A method and system for estimating rice leaf area index from dual-polarization SAR images based on dynamic constraints of canopy structure

Abstract

The invention discloses a rice leaf area index estimation method and system based on dual-polarization SAR images with dynamic constraints of canopy structure, comprising the following steps: obtaining dual-polarization SAR images, generating dual-polarization images of VV and VH after preprocessing the images, and calculating radar vegetation index; constructing a rice leaf area index estimation model based on dual-polarization SAR images with dynamic constraints of canopy structure; calculating the target rice growth days at the time of image acquisition; rice leaf area index estimation, bringing the calculated radar vegetation index and the obtained rice growth days into the established rice leaf area index estimation model based on dynamic constraints of canopy structure, and obtaining the target rice leaf area index at the current image acquisition time. The invention establishes a continuous relationship among vegetation index based on dynamic constraints of canopy, rice leaf area index and planting days, and has strong universality for cross-regional and cross-year prediction of rice leaf area index.

Description

Technical Field

[0001] The invention belongs to the technical field of quantitative remote sensing and rice monitoring, and in particular relates to a method and system for estimating rice leaf area index from dual-polarization SAR images based on dynamic constraints of canopy structure. Background Art

[0002] In modern agricultural production, LAI is an effective indicator for diagnosing rice growth and predicting yield. Satellite remote sensing technology has the characteristics of high observation frequency and wide coverage, providing people with large spatial scale and long time series observation information, and is an effective means to monitor crop growth dynamics. Among them, synthetic aperture radar (SAR) has more advantages than optical remote sensing under adverse weather conditions such as rain, clouds, haze and fog due to its all-weather continuous observation capability.

[0003] LAI estimation methods based on SAR images can be divided into two categories: model inversion method and empirical statistical method. The model inversion method describes the electromagnetic propagation process and energy radiation mechanism based on the physical model. The mathematical mechanism is complete and the inversion method is more universal. However, the physical modeling method is difficult to accurately describe the complex and changeable reality. An accurate and realistic model often requires a large number of parameters, which will bring huge uncertainty to the modeling. In contrast, the empirical statistical method uses the statistical relationship between ground LAI measurements and vegetation indices to perform linear or nonlinear modeling, which is a common method for satellite-based LAI retrieval. When the quality of the training samples is high enough, good inversion accuracy can be obtained, but in terms of universality, it is highly dependent on the representativeness of the regional dataset and shows significant variability in different locations. In addition, the inversion of crop biophysical parameters requires ground observation experiments synchronized with satellite imaging. The dataset used to develop the model defines the boundaries and limitations of the model and is not suitable for use in sites or time periods that are not involved in model parameterization and development. They are not enough to accurately predict LAI from an uncertain background.

[0004] The physical and biological properties of the vegetation canopy change as the phenological period progresses, affecting the relationship between LAI and the radar vegetation index (RVI). This makes models based on a single growth period perform poorly in reflecting rice growth characteristics in other growth periods. Combining information from multiple growth stages to address the data hierarchy phenomenon between LAI and RVI has been shown to be very effective. Deep learning and machine learning can be used to handle complex nonlinear relationships with multiple factors and have been widely used to predict agronomic traits with excellent results. However, these non-parametric methods rely on large training datasets that are expensive to collect and are not conducive to subsequent utilization.

[0005] Therefore, it is urgent to solve the above problems. Summary of the invention

[0006] Purpose of the invention: The first purpose of the present invention is to provide a method for estimating rice leaf area index from dual-polarization SAR images based on dynamic constraints of canopy structure. The estimation method establishes a continuous relationship among vegetation index based on dynamic constraints of canopy, rice leaf area index and planting days, and has strong universality for cross-regional and cross-year prediction of rice leaf area index.

[0007] The second object of the present invention is to provide a rice leaf area index estimation system based on dual-polarization SAR images with dynamic constraints on canopy structure.

[0008] Technical solution: To achieve the above objectives, the present invention discloses a method for estimating rice leaf area index from dual-polarization SAR images based on dynamic constraints of canopy structure, comprising the following steps:

[0009] (1) Obtain dual-polarization SAR images, generate dual-polarization images of VV and VH after image preprocessing, and calculate the radar vegetation index;

[0010] (2) Construct a rice leaf area index estimation model based on dual-polarization SAR images with dynamic constraints on canopy structure;

[0011] (3) Calculate the target rice growth days at the time of image acquisition;

[0012] (4) Rice leaf area index estimation: The radar vegetation index calculated in step (1) and the number of rice growing days obtained in step (3) are substituted into the dual-polarization SAR image rice leaf area index estimation model based on canopy structure dynamic constraints established in step (2) to obtain the target rice leaf area index at the current image acquisition time.

[0013] Wherein, step (1) specifically includes the following steps: acquiring a dual-polarization SAR image, performing orbit correction, radiation correction, image filtering, and terrain correction preprocessing on the image to generate dual-polarization images of VV and VH, and calculating the radar vegetation index RVI, the formula is:

[0014] (1)

[0015] In the formula and are the backscattering coefficients of VV and VH respectively.

[0016] Preferably, step (2) specifically includes the following steps:

[0017] (2.1) Construct a hierarchical linear model for estimating rice leaf area index;

[0018] Hierarchical linear models consist of two levels: a base level, which models the relationship between an outcome variable and the independent variables, and a subsequent level, in which the intercept and slope of each base-level independent variable are variables in the subsequent-level modeling;

[0019] (2)

[0020] In the formula is the leaf area index, is the base-level intercept, is the slope at the base level;

[0021] (3)

[0022] (4)

[0023] In the formula and are the polynomial parameters, is the degree of the polynomial, is the number of days for rice to grow;

[0024] The hierarchical linear model for estimating rice leaf area index is expressed as:

[0025] (5)

[0026] (2.2) Construct a canopy structure dynamic model to characterize the rice leaf area index;

[0027] The canopy structure dynamic model is used to simulate the temporal variability of the leaf area index. The canopy structure dynamic model is expressed as:

[0028] (6)

[0029] In the formula Describes the maximum leaf area, parameters Indicates the turning point The relative growth rate at is the aging rate, Indicates the time when all leaves have aged;

[0030] The polynomial function is used to fit the canopy structure dynamic model. The processed canopy structure dynamic model is expressed as:

[0031] (7)

[0032] In the formula is the polynomial parameter, and the polynomial degree is consistent with that in formula (5);

[0033] (2.3) Constructing the constraints of canopy dynamic structure model and hierarchical linear model;

[0034] The leaf area index estimated by the hierarchical linear model needs to conform to the growth trajectory represented by the canopy structure dynamic model. For polynomials, if two polynomials are equal, then the coefficients of all their terms of the same order are equal. The constraint condition is expressed as:

[0035] (8)

[0036] In the formula , , ;

[0037] (2.4) Construct a dynamic constraint model of the canopy structure based on the constraint conditions;

[0038] For the constraints, , and can be determined by least squares. RVI Available through satellite data; RVI When changes occur, if you want to satisfy formula (8), you need to adjust the parameters of the hierarchical linear model; for satellite time series, , ,in is the number of satellite passes; each time the satellite collects rice field information, the corresponding model parameters can be obtained, and the formula is:

[0039] (9)

[0040] In the formula is the intercept parameter corresponding to each satellite acquisition, for The polynomial parameters of ;

[0041] According to the intercept parameter corresponding to each satellite image, formula (3) is refitted to obtain the dynamic constraint model of canopy structure, which is the rice leaf area index estimation model of dual-polarization SAR image. The formula is:

[0042] (10).

[0043] Furthermore, in step (3), if the rice planting date is known, the target rice growth days are calculated based on the planting date and the image acquisition date. The formula is:

[0044] (11)

[0045] In the formula is the day of the year corresponding to the image acquisition date, The day of the year corresponding to the rice planting date.

[0046] Furthermore, in step (3), if the rice planting date is unknown, the rice planting date is determined by establishing a phenological characteristic constraint rule based on the characteristic signal changes before and after rice transplanting, thereby calculating the target rice growth days; the phenological characteristic constraint rule expression is:

[0047] (12)

[0048] In the formula is the slope of the RVI backscatter time series before transplantation, is the slope of the RVI backscatter time series after transplantation.

[0049] Based on the same inventive concept, the present invention discloses a rice leaf area index estimation system based on dual-polarization SAR images with dynamic constraints on canopy structure, comprising:

[0050] The vegetation index calculation module is used to obtain dual-polarization SAR images, generate dual-polarization images of VV and VH after preprocessing the images, and calculate the radar vegetation index;

[0051] The estimation model building module is used to build a rice leaf area index estimation model based on dual-polarization SAR images with dynamic constraints on canopy structure;

[0052] The growth days estimation module is used to estimate the growth days of the target rice at the time of image acquisition;

[0053] The index estimation module is used for estimating the rice leaf area index. The calculated radar vegetation index and the obtained rice growth days are brought into the established dual-polarization SAR image rice leaf area index estimation model based on the dynamic constraints of the canopy structure to obtain the target rice leaf area index at the current image acquisition moment.

[0054] The vegetation index calculation module obtains dual-polarization SAR images, generates dual-polarization images of VV and VH after orbit correction, radiation correction, image filtering, and terrain correction preprocessing, and calculates the radar vegetation index RVI. The formula is:

[0055] (1)

[0056] In the formula and are the backscattering coefficients of VV and VH respectively.

[0057] Preferably, a hierarchical linear model for estimating rice leaf area index is constructed in the estimation model construction module;

[0058] Hierarchical linear models consist of two levels: a base level, which models the relationship between an outcome variable and the independent variables, and a subsequent level, in which the intercept and slope of each base-level independent variable are variables in the subsequent-level modeling;

[0059] (2)

[0060] In the formula is the leaf area index, is the base-level intercept, is the slope at the base level;

[0061] (3)

[0062] (4)

[0063] In the formula and are the polynomial parameters, is the degree of the polynomial, is the number of days for rice to grow;

[0064] The hierarchical linear model for estimating rice leaf area index is expressed as:

[0065] (5)

[0066] A canopy structure dynamic model representing the rice leaf area index was constructed. The canopy structure dynamic model was used to simulate the temporal variability of the leaf area index. The canopy structure dynamic model was expressed as:

[0067] (6)

[0068] In the formula Describes the maximum leaf area, parameters Indicates the turning point The relative growth rate at is the aging rate, Indicates the time when all leaves have aged;

[0069] The polynomial function is used to fit the canopy structure dynamic model. The processed canopy structure dynamic model is expressed as:

[0070] (7)

[0071] In the formula is the polynomial parameter, and the polynomial degree is consistent with that in formula (5);

[0072] The constraints of the canopy dynamic structure model and the hierarchical linear model are constructed. The leaf area index estimated by the hierarchical linear model needs to conform to the growth trajectory represented by the canopy dynamic structure model. For polynomials, if two polynomials are equal, then the coefficients of all their terms of the same order are equal. The constraints can be expressed as:

[0073] (8)

[0074] In the formula , , ;

[0075] According to the constraints, a dynamic constraint model of canopy structure is constructed. , and can be determined by least squares. RVI Available through satellite data; RVI When changes occur, if you want to satisfy formula (8), you need to adjust the parameters of the hierarchical linear model; for satellite time series, , ,in is the number of satellite passes; each time the satellite collects rice field information, the corresponding model parameters can be obtained, and the formula is:

[0076] (9)

[0077] In the formula is the intercept parameter corresponding to each satellite acquisition, for The polynomial parameters of ;

[0078] According to the intercept parameter corresponding to each satellite image, formula (3) is refitted to obtain the dynamic constraint model of canopy structure, which is the rice leaf area index estimation model of dual-polarization SAR image. The formula is:

[0079] (10).

[0080] Furthermore, in the growth days estimation module, if the rice planting date is known, the target rice growth days are calculated based on the planting date and the image acquisition date. The formula is:

[0081] (11)

[0082] In the formula is the day of the year corresponding to the image acquisition date, The day of the year corresponding to the rice planting date.

[0083] Furthermore, in the growth days estimation module, if the rice planting date is unknown, the rice planting date is determined based on the characteristic signal changes before and after rice transplanting, and the phenological characteristic constraint rules are established to estimate the target rice growth days; the phenological characteristic constraint rule expression is:

[0084] (12)

[0085] In the formula is the slope of the RVI backscatter time series before transplantation, is the slope of the RVI backscatter time series after transplantation.

[0086] Beneficial effects: Compared with the prior art, the present invention has the following significant advantages: the present invention solves the problem of sample imbalance between measured data and satellite data, and also solves the problem that the hierarchical linear model ignores the temporal variation characteristics of LAI due to blind fitting; the present invention takes into account the canopy dynamic information and time constraints in the modeling process. On the one hand, the canopy structure dynamic model can provide the expected value of any growth stage by simulating the temporal variation of LAI, so that the satellite data during the crop growth period can obtain the corresponding reference point; on the other hand, the canopy structure dynamic model can provide the changing trend of LAI at any growth stage, so that the inversion model can accurately characterize the growth trajectory of LAI. BRIEF DESCRIPTION OF THE DRAWINGS

[0087] Figure 1 It is a schematic diagram of the process of the present invention;

[0088] Figure 2 is the collection date of the ground survey data and the satellite data in the embodiment of the present invention;

[0089] Figure 3 A schematic diagram of hierarchical linear model calibration according to an embodiment of the present invention;

[0090] Figure 4 A schematic diagram of calibration of a canopy dynamic constraint model according to an embodiment of the present invention;

[0091] Figure 5 This is a schematic diagram of the calculation of the number of growing days in an embodiment of the present invention, wherein the image is the extraction result of the number of rice planting dates in 2019 and 2020;

[0092] Figure 6 It is a schematic diagram of the model performance evaluation of an embodiment of the present invention, wherein the images are the LAI inversion accuracy of different models and their time series monitoring results;

[0093] Figure 7 It is a schematic diagram of the cross-year accuracy test of the model of an embodiment of the present invention, wherein the image shows the cross-year LAI inversion accuracy and its time series monitoring results. DETAILED DESCRIPTION

[0094] The technical solution of the present invention is further described below in conjunction with the accompanying drawings.

[0095] Example 1: The observation data of this example comes from the dual-polarization SAR data and supporting ground observation data collected in the Changshu Rice Experimental Area from 2019 to 2020. Among them, the data in 2019 is used to establish the model, and the data in 2020 is used for model testing. On the one hand, the acquisition time of the ground measured data and the satellite data in 2019 is synchronized the most times (up to 5 times), which can ensure that the acquired satellite data can accurately reflect the growth status of rice in the study area. On the other hand, the satellite data in 2019 is more intensive, which can provide data support for the constraints of model parameters. Figure 2 As shown, the collection dates of ground survey data and satellite data in Changshu area from 2019 to 2020 are summarized.

[0096] like Figure 1 As shown, this embodiment discloses a method for estimating rice leaf area index from dual-polarization SAR images based on dynamic constraints of canopy structure, comprising the following steps:

[0097] (1) Obtain dual-polarization SAR images, generate dual-polarization images of VV and VH after image preprocessing, and calculate the radar vegetation index;

[0098] Step (1) specifically includes the following steps: acquiring a dual-polarization SAR image, performing orbit correction, radiation correction, image filtering, and terrain correction preprocessing on the image to generate dual-polarization images of VV and VH, and calculating the radar vegetation index RVI, the formula is:

[0099] (1)

[0100] In the formula and are the backscattering coefficients of VV and VH respectively.

[0101] (2) Construct a rice leaf area index estimation model based on dual-polarization SAR images with dynamic constraints on canopy structure;

[0102] Step (2) specifically includes the following steps:

[0103] (2.1) Construct a hierarchical linear model for estimating rice leaf area index;

[0104] Hierarchical linear models consist of two levels: a base level, similar to least squares regression modeling, that models the relationship between an outcome variable and the independent variables, and a subsequent level, where the intercept and slope of each base-level independent variable are the variables in the subsequent-level modeling; e.g. Figure 3As shown, image (a) is the linear regression result of RVI-LAI at different growth stages, and image (b) is and The fitting results of

[0105] (2)

[0106] In the formula is the leaf area index, is the base-level intercept, is the slope at the base level;

[0107] (3)

[0108] (4)

[0109] In the formula and are the polynomial parameters, is the degree of the polynomial, is the number of days for rice to grow;

[0110] The hierarchical linear model for estimating rice leaf area index is expressed as:

[0111] (5)

[0112] (2.2) Construct a canopy structure dynamic model to characterize the rice leaf area index;

[0113] The canopy structure dynamic model is used to simulate the temporal variability of the leaf area index. The canopy structure dynamic model is expressed as:

[0114] (6)

[0115] In the formula Describes the maximum leaf area, parameters Indicates the turning point The relative growth rate at is the aging rate, Indicates the time when all leaves have aged;

[0116] The polynomial function is used to fit the canopy structure dynamic model, such as Figure 4 As shown, image (a) is the simulation result of the canopy dynamic model; image (b) is the result of model parameter adjustment; the processed canopy structure dynamic model is expressed as:

[0117] (7)

[0118] In the formula is the polynomial parameter, and the polynomial degree is consistent with that in formula (5);

[0119] (2.3) Constructing the constraints of canopy dynamic structure model and hierarchical linear model;

[0120] The leaf area index estimated by the hierarchical linear model needs to conform to the growth trajectory represented by the canopy structure dynamic model. For polynomials, if two polynomials are equal, then the coefficients of all their terms of the same order are equal. The constraint condition is expressed as:

[0121] (8)

[0122] In the formula , , ;

[0123] (2.4) Construct a dynamic constraint model of the canopy structure based on the constraint conditions;

[0124] For the constraints, , and can be determined by least squares. RVI Available through satellite data; RVI When changes occur, if you want to satisfy formula (8), you need to adjust the parameters of the hierarchical linear model; for satellite time series, , ,in is the number of satellite passes; each time the satellite collects rice field information, the corresponding model parameters can be obtained, and the formula is:

[0125] (9)

[0126] In the formula is the intercept parameter corresponding to each satellite acquisition, for The polynomial parameters of ;

[0127] According to the intercept parameter corresponding to each satellite image, formula (3) is refitted to obtain the dynamic constraint model of canopy structure, which is the rice leaf area index estimation model of dual-polarization SAR image. The formula is:

[0128] (10).

[0129] (3) Estimate the target rice growth days at the time of image acquisition, such as Figure 5 As shown;

[0130] In step (3), if the rice planting date is known, the target rice growth days are calculated based on the planting date and the image acquisition date. The formula is:

[0131] (11)

[0132] In the formula is the day of the year corresponding to the image acquisition date, The day of the year corresponding to the rice planting date.

[0133] In step (3), if the rice planting date is unknown, the rice planting date is determined by establishing phenological characteristic constraint rules based on the characteristic signal changes before and after rice transplanting, and the target rice growth days are calculated. The expression of the phenological characteristic constraint rules is:

[0134] (12)

[0135] In the formula is the slope of the RVI backscatter time series before transplantation, is the slope of the RVI backscatter time series after transplantation.

[0136] (4) Rice leaf area index estimation: The radar vegetation index calculated in step (1) and the number of rice growing days obtained in step (3) are substituted into the dual-polarization SAR image rice leaf area index estimation model based on canopy structure dynamic constraints established in step (2) to obtain the target rice leaf area index at the current image acquisition time.

[0137] In this example, the performance of the canopy structure dynamic constraint model was verified using experimental data from 2019, and the hierarchical linear model and the Gaussian process model were compared. The results are as follows: Figure 6 As shown in the figure, the comparison between the LAI estimation and the ground data shows that the hierarchical linear model (R 2 =0.7234, RMSE=0.9561) and Gaussian process model (R 2 =0.7143, RMSE=0.9717) are comparable in prediction accuracy. 2 and RMSE are better than the other two models (R 2 =0.7688, RMSE=0.8742). From the time series monitoring results of LAI of different models, the monitoring results of the canopy structure dynamic constraint model are highly consistent with the measured LAI. The monitoring results of the hierarchical linear model and the Gaussian process model in some key periods do not conform to the temporal changes of LAI.

[0138] In this embodiment, the experimental data of 2020 is used to perform a cross-year test, and the results are as follows: Figure 7 The results show that the canopy structure dynamic constraint model proposed in this invention can estimate rice LAI throughout the growth period, and the estimation results are stable and accurate (R 2=0.7242, RMSE=1.281). Therefore, the method proposed in the present invention has strong generalization and robustness in practical applications.

[0139] The rice leaf area index estimation model based on dual-polarization SAR images with dynamic constraints on canopy structure established by the present invention solves the problem of sample imbalance between measured data and satellite data, and also solves the problem that conventional models ignore the temporal variation characteristics of LAI due to blind fitting. This method can provide expected values ​​for any growth stage by simulating the temporal variation of LAI, so that satellite data during the crop growth period can obtain corresponding reference points, so that the estimation model can accurately depict the growth trajectory of LAI. In addition, many traits of crops during the growth period, such as plant height, dry matter weight, biomass, etc., will show regular quantitative changes as the growth period progresses. This means that the present invention can also be applied to the monitoring of other crop physiological parameters.

[0140] Embodiment 2: This embodiment discloses a rice leaf area index estimation system based on dual-polarization SAR images with dynamic constraints on canopy structure, comprising:

[0141] The vegetation index calculation module is used to obtain dual-polarization SAR images, generate dual-polarization images of VV and VH after preprocessing the images, and calculate the radar vegetation index;

[0142] The vegetation index calculation module obtains dual-polarization SAR images, generates dual-polarization images of VV and VH after orbit correction, radiation correction, image filtering, and terrain correction preprocessing, and calculates the radar vegetation index RVI. The formula is:

[0143] (1)

[0144] In the formula and are the backscattering coefficients of VV and VH respectively.

[0145] The estimation model building module is used to build a rice leaf area index estimation model based on dual-polarization SAR images with dynamic constraints on canopy structure;

[0146] In the estimation model building module, a hierarchical linear model for estimating rice leaf area index was constructed;

[0147] Hierarchical linear models consist of two levels: a base level, which models the relationship between an outcome variable and the independent variables, and a subsequent level, in which the intercept and slope of each base-level independent variable are variables in the subsequent-level modeling;

[0148] (2)

[0149] In the formula is the leaf area index, is the base-level intercept, is the slope at the base level;

[0150] (3)

[0151] (4)

[0152] In the formula and are the polynomial parameters, is the degree of the polynomial, is the number of days for rice to grow;

[0153] The hierarchical linear model for estimating rice leaf area index is expressed as:

[0154] (5)

[0155] A canopy structure dynamic model representing the rice leaf area index was constructed. The canopy structure dynamic model was used to simulate the temporal variability of the leaf area index. The canopy structure dynamic model was expressed as:

[0156] (6)

[0157] In the formula Describes the maximum leaf area, parameters Indicates the turning point The relative growth rate at is the aging rate, Indicates the time when all leaves have aged;

[0158] The polynomial function is used to fit the canopy structure dynamic model. The processed canopy structure dynamic model is expressed as:

[0159] (7)

[0160] In the formula is the polynomial parameter, and the polynomial degree is consistent with that in formula (5);

[0161] The constraints of the canopy dynamic structure model and the hierarchical linear model are constructed. The leaf area index estimated by the hierarchical linear model needs to conform to the growth trajectory represented by the canopy dynamic structure model. For polynomials, if two polynomials are equal, then the coefficients of all their terms of the same order are equal. The constraints can be expressed as:

[0162] (8)

[0163] In the formula , , ;

[0164] According to the constraints, a dynamic constraint model of canopy structure is constructed. , and can be determined by least squares. RVI Available through satellite data; RVI When changes occur, if you want to satisfy formula (8), you need to adjust the parameters of the hierarchical linear model; for satellite time series, , ,in is the number of satellite passes; each time the satellite collects rice field information, the corresponding model parameters can be obtained, and the formula is:

[0165] (9)

[0166] In the formula is the intercept parameter corresponding to each satellite acquisition, for The polynomial parameters of ;

[0167] According to the intercept parameter corresponding to each satellite image, formula (3) is refitted to obtain the dynamic constraint model of canopy structure, which is the rice leaf area index estimation model of dual-polarization SAR image. The formula is:

[0168] (10).

[0169] The growth days estimation module is used to estimate the growth days of the target rice at the time of image acquisition;

[0170] In the growth days calculation module, if the rice planting date is known, the target rice growth days are calculated based on the planting date and the image acquisition date. The formula is:

[0171] (11)

[0172] In the formula is the day of the year corresponding to the image acquisition date, The day of the year corresponding to the rice planting date.

[0173] In the growth days estimation module, if the rice planting date is unknown, the rice planting date is determined by establishing phenological characteristic constraint rules based on the characteristic signal changes before and after rice transplanting, so as to estimate the target rice growth days; the phenological characteristic constraint rule expression is:

[0174] (12)

[0175] In the formula is the slope of the RVI backscatter time series before transplantation, is the slope of the RVI backscatter time series after transplantation.

[0176] The index estimation module is used for estimating the rice leaf area index. The calculated radar vegetation index and the obtained rice growth days are brought into the established dual-polarization SAR image rice leaf area index estimation model based on the dynamic constraints of the canopy structure to obtain the target rice leaf area index at the current image acquisition moment.

Claims

1. A method for estimating rice leaf area index from dual-polarization SAR images based on dynamic constraints of canopy structure, characterized in that: The following steps are involved: (1) Obtain dual-polarization SAR images, generate dual-polarization images of VV and VH after image preprocessing, and calculate the radar vegetation index; The step (1) specifically includes the following steps: acquiring a dual-polarization SAR image, performing orbit correction, radiation correction, image filtering, and terrain correction preprocessing on the image to generate a dual-polarization image of VV and VH, and calculating the radar vegetation index RVI, the formula being: （1） In the formula and are the backscattering coefficients of VV and VH respectively; (2) A rice leaf area index estimation model based on dual-polarization SAR images with dynamic constraints on canopy structure was constructed. Satellite transit data synchronized with ground-based measured data were used to establish the model. The step (2) specifically includes the following steps: (2.1) Construct a hierarchical linear model for estimating rice leaf area index; Hierarchical linear models consist of two levels: a base level, which models the relationship between an outcome variable and the independent variables, and a subsequent level, in which the intercept and slope of each base-level independent variable are variables in the subsequent-level modeling; （2） In the formula is the leaf area index, which is the ground measured data; is the base-level intercept, is the slope at the base level; （3） （4） In the formula and are the polynomial parameters, is the degree of the polynomial, is the number of days for rice to grow; The hierarchical linear model for estimating rice leaf area index is expressed as: （5） (2.2) Construct a canopy structure dynamic model to characterize the rice leaf area index; The canopy structure dynamic model is used to simulate the temporal variability of the leaf area index. The canopy structure dynamic model is expressed as: （6） In the formula Describes the maximum leaf area, parameters Indicates the turning point The relative growth rate at is the aging rate, Indicates the time when all leaves have aged; The polynomial function is used to fit the canopy structure dynamic model. The processed canopy structure dynamic model is expressed as: （7） In the formula is the polynomial parameter, and the polynomial degree is consistent with that in formula (5); (2.3) Constructing the constraints of canopy dynamic structure model and hierarchical linear model; The leaf area index estimated by the hierarchical linear model needs to conform to the growth trajectory represented by the canopy structure dynamic model. For polynomials, if two polynomials are equal, then the coefficients of all their terms of the same order are equal. The constraint condition is expressed as: （8） In the formula , , ; (2.4) Construct a dynamic constraint model of the canopy structure based on the constraint conditions; For the constraints, , and can be determined by least squares. RVI It can be obtained through satellite data, which is satellite transit data; RVI When changes occur, if you want to satisfy formula (8), you need to adjust the parameters of the hierarchical linear model; for satellite time series, , ,in is the number of satellite passes; each time the satellite collects rice field information, the corresponding model parameters can be obtained, and the formula is: （9） In the formula is the intercept parameter corresponding to each satellite acquisition, for The polynomial parameters of ; According to the intercept parameter corresponding to each satellite image, formula (3) is refitted to obtain the dynamic constraint model of canopy structure, which is the rice leaf area index estimation model of dual-polarization SAR image. The formula is: （10）； (3) Calculate the target rice growth days at the time of image acquisition; In step (3), if the rice planting date is unknown, the rice planting date is determined by establishing a phenological characteristic constraint rule based on the characteristic signal changes before and after rice transplanting, thereby calculating the target rice growth days; the phenological characteristic constraint rule expression is: （12） In the formula is the slope of the RVI backscatter time series before transplantation, is the slope of the RVI backscatter time series after transplantation; In step (3), if the rice planting date is known, the target rice growth days are calculated based on the planting date and the image acquisition date. The formula is: （11） In the formula is the day of the year corresponding to the image acquisition date, is the day of the year corresponding to the rice planting date; (4) Rice leaf area index estimation: The radar vegetation index calculated in step (1) and the number of rice growing days obtained in step (3) are substituted into the dual-polarization SAR image rice leaf area index estimation model based on canopy structure dynamic constraints established in step (2) to obtain the target rice leaf area index at the current image acquisition time.

2. A rice leaf area index estimation system based on dual-polarization SAR images with dynamic constraints on canopy structure, characterized in that: include: The vegetation index calculation module is used to obtain dual-polarization SAR images, generate dual-polarization images of VV and VH after preprocessing the images, and calculate the radar vegetation index; The vegetation index calculation module obtains dual-polarization SAR images, generates dual-polarization images of VV and VH after orbit correction, radiation correction, image filtering, and terrain correction preprocessing, and calculates the radar vegetation index RVI. The formula is: （1） In the formula and are the backscattering coefficients of VV and VH respectively; The estimation model building module is used to build a rice leaf area index estimation model based on dual-polarization SAR images with dynamic constraints on canopy structure; A hierarchical linear model for estimating rice leaf area index is constructed in the estimation model construction module; Hierarchical linear models consist of two levels: a base level, which models the relationship between an outcome variable and the independent variables, and a subsequent level, in which the intercept and slope of each base-level independent variable are variables in the subsequent-level modeling; （2） In the formula is the leaf area index, is the base-level intercept, is the slope at the base level; （3） （4） In the formula and are the polynomial parameters, is the degree of the polynomial, is the number of days for rice to grow; The hierarchical linear model for estimating rice leaf area index is expressed as: （5） A canopy structure dynamic model representing the rice leaf area index was constructed. The canopy structure dynamic model was used to simulate the temporal variability of the leaf area index. The canopy structure dynamic model was expressed as: （6） In the formula Describes the maximum leaf area, parameters Indicates the turning point The relative growth rate at is the aging rate, Indicates the time when all leaves have aged; The polynomial function is used to fit the canopy structure dynamic model. The processed canopy structure dynamic model is expressed as: （7） In the formula is the polynomial parameter, and the polynomial degree is consistent with that in formula (5); The constraints of the canopy dynamic structure model and the hierarchical linear model are constructed. The leaf area index estimated by the hierarchical linear model needs to conform to the growth trajectory represented by the canopy dynamic structure model. For polynomials, if two polynomials are equal, then the coefficients of all their terms of the same order are equal. The constraints can be expressed as: （8） In the formula , , ; According to the constraints, a dynamic constraint model of canopy structure is constructed. , and can be determined by least squares. RVI Available through satellite data; RVI When changes occur, if you want to satisfy formula (8), you need to adjust the parameters of the hierarchical linear model; for satellite time series, , ,in is the number of satellite passes; each time the satellite collects rice field information, the corresponding model parameters can be obtained, and the formula is: （9） In the formula is the intercept parameter corresponding to each satellite acquisition, for The polynomial parameters of ; According to the intercept parameter corresponding to each satellite image, formula (3) is refitted to obtain the dynamic constraint model of canopy structure, which is the rice leaf area index estimation model of dual-polarization SAR image. The formula is: （10）； The growth days estimation module is used to estimate the growth days of the target rice at the time of image acquisition; In the growth days calculation module, if the rice planting date is known, the target rice growth days are calculated according to the planting date and the image acquisition date. The formula is: （11） In the formula is the day of the year corresponding to the image acquisition date, is the day of the year corresponding to the rice planting date; If the rice planting date is unknown in the growth days calculation module, the rice planting date is determined by establishing phenological characteristic constraint rules according to the characteristic signal changes before and after rice transplanting, so as to calculate the target rice growth days; the phenological characteristic constraint rule expression is: （12） In the formula is the slope of the RVI backscatter time series before transplantation, is the slope of the RVI backscatter time series after transplantation; The index estimation module is used for estimating the rice leaf area index. The calculated radar vegetation index and the obtained rice growth days are brought into the established dual-polarization SAR image rice leaf area index estimation model based on the dynamic constraints of the canopy structure to obtain the target rice leaf area index at the current image acquisition moment.

Images