CN103345707A - Crop maturation stage remote sensing prediction method based on multi-source remote sensing data - Google Patents

Abstract

The invention provides a crop maturation stage remote sensing prediction method based on multi-source remote sensing data. The method comprises the steps that high temporal-spatial resolution remote sensing data are generated through the multi-source remote sensing data in a blended mode so that requirements for cropland dimension carrying-out dynamic monitoring of crop maturation stage prediction can be met; then, according to the dynamic change law of moisture content and chlorophyll content of different parts of maturation stage plants of crops, digital expression of the change law is formed. On the basis of the coupling method of the change law of biochemical parameters of the crops and biochemical parameters of the crops of remote sensing inversion, the moisture content and the chlorophyll content, obtained in a remote sensing mode, of crop canopies are combined to the change law of the biochemical parameters of maturation near stages of the crops, and the crop maturation stage remote sensing prediction method is formed. By means of remote sensing prediction of maturation stages of the crops, the application range of remote sensing in the accurate agricultural field is further extended, and the newly added technique of remote sensing application in the agricultural field is excavated.

Description

A kind of crops maturity stage remote sensing prediction method based on RS data

Technical field

The present invention relates to technical field of remote sensing image processing, particularly a kind of crops maturity stage remote sensing prediction method based on RS data.

Background technology

Precision agriculture is based on the space parallax opposite sex of field-crop and environment, obtain the agricultural land information of different units in the farmland by various technological means, and utilize the variable technology to carry out the farmland optimum management thus, realize that production run becomes more meticulous, the agriculture management management system of accuracy.Precisely farming is the inevitable outcome of modern agricultural development, its objective is and boosts productivity, and optimizes harvest, the protection environment.Optimizing crop harvesting is an important step of precision agriculture; output, the quality of crop harvesting time to crop has significant effects; reasonably predict the crop harvesting time; help to improve the quality and yield of agricultural product; simultaneously can also instruct agricultural machinery reasonably to dispatch arrangement, this mechanized harvest to scale crop-planting zone is significant.Stone roller is made in harvesting in good time, avoids the adverse weather influence, is the key link in the agricultural production, and gathering in the crops too early or spending evening all can influence output, is unfavorable for that high yield increases income.Be example with the corn, gather in the crops in advance and will cause per unit area yield to reduce more than 10% in 10 days, and that for soybean ripe untimely results in back and exposing to the sun of causing split also can cause the production loss more than 15%.

Maturity stage refers to whole naturally time limit of giving birth to from generation to generation of crop, and on the agronomy meaning, maturation refers to a stage of crop growth, refers to that the fruit of crop grows to the stage that can gather in the crops.Traditional judgement to crop maturity stage and optimal harvest time mainly is that the crop features such as color, structure and canopy structure according to seed or blade are carried out subjective decipher; this method is difficult in widespread adoption; easily introduce the error of subjective judgement; this method can only be used for on-the-spot judgement simultaneously, does not possess the ability of prediction.At these problems, develop gradually and some crop maturity stage Forecasting Methodology and models, wherein the topmost crop maturity stage Forecasting Methodology that is based on meteorological statistics and based on the crop maturity stage Forecasting Methodology of crop growth model.

Forecasting Methodology supposition meteorological condition based on meteorological statistics is the main factor that influences process of crop growth and cause the variation of crop maturity stage, therefore can utilize the meteorological condition of crop different phenological or the time of occurrence in specific phenological period to carry out the prediction in crop maturity stage then.The model that these class methods develop mostly is simple and easy to usefulness, and only uses less driving data (as temperature, precipitation etc.) just can carry out prediction.But the weather data that with temperature is representative shows homogeneous in the larger context, can't reflect the field difference in piece yardstick maturity stage, thereby can't in small range, predict the maturation time of different fields piece and formulate the harvesting order of optimizing, simultaneously these models are based on that the mode of statistical regression sets up more, only be applicable to specific zone, the model of setting up in a zone can't be promoted the crop of other zones or other kinds.

Be the process that crop growth model can be described crop growth and output formation from the growth mechanism that the crop photosynthesis effect drives based on the theoretical foundation of the Forecasting Methodology of crop growth model, utilize crop growth model to turn to target structure cost function with the optimum of crop yield or quality (or both are comprehensive), oppositely the crop harvesting time of solving-optimizing, realize the prediction in crop maturity stage.The major part of growth model own is single-point (site specific) model, also there is not the not pervasive fully crop modeling that is used for the large scale simulation, this makes growth model before zones of different is used, and all needs to make its localization by the demarcation to model parameter.And that crop growth model itself relates to parameter is numerous, and the workload of Biao Dinging is huge one by one, and this has limited its applying in crop maturity stage prediction field to a certain extent.When utilizing crop growth model to carry out crop growth simulation in a big way, information such as required soil types, crop varieties all are difficult to obtain in addition, have brought certain difficulty also for the crop maturity stage prediction that utilizes crop growth model to carry out in a big way.

At present; the agricultural production pattern that has formed scale has been developed in Northeast China and northwest; further intensification along with " soil circulation "; other regional traditional small family's agriculturals of China also will change to scale and mechanized agriculture by subcontracting with form such as cooperative society; and the prediction in crop maturity stage and monitoring, the new demand that the enforcement of accurate mechanized farming proposes remote sensing technology under the scale agricultural just.And both at home and abroad also rare scholar utilize satellite remote sensing technology, carry out the crop degree of ripeness information of obtaining, optimize the research of harvesting order.

Summary of the invention

(1) technical matters that will solve

The problem to be solved in the present invention provides a kind of crops maturity stage remote sensing prediction method based on RS data, utilize remote sensing technology, in on a large scale the maturity stage of crops is carried out objective, high-precision prediction, and this Forecasting Methodology can be applied in zones of different.

(2) technical scheme

In order to address the above problem, the invention provides a kind of crops maturity stage remote sensing prediction method based on RS data, it is characterized in that this method may further comprise the steps:

(1) remotely-sensed data of ground spectral reflectance is carried out pre-service, these remotely-sensed datas derive from a plurality of satellite sensors;

(2) pretreated RS data is carried out data fusion;

(3) based on ground observation data extract crops maturity stage biochemical parameter Changing Pattern;

(4) according to Changing Pattern and the remotely-sensed data inverting crops biochemical parameter of the crops biochemical parameter that extracts;

(5) the crops biochemical parameter prediction crops maturity stage that obtains according to inverting.

Wherein, remotely-sensed data comprises MODIS data, MERSI data, HJ-1CCD data and HJ-1IRS data.The pretreated content of these data comprises radiation calibration, geometric correction and atmosphere correction.

Wherein, geometric correction adopts the quadratic polynomial method.

Wherein, the RS data fusion comprises that the space-time dimension data merges and the spectrum dimension data merges.The space-time dimension data merges employing space-time adaptability reflectivity Fusion Model, and described spectrum dimension data merges the employing wavelet transform fusion.

Wherein, wavelet transform fusion may further comprise the steps:

(1) image is carried out wavelet transformation, be about on the different characteristic territory of picture breakdown under the different frequency;

(2) low frequency component and the high fdrequency component after will decomposing merges by certain fusion rule.Fusion rule comprises high frequency fusion rule and low frequency fusion rule, and described high frequency fusion rule comprises method of substitution, method of weighted mean and based on the maximum value method, and described low frequency fusion rule comprises method of substitution, method of weighted mean and based on the edge reservation method.

Wherein, crops maturity stage biochemical parameter Changing Pattern adopts least square fitting to obtain.Biochemical parameter comprises canopy leaf water content and chlorophyll content.Canopy leaf water content is monitored by reflectance varies or the NDWI of short-wave infrared wave band, and chlorophyll content is monitored by reflectivity or the NDVI of near-infrared band.

Wherein, the remote-sensing inversion of biochemical parameter adopts the method for statistical model or the method for mechanism model.

Its Forecasting Methodology is: at the picture dot yardstick crop water of remote sensing appraising and the crop maturity stage after the chlorophyll content and functionization are closed on stage biochemical parameter Changing Pattern and be coupled, determine the time gap apart from the maturity stage, thereby realize the estimation in crop maturity stage.

(3) beneficial effect

The present invention is predicted as purpose with the crop maturity stage, to improving crop yield, quality and rationally arranging harvesting (especially under extensive mechanization condition) that important practical usage is arranged.

The method that the present invention develops has taken full advantage of the characteristics that remote sensing technology can reflect the farmland special heterogeneity, has higher spatial resolution, not only can reflect the maturity stage difference between field piece and the field piece, even can reflect the crop maturity stage difference of piece inside, same field.

The present invention is merged generation high-spatial and temporal resolution remotely-sensed data by RS data, both satisfied the demand that the prediction of crop maturity stage is carried out dynamic monitoring at the farmland yardstick, source data is free data simultaneously, has reduced the cost of using again, can ensure applying of invention.

Description of drawings

Fig. 1 is the crops maturity stage remote sensing prediction process flow diagram based on RS data;

Fig. 2 is remotely-sensed data figure;

Fig. 3 is small echo decomposing schematic representation successively;

Fig. 4 is canopy moisture Changing Pattern figure;

Fig. 5 is chlorophyll Changing Pattern figure;

Fig. 6 is the maturity stage figure that predicts the outcome.

Embodiment

Below in conjunction with drawings and Examples embodiments of the present invention are described in further detail.Following examples are used for explanation the present invention, but can not be used for limiting the scope of the invention.

As shown in Figure 1, the crops maturity stage remote sensing prediction method based on RS data may further comprise the steps:

1, the RS data of ground spectral reflectance is carried out pre-service

According to crop phenological period and combined ground observation data, in the ripe preceding remotely-sensed data that began to obtain monitoring section in 30 days of monitored area crop, as shown in Figure 2, the remotely-sensed data of obtaining comprises 4 classes, derive from Moderate Imaging Spectroradiomete (moderate-resolution imaging spectro-radiometer respectively, be called for short MODIS), intermediate resolution imaging frequency spectrograph (Medium Resolution Spectral Imager, be called for short MERSI), environment and disaster monitoring forecast moonlet constellation (are called for short " No. one, environment ", code name HJ-1) CCD camera and HJ-1 infrared camera (infrared scanner, be called for short IRS), these data all are the form storages with three-dimensional matrice, comprise the two dimension (X on the space, Y) and spectrum dimension.Wherein MODIS data and MERSI data are that frequency is obtained with the sky, and HJ-1CCD and HJ-1IRS are then according to as much as possible the obtaining of the situation of obtaining of data.

Select quality better the data of (do not have or only have a small amount of cloud to cover) as data source data are carried out pre-service, the pretreated content of data comprises operations such as radiation calibration, geometric exact correction, atmosphere correction.

(1) data radiation calibration

The radiation calibration of data is undertaken by following formula:

L=DN/g+L ₀ （1）

${\mathrm{\ρ}}_{\mathrm{TOA}}=\frac{\mathrm{\π}\·d^2\·L}{E_s\·\cos {\mathrm{\θ}}_s}---\left(2\right)$

Wherein, L is radiancy; DN is the digital signal value of all kinds of remotely-sensed datas; G and L ₀Be calibrating parameters, metadata that can be from the meta data file (as MODIS HDF(Hierarchical Data Format, a kind of data layout) of different data and the XML file in the environment sing data) in obtain; ρ _TOABe apparent reflectance (being the spectral reflectivity of Top Of Atmosphere); D is the solar distance correction factor; E _sBe the radiation flux of the sun on the atmospheric envelope top; θ _sBe sun altitude.

(2) geometric correction

The purpose of geometric correction is that remote sensing image data is projected on the plane, makes it meet the system of map projection.The geometric correction of data adopts the quadratic polynomial method to carry out in the present embodiment, use target area high-resolution remote sensing image or large scale (1:10 is more than ten thousand) topomap earlier, HJ-1CCD is carried out geometric correction, the HJ-1CCD data data (MODIS, MERSI and HJ-1IRS) relatively low to other three kinds of resolution that re-use after the correction are carried out geometric correction, during correction with error control in 0.5 pixel.

Geometric correction adopts the mode of polynomial expression registration, is shown below:

x _warp=a ₀×x _ref+b ₀×y _ref+c ₀ （3）

y _warp=a ₁×x _ref+b ₁×y _ref+c ₁

Wherein, x _Warp, y _WarpBe to wait to correct on the remote sensing image a bit; x _Ref, y _RefBe with reference to same place on the image; A, b, c are the registration coefficient, and the least square fitting by a plurality of corresponding point obtains.

(3) atmosphere is corrected

The various radiation energy that remote sensing utilizes all will interact with earth atmosphere (or scattering or absorption), and make energy attenuation, and spectral distribution is changed.Eliminate the processing procedure of atmosphere image, be called atmosphere and correct.Atmosphere correction main algorithm is as follows:

Wherein, ρ ₀Be the atmospheric path radiation reflectivity; ρ _sBe the earth surface reflection rate; T is transmitance; u _s, u _vBe respectively solar zenith angle cosine value and satellite zenith angle cosine value; S is the downward hemispherical reflectance of surface air.Wherein the atmospheric condition parameter in the input parameter can be obtained from MODIS data product MOD04.

2, pretreated RS data is merged

Be subjected to the restriction of technical conditions, when using remotely-sensed data, have to accept or reject in time and space resolution.The remotely-sensed data of high spatial resolution only covers less spatial dimension usually, causes its heavy visit cycle long; And the sensor of high time resolution carries out large-scale repeated accesses with the short heavily visit cycle, and spatial resolution is lower.One of solution to this problem merges the data that have different spatial and temporal resolution features on the different sensors exactly, generates to have the remotely-sensed data of high time and high spatial resolution simultaneously.

(1) the space-time dimension data merges

In the present embodiment, adopt space-time adaptability reflectivity Fusion Model, carry out the prediction of earth surface reflection rate in conjunction with the MODIS data of high time resolution in the spatial resolution of HJ-1CCD, this method is at first obtained same time t ₁MODIS and HJ-1CCD image, by calculating the difference of space distribution between image, in conjunction with another time t ₂The MODIS data carry out the prediction of corresponding time HJ-1CCD image.In forecasting process, use the method for moving window to reduce the influence on low resolution remotely-sensed data pixel border, when using moving window to carry out the calculating of center pixel value, space length, spectrum intervals and time interval from as weight.

Prediction algorithm can be described with following formula:

$L(x_{w/2},y_{w/2},t_2)=\underset{i=1}{\overset{w}{\mathrm{\Σ}}}\underset{j=1}{\overset{w}{\mathrm{\Σ}}}\underset{k=1}{\overset{n}{\mathrm{\Σ}}}{W}_{\mathrm{ijk}}\×(M(x_i,y_j,t_2)+L(x_i,y_j,t_1)-M(x_i,y_j,t_1))---\left(5\right)$

In the formula, L (x _W/2, y _W/2, T ₂) be the t of prediction ₂HJ-1CCD pixel value constantly; W is the size of moving window, only uses effective pixel to predict in the window; (x _W/2, y _W/2) be the center pixel of window; M (x _i, y _i, t ₂) be the window's position (x _i, y _i) be in t ₂The pixel value of moment MODIS; L (x _i, y _i, t ₁) be that HJ-1CCD is at t ₁Corresponding pixel value constantly; M (x _i, y _i, t ₁) be that MODIS is at t ₁Corresponding pixel value constantly; W _IjkBe the weight of each pixel in the window when the pre-measured center pixel.

Weighting function in the algorithm can calculate with following formula:

$W_{\mathrm{ijk}}=1/C_{\mathrm{ijk}}/\underset{i=1}{\overset{w}{\mathrm{\Σ}}}\underset{j=1}{\overset{w}{\mathrm{\Σ}}}\underset{k=1}{\overset{n}{\mathrm{\Σ}}}(1/C_{\mathrm{ijk}})---\left(6\right)$

In the formula, C _IjkIt is the product that spectrum intervals, time gap and space length according to other pixels (comprising the multidate data) in the prediction pixel of window center and the window calculate.

(2) the spectrum dimension data merges

Adopt in the present embodiment based on method of wavelet and carry out fusion between HJ-1CCD and the HJ-1IRS data.

At first, a width of cloth image is carried out wavelet transformation, be about to it and decompose on the different characteristic territory under the different frequency.As shown in Figure 3, for image C ₀, can it be decomposed into C with wavelet algorithm ₁, D ₁ ¹, D ₁ ², D ₁ ³Four components, wherein C ₁Shown C ₀Low frequency component, i.e. the low frequency part of image; D ₁ ¹The high fdrequency component that shows vertical direction, the i.e. horizontal edge of image; D ₁ ²Show C ₀The high fdrequency component of horizontal direction, the i.e. vertical edge of image; D ₁ ³The high fdrequency component that shows both direction, the i.e. edge, diagonal angle of image.And C ₁Can decompose again by wavelet transformation, and the like, can be with C ₀Successively decompose any n layer.Finally obtain C _n, D _n ¹, D _n ², D _n ³, D _N-1 ¹, D _N-1 ², D _N-1 ³..., D ₁ ¹, D ₁ ², D ₁ ³

Secondly, low frequency component and high fdrequency component after the decomposition of two images are merged by certain fusion rule:

■ high frequency fusion rule

In the wavelet decomposition of piece image, the bigger wavelet coefficient of absolute value changes big features such as edge corresponding to contrast in the image, and human eye is relatively more responsive for these features.So, for the high-frequency territory we based on the selective rule of maximum value.For image X, can be a variable S who weighs its conspicuousness of wavelet coefficient definition in its high-frequency territory:

$S_j^{\mathrm{\ϵ}}(X,p)=\underset{q\∈Q}{\max }\left(\right|D_j^{\mathrm{\ϵ}}(X,q)\left|\right)---\left(7\right)$

Wherein, j represents the number of plies of wavelet coefficient; ε=1,2, the sequence number of 3 expression frequency bands; P=(m, n) locus of expression wavelet coefficient; Q represents one 3 * 3 square window centered by p; Q be in the window more arbitrarily.

For the same definable S of wavelet coefficient corresponding among the image Y ^ε _j(Y, p).In order in last fused images, to keep in two width of cloth original images notable attribute, we in the wavelet coefficient of two width of cloth original images, select the bigger wavelet coefficient of S value as composograph in the wavelet coefficient of correspondence position.If use M* ^ε _j(X, p), M* ^ε _j(Y p) represents the value of the decision table on two width of cloth image relevant positions respectively, and above-mentioned thought just can be expressed as with mathematical formulae:

A kind of good image interfusion method point in zone of reply when selecting wavelet coefficient is taked identical selection scheme, so the decision table that reply obtains carries out consistency checking. and here, we adopt the majority voting principle. and the value that the back decision table is revised in order is M ^εJ (X, p), then:

$M_j^{\mathrm{\ϵ}}(Y,p)=1-M_j^{\mathrm{\ϵ}}(X,p)---\left(11\right)$

After obtaining the value of each point in the decision table, just can calculate the high frequency wavelet coefficient of fused images Z:

$D_j^{\mathrm{\ϵ}}(Z,p)=M_j^{\mathrm{\ϵ}}(X,p)\×D_j^{\mathrm{\ϵ}}(X,p)+M_j^{\mathrm{\ϵ}}(Y,p)\×D_j^{\mathrm{\ϵ}}(Y,p)---\left(12\right)$

■ low frequency fusion rule

For variable E of the low frequency coefficient of image X definition (X, p):

E(X,p)=(F ₁*C _j) ²(X,p)+(F ₂*C _j) ²(X,p)+(F ₃*C _j) ²(X,p)

$F_1=\left[\begin{array}{ccc}-1& -1& -1\\ 2& 2& 2\\ -1& -1& -1\end{array}\right],F_2=\left[\begin{array}{ccc}-1& 2& -1\\ -1& 2& -1\\ -1& 2& -1\end{array}\right],F_3=\left[\begin{array}{ccc}-1& 0& -1\\ 0& 4& 0\\ -1& 0& -1\end{array}\right]---\left(13\right)$

Wherein * represents convolution.Equally, for image Y, can defining variable E (Y, p).Variable E has reflected that to a certain extent image is in the marginal information of level, vertical and diagonal.Therefore in order to keep the details in the original image preferably, can calculate variable E to the low frequency coefficient of two width of cloth images, and select the bigger low frequency coefficient of E as the low frequency coefficient of composograph, so just can in fused images, farthest keep the marginal information of original image.Merging function is expressed as follows:

C _j(Z,p)=W(X,p)×C _j(X,p)+W(Y,p)×C _j(Y,p)

At last, the component after merging is carried out wavelet inverse transformation and can obtain merging the back image.

3, based on ground observation data extract crops maturity stage biochemical parameter Changing Pattern

Carrying out canopy leaf water in the crop maturation and the measurement of chlorophyll content the ripe previous moon of crop, survey frequency can be sky or per 2 days, wherein chlorophyllous measurement uses chlorophyll meter to measure, leaf water content then is to claim its weight in wet base in blade sampling back, claim its dry weight after the oven dry again, calculate water cut=(weight in wet base-dry weight)/weight in wet base then.

Utilize common mathematical function (as linear function, logarithmic function, exponential function, 2 order polynomials etc.) that match is carried out in the variation of chlorophyll and moisture, as Fig. 4 and Fig. 5, make up these two parameters at the function representation of crop maturity stage Changing Pattern.Adopt least square method to determine the functional form of match in the fit procedure.

Because linear function is the special case of polynomial function, and logarithmic function and exponential function all can turn to linear function analysis, so present embodiment is the principle that example illustrates least square fitting with the polynomial function.

Suppose given data point (x _i, y _i) (i=0,1 ..., m), (function class that the polynomial expression of n≤m) constitutes now asks one to Φ for all number of times are no more than n

Make following formula obtain minimum value:

$I=\underset{i=0}{\overset{m}{\mathrm{\Σ}}}{[p_n\left(x_i\right)-y_i]}^2=\underset{i=0}{\overset{n}{\mathrm{\Σ}}}{(\underset{k=0}{\overset{n}{\mathrm{\Σ}}}{a}_k{x}_i^k-y_i)}^2---\left(15\right)$

Obviously,

$I=\underset{i=0}{\overset{m}{\mathrm{\Σ}}}{(\underset{k=0}{\overset{n}{\mathrm{\Σ}}}{a}_k{x}_i^k-y_i)}^2$

Be a ₀, a ₁..., a _nThe multivariate function, so the problems referred to above are and ask I=I (a ₀, a ₁..., a _n) extreme-value problem.Ask the necessary condition of extreme value by the multivariate function:

$\frac{\∂I}{{\∂a}_j}=2\underset{i=0}{\overset{m}{\mathrm{\Σ}}}(\underset{k=0}{\overset{n}{\mathrm{\Σ}}}{a}_k{x}_i^k-y_i)x_i^j=0,$ j=0,1,…,n （16）

That is:

$\underset{k=0}{\overset{n}{\mathrm{\Σ}}}\left(\underset{i=0}{\overset{m}{\mathrm{\Σ}}}{x}_i^{j+k}\right)a_k=\underset{i=0}{\overset{m}{\mathrm{\Σ}}}{x}_i^j{y}_i$ j=0,1,…,n （17）

Wherein, (17) are about a ₀, a ₁..., a _nSystem of linear equations, be expressed in matrix as:

$\left[\begin{array}{cccc}m+1& \underset{i=0}{\overset{m}{\mathrm{\Σ}}}{x}_i& \·\·\·& \underset{i=0}{\overset{m}{\mathrm{\Σ}}}{x}_i^n\\ \underset{i=0}{\overset{m}{\mathrm{\Σ}}}{x}_i& \underset{i=0}{\overset{m}{\mathrm{\Σ}}}{x}_i^2& \·\·\·& \underset{i=0}{\overset{m}{\mathrm{\Σ}}}{x}_i^{n+1}\\ \·& \·& & \·\\ \·& \·& & \·\\ \·& \·& & \·\\ \underset{i=0}{\overset{m}{\mathrm{\Σ}}}{x}_i^n& \underset{i=0}{\overset{m}{\mathrm{\Σ}}}{x}_i^{n+1}& \·\·\·& \underset{i=0}{\overset{m}{\mathrm{\Σ}}}{x}_i^{2n}\end{array}\right]\left[\begin{array}{c}{a}_0\\ a_1\\ \·\\ \·\\ \·\\ a_n\end{array}\right]\left[\begin{array}{c}\underset{i=0}{\overset{m}{\mathrm{\Σ}}}{y}_i\\ \underset{i=0}{\overset{m}{\mathrm{\Σ}}}{x}_i{y}_i\\ \·\\ \·\\ \·\\ \underset{i=0}{\overset{m}{\mathrm{\Σ}}}{x}_i^n{y}_i\end{array}\right]---\left(18\right)$

Formula (17) or formula (18) are called normal equations group or normal equation group.

Can prove that the matrix of coefficients of system of equations (18) is a symmetric positive definite matrix, so there is unique solution.From formula (18), solve a _k(k=0,1 ..., n), thereby can get polynomial expression:

$p_n\left(x\right)=\underset{k=0}{\overset{n}{\mathrm{\Σ}}}{a}_k{x}^k---\left(19\right)$

Can prove the p in the formula (19) _n(x) satisfy formula (15), i.e. p _n(x) be the polynomial fitting of asking.

4, according to Changing Pattern and the remotely-sensed data inverting crops biochemical parameter of the crops biochemical parameter that extracts

At first calculate some indexes, use these indexes and reflectivity information to carry out the estimation of crop canopies leaf water and chlorophyll content then, in the estimation process, provide statistical model and mechanism model two kinds of methods simultaneously.Wherein statistical model is simple and easy to use, and only use less driving data (as single remote sensing index) just can carry out estimation and acquisition degree of precision, but this class model does not have theoretical foundation, and the model of setting up in a zone can't be promoted the crop of other zones or other kinds; Mechanism model has theoretical foundation preferably, but the model calculation complexity, the time that needs when calculating based on picture dot is longer, also needs more parameter as input simultaneously.

(1) index calculates

Because crop is when the maturity stage closes on, canopy moisture and chlorophyll have the clear regularity variation, studies show that in a large number simultaneously, the chlorophyllous variation of crop can be by reflectivity or normalized differential vegetation index (the Normalized Difference Vegetation Index of near-infrared band, be called for short NDVI) monitor, and the variation of canopy moisture can be monitored by the reflectance varies of short-wave infrared wave band or with its constructed normalization aqua index (Normal Differential Water Index is called for short NDWI).Present embodiment uses the reflectivity of the constructed NDVI of HJ-1A CCD and HJ-1B IRS short-wave infrared wave band and constructed NDWI thereof to make up winter wheat degree of ripeness forecast model.

$\left\{\begin{array}{c}\mathrm{NDVI}=(R_{\mathrm{NIR}}-R_R)/(R_{\mathrm{NIR}}+R_R)\\ \mathrm{NDWI}=(R_{\mathrm{NIR}}-R_{\mathrm{SWIR}})/(R_{\mathrm{NIR}}+R_{\mathrm{SWIR}})\end{array}\right.---\left(20\right)$

In the formula, R _NIRBe the reflectivity of crop at near-infrared band, R _RBe the reflectivity of crop at infrared band, R _SWIRBe the reflectivity of crop at the short-wave infrared wave band.

(2) statistical model

Be input based on the evaluation method of statistical model with wave band reflectivity and the constructed various indexes thereof of HJ-1CCD and IRS, wherein index comprises NDVI and NDWI.Study the correlativity of these indexes and different-waveband reflectivity and crop water and chlorophyll content by the method for regretional analysis, choose the higher factor of correlativity, extract major component in the factor with the method for principal component analysis (PCA) again, and make up monobasic or polynary recurrence appraising model.In definite process of model parameter, still use least square method.

(3) mechanism model

The blade optical model is based on biophysics mechanism, by describing scattering and the absorption of photon in blade, simulate the spectral characteristic of blade, its forward process all comprises biochemical component content usually, these parameters can't obtain analytical expression usually, but can obtain by reverse inverting.Further leaf model can be coupled in the canopy model, just can utilize the canopy spectra data inversion to obtain component concentration.Because physical model has been explained the mechanism of action of light and blade material, principle is clear, in the original hypothesis scope of model, is not subject to factors such as time place in addition, therefore becomes the important means that the plant biochemistry component parameter is extracted.

Present embodiment adopts PROSAIL physical model inverting crop water and chlorophyll content.The PROSAIL model is coupled to form by blade optical property model PROSPECT and canopy reflection model SAIL, had studies show that this model integrated the advantage of blade yardstick and canopy yardstick two class models, in the different spatial resolutions remotely-sensed data good stability is arranged.Utilize the foundation of PROSAIL model based on the data look-up table of chlorophyll content, leaf water content etc.The model inversion strategy is for minimizing the objective function method, the calculated difference function F.When difference functions F value is more little, the reflectance value of modeling reflectance value and actual measurement is more approaching, then difference functions F minimum or a hour corresponding analog result namely be considered to inversion result.The reflectivity of HJ-1CCD/IRS after selecting to merge is input quantity, and difference functions F formula is as follows:

$F=\underset{k=1}{\overset{n}{\mathrm{\Σ}}}\left(\right|{\mathrm{\ρ}}_{\mathrm{mod}}^k-{\mathrm{\ρ}}_{\mathrm{HJ}}^k\left|\right)---\left(21\right)$

In the formula,

Simulated reflectivity value for wavelength k;

Actual measurement reflectance value for wavelength k.Finally select the corresponding analog parameter value of F value of preceding 50 minimums to average, as picture dot crop water and the chlorophyll content of final inverting.Algorithm is intended realizing with the IDL Programming with Pascal Language under the ENVI environment.

5, the crops biochemical parameter prediction crops maturity stage that obtains according to inverting

At the picture dot yardstick crop water of remote sensing appraising and the crop maturity stage after the chlorophyll content and functionization are closed on stage biochemical parameter Changing Pattern and be coupled, determine the time gap apart from the maturity stage, thereby realize the estimation in crop maturity stage.The ripe date of estimation is represented with cromogram, as shown in Figure 6, different different ripe date of color showing among the figure.

Embodiments of the invention provide for example with for the purpose of describing, and are not exhaustively or limit the invention to disclosed form.Many modifications and variations are apparent for the ordinary skill in the art.Selecting and describing embodiment is for better explanation principle of the present invention and practical application, thereby and those of ordinary skill in the art can understand the various embodiment that have various modifications that the present invention's design is suitable for special-purpose.

Claims (13)

1. crops maturity stage remote sensing prediction method based on RS data is characterized in that this method may further comprise the steps:

(1) remotely-sensed data of ground spectral reflectance is carried out pre-service, these remotely-sensed datas derive from a plurality of satellite sensors;

(2) pretreated RS data is carried out data fusion;

(3) based on ground observation data extract crops maturity stage biochemical parameter Changing Pattern;

(4) according to Changing Pattern and the remotely-sensed data inverting crops biochemical parameter of the crops biochemical parameter that extracts;

(5) the crops biochemical parameter prediction crops maturity stage that obtains according to inverting.

2. the crops maturity stage remote sensing prediction method based on RS data according to claim 1 is characterized in that described remotely-sensed data comprises MODIS data, MERSI data, HJ-1CCD data and HJ-1IRS data.

3. the crops maturity stage remote sensing prediction method based on RS data according to claim 1 is characterized in that, the pretreated content of described remotely-sensed data comprises radiation calibration, geometric correction and atmosphere correction.

4. the crops maturity stage remote sensing prediction method based on RS data according to claim 3 is characterized in that, described geometric correction adopts the quadratic polynomial method.

5. the crops maturity stage remote sensing prediction method based on RS data according to claim 1 is characterized in that, described RS data fusion comprises that the space-time dimension data merges and the spectrum dimension data merges.

6. the crops maturity stage remote sensing prediction method based on RS data according to claim 5, it is characterized in that, described space-time dimension data merges employing space-time adaptability reflectivity Fusion Model, and described spectrum dimension data merges the employing wavelet transform fusion.

7. the crops maturity stage remote sensing prediction method based on RS data according to claim 6 is characterized in that described wavelet transform fusion may further comprise the steps:

(1) image is carried out wavelet transformation, be about on the different characteristic territory of picture breakdown under the different frequency;

(2) low frequency component and the high fdrequency component after will decomposing merges by certain fusion rule.

8. the crops maturity stage remote sensing prediction method based on RS data according to claim 7, it is characterized in that, described fusion rule comprises high frequency fusion rule and low frequency fusion rule, described high frequency fusion rule comprises method of substitution, method of weighted mean and based on the maximum value method, and described low frequency fusion rule comprises method of substitution, method of weighted mean and based on the edge reservation method.

9. the crops maturity stage remote sensing prediction method based on RS data according to claim 1 is characterized in that, described crops maturity stage biochemical parameter Changing Pattern adopts least square fitting to obtain.

10. the crops maturity stage remote sensing prediction method based on RS data according to claim 9 is characterized in that described biochemical parameter comprises canopy leaf water content and chlorophyll content.

11. the crops maturity stage remote sensing prediction method based on RS data according to claim 10, it is characterized in that, described canopy leaf water content is monitored by reflectance varies or the NDWI of short-wave infrared wave band, and described chlorophyll content is monitored by reflectivity or the NDVI of near-infrared band.

12. the crops maturity stage remote sensing prediction method based on RS data according to claim 1 is characterized in that, the remote-sensing inversion of described biochemical parameter adopts the method for statistical model or the method for mechanism model.

13. according to each described crops maturity stage remote sensing prediction method based on RS data of claim 1～12, it is characterized in that, at the picture dot yardstick crop water of remote sensing appraising and the crop maturity stage after the chlorophyll content and functionization being closed on stage biochemical parameter Changing Pattern is coupled, determine the time gap apart from the maturity stage, thereby realize the estimation in crop maturity stage.

Images