based on the measured data from the sample points. The accuracy of the training set is shown in Table 4. Figure 9 Overall, the training accuracy for each phenological stage is good, and the TPAI predicted values ​​show a better fit with the actual values. R 2 Both values ​​were greater than 0.65 and passed the significance test (P<0.001), with RMSE and MAE both less than 0.6. This indicates that RF effectively captured the relationship between radar characteristics and TPAI.

[0088] Table 4. Optimal hyperparameters of radiometric rendering (RF) for each phenological stage of rapeseed.

[0089] The importance ranking of features in each period of RF output is as follows: Figure 10 Compared to the backscattering coefficient of a single polarization mode, the polarization ratios (HH / VV and HH / HV) show greater importance. This indicates that the polarization ratios provide more plant information than single polarization, which is consistent with the conclusion in existing studies that the polarization ratios have great potential for retrieving crop parameters.

[0090] The trained RF model was used to retrieve the TPAI (Transient Photonic Aspect Ratio) of the area covered by Radarsat-2 image pixel by pixel. Based on the spatially interpolated rapeseed microwave scattering ratio, the microwave scattering coefficients of the stem-leaf layer and pod layer of rapeseed at different stages were calculated from the SAR backscattering coefficient. These coefficients were then input into the trained RF model to retrieve the TPAI pixel by pixel, and the TPAI retrieval results for the region were obtained based on a rapeseed distribution data mask. Figure 11 The TPAI value at the six-leaf stage ranged from 4.29 to 6.86, with higher values ​​in the northeastern part of the study area. During the flowering stage, the TPAI value ranged from 3.98 to 6.27, with a relatively even distribution across the region, although some areas showed higher values. In most areas, the TPAI value was concentrated between 3.98 and 5.13. Upon entering the early maturity stage, the TPAI value increased significantly, ranging from 6.21 to 9.07, with a relatively consistent spatial distribution, and higher values ​​appearing in the central region. At maturity, as the leaves completely fell, the TPAI value decreased significantly, ranging from 2.63 to 3.12, with a relatively consistent spatial distribution. The regional TPAI results were validated based on the validation points, and the accuracy was as follows: Figure 12 As shown. R in each period 2 The values ​​were 0.62, 0.59, 0.55, and 0.71, respectively, all greater than 0.50 and passing the significance test (P<0.001). The average RMSE was 0.44, and the average MAE was 0.35. This indicates that the RF model performed well on the validation set, demonstrating good generalization ability and effectively retrieving the TPAI values ​​of rapeseed at different growth stages in the region. The TPAI inversion R of rapeseed throughout its entire growth period... 2 =0.92, RMSE=0.48, MAE=0.35 ( Figure 13 This indicates that the proposed model can capture the plant characteristics of rapeseed at various phenological stages well, establish a correlation with radar characteristics, and thus accurately retrieve the TPAI value of rapeseed throughout its entire growth period.

[0091] (5) Comparison of TPAI accuracy for direct inversion of SAR image backscattering coefficients To verify the effectiveness and accuracy improvement performance of the constructed MLMSTM for extracting plant layer scattering information and retrieving TPAI, this embodiment of the invention calculates the accuracy of directly retrieving TPAI from SAR image backscattering coefficients using the same training and validation sets and RF inversion steps, based on sample SAR image backscattering coefficients. Figure 14 The inversion of TPAI based on MLMSTM at the six-leaf stage is superior to the inversion of R based on SAR image backscattering coefficients. 2 Increased by 0.31, RMSE decreased by 0.18; R during flowering period 2 Increased by 0.41, RMSE decreased by 0.19; R at the initial maturity stage 2 An increase of 0.38 resulted in a decrease of 0.24 in RMSE; R at maturity... 2The accuracy of TPAI retrieval increased by 0.36, while the RMSE decreased by 0.05. Overall, the extraction of plant layer scattering information based on MLMSTM eliminated the influence of underlying surface scattering, effectively improving the TPAI retrieval accuracy for each phenological stage of rapeseed, further demonstrating the feasibility and effectiveness of constructing the rapeseed MLMSTM model in this study. It is worth mentioning that although the RMSE of TPAI retrieval based on SAR image backscattering coefficients was improved by 0.36, the accuracy of TPAI retrieval decreased by 0.05. 2 Lower, but from Figure 14 The line graph on the right shows that both models captured the overall trend of TPAI well, and the RMSE was within an acceptable range, indicating that the RF model learned the overall TPAI change characteristics of rapeseed samples at various phenological stages well in both data sources.

[0092] Throughout the phenological period, rapeseed undergoes significant changes in canopy composition and vertical structure, with photosynthetic organs transitioning from leaves to siliques. Rapeseed growth monitoring requires comprehensive consideration of both LAI (Layered Area Intake) and PAI (Plant-Based Area Intake). SAR (Radiative Resonance Analysis) can penetrate vegetation and contains information from various plant layers. The backscattering coefficient obtained from remote sensing images is affected not only by plant layer scattering but also by interference from underlying surface scattering. To accurately retrieve canopy parameters, microwave scattering information from the crop layer needs to be extracted to reduce the influence of underlying surface scattering. Based on this, this invention, using vector radiative transfer theory, constructs an MLMSTM (Multi-Layered Radiative Transfer Model) to extract rapeseed plant layer scattering information, comprehensively considering plant characteristics and major photosynthetic organs at different phenological stages. The six-leaf stage uses a leaf-soil layered scattering model; the flowering and silique stages use flower-leaf-soil and fruit-leaf-soil layered scattering models; and the maturity stage uses a fruit-soil layered scattering model. LAI is mainly related to stem-leaf layer scattering information, while P... od The TPAI (Temperature Backscattering Aspect Ratio) is related to the silique layer scattering information. Then, the ratio of plant layer scattering to image backscattering coefficient is calculated. Taking the rapeseed microwave scattering ratio as the starting point, spatial interpolation is used to extend to the region, thereby obtaining the regional rapeseed plant layer scattering information. Finally, the TPAI of the region's rapeseed is retrieved using an RF model. The results show that the TPAI retrieval accuracy for the entire rapeseed growth period is good, and the R... 2 =0.92, RMSE=0.48, MAE=0.35. The accuracy of TPAI retrieved based on MLMSTM for each phenological period was compared with that retrieved based on SAR image backscattering coefficients. 2 The average improvement of 0.37 and the average decrease of RMSE of 0.17 demonstrate the feasibility and necessity of the method proposed in this patent.

[0093] The embodiments described above are merely preferred embodiments of the present invention. The terms "in one embodiment," "in another embodiment," "in yet another embodiment," or "in still another embodiment" used in this specification all refer to one or more of the same or different embodiments according to this disclosure. Ordinary variations and substitutions made by those skilled in the art within the scope of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for retrieving the total photosynthetic area index of rapeseed, comprising: A multilayer microwave scattering mechanism model (MLMSTM) for rapeseed was constructed based on vector radiative transfer theory to obtain the total backscattering coefficients of HH, VV, VH, and HV, and the plant layer backscattering coefficient, wherein the plant layer backscattering coefficient includes the stem and leaf layer backscattering coefficient and the silique layer backscattering coefficient. Based on the validation of the effectiveness of the MLMSTM using LAMP data from the microwave target characteristic measurement laboratory, the corrected MLMSTM was determined according to the Sobol parameter sensitivity analysis. Based on the corrected MLMSTM, the microwave scattering ratio of each pixel in the Radarsat-2 image region is determined using a spatial interpolation method. Based on the pixel-by-pixel microwave scattering ratio, the total photosynthetic area index (TPAI) of rapeseed is inverted using a trained machine learning model, wherein the machine learning model is a random forest (RF) model.

2. The method for retrieving the total photosynthetic area index of rapeseed according to claim 1, wherein the multilayer microwave scattering mechanism model (MLMSTM) of rapeseed based on vector radiative transfer theory comprises: Based on the plant characteristics of rapeseed at different key phenological stages, the phase matrix and extinction matrix of rapeseed in different media were determined based on the microwave scattering amplitude function of single scattering particles. Different media refer to leaves, flowers and / or siliques. Based on the LAMP data, the Muller matrix of the soil layer was determined on the basis of the improved Oh model; Based on the phase matrix and extinction matrix of the different media and the Muller matrix of the soil layer, the total Muller matrix is ​​determined, thereby constructing the MLMSTM of rapeseed.

3. The method for retrieving the total photosynthetic area index of rapeseed according to claim 2, wherein determining the Muller matrix of the soil layer based on the improved Oh model using the LAMP data includes: Soil moisture content, root mean square height, k value, and true values ​​of bare soil backscattering coefficients at three incident angles were collected as fitting datasets for five phenological stages. The five phenological stages refer to the six-leaf stage, 30% flowering stage, 90% flowering stage, early maturity stage, and mature stage. For any given frequency, the Oh model is calibrated based on the fitted dataset to obtain an improved Oh model; Based on the improved Oh model, the Muller matrix of the soil layer is determined according to the root mean square height of the soil and the soil moisture content.

4. The method for retrieving the total photosynthetic area index of rapeseed according to claim 2 or 3, wherein the MLMSTM of rapeseed comprises: A leaf-soil stratified scattering model at the six-leaf stage; A flower-leaf-soil layered scattering model at 30% flowering period; A layered scattering model of flower / fruit-leaf-soil during 90% of the flowering period; A layered scattering model of fruit-leaf-soil at the early ripening stage; A fruit-soil stratified scattering model at maturity.

5. The method for retrieving the total photosynthetic area index of rapeseed according to claim 1, wherein determining the corrected MLMSTM based on Sobol parameter sensitivity analysis includes: m parameters and n sample points are determined from ground-collected data, where n is greater than 65 times m. The parameters include crop parameters and soil parameters. Determine the variance produced by a single parameter, the variance produced by the interaction of multiple parameters, and the variance produced by the combined effect of all parameters. The sensitivity of each parameter is determined based on the influence of a single parameter and the interaction of multiple parameters on each backscattering coefficient. Based on the aforementioned sensitivity, key crop and soil parameters are extracted to obtain the corrected MLMSTM.

6. The method for retrieving the total photosynthetic area index of rapeseed according to claim 5, wherein the crop parameters include the leaf length, leaf thickness, leaf width of rapeseed and the diameter and height of rapeseed flowers, and the soil parameters include soil roughness and moisture content.

7. The method for retrieving the total photosynthetic area index of rapeseed according to claim 1, wherein determining the microwave scattering ratio per pixel of the Radarsat-2 image region based on the corrected MLMSTM using a spatial interpolation method includes: Based on the corrected MLMSTM and the parameters in the ground-acquired data, the backscattering coefficient of the plant layer is determined; Extract the SAR image backscattering coefficients of the first number of samples in the Radarsat-2 image region; The microwave scattering ratio of each sample point is determined based on the backscattering coefficient of the plant layer and the backscattering coefficient of the SAR image. The microwave scattering ratio of each pixel in the Radarsat-2 image region is determined by using a spatial interpolation method based on the microwave scattering ratio of each sample point.

8. The method for retrieving the total photosynthetic area index (TPAI) of rapeseed according to claim 7, wherein the retrieval of the TPAI of rapeseed based on the pixel-by-pixel microwave scattering ratio using a trained machine learning model (random forest RF algorithm) includes: Leaf area index and silique pericarp area index were obtained from ground-collected data; The leaf area index and angular pericarp area index of the second number of samples are used to train the RF model and determine the optimal parameters of the RF model, wherein the second number is less than the first number; Based on the optimal parameters of the RF model, the TPAI value of rapeseed is retrieved pixel by pixel.

9. The method for retrieving the total photosynthetic area index of rapeseed according to claim 8 further includes: The RF model was validated using the leaf area index and angular pericarp area index of the third number of samples, where the sum of the second and third numbers equaled the first number.

10. A system for inverting the total photosynthetic area index of rapeseed, characterized in that, It includes a memory and a processor, wherein the memory stores a computer-executable program, and the processor is capable of running the computer program to implement the method as described in any one of claims 1-9.

Images