CN102682430B - Non-uniform correction method combining spectrum and spatial information for high-spectral data - Google Patents

Abstract

The invention provides a non-uniform correction method combining a spectrum and spatial information for high-spectral data. The non-uniform correction method comprises the following six steps: step 1, the high-spectral data are read in; step 2, edges with non-uniform characteristics in each single-band image are calculated by using edge detection operators; step 3, image segmentation is realized by using the edges with the non-uniform characteristics, which are extracted in the step 2, so as to judge the position of a non-uniform response detector cell; the position of a reference detector cell is determined according to the position of the non-uniform response detector cell, which is obtained in the step 3, and a corresponding spectral dimension ratio is calculated; step 5, the correction coefficient of the non-uniform response detector cell is calculated according to the spectral dimension ratio of the reference detector cell, which is obtained in the step 4; and step 6, a non-uniform response pixel value is updated by using the correction coefficient, thereby realizing non-uniform correction. By adopting the non-uniform correction method, the non-uniform characteristics in the images can be separated more steadily, and a more reliable correction result can be obtained through combining the spectrum and the spatial information, so that the non-uniform correction method has a good practical value and a broad application prospect in the field of spectral image preprocessing.

Description

The non-homogeneous bearing calibration of high-spectral data of a kind of spectrum and spatial information combination

(1) technical field

The present invention relates to the non-homogeneous bearing calibration of high-spectral data of a kind of spectrum and spatial information combination, belong to the remote sensing technology application in engineering science technology, be applicable to high-spectral data pre-service.

(2) background technology

Hyperspectral imager is a kind of novel remote sensing load, its spectrum has tight, continuous feature, can record spectrum and the spatial information feature of tested atural object simultaneously, the material that originally can not survey in broadband remote sensing can be detected in high-spectrum remote-sensing.Due to the heterogeneity of detector cells (visiting unit) response, or the heterogeneity of multiple sensing circuits in electronic link, or radiation calibration introduce uncertain, the spatial domain of data all can present heterogencity, view data quality is reduced, and market demand ability weakens.

Current non-homogeneous bearing calibration is mainly to visit first hypothesis not changing with imaging time with the first statistical information amount of adjacent spy based on non-homogeneous, the statistical information amount that non-homogeneous is listed as to (OK) in image space territory is carried out certain computing, make it flux matched with the statistical information of reference picture, thereby but in the more complicated gradation of image of atural object situation pockety, the correction coefficient that the method calculates still comprises the contribution of atural object difference, can not embody the difference of visiting between unit completely, correction coefficient is carried out after non-homogeneous correction image accordingly, can produce in image space territory light and shade discontinuous " band effect " phenomenon.

The object that image is cut apart is image to be divided into the region of each tool characteristic and to extract interested target.Image cut apart in the most general method be that it depends on the image border of being found by image gradient based on the cutting apart of edge, these edges have indicated discontinuous position, picture characteristics aspect.If f (x, y) is input picture, its gradient vector at position (x, y) is defined as follows:

$\▿f=\left[\begin{array}{cc}\frac{\∂f}{\∂x}& \frac{\∂f}{\∂y}\end{array}\right]$

From vector analysis, the maximum rate of change direction of the f that gradient vector is pointed at coordinate (x, y), and the amplitude of gradient vector

provide

the maximum rate of change that in direction, after every increase unit distance, f (x, y) value increases.Easy for algorithm, easily realize, generally by each component of edge detection operator convolutional calculation gradient vector.But in non-homogeneous bearing calibration, the image result based on rim detection is discontinuous edge, must, in conjunction with the formation mechanism of non-homogeneous feature, discontinuity edge be merged into continuous boundary chain.

(3) summary of the invention

The object of this invention is to provide the non-homogeneous bearing calibration of high-spectral data of a kind of spectrum and spatial information combination, it has overcome the deficiency that existing bearing calibration is only carried out non-homogeneous correction from data space territory, effectively utilize the feature of high-spectral data collection of illustrative plates unification, suppressed the impact of complicated atural object, it is the non-homogeneous bearing calibration of high-spectral data that a kind of stability is strong, reliability is high, degree of accuracy is high.

Technical solution of the present invention is: based on Image Segmentation Theory, utilize the formation mechanism of directivity edge detection operator in conjunction with non-homogeneous feature in high-spectral data, realization character separates with background, determine that non-homogeneous visits first position, tie up ratio by spectrum and calculate non-homogeneous correction coefficient, finally upgrade by pixel the non-homogeneous correction that realizes high-spectral data.

The non-homogeneous bearing calibration of high-spectral data of a kind of spectrum of the present invention and spatial information combination, its step is as follows:

Step 1: the reading in of high-spectral data;

Step 2: the edge that is calculated non-homogeneous feature in each single band image by edge detection operator;

Step 3: the non-homogeneous edge feature being extracted by step 2 is realized image and cut apart, judges that the non-homogeneous of response visits first position;

Step 4: respond non-homogeneous by step 3 and visit first location positioning with reference to visiting first position, and calculate corresponding spectrum dimension ratio;

Step 5: visit the non-homogeneous of first spectrum dimension ratio calculated response by step 4 and visit first correction coefficient;

Step 6: realized the renewal of the non-homogenous pixel ' value of response by correction coefficient, realize heterogencity and proofread and correct.

Wherein, the edge detection operator described in step 2 is Sobel operator, is defined as follows:

$h_1=\left|\begin{array}{ccc}-1& 0& 1\\ -2& 0& 2\\ -1& 0& 1\end{array}\right|,$ $h_2=\left|\begin{array}{ccc}1& 2& 1\\ 0& 0& 0\\ -1& -2& -1\end{array}\right|$

In formula, h ₁, h ₂it is respectively the edge detection operator of vertical direction and horizontal direction.According to the edge detection operator of the non-homogeneous feature selecting respective direction in high-spectral data.

Wherein, the image of realizing described in step 3 is cut apart, and judges that responding non-homogeneous visits first position, and decision criteria is as follows:

In formula, Loc is the non-homogeneous of this spy unit response of 1 expression, is 0 expression response homogeneous; R is the number that is labeled as marginal point on non-homogeneous edge feature, and P is the number of maximum continuous boundary point; M is the line number of image; Res is the spatial resolution of data.

Wherein, described in step 4, refer to reference to visiting unit the homogeneous spy unit of visiting first arest neighbors with non-homogeneous, spectrum dimension ratio is obtained by the spoke brightness calculation of different-waveband:

$r(x_0,y_j,k_0,n)=\frac{L(x_0,y_j,k_0+n)}{L(x_0,y_j,k_0)}=[r(x_0,y_1,k_0,n),......,r(x_0,y_M,k_0,n)]$

$\stackrel{\‾}{r}({\stackrel{\‾}{x}}_0,y_j,k_0,n)=\frac{L({\stackrel{\‾}{x}}_0,y_j,k_0+n)}{L({\stackrel{\‾}{x}}_0,y_j,k_0)}=[\stackrel{\‾}{r}({\stackrel{\‾}{x}}_0,y_1,k_0,n),......,\stackrel{\‾}{r}({\stackrel{\‾}{x}}_0,y_M,k_0,n)]$

In formula, r (x ₀, y _j, k ₀, n) represent to visit (the x of unit ₀, k ₀) spectrum dimension ratio,

represent with reference to visiting unit

spectrum dimension ratio; L (x ₀, y _j, k ₀+ n), L (x ₀, y _j, k ₀) respectively represent visit unit (x ₀, k ₀) at k ₀the spoke brightness value of+n wave band, K-band,

represent respectively to visit unit

at k ₀the spoke brightness value of+n wave band, K-band, wherein y _jthe row sequence number of space dimension, j=1,2 ..., M, and M is the line number of image.

Wherein, described in step 5, asking for the non-homogeneous of response, to visit first correction factor calculation formula as follows:

$a(x_0,k_0)=\frac{\mathrm{Std}(x_0,k_0)}{\mathrm{Std}({\stackrel{\‾}{x}}_0,k_0)}$

$b(x_0,k_0)=\mathrm{Mean}(x_0,k_0)-\frac{\mathrm{Std}(x_0,k_0)}{\mathrm{Std}({\stackrel{\‾}{x}}_0,k_0)}\mathrm{Mean}({\stackrel{\‾}{x}}_0,k_0)$

In formula, a (x ₀, k ₀), b (x ₀, k ₀) be that non-homogeneous is visited the (x of unit ₀, k ₀) correction coefficient, respectively characterize gain and biasing; Std (x ₀, k ₀),

represent respectively non-homogeneous visit first spectrum dimension ratio, with reference to the standard deviation of visiting first spectrum dimension ratio; Mean (x ₀, k ₀),

represent respectively non-homogeneous visit first spectrum dimension ratio, with reference to the average of visiting first spectrum dimension ratio;

Wherein, respond the renewal of non-homogenous pixel ' value described in step 6, update mode is as follows:

L*(x ₀，y _j，k ₀)＝a(x ₀，k ₀)L(x ₀，y _j，k ₀)+b(x ₀，k ₀)

In formula, (x ₀, k ₀) represent that the non-homogeneous spy of response is first, L (x ₀, y _j, k ₀) expression k ₀wave band pixel (x ₀, y _j) spoke brightness before renewal, L ^*(x ₀, y _j, k ₀) be the spoke brightness after upgrading.

The present invention's advantage is compared with prior art: the limitation that has overcome the non-homogeneous bearing calibration of traditional high-spectral data and carry out from the single aspect of spatial domain data processing, spectrum and spatial information that this method has utilized high-spectral data to provide simultaneously, realized the correction of heterogencity.It has advantages of following: the theory that (1) is cut apart based on image, and the spectrum and the spatial information that have utilized high-spectral data to provide have suppressed the impact of the factors such as complicated atural object on non-homogeneous correction result simultaneously effectively, have improved the reliability of algorithm; (2) by cut apart the formation mechanism in conjunction with non-homogeneous feature in data with the edge of directional information, realization character separates with the effective of background.

(4) accompanying drawing explanation

Fig. 1 is FB(flow block) of the present invention

(5) embodiment

See Fig. 1, take the spaceborne high-spectral data of Hyperion as example, the non-homogeneous bearing calibration of the high-spectral data of a kind of spectrum of the present invention and spatial information combination, the method concrete steps are as follows:

Step 1: the reading in of high-spectral data: read in Hyperion high-spectral data;

Step 2: the edge that is calculated non-homogeneous feature in each single band image by edge detection operator;

Because Hyperion is dispersion push-broom type hyperspectral imager, non-homogeneous mark sheet, now along rail direction, selects the Sobel of vertical direction to detect son, is defined as follows:

$h=\left|\begin{array}{ccc}-1& 0& 1\\ -2& 0& 2\\ -1& 0& 1\end{array}\right|$

Step 3: the non-homogeneous edge feature being extracted by step 2 is realized image and cut apart, judges that the non-homogeneous of response visits first position; The non-homogeneous edge feature that step 2 is extracted, its number that is labeled as marginal point is R, and the number of maximum continuous boundary point is P, and the line number of Hyperion data is M, and spatial resolution is 30m, utilizes decision criteria to determine that non-homogeneous visits first position:

In formula, Loc is the non-homogeneous of this spy unit response of 1 expression, is 0 expression response homogeneous.

Step 4: respond non-homogeneous by step 3 and visit first location positioning with reference to visiting first position, and calculate corresponding spectrum dimension ratio;

Visit first position and with reference to visiting first position, calculate spectrum dimension ratio according to the non-homogeneous of determining:

$r(x_0,y_j,k_0,n)=\frac{L(x_0,y_j,k_0+n)}{L(x_0,y_j,k_0)}=[r(x_0,y_1,k_0,n),......,r(x_0,y_M,k_0,n)]$

$\stackrel{\‾}{r}({\stackrel{\‾}{x}}_0,y_j,k_0,n)=\frac{L({\stackrel{\‾}{x}}_0,y_j,k_0+n)}{L({\stackrel{\‾}{x}}_0,y_j,k_0)}=[\stackrel{\‾}{r}({\stackrel{\‾}{x}}_0,y_1,k_0,n),......,\stackrel{\‾}{r}({\stackrel{\‾}{x}}_0,y_M,k_0,n)]$

In formula, r (x ₀, y _j, k ₀, n) represent to visit (the x of unit ₀, k ₀) spectrum dimension ratio,

represent with reference to visiting unit

spectrum dimension ratio; ; L (x ₀, y _j, k ₀+ n), L (x ₀, y _j, k ₀) respectively represent visit unit (x ₀, k ₀) at k ₀the spoke brightness value of+n wave band, K-band, represent respectively to visit unit

at k ₀the spoke brightness value of+n wave band, K-band, wherein y _jthe row sequence number of space dimension, j=1,2 ..., M, and M is the line number of image.

Step 5: visit the non-homogeneous of first spectrum dimension ratio calculated response by step 4 and visit first correction coefficient;

If Std is (x ₀, k ₀), represent respectively non-homogeneous visit first spectrum dimension ratio, with reference to the standard deviation of visiting first spectrum dimension ratio, Mean (x ₀, k ₀),

represent respectively non-homogeneous visit first spectrum dimension ratio, with reference to the average of visiting first spectrum dimension ratio, non-homogeneous is visited the (x of unit ₀, k ₀) correction coefficient be:

$a(x_0,k_0)=\frac{\mathrm{Std}(x_0,k_0)}{\mathrm{Std}({\stackrel{\‾}{x}}_0,k_0)}$

$b(x_0,k_0)=\mathrm{Mean}(x_0,k_0)-\frac{\mathrm{Std}(x_0,k_0)}{\mathrm{Std}({\stackrel{\‾}{x}}_0,k_0)}\mathrm{Mean}({\stackrel{\‾}{x}}_0,k_0)$

In formula, a (x ₀, k ₀), b (x ₀, k ₀) characterize gain and biasing respectively;

Step 6: realized the renewal of the non-homogenous pixel ' value of response by correction coefficient, realize heterogencity and proofread and correct;

Visit (the x of unit if respond non-homogeneous ₀, k ₀), L (x ₀, y _j, k ₀) be k ₀wave band pixel (x ₀, y _j) spoke brightness value, the spoke brightness value after renewal is L* (x ₀, y _j, k ₀), update mode is as follows:

L*(x ₀，y _j，k ₀)＝a(x ₀，k ₀)L(x ₀，y _j，k ₀)+b(x ₀，k ₀)。

Claims (3)

1. the non-homogeneous bearing calibration of the high-spectral data of spectrum and spatial information combination, is characterized in that: the method concrete steps are as follows:

Step 1: the reading in of high-spectral data;

Step 2: the edge that is calculated non-homogeneous feature in each single band image by edge detection operator;

Step 3: the non-homogeneous edge feature being extracted by step 2 is realized image and cut apart, judges that the non-homogeneous of response visits

The position of unit;

Step 4: respond non-homogeneous by step 3 and visit first location positioning with reference to visiting first position, and calculate corresponding light

Spectrum dimension ratio;

Step 5: visit the non-homogeneous of first spectrum dimension ratio calculated response by step 4 and visit first correction coefficient;

Step 6: realized the renewal of the non-homogenous pixel ' value of response by correction coefficient, realize heterogencity and proofread and correct;

Wherein, the image of realizing described in step 3 is cut apart, and judges that responding non-homogeneous visits first position, and decision criteria is as follows:

In formula, Loc is the non-homogeneous of this spy unit response of 1 expression, is 0 expression response homogeneous; R is the number that is labeled as marginal point on non-homogeneous edge feature, and P is the number of maximum continuous boundary point; M is the line number of image; Res is the spatial resolution of data;

Wherein, described in step 4, refer to reference to visiting unit the homogeneous spy unit of visiting first arest neighbors with non-homogeneous, spectrum dimension ratio is obtained by the spoke brightness calculation of different-waveband:

$r(x_0,y_j,k_0,n)=\frac{L(x_0,y_j,k_0+n)}{L(x_0,y_j,k_0)}$

$\stackrel{\‾}{r}({\stackrel{\‾}{x}}_0,y_j,k_0,n)=\frac{L({\stackrel{\‾}{x}}_0,y_j,k_0+n)}{L({\stackrel{\‾}{x}}_0,y_j,k_0)}$

In formula, r (x ₀, y _j, k ₀, n) represent to visit (the x of unit ₀, k ₀) spectrum dimension ratio,

represent with reference to visiting unit spectrum dimension ratio; L (x ₀, y _j, k ₀+ n), L (x ₀, y _j, k ₀) respectively represent visit unit (x ₀, k ₀) at k ₀+ n wave band, k ₀the spoke brightness value of wave band,

represent respectively to visit unit

at k ₀+ n wave band, k ₀the spoke brightness value of wave band, wherein y _jthe row sequence number of space dimension, j=1,2 ..., M, and M is the line number of image;

Wherein, in step 5, asking for the non-homogeneous of response, to visit first correction factor calculation formula as follows:

$a(x_0,k_0)=\frac{\mathrm{Std}(x_0,k_0)}{\mathrm{Std}({\stackrel{\‾}{x}}_0,k_0)}$

$b(x_0,k_0)=\mathrm{Mean}(x_0,k_0)-\frac{\mathrm{Std}(x_0,k_0)}{\mathrm{Std}({\stackrel{\‾}{x}}_0,k_0)}\mathrm{Mean}({\stackrel{\‾}{x}}_0,k_0)$

In formula, a (x ₀, k ₀), b (x ₀, k ₀) be that non-homogeneous is visited the (x of unit ₀, k ₀) correction coefficient, respectively characterize gain and biasing; Std (x ₀, k ₀), represent respectively non-homogeneous visit first spectrum dimension ratio, with reference to the standard deviation of visiting first spectrum dimension ratio; Mean (x ₀, k ₀), represent respectively non-homogeneous visit first spectrum dimension ratio, with reference to the average of visiting first spectrum dimension ratio.

2. the non-homogeneous bearing calibration of the high-spectral data of a kind of spectrum according to claim 1 and spatial information combination, is characterized in that: the edge detection operator described in step 2 is Sobel operator, is defined as follows:

$h_1=\left|\begin{array}{ccc}-1& 0& 1\\ -2& 0& 2\\ -1& 0& 1\end{array}\right|,h_2=\left|\begin{array}{ccc}1& 2& 1\\ 0& 0& 0\\ -1& -2& -1\end{array}\right|$

In formula, h ₁, h ₂it is respectively the edge detection operator of vertical direction and horizontal direction; According to the edge detection operator of the non-homogeneous feature selecting respective direction in high-spectral data.

3. the non-homogeneous bearing calibration of the high-spectral data of a kind of spectrum according to claim 1 and spatial information combination, is characterized in that: described in step 6, respond the renewal of non-homogenous pixel ' value, update mode is as follows:

L ^*(x ₀,y _j,k ₀)=a(x ₀,k ₀)L(x ₀,y _j,k ₀)+b(x ₀,k ₀)

In formula, (x ₀, k ₀) represent that the non-homogeneous spy of response is first, L (x ₀, y _j, k ₀) expression k ₀wave band pixel (x ₀, y _j) spoke brightness before renewal, L ^*(x ₀, y _j, k ₀) be the spoke brightness after upgrading; A (x ₀, k ₀), b (x ₀, k ₀) be that non-homogeneous is visited the (x of unit ₀, k ₀) correction coefficient, respectively characterize gain and biasing.

Images