CN103218779A - Method for detecting and correcting hyperspectral data dead pixels of interference imaging spectrometer - Google Patents

Abstract

The invention discloses a method for detecting and correcting hyperspectral data dead pixels of an interference imaging spectrometer. The method comprises the following steps of: selecting a pixel serving as a point to be judged, and calculating a spectral angle and an Euclidean distance between the spectrum of the point to be judged and the mid-value spectrum of an adjacent pixel till all pixels are traversed and calculated and the spectral angle matrix and Euclidean distance matrix of all the pixels are obtained; calculating the threshold value of the spectral angle matrix and/or the Euclidean distance matrix respectively, and judging the dead pixels of all the pixels to generate a dead pixel marking matrix; and correcting dead pixels in the dead pixel marking matrix. According to the method, the threshold value has higher accuracy, and the detection and correction efficiency of dead pixels are increased.

Description

The bad point of a kind of high-spectral data of inteference imaging spectrometer detects and modification method

Technical field

The invention belongs to the high-spectrum remote sensing data process field, the bad point of high-spectral data that relates in particular to a kind of inteference imaging spectrometer detects and modification method.

Background technology

The lunar exploration satellite Chang'e I (CE-1) of first independent research of China launched on October 24th, 2007, after going through 495 days lunar orbiting exploration, completed successfully on March 1st, 2009 and to hit the moon, finished to survey mission.One of eight big useful load that inteference imaging spectrometer (IIM) carries as CE-1 are being born and are being obtained the moonscape spectral information and analyze the scientific goal that the menology material is formed.

IIM science data collection is responsible for handling and distribution by Nation Astronomical Observatory lunar orbiting exploration engineering Ground Application system, but uses free download per family.Can be 2 grades of data for the IIM data of user's download at present, be the spoke brightness data, also need further processing could satisfy the needs of scientific research.Because there are bad some phenomenon in moon exploration circumstance complication, IIM image data, the existence of bad point has increased the variance of image, can have influence on the subsequent treatment of IIM data, thus need detect and revise bad point, to obtain the higher high-spectral data of precision.

For the detection and the correction of the bad point of remote sensing image, mainly be to utilize point to be determined and adjacent picture elements difference spatially in the past, judged bad point by respective threshold is set.But IIM high-spectral data wave band is many, as utilizes the bad point of tradition decision method, and the threshold value setting is comparatively complicated.Data difference is big between each wave band of IIM, and the IIM data coverage is wide, and it is various to contain landform, is difficult to find a threshold value to be applicable to all wave bands; As pursuing the band setting threshold value, defining of threshold value takes time and effort, and precision is low, and these factors have greatly limited bad point and judged and the efficient of revising.

Summary of the invention

The present invention has overcome that the efficient that detects bad point in the prior art is lower, the calculated threshold precision is low and defective such as required time is long, and the bad point of high-spectral data that has proposed a kind of inteference imaging spectrometer detects and modification method.

The bad point of high-spectral data that the present invention proposes a kind of inteference imaging spectrometer detects and modification method, may further comprise the steps:

Step 1: choose a pixel as point to be determined, calculate spectrum angle and Euclidean distance between described to be determined spectrum and the contiguous pixel intermediate value spectrum, and choose next pixel and calculate as point to be determined, calculate spectrum angular moment battle array and the Euclidean distance matrix that obtains all pixels until all pixels of traversal;

Step 2: the threshold value of calculating described spectrum angular moment battle array and/or Euclidean distance matrix is respectively gone bad some judgement to described all pixels, generates bad some mark matrix;

Step 3: the bad point in the described bad some mark matrix is revised.

Wherein, further comprise in the described step 1: be lower than 18 wave band data by signal to noise ratio (S/N ratio) in the described high-spectral data of wave band investigation eliminating.

Wherein, described intermediate value spectrum is the intermediate value of the spectral value of the point non-to be determined in the window area that is the center with described point to be determined on each wave band.

Wherein, the following expression of computing formula at described spectrum angle:

$\cos \mathrm{\α}=\frac{A*B}{\left|A\right|*\left|B\right|}=\frac{{\mathrm{\Σ}}_{i=1}^N{A}_i*B_i}{\sqrt{{\mathrm{\Σ}}_{i=1}^N{A}_i*A_i}*\sqrt{{\mathrm{\Σ}}_{i=1}^N{B}_i*B_i}};$

In the formula, α represents the spectrum angle, and A and B represent two curves of spectrum respectively, and N represents wave band number, A _iAnd B _iBe illustrated respectively in the spectral value of i wave band.

Wherein, the following expression of the computing formula of described Euclidean distance:

$d=\sqrt{\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{(x_{i1}-x_{i2})}^2};$

In the formula, d represents Euclidean distance, x _I1Expression article one curve of spectrum is at the spectral value of i wave band, x _I2The expression second curve of spectrum is at the spectral value of i wave band, and N represents the wave band number.

Wherein, in the described step 2, the process that generates bad some mark matrix comprises:

Steps A 1: described spectrum angular moment battle array or described Euclidean distance matrix are got subclass with the moving window method, calculate the intermediate value of described subclass;

Steps A 2: calculate in the described subclass each element and the absolute value of the difference between the intermediate value of described subclass;

Steps A 3:, calculate the intermediate value of the absolute value of described difference according to the absolute value of described difference;

Steps A 4:, calculate and obtain threshold value according to the intermediate value of the absolute value of described difference;

Steps A 5: judge according to described threshold value, if the element in the described subclass judges that then described element is a bad point, otherwise be judged to be normal point that judgement finishes the back and generates bad some mark matrix greater than described threshold value.

Wherein, the following expression of the computing formula of described threshold value:

Threshold＝β*Med_AbsDev_A；

In the formula, Threshold represents threshold value, and Med_AbsDev_A represents the intermediate value of absolute value of the difference of subclass A, and β represents to adjust coefficient.

Wherein, in the process of in the described step 3 of execution in step, described bad point being revised, choose the modified value of the mean value of non-bad some pixel in the described bad some region as described bad point; The following expression of the computing formula of described makeover process:

$R_{\mathrm{bp}}=\frac{{\mathrm{\Σ}}_{i=1}^n{R}_i}{n};$

In the formula, R _BpThe back spectral value is revised in expression, non-bad some number in the scope around n represents at bad, R _iBe non-bad some pixel value.

Utilization of the present invention technical scheme be, utilize the principle of vector, the multidimensional spectroscopic data of high spectrum is mapped in the geometry hyperspace, and promptly the spectral information of certain point is corresponding to a point in the geometry hyperspace on the space, and this point can constitute a non-vanishing vector with the initial point in the hyperspace.Go bad a little judgement by the difference between the intermediate value spectrum of measuring spectrum to be determined and contiguous pixel among the present invention, decision process is mapped as two points in the hyperspace for the intermediate value spectrum with the spectrum of point to be determined and contiguous pixel, these two points respectively with hyperspace in initial point constitute two vectors, judge by calculating the angle (this angle is called as the spectrum angle in spectroscopy) and the Euclidean distance of two points in hyperspace of two vectors in hyperspace whether point to be determined is bad point.The present invention has higher efficient for the method that bad point detects, and the IIM high-spectral data has 32 wave bands, adopt tradition to judge that by wave band bad point methods need be provided with 32 threshold values, and this method only need be provided with 2 threshold values, and efficient improves obviously; And bad point judges that precision is higher, and False Rate has reduced by 50% at least.

Description of drawings

Fig. 1 represents that the bad point of the high-spectral data of inteference imaging spectrometer detects and the process flow diagram of modification method;

Fig. 2 represents the rail image thumbnail and the segment map detail drawing of inteference imaging spectrometer among the embodiment;

Fig. 3 represents to utilize rail image data thumbnail and the segment map detail drawing among the revised Fig. 2 of the present invention.

Embodiment

In conjunction with following specific embodiments and the drawings, the present invention is described in further detail.Implement process of the present invention, condition, experimental technique etc., except that the following content of mentioning specially, be the universal knowledege and the common practise of this area, the present invention is not particularly limited content.

As shown in Figure 1, the bad point of the high-spectral data of inteference imaging spectrometer of the present invention detects and modification method, may further comprise the steps:

Step 1: choose a pixel as point to be determined, calculate spectrum angle and Euclidean distance between to be determined spectrum and the contiguous pixel intermediate value spectrum, and choose next pixel and calculate as point to be determined, calculate spectrum angular moment battle array and the Euclidean distance matrix that obtains all pixels until all pixels of traversal;

In the step 1,, can not use,, only consider 6-31 totally 26 wave bands so do not consider these wave bands in the algorithm implementation process because IIM (inteference imaging spectrometer) high-spectral data is bigger at 1-5 wave band and the 32nd band noise.After the wave band investigation, calculate the high spectrum angular data of 26 wave bands, with point to be determined is the center, adopt the 5*5 moving window, calculate spectrum angle and Euclidean distance between the intermediate value spectrum of other pixels in to be determined spectrum and the window, the value of intermediate value spectrum on each wave band equals the intermediate value of other pixel spectral values in this wave band moving window, calculates the spectrum angle and the Euclidean distance of point to be determined and intermediate value spectrum, and the computing formula at spectrum angle is as follows:

$\cos \mathrm{\α}=\frac{A*B}{\left|A\right|*\left|B\right|}=\frac{{\mathrm{\Σ}}_{i=1}^N{A}_i*B_i}{\sqrt{{\mathrm{\Σ}}_{i=1}^N{A}_i*A_i}*\sqrt{{\mathrm{\Σ}}_{i=1}^N{B}_i*B_i}};$

Wherein, α represents the spectrum angle, and A and B represent two curves of spectrum respectively, and N represents wave band number, A _iAnd B _iBe illustrated respectively in the spectral value of i wave band.

The following expression of the computing formula of Euclidean distance:

$d=\sqrt{\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{(x_{i1}-x_{i2})}^2};$

In the formula, d represents Euclidean distance, x _I1Expression article one curve of spectrum is at the spectral value of i wave band, x _I2The expression second curve of spectrum is at the spectral value of i wave band, and N represents the wave band number.

Calculate the spectrum angle and Euclidean distance of a pixel of acquisition according to above-mentioned steps after, choose next pixel and calculate, behind all pixels of traversal, calculate the spectrum angular moment battle array and the Euclidean distance matrix of all pixels as point to be determined

Step 2: the threshold value of calculating spectrum angular moment battle array and/or Euclidean distance matrix is respectively gone bad some judgement to all pixels, generates bad some mark matrix.

In step 2, spectrum angular moment battle array and Euclidean distance matrix are got subclass with the 8*8 moving window carry out computing, suppose that subclass is A.

Steps A 1: spectrum angular moment battle array and Euclidean distance matrix are got subclass A with the moving window method, the intermediate value Med_A of subset of computations;

Steps A 2: in the subset of computations each element and the absolute value AbsDev_A of the difference between the intermediate value Med_A of subclass;

Steps A 3: according to the absolute value AbsDev_A of difference, the intermediate value Med_AbsDev_A of the absolute value of calculated difference;

Steps A 4: the intermediate value Med_AbsDev_A according to the absolute value of difference, calculate and obtain threshold value Threshold; The following expression of the computing formula of threshold value:

Threshold＝β*Med_AbsDev_A；

In the formula, Threshold represents threshold value, and Med_AbsDev_A represents the intermediate value of absolute value of the difference of subclass A, and β represents to adjust coefficient.

Steps A 5: judge Threshold according to threshold value, greater than threshold value, then decision element is a bad point, otherwise is judged to be normal point as if the element in the subclass.

Step 3: the bad point in the bad some mark matrix is revised.

Utilize the bad some mark matrix that step 2 calculates, bad point pursue the wave band correction, be specially and adopt around the bad point that the average of pixel substitutes bad some initial value in the 5*5 scope, to the influence of average, adopt following formula correction when avoiding occur continuously at bad:

$R_{\mathrm{bp}}=\frac{{\mathrm{\Σ}}_{i=1}^n{R}_i}{n}$

Wherein, R _BpFor revising the back spectral value, n is non-bad some number in the 5*5 scope around the bad point, R _iBe non-bad some pixel value.

Present embodiment adopts CE-1IIM2A level data the 2613rd rail image, and to be determined with these rail image the 76th row, the pixel of the 275th row is an example.The IIM data can be obtained from Nation Astronomical Observatory lunar orbiting exploration engineering Ground Application system.The 2613rd rail data are referring to Fig. 2, and Fig. 2 shows 2613 rail data thumbnails and part detail drawing.The CE-1IIM partial parameters sees Table 1.

Table 1 inteference imaging spectrometer IIM important technological parameters

The bad point of IIM high-spectral data detects and revises process flow diagram referring to Fig. 1, and specific implementation process is as follows:

These data are carried out the wave band investigation, because the IIM high-spectral data is lower than 18 at 1-5 wave band and the 32nd wave band signal to noise ratio (S/N ratio), noise is bigger, can not use, so do not consider these wave bands in the algorithm implementation process, only considers 6-31 totally 26 wave bands.

Calculating the spectrum angular data, is the center with point to be determined, adopts the 5*5 moving window, calculates spectrum angle and Euclidean distance between the intermediate value spectrum of interior other pixels of to be determined spectrum and window.The value of intermediate value spectrum on each wave band equals the intermediate value of other pixel spectral values in this wave band moving window.

For example, with the row of the 76th in the IIM2613 rail data, the 275th row pixel is a point to be determined, calculate the spectrum angle and the Euclidean distance of this point and intermediate value spectrum, this to be determined some spectral sequence is (0.17961,0.17969,0.06784,0.02099,0.06118,0.03497,0.07714,0.13898,0.15992,0.06382,0.02031,0.03634,0.04539,0.00076,0.01389,0.05145,0.10627,0.07615,0.06551,0.07539,0.03543,0.02360,0.02928,0.03907,0.01836,0.03687); The moving window of this moment is 74 to 78 row, 273 to 277 row, and the intermediate value spectral sequence calculates and is (0.01369,0.01457,0.01547,0.02099,0.01900,0.01709,0.01774,0.02108,0.01947,0.01893,0.01878,0.01917,0.01715,0.01813,0.01720,0.01732,0.01689,0.01614,0.01569,0.01643,0.01523,0.01525,0.01482,0.01306,0.01371,0.00685); The spectrum angle that calculates between this point to be determined and the intermediate value spectrum according to spectrum angle and Euclidean distance formula is 38.3536, and Euclidean distance is 0.3482.

The spectrum angular moment battle array that calculates and Euclidean distance matrix are got subclass with the 8*8 moving window carry out computing, suppose that subclass is A.Subclass A gets the row of the 76th in the 2613 rail data herein, the 275th row pixel place moving window, and this moment, moving window was the 73-80 row, 273-280 is capable, calculating is example with the spectrum angle, and subclass A value is (2.00877,2.13469,2.85821,1.90362,2.74003,2.56796,2.97947,2.42689,1.85983,1.95835,2.55956,1.06091,1.96543,1.47618,1.62136,2.78853,2.16193,1.96354,2.0211,38.3536,2.9071,2.36521,1.37416,1.98574,2.34368,2.67034,1.84917,1.56398,2.28508,1.79559,2.28705,1.43994,1.84981,2.01596,2.36645,1.92966,2.98282,1.68864,1.9438,2.07424,2.24394,1.8344,1.47605,1.8426,2.18131,2.27245,1.64794,2.09667,1.59704,1.78037,2.51855,1.58979,3.52342,2.55167,1.40165,2.52964,1.27871,1.63793,1.70754,2.3643,2.46514,1.70353,1.73404,2.66998).

1) the intermediate value Med_A=2.0123 of subset of computations A;

2) the absolute value AbsDev_A=(0.00360,0.12233,0.84585,0.10875,0.72766 of the difference of subset of computations A and intermediate value Med_A, 0.55560,0.96710,0.41452,0.15253,0.05402,0.54720,0.95146,0.04693,0.53618,0.39100,0.77617,0.14957,0.04883,0.00874,36.34120,0.89474,0.35285,0.63821,0.02663,0.33132,0.65798,0.16319,0.44838,0.27271,0.21678,0.27468,0.57242,0.16256,0.00360,0.35409,0.08271,0.97045,0.32372,0.06856,0.06188,0.23157,0.17796,0.53632,0.16976,0.16895,0.26009,0.36443,0.08431,0.41533,0.23199,0.50619,0.42257,1.51106,0.53931,0.61071,0.51728,0.73366,0.37443,0.30482,0.35194,0.45277,0.30884,0.27832,0.65761);

3) the intermediate value Med_AbsDev_A=0.3524 of calculating AbsDev_A;

4) obtain judgment threshold, Threshold=β * Med_AbsDev_A, for the spectrum angle, β gets 9; For Euclidean distance, β gets 80, at this moment Threshold=3.1715;

5) compare the value of each element among the subclass A and the size of Threshold, if greater than Threshold, then be labeled as bad point, otherwise be normal point.Must badly put the mark matrix and be this moment:

(0，0，0，0，0，0，0，0，0，0，0，0，0，0，0，0，0，0，0，1，0，0，0，0，0，0，0，0，0，0，0，0，0，0，0，0，0，0，0，0，0，0，0，0，0，0，0，0，0，0，0，0，1，0，0，0，0，0，0，0，0，0，0，0)。

By bad some mark matrix as can be known, the point to be determined of the 76th row 275 row is a bad point.Adopt said method all pixels all to be gone bad some judgement, thereby obtain the bad some mark matrix of all pixels with the 8*8 moving window.

In conjunction with the bad some mark matrix that obtains, utilize the average method of substitution that the bad point in all pixels is revised, promptly adopt the average of the interior pixel of 5*5 scope around bad to substitute a bad initial value, use following formula to revise, makeover process carries out all wave bands of IIM:

$R_{\mathrm{bp}}=\frac{{\mathrm{\Σ}}_{i=1}^n{R}_i}{n}$

For example, spectrum is (0.01935,0.02027,0.015841 after the decision-point correction of revised the 76th row of following formula 275 row, 0.01635,0.01457,0.01369,0.01475,0.01521,0.02061,0.01897,0.01716,0.01744,0.02056,0.01922,0.01862,0.01863,0.01864,0.01684,0.01811,0.01711,0.01733,0.01655,0.01596,0.01553,0.01626,0.01498,0.01496,0.01449,0.01284,0.01347,0.00674,0.00466).

What Fig. 2 showed is inteference imaging spectrometer rail image thumbnail and segment map detail drawing before revising.Revised design sketch is referring to Fig. 3.Can find by comparison diagram 2 and Fig. 3, the arithmetic accuracy height, bad point is all correctly revised among the figure.

Protection content of the present invention is not limited to above embodiment.Under the spirit and scope that do not deviate from inventive concept, variation and advantage that those skilled in the art can expect all are included among the present invention, and are protection domain with the appending claims.

Claims (8)

1. the bad point of the high-spectral data of an inteference imaging spectrometer detects and modification method, it is characterized in that, may further comprise the steps:

Step 1: choose a pixel as point to be determined, calculate spectrum angle and Euclidean distance between described to be determined spectrum and the contiguous pixel intermediate value spectrum, and choose next pixel and calculate as point to be determined, calculate spectrum angular moment battle array and the Euclidean distance matrix that obtains all pixels until all pixels of traversal;

Step 2: the threshold value of calculating described spectrum angular moment battle array and/or Euclidean distance matrix is respectively gone bad some judgement to described all pixels, generates bad some mark matrix;

Step 3: the bad point in the described bad some mark matrix is revised.

2. the bad point of the high-spectral data of inteference imaging spectrometer as claimed in claim 1 detects and modification method, it is characterized in that, further comprises in the described step 1: be lower than 18 wave band data by signal to noise ratio (S/N ratio) in the described high-spectral data of wave band investigation eliminating.

3. the bad point of the high-spectral data of inteference imaging spectrometer as claimed in claim 1 detects and modification method, it is characterized in that described intermediate value spectrum is the intermediate value of the spectral value of the point non-to be determined in the window area that is the center with described point to be determined on each wave band.

4. the bad point of the high-spectral data of inteference imaging spectrometer as claimed in claim 1 detects and modification method, it is characterized in that the following expression of computing formula at described spectrum angle:

$\cos \mathrm{\α}=\frac{A*B}{\left|A\right|*\left|B\right|}=\frac{{\mathrm{\Σ}}_{i=1}^N{A}_i*B_i}{\sqrt{{\mathrm{\Σ}}_{i=1}^N{A}_i*A_i}*\sqrt{{\mathrm{\Σ}}_{i=1}^N{B}_i*B_i}};$

In the formula, α represents the spectrum angle, and A and B represent two curves of spectrum respectively, and N represents wave band number, A _iAnd B _iBe illustrated respectively in the spectral value of i wave band.

5. the bad point of the high-spectral data of inteference imaging spectrometer as claimed in claim 1 detects and modification method, it is characterized in that the following expression of the computing formula of described Euclidean distance:

$d=\sqrt{\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{(x_{i1}-x_{i2})}^2};$

In the formula, d represents Euclidean distance, x _I1Expression article one curve of spectrum is at the spectral value of i wave band, x _I2The expression second curve of spectrum is at the spectral value of i wave band, and N represents the wave band number.

6. the bad point of the high-spectral data of inteference imaging spectrometer as claimed in claim 1 detects and modification method, it is characterized in that, in the described step 2, the process that generates bad some mark matrix comprises:

Steps A 1: described spectrum angular moment battle array or described Euclidean distance matrix are got subclass with the moving window method, calculate the intermediate value of described subclass;

Steps A 2: calculate in the described subclass each element and the absolute value of the difference between the intermediate value of described subclass;

Steps A 3:, calculate the intermediate value of the absolute value of described difference according to the absolute value of described difference;

Steps A 4:, calculate and obtain threshold value according to the intermediate value of the absolute value of described difference;

Steps A 5: judge according to described threshold value, if the element in the described subclass judges that then described element is a bad point, otherwise be judged to be normal point that judgement finishes the back and generates bad some mark matrix greater than described threshold value.

7. the bad point of the high-spectral data of inteference imaging spectrometer as claimed in claim 6 detects and modification method, it is characterized in that the following expression of the computing formula of described threshold value:

Threshold＝β*Med_AbsDev_A；

In the formula, Threshold represents threshold value, and Med_AbsDev_A represents the intermediate value of absolute value of the difference of subclass A, and β represents to adjust coefficient.

8. the bad point of the high-spectral data of inteference imaging spectrometer as claimed in claim 1 detects and modification method, it is characterized in that, in the process of in the described step 3 of execution in step, described bad point being revised, choose the modified value of the mean value of non-bad some pixel in the described bad some region as described bad point; The following expression of the computing formula of described makeover process:

$R_{\mathrm{bp}}=\frac{{\mathrm{\Σ}}_{i=1}^n{R}_i}{n};$

In the formula, R _BpThe back spectral value is revised in expression, non-bad some number in the scope around n represents at bad, R _iBe non-bad some pixel value.

Images