CN105530507B - A kind of method according to in-orbit parameter prediction picture quality - Google Patents

Abstract

The present invention provides a kind of method according to in-orbit parameter prediction picture quality, by the in-orbit imaging of computer simulation remote sensor, emulation obtains multigroup emulating image under different in-orbit parameters, calculate image parameter and selected based on correlation analysis and sensitiveness and participate in the image parameter that picture quality comprehensive parameters build, then build respectively with 10 relational models (positive and reverse) between image parameter and picture quality comprehensive parameters, training parameter is constructed on this basis, then linear regression is carried out, construct the collective model of image quality parameter and in-orbit parameter, based on this collective model, according to the in-orbit parameter of remote sensor, the collective model of the input construction of abovementioned steps 7, using the picture quality that the output of the collective model shoots as remote sensor.Using method proposed by the present invention, the rapid Estimation of picture quality is capable of achieving, the calibration that line parameter is entered for remote sensor is in-orbit provides theoretical foundation and technical support, while greatly reducing manpower and time cost.

Description

A kind of method according to in-orbit parameter prediction picture quality

Technical field

The present invention relates to remote sensing satellite image data processing and image quality evaluation technical field, in particular to one kind According to the method for in-orbit parameter prediction picture quality.

Background technology

With many sequential transmissions of high-resolution commercial satellite of China, remote sensing images product is also presented a kind of blowout Growth state, this all brings great convenience for the related work of industry-by-industry.But remote sensor is just as common mechanical device Part is the same, and with the work of long-time high intensity, device and circuit can produce aging, such as examination and maintenance not in time, certainly will The decline of its working condition and shooting image quality can be caused.But remote sensor, can again unlike one, the camera that people use The examination and maintenance of state are carried out at any time, and it flies in space high speed, and this causes that carrying out artificial maintenance to it is almost not It is possible.Fortunately staff can on ground by adjusting the in-orbit parameter of remote sensor, making up it because of mechanical aging or The imaging error that pose adjustment is caused.

Current way is that the staff on ground inspects periodically to remote sensor imaging contexts, according to remote sensing images The in-orbit parameter of quality condition manual adjustment remote sensor, to make up the error caused by the reasons such as long-time mechanical aging.It is this Adjustment mode needs the result of remote sensing expert multiple image according to the observation, and by virtue of experience the in-orbit parameter to remote sensor is made suitably Adjustment, and after the adjustment, in addition it is also necessary to generate the quality condition of image again according to remote sensor to judge that this adjustment is No success.This method had both consumed substantial amounts of artificial, and certain imaging cycle is needed again to judge to adjust the reasonability of result.

So urgent need is a kind of can be according to the method for the in-orbit parameter prediction picture quality of remote sensor.But the in-orbit parameter of remote sensor The acquisition of relation is faced with very big challenge between change and picture quality.

The content of the invention

Asked present invention aim at the quality that cannot accurately estimate image automatically according to in-orbit parameter for prior art Topic, there is provided one kind obtains the relation between in-orbit parameter and picture quality by emulating image, realizes according to in-orbit parameter prediction The method of picture quality.

Above-mentioned purpose of the invention realized by the technical characteristic of independent claims, and dependent claims are selecting else or have The mode of profit develops the technical characteristic of independent claims.

To reach above-mentioned purpose, the technical solution adopted in the present invention is as follows：

A kind of method according to in-orbit parameter prediction picture quality, comprises the following steps：

Step 1, by the in-orbit imaging of computer simulation remote sensor, emulation obtains multigroup emulation under different in-orbit parameters Image；

Step 2, the multiple images parameter for calculating each width emulating image, including contrast, comentropy, variance, definition, Signal to noise ratio, information capacity, details energy, angular second moment, image is related, image average, image power spectrum, edge energy, radiation Precision steepness；

Step 3, the foregoing image parameter being calculated is normalized, is then based on the shifting of correlation analysis result Except wherein disturbing the data of modeling, then the effective image parameter to in-orbit parameter sensitivity is chosen, image is participated in using its average value The structure of quality comprehensive parameter；

Mathematical Modeling between step 4, structure picture quality comprehensive parameters and single in-orbit parameter, respectively with focal length, as Move, forward direction linear movement, angular speed and integration series are variable, with picture quality comprehensive parameters as dependent variable, build five numbers Learn model；

Step 5：The Mathematical Modeling between single in-orbit parameter and picture quality comprehensive parameters is built, with picture quality synthesis Parameter is variable, then respectively with focal length, as moving, and forward direction linear movement, angular speed and integration series are dependent variable, build five Mathematical Modeling；

Step 6, according to step 4 and 5 set up 10 Mathematical Modelings of mathematics, for picture quality comprehensive parameters, in [0,1] In the range of take N number of value at equal intervals, according to the Mathematical Modeling that step 5 is obtained, calculate respectively comprehensive corresponding to each picture quality Close the value of the in-orbit parameter of parameter；Then the Mathematical Modeling for being obtained further according to step 4 calculates the corresponding image of in-orbit parameter again The value of quality comprehensive parameter, training data is built with this；

Step 7, the collective model for building picture quality comprehensive parameters and in-orbit parameter, with the image of N number of grade of setting The in-orbit parameter of single image that quality comprehensive parameter is calculated as response vector, using foregoing model is linearly returned as input Return, construct the collective model of image quality parameter and in-orbit parameter；

Step 8, the in-orbit parameter according to remote sensor, the collective model of the input construction of abovementioned steps 7, with the collective model Export the picture quality shot as remote sensor.

From the above technical solution of the present invention shows that, the beneficial effects of the present invention are：

Adjustment to the in-orbit parameter of remote sensor in the past is required for remote sensing expert rule of thumb to complete, and utilizes this method to set up Mathematical Modeling, picture quality can just be estimated according to the in-orbit parameter of remote sensor, such that it is able to instruct remote sensor to make appropriate tune The whole remote sensing images to obtain high-quality, the calibration that line parameter is entered for remote sensor is in-orbit provides theoretical foundation and technical support, Greatly reduce manpower and time cost simultaneously.

Brief description of the drawings

Fig. 1 realizes schematic flow sheet for an embodiment of the present invention according to in-orbit parameter prediction picture quality.

Fig. 2 a-2e are in-orbit parameter and parts of images parameter (contrast (curve 1), definition (curve 2), details energy (curve 3), comentropy (curve 4)) relation schematic diagram, wherein：2a is the relation schematic diagram of focal length, and 2b is as the relation moved is shown It is intended to, 2c is the preceding relation schematic diagram to linear movement, 2d is the relation schematic diagram for integrating series, and 2e shows for the relation of angular speed It is intended to.

Specific embodiment

In order to know more about technology contents of the invention, especially exemplified by specific embodiment and institute's accompanying drawings are coordinated to be described as follows.

On the whole, the method according to in-orbit parameter prediction picture quality that the present embodiment is proposed, for focal length, as moving, Forward direction linear movement, integrates sum of series five in-orbit parameters of angular speed, is simulated respectively by computer in different in-orbit parameters The image that lower remote sensor is generated, then calculates multiple mass parameters of image, and the situation of change according to these parameters extracts one Kind it is capable of the comprehensive parameters of picture quality anyway.Then multiple grades and the comprehensive matter of emulating image for each in-orbit parameter Amount parameter carries out regression analysis, the multiple Mathematical Modelings between in-orbit parameter and picture quality comprehensive parameters is set up, finally these Model combines the collective model for constituting in-orbit parameter and image quality parameter by least square method, passes through so as to reach The purpose of in-orbit parameter prediction picture quality.

As shown in figure 1, preferred embodiment of the invention, a kind of method according to in-orbit parameter prediction picture quality, Comprise the following steps：

Step 1, by the in-orbit imaging of computer simulation remote sensor, emulation obtains multigroup emulation under different in-orbit parameters Image；

Step 2, the multiple images parameter for calculating each width emulating image, including contrast, comentropy, variance, definition, Signal to noise ratio, information capacity, details energy, angular second moment, image is related, image average, image power spectrum, edge energy, radiation Precision steepness；

Step 3, the foregoing image parameter being calculated is normalized, is then based on the shifting of correlation analysis result Except wherein disturbing the data of modeling, then the effective image parameter to in-orbit parameter sensitivity is chosen, image is participated in using its average value The structure of quality comprehensive parameter；

Mathematical Modeling between step 4, structure picture quality comprehensive parameters and single in-orbit parameter, respectively with focal length, as Move, forward direction linear movement, angular speed and integration series are variable, with picture quality comprehensive parameters as dependent variable, build five numbers Learn model；

Step 5：The Mathematical Modeling between single in-orbit parameter and picture quality comprehensive parameters is built, with picture quality synthesis Parameter is variable, then respectively with focal length, as moving, and forward direction linear movement, angular speed and integration series are dependent variable, build five Mathematical Modeling；

Step 6, according to step 4 and 5 set up 10 Mathematical Modelings of mathematics, for picture quality comprehensive parameters, in [0,1] In the range of take N number of value at equal intervals, according to the Mathematical Modeling that step 5 is obtained, calculate respectively comprehensive corresponding to each picture quality Close the value of the in-orbit parameter of parameter；Then the Mathematical Modeling for being obtained further according to step 4 calculates the corresponding image of in-orbit parameter again The value of quality comprehensive parameter, training data is built with this；

Step 7, the collective model for building picture quality comprehensive parameters and in-orbit parameter, with the image of N number of grade of setting The in-orbit parameter of single image that quality comprehensive parameter is calculated as response vector, using foregoing model is linearly returned as input Return, construct the collective model of image quality parameter and in-orbit parameter；

Step 8, the in-orbit parameter according to remote sensor, the collective model of the input construction of abovementioned steps 7, with the collective model Export the picture quality shot as remote sensor.

Below in conjunction with the accompanying drawings shown in 1, the specific implementation of above steps is described in detail.

Step 1, acquisition emulating image

P width image using a certain satellite in different latitude as target image, then according to modulation transfer function model And ray trace model, by the visual remote sensing device of computer simulation satellite imaging process in orbit, emulation is obtained P*Q width focal length emulating images under different focal grade Q.

For example, using 17 width images of No. three satellites of resource in different latitude as target image, being obtained by software emulation To the experimental image under different in-orbit parameters, for setting up the Mathematical Modeling between in-orbit parameter and image quality parameter.Root According to modulation transfer function model and ray trace model, by No. 3 in-orbit fortune of satellite visible remote sensor of computer simulation resource Capable imaging process, emulation obtains image.During emulation, the different Parameters variation grade for different in-orbit parameter settings, Substantial amounts of emulation data are generated in case relationship modeling is used.

Step 2, calculating image parameter

Calculate the multiple images parameter of each width emulating image, including contrast, comentropy, variance, definition, noise Than information capacity, details energy, angular second moment, image is related, image average, image power spectrum, edge energy, Electrodynamic radiation Steepness.

The present embodiment, can calculate aforesaid plurality of parameter, (such as Ratael C.Gonzalez using conventional method Write with Richard E.Woods,《Digital Image Processing》The second edition, the computational methods proposed in Electronic Industry Press), This is repeated no more.

Step 3, structure picture quality comprehensive parameters

In this step, the foregoing image parameter being calculated is normalized, is then based on correlation analysis knot Fruit removes the data for wherein disturbing modeling, then chooses the effective image parameter to in-orbit parameter sensitivity, is participated in using its average value The structure of picture quality comprehensive parameters.

Due to all of image parameter that step 2 is calculated might not be suitable for modeling because they and differ It is fixed all sensitive to all of in-orbit parameter.Therefore, in this step, to image parameter, (1) is normalized place according to the following formula first Reason, then according to it with the Changing Pattern of in-orbit parameter, to five in-orbit parameters, (focal length, as moving, forward direction is linearly transported for selection It is dynamic, angular speed and integration series) while the image parameter of sensitivity participates in the structure of picture quality comprehensive parameters.

Wherein：x_ijRepresent the value of the image parameter of j-th grade of the i-th width target image

In this step, using SPSS softwares (statistical product and service solution software) to foregoing 13 image parameters (contrast, comentropy, variance, definition, signal to noise ratio, information capacity, details energy, angular second moment, image is related, and image is equal Value, image power spectrum, edge energy, Electrodynamic radiation steepness) correlation analysis are carried out, remove the data of interference modeling.

Specifically, correlation analysis are carried out with 13 image parameters respectively for each in-orbit parameter, weeds out phase relation , less than the image parameter of 0.95 (i.e. correlation is not strong), (focal length, as moving, forward direction is linear then to choose five in-orbit parameters for number Motion, angular speed and integration series) image parameter that includes carrys out the final structure for participating in picture quality comprehensive parameters.It is i.e. in-orbit Parameter determines the final image parameter for participating in the foundation of picture quality comprehensive parameters.

As an example, as shown in Fig. 2 finally choosing contrast (curve 1), definition (curve 2), details energy (curve 3), four image parameters of comentropy (curve 4) participate in the structure of picture quality comprehensive parameters, because this four image parameters exist In the range of [0,1], four curve shapes are similar, and change very regular, so in the present embodiment, with contrast, clearly Degree, the average value after details energy, four image parameter normalization of comentropy is used as picture quality comprehensive parameters.

Following table is the image synthesis parameter and five Spearman coefficient correlations of in-orbit parameter.

Table 1：Image synthesis parameter and five Spearman coefficient correlations of in-orbit parameter

Can be seen that with the average of aforementioned four image parameter be rational as image synthesis parameter according to data in table.

Mathematical Modeling between step 4, structure picture quality comprehensive parameters and single in-orbit parameter

In order to obtain the relation between in-orbit parameter and picture quality comprehensive parameters, it is necessary to substantial amounts of data, and emulate mould Type is merely able to image of the simulation in single in-orbit parameter remote device generation, in order to build a parameters relationship model for synthesis, institute To need to go out substantial amounts of training data with the Construction of A Model of single in-orbit parameter by picture quality comprehensive parameters in advance, in case building Vertical collective model is used.

In the present embodiment, respectively with focal length, as moving, forward direction linear movement, angular speed and integration series are variable, with image Quality comprehensive parameter is dependent variable, using the fitting function of SPSS softwares, builds five Mathematical Modelings：

Y=e^0.123-27.183x₁ (2)

Wherein：x₁Represent as moving, x₂Represent focal length, x₃To linear movement, x before representing₄Represent angular speed, x₅Represent integration stages Number, y represents picture quality comprehensive parameters.

Step 5：Build the Mathematical Modeling between single in-orbit parameter and picture quality comprehensive parameters

Purpose with step 4 is similar, and the relational model built between single in-orbit parameter and picture quality comprehensive parameters is Think that structure collective model is used to generate different in-orbit parameters respectively under unified image quality parameter.

In the present embodiment, with picture quality comprehensive parameters as variable, then respectively with focal length, as moving, forward direction linear movement, Angular speed and integration series are dependent variable, using the fitting function of SPSS softwares, then build five Mathematical Modelings：

x₁=0.005-0.037 × lny (7)

x₂=e^{-2.347+2.901y} (8)

x₃=-0.492 × y³+1.018×y²-0.772×y+0.240 (9)

x₄=-0.231 × y³+0.293×y²-0.299×y+0.249 (10)

x₅=24.884y³-78.859×y²-64.214×y+126.622 (11)

Wherein, the implication of each parameter with it is consistent involved by step 4, i.e.,：x₁Represent as moving, x₂Represent focal length, x₃Before expression To linear movement, x₄Represent angular speed, x₅Integration series is represented, y represents picture quality comprehensive parameters.

Step 6, construction training data

According to step 4 and 5 10 Mathematical Modelings of mathematics set up, for picture quality comprehensive parameters, in the scope of [0,1] N number of value is inside taken at equal intervals, according to the Mathematical Modeling that step 5 is obtained, is calculated respectively corresponding to each picture quality comprehensive parameters In-orbit parameter value；Then it is comprehensive that the Mathematical Modeling for being obtained further according to step 4 calculates the corresponding picture quality of in-orbit parameter again The value of parameter is closed, training data is built with this.

Specifically, building comprehensive parameters model needs to obtain training data X and Y, Y corresponding to picture quality comprehensive parameters, X It is that the functional value being calculated by the model of step 5 of five corresponding in-orbit parameters of each element passes through certain in Y Conversion (such as following formula 13) after obtain.

Take (such as degeneration of each in-orbit parameter setting when 50, N is typically with emulation of N number of value at equal intervals between 0 to 1 Grade is close, such as 39 focal length grades), as Y.And X is according to five transformation ranges of in-orbit parameters simulation image parameter Size and the ratio decision of the transformation range size of focal length emulating image parameter, use r_kRepresent ratio：

Then with each value y in Y_iIt is correspondingIt is expressed as：

Wherein：f_kIt is in-orbit parameter k corresponding Mathematical Modeling, g in step 4_kFor in-orbit parameter k is corresponding in steps of 5 Mathematical Modeling.

According to above method, five r are calculated_kRespectively：0.998,1,0.7817,0.4660,0.0818, Ran Hou 51 values are equidistantly chosen in the interval [0,1] of image synthesis parameter y, then according to being calculated corresponding xi respectively, Data are as shown in the table：

The training data of table 2

Step 7, the collective model for building picture quality comprehensive parameters and in-orbit parameter

According to 5 Mathematical Modeling (f of calculating picture quality that step 4 is set up_iRepresent i-th mathematical function) understand, root The comprehensive parameters of picture quality can be calculated according to single in-orbit parameter, but in order to set up multiple in-orbit parameters and image matter The relation of comprehensive parameters is measured, what is selected in the present embodiment is average weighted mode, proposed Liru hypograph quality comprehensive parameter With the collective model of in-orbit parameter, a_kAs weight, has weighed the degree of each in-orbit parameter influence picture quality：

Y=a₁×f₁(x₁)+a₂×f₂(x₂)+a₃×f₃(x₃)+a₄×f₄(x₄)+a₅×f₅(x₅) (14)

Wherein：f_iI-th function model in step 4 is represented, y represents picture quality comprehensive parameters, a_kExpression is directed to often The coefficient of individual in-orbit parameter k, and and be 1.The form for being write as matrix is：

Y=α^Tβ (15)

Wherein： And there is a₁+a₂+a₃+a₄+a₅=1.

To solve above-mentioned β, following object function need to be optimized：

s.t.a₁+a₂+a₃+a₄+a₅=1

Introduce glug and draw day multiplier λ, above-mentioned optimization problem is converted into following problems

It is 0 to make its gradient

That is,：

Above formula is merged into matrix to represent：

Represented with ARepresented with BSoSo as to solve, model is finally obtained β：

Y=1.0616f₁-0.0350f₂-0.0069f₃-0.0022f₄-0.0175f₅ (21)

f_iI-th function model in step 4 is represented, y represents picture quality comprehensive parameters；

Step 8, estimate picture quality

According to the in-orbit parameter of remote sensor, the collective model of the input construction of abovementioned steps 7 is made with the output of the collective model For the picture quality that remote sensor shoots.

Although the present invention is disclosed above with preferred embodiment, so it is not limited to the present invention.Skill belonging to of the invention Has usually intellectual in art field, without departing from the spirit and scope of the present invention, when can be used for a variety of modifications and variations.Cause This, protection scope of the present invention ought be defined depending on those as defined in claim.

Claims (7)

1. a kind of method according to in-orbit parameter prediction picture quality, it is characterised in that comprise the following steps：

Step 1, by the in-orbit imaging of computer simulation remote sensor, emulation obtains multigroup analogous diagram under different in-orbit parameters Picture；

Step 2, the multiple images parameter for calculating each width emulating image, including contrast, comentropy, variance, definition, noise Than information capacity, details energy, angular second moment, image is related, image average, image power spectrum, edge energy, Electrodynamic radiation Steepness；

Step 3, the foregoing image parameter being calculated is normalized, is then based on correlation analysis result and removes it The data of middle interference modeling, then the effective image parameter to in-orbit parameter sensitivity is chosen, participate in picture quality using its average value The structure of comprehensive parameters；

Mathematical Modeling between step 4, structure picture quality comprehensive parameters and single in-orbit parameter, it is preceding as moving respectively with focal length To linear movement, angular speed and integration series are variable, with picture quality comprehensive parameters as dependent variable, build five mathematical modulos Type；

Step 5：The Mathematical Modeling between single in-orbit parameter and picture quality comprehensive parameters is built, with picture quality comprehensive parameters It is variable, then respectively with focal length, as moving, forward direction linear movement, angular speed and integration series are dependent variable, build five mathematics Model；

Step 6, according to step 4 and 5 set up 10 Mathematical Modelings, for picture quality comprehensive parameters, in the range of [0,1] N number of value is taken at equal intervals, according to the Mathematical Modeling that step 5 is obtained, is calculated respectively corresponding to each picture quality comprehensive parameters The value of in-orbit parameter；Then the Mathematical Modeling for being obtained further according to step 4 calculates the corresponding picture quality synthesis of in-orbit parameter again The value of parameter, training data is built with this；

Step 7, the collective model for building picture quality comprehensive parameters and in-orbit parameter, with the picture quality of N number of grade of setting The in-orbit parameter of single image that comprehensive parameters are calculated as response vector, using foregoing model carries out linear regression, structure as input Produce the collective model of image quality parameter and in-orbit parameter；

Step 8, the in-orbit parameter according to remote sensor, the collective model of the input construction of abovementioned steps 7, with the output of the collective model As the picture quality that remote sensor shoots.

2. the method according to in-orbit parameter prediction picture quality according to claim 1, it is characterised in that abovementioned steps 1 Implementation comprise the following steps：

P width image using a certain satellite in different latitude as target image, then according to modulation transfer function model and light Line trace model, by the visual remote sensing device of computer simulation satellite imaging process in orbit, emulation obtains difference P*Q width focal length emulating images under focal length grade Q.

3. the method according to in-orbit parameter prediction picture quality according to claim 1, it is characterised in that abovementioned steps 3 Implementation include：

1) image parameter being calculated to step 2 is normalized according to the following formula：

$x_{ij}=\frac{x_{ij}-\min \left(x_{i1},x_{i2},\mathrm{...},x_{in}\right)}{\max \left(x_{i1},x_{i2},\mathrm{...},x_{in}\right)-\min \left(x_{i1},x_{i2},\mathrm{...},x_{in}\right)}---\left(1\right)$

Wherein：x_ijRepresent the value of the image parameter of j-th grade of the i-th width target image；

2) it is contrast to foregoing 13 image parameters using SPSS softwares, comentropy, variance, definition, signal to noise ratio, information is held Amount, details energy, angular second moment, image is related, and image average, image power spectrum, edge energy, Electrodynamic radiation steepness is carried out Correlation analysis, weed out image parameter of the coefficient correlation less than 0.95；

3) and then to choose five in-orbit parameters be focal length, as moving, forward direction linear movement, angular speed and the integration figure that includes of series As parameter participates in the image parameter that picture quality comprehensive parameters build as final, after selected image parameter normalization Average value is used as picture quality comprehensive parameters.

4. the method according to in-orbit parameter prediction picture quality according to claim 1, it is characterised in that abovementioned steps 4 Realization specifically include following steps：

Respectively with focal length, as moving, forward direction linear movement, angular speed and integration series are variable, are with picture quality comprehensive parameters Dependent variable, using the fitting function of SPSS softwares, builds five Mathematical Modelings：

$y=e^{0.123-27.183}{x}_1---\left(2\right)$

$y=0.228\×x_2^3-1.079\×x_2^2+1.815\×x_2-0.118---\left(3\right)$

$y=-112.941\×x_3^3+61.560\×x_3^2-12.421\×x_3+0.990---\left(4\right)$

$y=69.263\×x_4^3-31.989\×x_4^2-0.362\×x_4+0.998---\left(5\right)$

$y=-2.548e-7\×x_5^3+2.937e-5\×x_5^2-0.008\×x_5+1.071$ $\left(6\right)$

Wherein：x₁Represent as moving, x₂Represent focal length, x₃To linear movement, x before representing₄Represent angular speed, x₅Represent integration series, y Represent picture quality comprehensive parameters.

5. the method according to in-orbit parameter prediction picture quality according to claim 4, it is characterised in that abovementioned steps 5 Realization specifically include following steps：

With picture quality comprehensive parameters as variable, then respectively with focal length, as shifting, forward direction linear movement, angular speed and integration stages Number is dependent variable, using the fitting function of SPSS softwares, builds five Mathematical Modelings：

x₁=0.005-0.037 × lny (7)

x₂=e^{-2.347+2.901y} (8)

x₃=-0.492 × y³+1.018×y²-0.772×y+0.240 (9)

x₄=-0.231 × y³+0.293×y²-0.299×y+0.249 (10)

x₅=24.884y³-78.859×y²-64.214×y+126.622 (11)

Wherein, x₁Represent as moving, x₂Represent focal length, x₃To linear movement, x before representing₄Represent angular speed, x₅Represent integration series, y Represent picture quality comprehensive parameters.

6. the method according to in-orbit parameter prediction picture quality according to claim 5, it is characterised in that abovementioned steps 6 Realization specifically include following steps：

Build comprehensive parameters model and need to obtain training data X and Y, Y and correspond to picture quality comprehensive parameters, X be in Y each The functional value being calculated by the Mathematical Modeling of abovementioned steps 5 of five corresponding in-orbit parameters of element is become through following formula (13) The value for alternatively obtaining afterwards, specifically：

N number of value is taken at equal intervals between 0 to 1, as Y, conversion of the X according to the emulating image parameter of foregoing five in-orbit parameters Range size and the ratio decision of the transformation range size of focal length emulating image parameter, use r_kRepresent ratio：

Then, with each value y in Y_iIt is correspondingIt is expressed as：

$x_i^k=\left\{\begin{array}{cc}{f}_k\left(g_k\left(\frac{y_i-\left(1-r_k\right)}{r_k}\right)\right)& y_i\&GreaterEqual;\left(1-r_k\right)\\ f_k\left(g_k\left(0\right)\right)& y_i<\left(1-r_k\right)\end{array}\right.---\left(13\right)$

Wherein：f_kIt is in-orbit parameter k corresponding Mathematical Modeling, g in step 4_kIt is in-orbit parameter k Mathematical Modelings in steps of 5；

According to above formula (12), (13), five ratio rs are calculated_k, then image synthesis parameter y interval [0, 1] 51 values are equidistantly chosen in, then according to being calculated corresponding x respectively_i, so as to construct complete training data.

7. the method according to in-orbit parameter prediction picture quality according to claim 6, it is characterised in that abovementioned steps 7 Realization comprise the following steps：

According to 5 Mathematical Modelings of calculating picture quality comprehensive parameters that step 4 is set up, below figure is set up as quality comprehensive parameter With the collective model of in-orbit parameter：

Y=a₁×f₁(x₁)+a₂×f₂(x₂)+a₃×f₃(x₃)+a₄×f₄(x₄)+a₅×f₅(x₅) (14)

Wherein：f_iI-th function model in step 4 is represented, y represents picture quality comprehensive parameters, a_kExpression be directed to each The coefficient of rail parameter k, and and be 1；

The expression matrix form of formula (14) is：

Y=α^Tβ (15)

Wherein：And there is a₁+a₂+a₃+a₄+a₅=1；

To solve above-mentioned β, following object function need to be optimized：

$\min \frac{1}{2}\left|\right|Y-X\mathrm{\&beta;}|{|}^2$

s.t. a₁+a₂+a₃+a₄+a₅=1 (16)

Introduce glug and draw day multiplier λ, above-mentioned optimization problem is converted into following problems

$\min \frac{1}{2}\left|\right|Y-X\mathrm{\&beta;}|{|}^2+\mathrm{\&lambda;}(a_1+a_2+a_3+a_4+a_5-1)---\left(17\right)$

It is 0 to make its gradient, then：

$\frac{\&part;}{\&part;\mathrm{\&beta;}}\left(\frac{1}{2}\left|\right|Y-X\mathrm{\&beta;}|{|}^2+\mathrm{\&lambda;}\left(a_1+a_2+a_3+a_4+a_5-1\right)\right)=0---\left(18\right)$

That is,：

$\left\{\begin{array}{c}{X}^T(X\mathrm{\&beta;}-Y)+\left[\begin{array}{c}\mathrm{\&lambda;}\\ \mathrm{\&lambda;}\\ \mathrm{\&lambda;}\\ \mathrm{\&lambda;}\\ \mathrm{\&lambda;}\end{array}\right]=0\\ \left[\begin{array}{ccccc}1& 1& 1& 1& 1\end{array}\right]\left[\begin{array}{c}{a}_1\\ a_2\\ a_3\\ a_4\\ a_5\end{array}\right]+0\left[\begin{array}{c}\mathrm{\&lambda;}\\ \mathrm{\&lambda;}\\ \mathrm{\&lambda;}\\ \mathrm{\&lambda;}\\ \mathrm{\&lambda;}\end{array}\right]=1\end{array}\right.---\left(19\right)$

Above formula is merged into matrix to represent：

$\left[\begin{array}{cc}{X}^{T}X& 1\\ 1& 0\end{array}\right]\left[\begin{array}{c}{a}_1\\ a_2\\ a_3\\ a_4\\ a_5\end{array}\right]=\left[\begin{array}{c}{X}^{T}Y\\ 1\end{array}\right]---\left(20\right)$

Represented with ARepresented with BSoβ tries to achieve foregoing a so as to solve₁、a₂、a₃、 a₄、a₅Numerical value, substituted into the comprehensive mould that formula (14) can obtain final picture quality comprehensive parameters and in-orbit parameter Type.