CN102324098B - Relative radiation calibration method in combination with laboratory calibration and even landscape statistics - Google Patents

Abstract

The invention relates to a relative radiation calibration method in combination with laboratory calibration and even landscape statistics, which comprises steps of: 1) according to acquired laboratory calibration data, obtaining the radiance and the corresponding DN value of a lamp at each level and seeking the calibration coefficient of each pixel; 2) according to images acquired at the orbital running period of a satellite, selecting areas which are even on the whole to generate the calibration coefficient of each pixel through an even landscape statistic method; and 3) conducting joint calibration according to the step 1 and the step 2 to obtain new calibration coefficients. The method provided by the invention overcomes the defects of the prior art. In combination with the advantages of the laboratory calibration and the ground even landscape statistics, according to the characteristics of satellite running and camera response, new calibration coefficients are dynamically formed and the relative radiation calibration accuracy is ensured.

Description

A kind of relative radiometric calibration method in conjunction with Laboratory Calibration and even landscape statistics

Technical field

The present invention relates to a kind of remote sensing images relative radiometric calibration method.

Background technology

China succeeds in sending up China-Brazil Earth Resources Satellite (CBERS) and environment mitigation satellite (HJ-1) and in succession changes the businessization operation over to, is through with and can only relies on for a long time the history of the external remote sensing satellite data of import.CBERS and HJ-1 two large serial satellite datas have been widely used in the fields such as agricultural resource investigation, Crop Estimation, ECOLOGICAL ENVIRONMENTAL MONITORING, city planning, survey of territorial resources and disaster monitoring and military affairs, indicate that China's spacer remote sensing application has entered a brand-new stage.

Relatively radiant correction is called homogenising and proofreaies and correct or normalized, and it is in order to proofread and correct each pixel and optical system response difference in the remote sensor the original densitometer numerical value (DN) that obtains to be carried out a kind of processing procedure that " again " quantizes.Radiant correction is as the pretreated important step of remotely-sensed data relatively, and key is to obtain effective calibration coefficient, i.e. normalized gain (NG) and biasing (B).Domestic and international main flow remote sensing satellite, for example domestic CBERS, HJ-1 and external Landsat, SPOT satellite adopt scaling method and ground statistic law to obtain calibration coefficient.Scaling method uses calibration lamp data as input, passes through on this basis the calibration coefficient of different algorithm basis of formations.Generally speaking, the calibration lamp comprises the uniform light source of height of some different spoke brightness degrees, or when the brightness of light source spoke can not be satisfied homogeneous condition fully, the profile of its spoke brightness can accurately be measured.Because it is higher to extract the correction coefficient precision from calibration lamp data, so satellite launch needs to carry out Laboratory Calibration as last.Satellite is through after a while operation, certain variation will occur in the sensor performance that satellite carries, when spaceborne internal calibration thrashing or use at need, statistic law is as a kind of alternative method of calibration, by analyzing and statistics calculating calibration correction coefficient from ground image, this method is suitable for imaging form polynary and that sweep.In addition, the QuickBird satellite that represents international remote sensing advanced level then adopts the method for linear array 90-degree rotation, make the direction of scanning be parallel to pixel linear array direction fully, along with satellite motion, the terrain object of each pixel experience is consistent, the even scenery that this has just gathered different spoke intensity levels effectively dynamically obtains the radiation calibration coefficient, has guaranteed the precision of irradiation treatment.Because this calibration mode needs the rotary coke plane, and design of satellites has been proposed higher requirement, domsat does not adopt this technology.

Because satellite launch lift-off ascent stage inclement condition, there are larger difference in satellite actual motion environment and surface state, and the factors such as the state after image formation state and the lift-off of camera is inconsistent during ground test, may make the calibration data of ground acquisition and analysis can not be fit to the requirement of high-precision correction fully, need to carry out the follow-up improvement in ground and process.

Summary of the invention

Technology of the present invention is dealt with problems and is: overcome the deficiencies in the prior art, a kind of remote sensing images relative radiometric calibration method is provided, the method is in conjunction with Laboratory Calibration and ground even landscape Statistical Superiority, characteristic according to satellite transit and camera response, the calibration coefficient that dynamic formation is new guarantees relative radiant correction precision.

Technical solution of the present invention is: a kind of relative radiometric calibration method in conjunction with Laboratory Calibration and even landscape statistics, and step is as follows:

(1) according to the Laboratory Calibration data that gather, obtains spoke brightness and the corresponding DN value of every grade of lamp, and ask for the calibration coefficient of each pixel;

(2) image that gathers in orbit according to satellite is chosen overall uniformly zone, generates the calibration coefficient of each pixel by the even landscape statistic law;

(3) result according to step (1), (2) unites calibration, and detailed process is as follows:

(3.1) associating pixel design: m pixel is combined into an associating pixel, and namely array is comprised of the associating pixel, and each associating pixel is on average obtained by m pixel, for the push-scanning image pattern, and 50＜m＜200;

(3.2) according to the result of calculation of step (1), ask for the response coefficient LNG of laboratory associating pixel _j

(3.3) according to the result of calculation of step (3), ask for the response coefficient HNG of ground even landscape associating pixel _j

(3.4) ask for associating pixel correction factor COR _j, COR _j=LGN _j/ HGN _j

(3.5) correction factor equation model: take associating pixel sequence number as horizontal ordinate, correction factor is ordinate, adopts second-degree parabola equation model correction factor curve;

(3.6) correction factor interpolation: the correction factor curve to above-mentioned match disperses, and obtains one group for the correction factor cor of single pixel _i

(3.7) new calibration coefficient nNG _i: adopt the result of calculation of step (1) or step (2) to be in correction factor, can obtain the new calibration coefficient of this array.

At first adopt even landscape statistic law, histogram equalization method, three kinds of methods of least square method to ask for the calibration coefficient of each pixel in the described step (1), then choose the result of calculation of three kinds of optimal algorithms in the algorithm as the result of calculation of step (1).

The first-selection of choosing of described optimal algorithm is chosen homogeneous area large as far as possible in the image, calculates the gray average of heading image and this regional gray average, calculates the error of above-mentioned two gray averages, when error hour, this algorithm is optimal algorithm.

The present invention compared with prior art beneficial effect is:

(1) the inventive method elimination aberration between remote sensing images sheet intra-striate and sheet, guaranteed preferably the radiation quality of image to have solved large format remote sensing images radiant correction problem, the data of other earth observation satellite are processed provides reference.

(2) the present invention unites a kind of effective ways of bringing into play Laboratory Calibration and even landscape statistic for receiver advantage of calibration.Laboratory Calibration has guaranteed the stationarity of image local transition in the trimming process; And the even landscape statistics has overcome the jumping characteristic of array splicing.

(3) Laboratory Calibration of the present invention adopts three kinds of methods to process, and by broad sense noise rating, preferred calibration coefficient, as basic data, is the subsequent treatment service.

Description of drawings

Fig. 1 is overview flow chart of the present invention;

Fig. 2 is algorithm flow chart of the present invention;

Fig. 3 is that the present invention unites the pixel mapping graph.

Embodiment

1 overview flow chart, Fig. 2 algorithm flow chart and Fig. 3 unite the pixel design drawing by reference to the accompanying drawings, and the below is described as follows the specific embodiment of the present invention:

Step 1: Laboratory Calibration

(1.1) read level B at the bottom of the i pixel _i

(1.2) calculated product bulb separation k level spoke brightness L _k

$L_k=\frac{{\∫}_0^{\∞}{R}_k\left(\mathrm{\λ}\right)\·S\left(\mathrm{\λ}\right)\mathrm{d\λ}}{{\∫}_0^{\∞}S\left(\mathrm{\λ}\right)\mathrm{d\λ}}$ ②

Wherein, S (λ) is the camera spectral response functions, R _k(λ) be light source light spectrum spoke luminance function, λ is wavelength.

(1.3) read the DN value Q of i pixel under the brightness of k level spoke _{I, k}

(1.4) use the even landscape statistic law to calculate i pixel calibration coefficient lNG _i

${\mathrm{DN}}_i=\frac{1}{n}\underset{k=1}{\overset{n}{\mathrm{\Σ}}}(Q_{i,k}-B_i)$ ③

${\mathrm{lNG}}_i=\frac{{\mathrm{DN}}_i}{\frac{1}{N}\underset{j=1}{\overset{N}{\mathrm{\Σ}}}{\mathrm{DN}}_j}$ ④

Wherein, n is calibration lamp spoke brightness progression, and N is the pixel sum.

(1.5) use the histogram equalization method to calculate i pixel calibration coefficient lNG _i

Calculate the standard deviation of i pixel image DN value _i

Calculate the mean value σ of entire image standard deviation _R

${\mathrm{\σ}}_R=\frac{1}{N}\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{\mathrm{\σ}}_i$ ⑤

lNG _i＝σ _i/σ _R ⑥

Wherein, N is the pixel sum.

(1.6) use the histogram equalization method to calculate i pixel calibration coefficient B _i

Calculate the average μ of i pixel image DN value _i

Calculate the average μ of entire image DN value _R

${\mathrm{\μ}}_R=\frac{1}{n}\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{\mathrm{\μ}}_i$ ⑦

$B_i={\mathrm{\&mu;}}_i-\frac{{\mathrm{\&sigma;}}_i\&CenterDot;{\mathrm{\&mu;}}_R}{{\mathrm{\&sigma;}}_{\mathrm{<figure-callout\; id="8"\; label="R"\; filenames="HSA00000562069200011.png"\; state="\{\{state\}\}">R</figure-callout>}}}$ ⑧

(1.7) use least square method to calculate i pixel calibration coefficient lNG _i

$G_i=\frac{\underset{k=1}{\overset{n}{\mathrm{\&Sigma;}}}(Q_{i,k}-B_i)L_k}{\underset{k=1}{\overset{\mathrm{<figure-callout\; id="2"\; label="n\; L\; k"\; filenames="HSA00000562069200011.png,HSA00000562069200022.png"\; state="\{\{state\}\}">n</figure-callout>}}{\mathrm{\&Sigma;}}}{L}_k^{}{}_2^{}}$ ⑨

${\mathrm{lNG}}_i=\frac{G_i}{\frac{1}{N}\underset{j=1}{\overset{\mathrm{<figure-callout\; id="10"\; label="N\; G\; j"\; filenames="HSA00000562069200011.png"\; state="\{\{state\}\}">N</figure-callout>}}{\mathrm{\&Sigma;}}}{G}_j{}_{}}$ ⑩

(1.8) by the above-mentioned three kinds of Algorithm Performances of broad sense noise rating, preferred optimal algorithm.

Its algorithm can be summarized as: choose large as far as possible homogeneous area, calculate gray average and this regional gray average of every row (heading) image, calculate both errors.When error hour, this algorithm is optimal algorithm.

Step 2: even landscape calibration

(2.1) read level B at the bottom of the i pixel _i

(2.2) read the DN value Q of k width of cloth even landscape i pixel _{I, k}

(2.3) use the even landscape statistic law to calculate i pixel calibration coefficient hNG _i

${\mathrm{DN}}_i=\frac{1}{n}\underset{k=1}{\overset{n}{\mathrm{\&Sigma;}}}(Q_{i,k}-B_i)$

${\mathrm{hNG}}_i=\frac{{\mathrm{DN}}_i}{\frac{1}{N}\underset{j=1}{\overset{N}{\mathrm{\&Sigma;}}}{\mathrm{DN}}_j}$

(2.4) estimate the optimizing uniform scape

According to result, the increase and decrease even landscape, by visual interpretation, preferred selected even scenery reaches the purpose that improves calibration precision.

Step 3: unite calibration

(3.1) associating pixel design

As shown in Figure 3, m pixel is combined into an associating pixel, namely array is comprised of the associating pixel, and each associating pixel is on average obtained by m pixel, guarantees on this basis the uniform properties of scenery between the associating pixel.Consider present push-scanning image system, the pixel sum is generally thousands of more than tens thousand of, takes into account the consistance of atural object homogeneity and adjacent picture elements response, the span of m: 50＜m＜200.

(3.2) ask for the response coefficient LNG that pixel is united in the laboratory _j

${\mathrm{LGN}}_j=\frac{1}{m}\underset{i=1}{\overset{m}{\mathrm{\&Sigma;}}}{\mathrm{lGN}}_{j*m+i}$

(3.3) ask for the response coefficient HNG that the ground even landscape is united pixel _j

${\mathrm{HGN}}_j=\frac{1}{m}\underset{i=1}{\overset{m}{\mathrm{\&Sigma;}}}{\mathrm{hGN}}_{j*m+i}$

(3.4) ask for associating pixel correction factor COR _j

COR _j＝LGN _j/HGN _j

(3.5) match of correction factor equation;

Take associating pixel sequence number as horizontal ordinate, correction factor is ordinate, adopts second-degree parabola equation model correction factor curve.

(3.6) interpolation of correction factor;

In order to finish the associating pixel to the mapping of single pixel, need to disperse to this continuous para-curve, obtain one group for the correction factor cor of single pixel _i

(3.7) new calibration coefficient nNG _i

If divided by correction factor, can obtain the new calibration coefficient of this array with the Laboratory Calibration coefficient.

nNG _i＝lNG _i/cor _i

Embodiment

The CBERS Satellite Payloads comprises the cameras such as CCD and WFI.The CCD camera resolution is 19.5 meters, and fabric width is 113 kilometers, has sought compromise between resolution and fabric width, and this image user uses the most extensive; The WFI camera resolution is 258 meters, and fabric width reaches 890 kilometers, realizes easily heavily visiting fast target.In order to guarantee to observe the visual field, CCD and WFI camera have adopted the optics splicing.No matter the CBERS-02 satellite in the process, adopts Laboratory Calibration or even landscape statistical scaling in orbit, and there are certain problem in WFI and ccd image calibration result, tone transition vestige occurs in array splicing place especially, affect improvement of visual effect.In conjunction with Laboratory Calibration and even landscape statistics, namely adopt the calibration of uniting of laboratory and even landscape statistics, generate new calibration coefficient, use this Data correction original image, picture quality obtains larger improvement, has satisfied the remote sensing application demand.

The CBERS-02 satellite WFI calibration data of at first utilizing the laboratory to gather are carried out relative radiant correction to the WFI image, proofread and correct the result and show that there is obvious aberration in WFI two chip arrays at overlap, but the aberration transition of image is better in the array chip.When adopting the even landscape statistical method, in order to guarantee the uniform properties of atural object, choose overall uniform many scapes WFI image, generate one group of calibration coefficient.Use this coefficient that the WFI image is carried out relative radiant correction, proofread and correct the result and show that two array lap-joint aberration of WFI eliminate fully, but there is comparatively serious longitudinal stripe in image in some position in the array chip.The uniformity index that the WFI image is described is difficult to satisfy the calibration requirement, has equally error based on the calibration data of even landscape method.

Adopt method of the present invention to proofread and correct, proofread and correct the result and show that the inner striped of pattern matrix disappears substantially, aberration and the striped of left and right sides array are eliminated fully, and the image radiation correction mass is greatly improved, and solve preferably large format remote sensing images radiant correction problem.

There is the problem of light and shade transition in CCD camera B4 band image, and by the inventive method, the problems referred to above are equally effectively solved, and after revising the calibration coefficient profile, is difficult to find image light and shade transient.

Above-mentioned is an explanation of carrying out as example take CBERS satellite WFI and CCD camera, and the inventive method extends to other satellites or sensor.

The unspecified part of the present invention belongs to general knowledge as well known to those skilled in the art.

Claims (3)

1. relative radiometric calibration method in conjunction with Laboratory Calibration and even landscape statistics is characterized in that step is as follows:

(1) according to the Laboratory Calibration data that gather, obtains spoke brightness and the corresponding original densitometer numerical value of every grade of lamp, and ask for the calibration coefficient of each pixel;

(2) image that gathers in orbit according to satellite is chosen overall uniformly zone, generates the calibration coefficient of each pixel by the even landscape statistic law;

(3) result according to step (1), (2) unites calibration, and detailed process is as follows:

(3.1) associating pixel design: m pixel is combined into an associating pixel, and namely array is comprised of the associating pixel, and each associating pixel is on average obtained by m pixel, for the push-scanning image pattern, and 50＜m＜200;

(3.2) according to the result of calculation of step (1), ask for the response coefficient LGN of laboratory associating pixel _j

(3.3) according to the result of calculation of step (3), ask for the response coefficient HGN of ground even landscape associating pixel _j

(3.4) ask for associating pixel correction factor COR _j, COR _j=LGN _j/ HGN _j

(3.5) correction factor equation model: take associating pixel sequence number as horizontal ordinate, correction factor is ordinate, adopts second-degree parabola equation model correction factor curve;

(3.6) correction factor interpolation: the correction factor curve to above-mentioned match disperses, and obtains one group for the correction factor cor of single pixel _i

(3.7) new calibration coefficient nNG _i: adopt the result of calculation of step (1) or step (2) divided by correction factor, can obtain the new calibration coefficient of this array.

2. a kind of relative radiometric calibration method in conjunction with Laboratory Calibration and even landscape statistics according to claim 1, it is characterized in that: at first adopt even landscape statistic law, histogram equalization method, three kinds of methods of least square method to ask for the calibration coefficient of each pixel in the described step (1), then choose the result of calculation of three kinds of optimal algorithms in the algorithm as the result of calculation of step (1).

3. a kind of relative radiometric calibration method in conjunction with Laboratory Calibration and even landscape statistics according to claim 2, it is characterized in that: the first-selection of choosing of described optimal algorithm is chosen homogeneous area large as far as possible in the image, calculate gray average and this regional gray average of heading image, calculate the error of above-mentioned two gray averages, when error hour, this algorithm is optimal algorithm.

Images