CN105389466A - Middle and low resolution remote sensing product true value acquisition method for correcting scaling effect - Google Patents

Abstract

The present invention discloses a middle and low resolution remote sensing product true value acquisition method for correcting a scaling effect. The method comprises the following five steps: quantifying production of remote sensing products; performing model second derivative estimation; performing spatial heterogeneity estimation; performing spatial scaling effect estimation; and performing spatial scaling effect correction. According to the method provided by the present invention, constraints of demanding for matching high resolution data in the remote sensing product true value acquisition process are removed, the restriction of demanding for a continuously derivable remote sensing inverse model is broken through, and the scaling effect problem caused by in-class heterogeneity and inter-class heterogeneity is considered, so that acquisition accuracy of a middle and low resolution remote sensing product pixel scale true value is further improved, and simple and precise acquisition of the middle and low resolution remote sensing product pixel scale true value is finally implemented.

Description

Method for acquiring true value of medium-low resolution remote sensing product for correcting scale effect

Technical Field

The invention relates to a method for acquiring a true value of a medium-low resolution remote sensing product, in particular to a method for acquiring a true value of a medium-low resolution remote sensing product for correcting scale effect.

Background

In the sixty-seven decades of the 20 th century, with the continuous development of remote sensing technology, the remote sensing science provides a powerful means for regional measurement and estimation of surface elements, which can be said to be a leap from the traditional 'point' measurement to 'surface' measurement, and the remote sensing technology gradually becomes the only effective means for obtaining regional scale surface information at present. However, it is still not easy to hide the real value acquisition of the medium and low resolution remote sensing products behind the bright and bright background, especially for the non-uniform ground surface. The reason is mainly that most of models used in the remote sensing inversion process are established by analyzing the relationship between actual surface parameters and physical quantities measured by sensors under specific wavelengths. In short, most of these remote sensing models are developed from point source observation, and are generally only suitable for the condition that the subsurface is uniform. In practical application, most of the medium-low resolution pixels are mixed pixels due to the unevenness of the earth surface. The remote sensing models are used for inverting the surface parameters without correction, so that a large inversion error is brought by the change of the scale, the problem of scale effect is caused, the high-precision acquisition of the surface truth value of the remote sensing product is influenced, and the requirement of various industry decisions on the remote sensing inversion precision of the surface parameters cannot be met.

The existing acquisition method has the defects that:

the estimation of the truth value of the medium-low resolution remote sensing product can not be separated from the high resolution data. The matched high-resolution data is troublesome to obtain and has higher cost. In addition, conditions such as imaging time, observation geometry and the like of the high-resolution data and the medium-low resolution data are required to be approximately consistent, otherwise, certain deviation can be brought by estimation of spatial heterogeneity, and the accuracy of obtaining the truth value of the medium-low resolution remote sensing product is further influenced. The limitation results in that the real value of the remote sensing product with medium and low resolution cannot be effectively and accurately acquired due to the lack of high resolution data in the practical application process.

In the mathematical derivation of the spatial scale effect, the spatial scale effect is quantitatively estimated by using a Taylor series expansion formula. This process requires that the remote sensing inversion model is continuously derivable, so when the assumption is violated, that is, the selected remote sensing inversion model is discontinuous or non-derivable, the spatial scale effect cannot be estimated, and thus the real value of the remote sensing product with medium and low resolution cannot be obtained.

The spatial heterogeneity determines the size of the spatial scale effect, but the prior art only considers the scale effect caused by the heterogeneity within a class and ignores the scale effect caused by the heterogeneity between classes. However, the obvious inter-class heterogeneity between different types of tables is a key factor for determining the spatial scale effect, and if the influence of the inter-class heterogeneity cannot be effectively described, the precision of scale effect correction cannot be guaranteed, so that the accuracy of obtaining the truth value of the medium-low resolution remote sensing product is influenced.

Disclosure of Invention

In order to solve the defects of the technology, the invention provides a method for acquiring the true value of a medium-low resolution remote sensing product for correcting the scale effect.

In order to solve the technical problems, the invention adopts the technical scheme that: a method for acquiring a true value of a medium-low resolution remote sensing product for correcting scale effect comprises the following steps:

(1) and producing quantitative remote sensing products:

the production of the quantitative remote sensing product is specifically divided into 5 steps: firstly, acquiring kilometer-level medium-low resolution remote sensing data r, wherein r is observed values of a plurality of channels of a medium-low resolution remote sensing sensor; second, the component proportion omega of different types of tables is estimated through the classification chart of the matched subsurface_d(ii) a Thirdly, selecting a proper remote sensing inversion model f according to the classification diagram of the subsurface bedding surface and aiming at different subsurface bedding surfaces d_dGenerally, the types of subsurface bedding surfaces are different, and the forms or model coefficients of remote sensing inversion models are also different; fourthly, inputting the remote sensing data r into a corresponding remote sensing inversion model f_dIn the method, quantitative remote sensing products p corresponding to different subsurface pad surface components are directly obtained_dNamely: p is a radical of_d＝f_d(r); fifthly, acquiring a rough value of the medium-low resolution remote sensing product in an area weighting modeNamely:

$f\left(\stackrel{\‾}{r}\right)=\underset{d=1}{\overset{c}{\Σ}}{\mathrm{\ω}}_d{f}_d\left(r\right)---\left(1\right)$

wherein c is the total number of different underlying surface types of the ground surface; omega_dComponent proportions for different table types; f. of_d(r) represents the corresponding quantitative remote sensing product;

(2) estimating a model second derivative:

the remote sensing inversion function is taken as a black box, and the second-order derivative and the second-order partial derivative k of the model are estimated in a discrete solving mode_i，jNamely:

$k_{i,j}=\frac{{\∂}^{2}f}{\∂r_i\∂r_j}=\frac{\frac{\∂f}{\∂r_i}(r+{\mathrm{\Δr}}_j)-\frac{\∂f}{\∂r_i}(r-{\mathrm{\Δr}}_j)}{2{\mathrm{\Δr}}_j}---\left(2\right)$

wherein,is a first derivative of the model and can be expressed asr is observed values of a plurality of channels of the medium-low resolution remote sensing sensor; Δ r_iAnd Δ r_jAre all in sufficiently small increments, e.g. 10^-3；Represents mathematically solving the partial derivative of the function; f represents an abstract remote sensing inversion model;

(3) and spatial heterogeneity estimation:

firstly, according to a classification diagram of a matched subsurface mat surface, aiming at each surface type, a layered random sampling mode is adopted, sampling points are randomly distributed, and a sample is obtained by depending on a ground observation experiment; the number of sampling points is determined according to standard deviation, sampling error and the prior knowledge of confidence coefficient; within each surface type, a variance function γ (h) is obtained by statistically comparing the differences of observed variables separated by a certain lag distance, calculating the lag distance h, and then observing the variability of the variables, i.e.:

$\mathrm{\γ}\left(h\right)=\frac{1}{2N_h}\underset{0}{\overset{N_h}{\Σ}}{(r(x+h)-r\left(x\right))}^2---\left(3\right)$

wherein N is_hIs the number of observation variable data pairs separated by a lag distance h; r (x) is an observed variable value;

in order to characterize the multi-scale properties of the data, the variogram is considered as a linear combination of a spherical model and an exponential model; through fitting of a theoretical variation function model, important parameters for describing the variation function are obtained: weights, base station values and variation of the spherical model and the exponential model; corresponding intra-class heterogeneityCan be expressed as:

${\mathrm{\ν}}_{i,j}^s=\frac{1}{|\mathrm{\ν}{|}^2}\underset{x\∈\mathrm{\ν}}{\∫}\underset{y\∈\mathrm{\ν}}{\∫}\mathrm{\γ}\left|\right|x-y\left|\right|dxdy---\left(4\right)$

wherein | v | is the area of a certain earth surface type within a pixel dimension s obtained according to an earth surface underlying surface classification diagram, x and y are coordinates of spatial positions in the pixel in the horizontal and vertical directions, and meanwhile, the inter-class heterogeneity directly utilizes the component proportion omega_dTo measure; the spatial heterogeneity of the whole pixel is jointly represented by intra-class heterogeneity and inter-class heterogeneity;

(4) and estimating the spatial scale effect:

by continuing to analyze the differences between high resolution data and medium low resolution data in the same region to express the same objective thing, combining with the mathematically simplified approximation and derivation of the spatial scale effect in prior art methods, the scale effectCan be expressed as:

${\mathrm{\Δ}}_{scale}^{New}=\frac{1}{2}\underset{d=1}{\overset{c}{\Σ}}{\mathrm{\ω}}_d\left(\underset{i=1}{\overset{n}{\Σ}}\underset{j=1}{\overset{n}{\Σ}}{k}_{d,i,j}{\mathrm{\ν}}_{d,i,j}^s\right)---\left(5\right)$

wherein k is_d，i，jIs the model second order partial derivative of the underlying surface d to channels i and j,the heterogeneity between the i channel and the j channel of the underlying surface d under the pixel dimension s; component ratio omega in the above formula_dExplains the influence of inter-class heterogeneity on the scale effect, k_d，i，jAndexplaining the influence of the model nonlinearity degree and the intra-class heterogeneity on the scale effect;

(5) and correcting the spatial scale effect:

by combining formulas (1) and (5), the spatial scale effect correction can be carried out on the remote sensing product obtained by medium-low resolution inversion by utilizing the component proportion in the pixel, and the pixel scale true value of the medium-low resolution remote sensing product is obtained

$f_{corr}^{New}=f\left(\stackrel{\‾}{r}\right)+{\mathrm{\Δ}}_{scale}^{New}\underset{d=1}{\overset{c}{\Σ}}{\mathrm{\ω}}_d\left(f_d\right(r)+\frac{1}{2}\underset{i=1}{\overset{n}{\Σ}}\underset{j=1}{\overset{R}{\Σ}}{k}_{d,i,j}{v}_{d,i,j}^s)---\left(6\right)$

When the remote sensing inversion model is discontinuous or non-conductive in the whole input parameter distribution interval, the input parameter distribution interval can be divided into a plurality of subintervals again according to the discontinuous and non-conductive positions of the remote sensing inversion model, so that the remote sensing inversion model is continuous and conductive in each subinterval, the proportion and the equivalent parameter value of each subinterval are estimated simultaneously, and the pixel scale truth value of the medium-low resolution remote sensing product is estimated by combining the formula (6); n is the number of channels.

The method gets rid of the constraint on the requirement of the matched high-resolution data in the process of acquiring the real value of the remote sensing product, breaks through the limitation that the remote sensing inversion model in the prior art is required to be continuously guidable, simultaneously considers the problem of scale effect caused by intra-class heterogeneity and inter-class heterogeneity, further improves the acquisition precision of the pixel scale real value of the medium-low resolution remote sensing product, and finally realizes the simple, convenient and accurate acquisition of the pixel scale real value of the medium-low resolution remote sensing product.

Drawings

The present invention will be described in further detail with reference to the accompanying drawings and specific embodiments.

FIG. 1 is a flow chart of the present invention.

Detailed Description

The method for acquiring the truth value of the medium-low resolution remote sensing product is still realized by combining remote sensing model inversion and scale effect correction. But for the correction of the scale effect, the surface space heterogeneity is estimated by using the variation function in the earth statistics in a mode of arranging the points in different table types by combining the classification maps of the underlying surface coverage types. And finally, combining the component proportions of different types of the surface, and correcting scale effects caused by intra-class and inter-class heterogeneity simultaneously to further obtain a pixel scale true value of the medium and low resolution remote sensing product. The general technical flow chart of the invention is shown in fig. 1, and comprises the following five steps: 1. producing a quantitative remote sensing product; 2. estimating a second derivative of the model; 3. estimating spatial heterogeneity; 4. estimating a spatial scale effect; 5. and correcting the spatial scale effect.

1. Production of quantitative remote sensing products:

the method comprises the following 5 steps: firstly, acquiring kilometer-level medium-low resolution remote sensing data r, wherein r is observed values of a plurality of channels of a medium-low resolution remote sensing sensor; second, the component proportion omega of different types of tables is estimated through the classification chart of the matched subsurface_d(ii) a Thirdly, selecting a proper remote sensing inversion model f according to the classification diagram of the subsurface bedding surface and aiming at different subsurface bedding surfaces d_d(usually, the types of subsurface bedding surfaces are different, and the forms or model coefficients of remote sensing inversion models are also different); fourthly, inputting the remote sensing data r into a corresponding remote sensing inversion model f_dIn the method, quantitative remote sensing products p corresponding to different subsurface pad surface components are directly obtained_dNamely: p is a radical of_d＝f_d(r); fifthly, acquiring a rough value of the medium-low resolution remote sensing product in an area weighting modeNamely:

$f\left(\stackrel{\‾}{r}\right)=\underset{d=1}{\overset{c}{\Σ}}{\mathrm{\ω}}_d{f}_d\left(r\right)---\left(1\right)$

wherein c is the total number of different underlying surface types of the ground surface; omega_dComponent proportions for different table types; f. of_dAnd (r) represents the corresponding quantitative remote sensing product.

2. Estimating a model second derivative:

the remote sensing inversion function is taken as a black box, and the second-order derivative and the second-order partial derivative k of the model are estimated in a discrete solving mode_i，jNamely:

$k_{i,j}=\frac{{\∂}^{2}f}{\∂r_i\∂r_j}=\frac{\frac{\∂f}{\∂r_i}(r+{\mathrm{\Δr}}_j)-\frac{\∂f}{\∂r_i}(r-{\mathrm{\Δr}}_j)}{2{\mathrm{\Δr}}_j}---\left(2\right)$

wherein,is a first derivative of the model and can be expressed asr is observed values of a plurality of channels of the medium-low resolution remote sensing sensor; Δ r_iAnd Δ r_jAre all in sufficiently small increments, e.g. 10^-3；Represents mathematically solving the partial derivative of the function; f represents an abstract remote sensing inversion model;

3. spatial heterogeneity estimation:

firstly, according to a classification diagram of a matched subsurface mat surface, aiming at each surface type, a layered random sampling mode is adopted, sampling points are randomly distributed, and a sample is obtained by depending on a ground observation experiment. The number of sampling points is determined according to the standard deviation, sampling error and the prior knowledge of confidence coefficient. Within each surface type, a variance function γ (h) is obtained by statistically comparing the differences of observed variables separated by a certain lag distance, calculating the lag distance h, and then observing the variability of the variables, i.e.:

$\mathrm{\γ}\left(h\right)=\frac{1}{2N_h}\underset{0}{\overset{N_h}{\Σ}}{(r(x+h)-r\left(x\right))}^2---\left(3\right)$

wherein N is_hIs the number of observation variable data pairs separated by a lag distance h; r (x) is an observed variable value; to characterize the multi-scale properties of the data, the variogram is considered to be a linear combination of a spherical model and an exponential model. Through fitting of a theoretical variation function model, important parameters for describing the variation function are obtained: weights, base station values, and variables for the spherical model and the exponential model. Corresponding intra-class heterogeneityCan be expressed as:

${\mathrm{\ν}}_{i,j}^s=\frac{1}{|\mathrm{\ν}{|}^2}\underset{x\∈\mathrm{\ν}}{\∫}\underset{y\∈\mathrm{\ν}}{\∫}\mathrm{\γ}\left|\right|x-y\left|\right|dxdy---\left(4\right)$

wherein | v | is the area of a certain earth surface type within a pixel dimension s obtained according to an earth surface underlying surface classification diagram, x and y are coordinates of spatial positions in the pixel in the horizontal and vertical directions, and meanwhile, the inter-class heterogeneity directly utilizes the component proportion omega_dTo measure. The spatial heterogeneity of the whole pixel is collectively represented by intra-class heterogeneity and inter-class heterogeneity.

4. Estimating the spatial scale effect:

by continuing to analyze the differences between high resolution data and medium low resolution data in the same region to express the same objective thing, combining with the mathematically simplified approximation and derivation of the spatial scale effect in prior art methods, the scale effectCan be expressed as:

${\mathrm{\Δ}}_{scale}^{New}=\frac{1}{2}\underset{d=1}{\overset{c}{\Σ}}{\mathrm{\ω}}_d\left(\underset{i=1}{\overset{n}{\Σ}}\underset{j=1}{\overset{n}{\Σ}}{k}_{d,i,j}{\mathrm{\ν}}_{d,i,j}^s\right)---\left(5\right)$

wherein k is_d，i，jIs the model second order partial derivative of the underlying surface d to channels i and j,is the inter-class heterogeneity of i and j channels at pixel scale s for underlying surface d. Component ratio omega in the above formula_dExplains the influence of inter-class heterogeneity on the scale effect, k_d，i，jAndexplains the influence of model nonlinearity degree and intra-class heterogeneity on scale effect.

5. And (3) correcting the spatial scale effect:

by combining formulas (1) and (5), the spatial scale effect correction can be carried out on the remote sensing product obtained by medium-low resolution inversion by utilizing the component proportion in the pixel, and the pixel scale true value of the medium-low resolution remote sensing product is obtained

$f_{corr}^{New}=f\left(\stackrel{\‾}{r}\right)+{\mathrm{\Δ}}_{scale}^{New}\underset{d=1}{\overset{c}{\Σ}}{\mathrm{\ω}}_d\left(f_d\right(r)+\frac{1}{2}\underset{i=1}{\overset{n}{\Σ}}\underset{j=1}{\overset{R}{\Σ}}{k}_{d,i,j}{v}_{d,i,j}^s)---\left(6\right)$

When the remote sensing inversion model is discontinuous or non-conductive in the whole input parameter distribution interval, the input parameter distribution interval can be divided into a plurality of subintervals again according to the discontinuous and non-conductive positions of the remote sensing inversion model, so that the remote sensing inversion model is continuous and conductive in each subinterval, the proportion and the equivalent parameter value of each subinterval are estimated simultaneously, and the pixel dimension true value of the medium-low resolution remote sensing product is estimated by combining the formula (6). n is the number of channels. The same symbols in the above formulas of the present invention represent the same meanings.

The above embodiments are not intended to limit the present invention, and the present invention is not limited to the above examples, and those skilled in the art may make variations, modifications, additions or substitutions within the technical scope of the present invention.

Claims (1)

1. A method for acquiring a true value of a medium-low resolution remote sensing product for correcting scale effect is characterized by comprising the following steps: the acquisition method comprises the following steps:

(1) and producing quantitative remote sensing products:

the production of the quantitative remote sensing product is specifically divided into 5 steps: firstly, acquiring kilometer-level medium-low resolution remote sensing data r, wherein r is observed values of a plurality of channels of a medium-low resolution remote sensing sensor; second, the component proportion omega of different types of tables is estimated through the classification chart of the matched subsurface_d(ii) a Third, according to the subsurface bedding surfaceSelecting a proper remote sensing inversion model f according to different subsurface bedding surfaces d_dGenerally, the types of subsurface bedding surfaces are different, and the forms or model coefficients of remote sensing inversion models are also different; fourthly, inputting the remote sensing data r into a corresponding remote sensing inversion model f_dIn the method, quantitative remote sensing products p corresponding to different subsurface pad surface components are directly obtained_dNamely: p is a radical of_d＝f_d(r); fifthly, acquiring a rough value of the medium-low resolution remote sensing product in an area weighting modeNamely:

$f\left(\stackrel{\‾}{r}\right)=\underset{d=1}{\overset{c}{\Σ}}{\mathrm{\ω}}_d{f}_d\left(r\right)---\left(1\right)$

wherein c is the total number of different underlying surface types of the ground surface; omega_dComponent proportions for different table types; f. of_d(r) represents the corresponding quantitative remote sensing product;

(2) estimating a model second derivative:

the remote sensing inversion function is taken as a black box, and the second-order derivative and the second-order partial derivative k of the model are estimated in a discrete solving mode_i，jNamely:

$k_{i,j}=\frac{{\∂}^{2}f}{\∂r_i\∂r_j}=\frac{\frac{\∂f}{\∂r_i}(r+{\mathrm{\Δr}}_j)-\frac{\∂f}{\∂r_i}(r-{\mathrm{\Δr}}_j)}{2{\mathrm{\Δr}}_j}---\left(2\right)$

wherein,is a first derivative of the model and can be expressed as $\frac{\∂f}{\∂r_i}=\left(f\right(r+{\mathrm{\Δr}}_i)-f(r-{\mathrm{\Δr}}_i\left)\right)/\left(2{\mathrm{\Δr}}_i\right);$ r is observed values of a plurality of channels of the medium-low resolution remote sensing sensor; Δ r_iAnd Δ r_jAre all in sufficiently small increments, e.g. 10^-3；Represents mathematically solving the partial derivative of the function; f represents an abstract remote sensing inversion model;

(3) and spatial heterogeneity estimation:

firstly, according to a classification diagram of a matched subsurface mat surface, aiming at each surface type, a layered random sampling mode is adopted, sampling points are randomly distributed, and a sample is obtained by depending on a ground observation experiment; the number of sampling points is determined according to standard deviation, sampling error and the prior knowledge of confidence coefficient; within each surface type, a variance function γ (h) is obtained by statistically comparing the differences of observed variables separated by a certain lag distance, calculating the lag distance h, and then observing the variability of the variables, i.e.:

$\mathrm{\γ}\left(h\right)=\frac{1}{2N_h}\underset{0}{\overset{N_h}{\Σ}}{\left(r\right(x+h)-r(x\left)\right)}^2---\left(3\right)$

wherein N is_hIs the number of observation variable data pairs separated by a lag distance h; r (x) is an observed variable value;

in order to characterize the multi-scale properties of the data, the variogram is considered as a linear combination of a spherical model and an exponential model; through fitting of a theoretical variation function model, important parameters for describing the variation function are obtained: weights, base station values and variation of the spherical model and the exponential model; corresponding intra-class heterogeneityCan be expressed as:

$v_{i,j}^s=\frac{1}{|v{|}^2}\underset{x\∈v}{\∫}\underset{y\∈v}{\∫}\mathrm{\γ}\left|\right|x-y\left|\right|dxdy---\left(4\right)$

wherein | v | is the area of a certain earth surface type within a pixel dimension s obtained according to an earth surface underlying surface classification diagram, x and y are coordinates of spatial positions in the pixel in the horizontal and vertical directions, and meanwhile, the inter-class heterogeneity directly utilizes the component proportion omega_dTo measure; the spatial heterogeneity of the whole pixel is jointly represented by intra-class heterogeneity and inter-class heterogeneity;

(4) and estimating the spatial scale effect:

by continuing to analyze the differences between high resolution data and medium low resolution data in the same region to express the same objective thing, combining with the mathematically simplified approximation and derivation of the spatial scale effect in prior art methods, the scale effectCan be expressed as:

${\mathrm{\Δ}}_{scale}^{New}=\frac{1}{2}\underset{d=1}{\overset{c}{\Σ}}{\mathrm{\ω}}_d\left(\underset{i=1}{\overset{n}{\Σ}}\underset{j=1}{\overset{n}{\Σ}}{k}_{d,i,j}{v}_{d,i,j}^s\right)---\left(5\right)$

wherein k is_d，i，jIs the model second order partial derivative of the underlying surface d to channels i and j,the heterogeneity between the i channel and the j channel of the underlying surface d under the pixel dimension s; component ratio omega in the above formula_dExplains the influence of inter-class heterogeneity on the scale effect, k_d，i，jAndexplaining the influence of the model nonlinearity degree and the intra-class heterogeneity on the scale effect;

(5) and correcting the spatial scale effect:

by combining formulas (1) and (5), the spatial scale effect correction can be carried out on the remote sensing product obtained by medium-low resolution inversion by utilizing the component proportion in the pixel, and the pixel scale true value of the medium-low resolution remote sensing product is obtained

$f_{corr}^{New}=f\left(\stackrel{\‾}{r}\right)+{\mathrm{\Δ}}_{scale}^{New}=\underset{d=1}{\overset{c}{\Σ}}{\mathrm{\ω}}_d\left(f_d\right(r)+\frac{1}{2}\underset{i=1}{\overset{n}{\Σ}}\underset{j=1}{\overset{n}{\Σ}}{k}_{d,i,j}{v}_{d,i,j}^s)---\left(6\right)$

When the remote sensing inversion model is discontinuous or non-conductive in the whole input parameter distribution interval, the input parameter distribution interval can be divided into a plurality of subintervals again according to the discontinuous and non-conductive positions of the remote sensing inversion model, so that the remote sensing inversion model is continuous and conductive in each subinterval, the proportion and the equivalent parameter value of each subinterval are estimated simultaneously, and the pixel scale truth value of the medium-low resolution remote sensing product is estimated by combining the formula (6); n is the number of channels.