CN103714546A - Data processing method of imaging spectrometer - Google Patents

Abstract

The invention discloses a data processing method of a temporally-spatially modulated interference imaging spectrometer. The data processing method comprises the following steps that firstly, the spatio-temporally modulated interference imaging spectrometer is used for obtaining a series of image sequences of an interested target point; secondly, the position of the interested target point in a continuous push-scan interference image is obtained by means of image registration; thirdly, the interference strength of the interested target point corresponding to a certain optical path difference is obtained; fourthly, an interference image, composed of the interference strength of different optical path differences, of the interested target point is obtained; fifthly, the interference image of the interested target point is recovered by means of non-uniform Fourier transformation so as to obtain the spectrum of the interested target point. The data processing method is suitable for the airborne temporally-spatially modulated interference imaging spectrometer, effectively solves the problem that it is difficult for an existing airborne stabilized platform to meet the requirement for platform stability of the temporally-spatially modulated interference imaging spectrometer, and enables the temporally-spatially modulated interference imaging spectrometer to be capable of being effectively applied to the airborne platform.

Description

A kind of data processing method of imaging spectrometer

Technical field

The present invention relates to image processing techniques and signal processing technology, belong to remote sensing image data processing technology field, be specifically related to a kind of data processing method of imaging spectrometer.

Background technology

Space-time unite interferometric modulator imaging spectrometer is as a kind of novel imaging spectrometer, has that spectral resolution is high, signal to noise ratio (S/N ratio) is high, an advantage such as high flux and high stability.The utilization of space-time unite interferometric modulator imaging spectrometer pushes away the mode of sweeping and scans object scene, not in the same time target imaging at detector diverse location, thereby obtain the complete point interference of impact point, then by Fourier transform, restore by point interference the spectrum that produces this target.Because space-time unite interferometric modulator imaging spectrometer is by pushing away inswept journey to obtain the complete interferogram of target to target, once the attitude of carrying platform changes in pushing away inswept journey, as sidewinder, pitching, driftage etc., will cause the interferogram distortion of traditional method for extracting, restore the spectrum and just cannot reflect the real property of scenery target.In order to realize high-precision spectrum recovering, after need proofreading and correct target location, carry out spectrum recovering.

Data processing method for spaceborne space-time unite imaging spectrometer mainly comprises based on the relevant image registration correcting algorithm relevant to normalizing eliminate indigestion of phase place.According to the rotation translation feature of Fourier transform, by the phase place related function between computed image, utilize FFT to realize the high-precision correction of image sequence rotation distortion, employing realizes the correction of image translation distortion based on the relevant method of normalizing eliminate indigestion, finally carry out point interference nonuniform sampling, Fourier transform recovery spectrum.The method is not suitable for the correction that attitude changes comparatively violent airborne imaging spectrum instrument image, there will be ground object target point to extract inaccurate and cannot accurately restore atural object spectrum information.

Document [1]: Yann Ferrec and et al., " Experimental results from an airborne static Fourier transform imaging spectrometer; " Applied Optics, 50 (30), 2011. utilize phase place method relevant and characteristic matching to carry out image registration determines target location, interferogram is interpolated to uniformly-spaced interferogram in spatial domain, directly carries out Fourier transform and restore the spectrum picture that obtains target.But characteristic matching method for registering can only utilize the region that interference fringe is inviolent.

Document [2]: L.Su, Y.Yuan, and et al., " Spectrum Reconstruction Method for Airborne Temporally-Spatially Modulated Fourier Transform Imaging Spectrometers; " IEEE Transactions on Geoscience and Remote Sensing, 2013, utilize data that POS measures to follow the trail of pushing away continuously the position of sweeping image target, and then carry out the extraction of target interferogram and recovery.But, need to install related data is provided with POS system combination, there is alignment error and need and correct, require POS and imaging spectrometer to synchronize and carry out when data acquisition, increased system applies difficulty.

Summary of the invention

To the object of the invention is in order addressing the above problem, to propose a kind of interferogram that is applicable to the based target location tracking of space-time unite interferometric modulator imaging spectrometer and extract and spectrum recovering method.

A data processing method for space-time unite interferometric modulator imaging spectrometer, comprises following step:

Step 1: by space-time unite interferometric modulator imaging spectrometer, obtain the image that comprises interesting target point;

Step 2: by image registration, obtain interesting target point A in the position that pushes away continuously the interference image of sweeping;

Step 3: the interference strength that obtains the corresponding a certain optical path difference of interesting target point;

Step 4: obtain the interferogram that interesting target point is comprised of the interference strength of different optical path differences;

Step 5: the interferogram of interesting target point is restored to the spectrum that obtains this interesting target point by Nonuniform fast Fourier transform.

The invention has the advantages that:

(1) the present invention is based on displacement relation between image and carry out registration, without introducing POS system surveying instrument attitude data, assist, reduced the complexity of system;

(2) method proposed by the invention is applicable to airborne space-time unite interferometric modulator imaging spectrometer, effectively overcome current airborne stable platform difficulty and reached the difficult point that space-time unite interferometric modulator imaging spectrometer requires platform stable, made space-time unite interferometric modulator imaging spectrometer can effectively be applied to airborne platform;

(3) method for registering images that the present invention adopts is the displacement relation between estimated image accurately, and reaches sub-pixel; Algorithm based on Fourier's frequency domain more can effectively suppress interference fringe and the impact of noise on registration effect than the algorithm based on characteristic matching; Bearing calibration in registration can filter out the unusual registration value producing in traditional method for registering and proofread and correct.

Accompanying drawing explanation

Fig. 1 is method flow schematic diagram of the present invention.

Fig. 2 is the particular flow sheet of step of registration.

Embodiment

Below in conjunction with drawings and Examples, the present invention is described in further detail.

The present invention is a kind of data processing method of space-time unite interferometric modulator imaging spectrometer, and flow process as shown in Figure 1, comprises following step:

Step 1: by space-time unite interferometric modulator imaging spectrometer, obtain a series of images sequence that comprises interesting target point.

Space-time unite interferometric modulator imaging spectrometer is in common photographic system, to add interferometer to realize, so, the image that the space-time unite interferometric modulator imaging spectrometer that the present invention adopts obtains is no longer the direct image of target scene, but the target that superposeed is interfered the interference image of strength information, image is swept direction and is changed more sharp-pointed along pushing away.

According to Shannon's sampling theorem, determine the quantity M that gathers image, M represents to gather the quantity of image.For interesting target point, the corresponding specific optical path difference of every piece image, corresponding M the different optical path difference of M width image;

Restore the curve of spectrum of impact point, need to obtain the abundant interference data of the corresponding different optical path differences of impact point.Required sampling number can be determined according to Shannon's sampling theorem.Thereby need to be from sequence image abundant width image extracts the gray-scale value of interesting target point, obtain a series of interference strength data of the corresponding different optical path differences of interesting target point.

Step 2: obtain interesting target point Q in the position that pushes away continuously the interference image of sweeping by image registration.

The interference image sequence obtaining by space-time unite interferometric modulator imaging spectrometer is divided into N image sub-sequence, in each subsequence, has K _nwidth image, the first two field picture in each subsequence is set to reference picture, and remaining image is set to image subject to registration.

Method based on Fourier transform is asked for the mutual displacement relation of reference picture and image subject to registration.

The theoretical foundation of the method is the time shift theorem of Fourier transform, and a signal moves k sampling in time domain, and in frequency domain, the amplitude spectrum of this signal is constant so, but will change-ω of phase spectrum is k.If

$x\left(n\right)\stackrel{F}{\↔}X\left(\mathrm{\ω}\right)$

,

$x(n-k)\stackrel{F}{\↔}{e}^{-\mathrm{j\ωk}}X\left(\mathrm{\ω}\right)---\left(1\right)$

Wherein: x represents spatial domain signal, X represents the frequency spectrum of spatial domain signal x, and n is Space domain sampling point, and ω is Space Angle frequency.

Therefore, for image registration, given two width image f and the g that have (u, v) translation,

g(x,y)＝f(x-u,y-v)

Can obtain,

$G({\mathrm{\ω}}_x,{\mathrm{\ω}}_y)=e^{-j({\mathrm{\ω}}_{x}u+{\mathrm{\ω}}_{y}v)}F({\mathrm{\ω}}_x,{\mathrm{\ω}}_y)---\left(2\right)$

Wherein: F is the frequency spectrum of image f, G is the frequency spectrum of image g, and (x, y) is the coordinate of aerial image vegetarian refreshments, (ω _x, ω _y) be Space Angle frequency.

As shown in Figure 2, concrete registration process is as follows:

A. integer level pixel displacement is estimated:

The first two field picture in original sequence is made as to original reference image, step 2(E) in gained moving displacement estimated value be that image subject to registration is with respect to the relative displacement of original reference image.Image sequence is divided into N image sub-sequence successively by data acquisition order, establish again the reference picture that the first two field picture in each subsequence is this subsequence, other images are image subject to registration, steps A, B, in C, acquired results is the interior image subject to registration of subsequence with respect to the motion estimated values of the reference picture of this subsequence.

Image in antithetical phrase interference image sequence carries out moving displacement parameter to be estimated, obtains the interior arbitrary image of this subimage sequence with respect to the motion estimated values of the integer level pixel of the reference picture of this subimage sequence.According to formula 2, can obtain computing formula 3:

$\exp \left(j({\mathrm{\ω}}_x{u}_c+{\mathrm{\ω}}_y{v}_c)\right)=\frac{F({\mathrm{\ω}}_x,{\mathrm{\ω}}_y)G^{*}({\mathrm{\ω}}_x,{\mathrm{\ω}}_y)}{\left|F({\mathrm{\ω}}_x,{\mathrm{\ω}}_y)G^{*}({\mathrm{\ω}}_x,{\mathrm{\ω}}_y)\right|}---\left(3\right)$

Wherein: F is image spectrum subject to registration, G is reference picture frequency spectrum, G ^*for the conjugation spectrum of G, F ^*for the conjugation spectrum of F, the motion estimated values of horizontal direction is u _c, the motion estimated values on vertical direction is v _c, known u _cand v _cbe 1 * K _ncolumn vector.This cross-power spectrum is carried out to Fourier's inversion, can obtain the cross correlation function between reference picture and image subject to registration, the corresponding numerical value of cross-correlation function value peak be the whole Pixel-level Displacement Estimation value between image subject to registration and reference picture.

B. singular value screening and correction;

Due to the impact of the factors such as noise, A step gained integer pixel displacement estimated result might not be the true value of cross correlation function peak value.The handled image of the present invention is to push away continuously the target scene interference image of sweeping acquisition, so moving displacement parameter has certain rule, can by the method for setting threshold, filter out singular value and to its correction.

As follows to the judgement of singular value and calibration steps:

1) setting threshold T, according to the whole Pixel-level Displacement Estimation of steps A gained result, compares interframe displacement and threshold value T, if be greater than T, thinks that the moving displacement parameter of this image needs to proofread and correct, if be less than or equal to T, enters step C; .

2) sequence number of the image of extract to need proofreading and correct, take this image former frame image is reference picture, 1 in repeating step A, step B).

C. sub-pixel Displacement Estimation;

According to previous step estimated result, by image reversal translation subject to registration, the sub-pixel displacement between the image subject to registration after the reference picture in the method calculating subimage sequence of use phase differential matching and oppositely translation.

Concrete steps are as follows:

1) use Fourier transform, with reference to image and image subject to registration, transform from a spatial domain to frequency domain.

2) for preventing that frequency aliasing and noise from impacting estimated result, extract the low frequency region of reference picture and image subject to registration, calculate both phase differential ∠ (F _{n * n}/ G _{n * n}).∠ (.) represents that two spectral phases between signal are poor, and n * n is selected low frequency region, and wherein n is zone radius.

3) use least square method to phase differential ∠ (F _{n * n}/ G _{n * n}) carry out matching, calculated level direction motion estimated values u _swith vertical direction motion estimated values v _s.

D. because gained motion estimated values in steps A, B, C is image subject to registration with respect to the moving displacement estimated value of the reference picture of its place subimage sequence, for obtaining image subject to registration with respect to the moving displacement estimated value of original reference image, need repeating step A, B and C to carry out kinematic parameter estimation to the reference picture of adjacent subimage sequence (the first two field picture), obtain the position relationship between the reference picture of adjacent subimage sequence, make ε _ifor the motion estimated values in horizontal direction, τ _ifor the motion estimated values on vertical direction.

Reference picture can be any piece image of this subimage sequence, in the present invention, for convenience of deriving and illustrating, has selected the first two field picture of subimage sequence as the reference picture of this subimage sequence.

E. obtain image subject to registration with respect to the relative displacement of original reference image;

N (n=1 ..., N) k in number of sub images sequence (k=1 ..., K _n) two field picture is as follows with respect to the relative displacement calculation expression of original reference image

$U_{n,k}=u_{c,k}+u_{s,k}+\underset{i=0}{\overset{n-1}{\mathrm{\Σ}}}{\mathrm{\ϵ}}_i---\left(4\right)$

$V_{n,k}=v_{c,k}+v_{s,k}+\underset{i=0}{\overset{n-1}{\mathrm{\Σ}}}{\mathrm{\τ}}_i---\left(5\right)$

Wherein: U _n,kand V _n,kbe respectively image subject to registration with respect to horizontal direction and the vertical direction Displacement Estimation value of original reference image, u _c,kand v _c,kbe respectively image subject to registration in n number of sub images sequence with respect to horizontal direction and the whole pixel displacement estimated value of vertical direction of the reference picture of this subimage sequence, u _s,kand v _s,kbe respectively image subject to registration in n number of sub images sequence with respect to horizontal direction and the vertical direction Displacement estimated value of the reference picture of this subimage sequence, ε _iand τ _ibe respectively the reference picture of i number of sub images sequence and the relative displacement between i+1 number of sub images sequence reference picture, known ε ₀and τ ₀be 0.

F. the position of target setting on consecutive image;

According to interesting target position (x, y) on k two field picture o'clock in n number of sub images sequence, can obtain the estimating target position on the k ' two field picture (x ', y ') in the n ' number of sub images sequence:

x′＝x+U _n′,k′-U _n,k (6)

y′＝y+V _n′,k′-V _n,k (7)

Wherein, (U _n,k, V _n,k) and (U _{n ', k '}, V _{n ', k '}) be respectively the registration displacement parameter of the original reference Image Acquisition with respect to complete image sequence of the k ' two field picture in k two field picture and the n ' number of sub images sequence in n number of sub images sequence.

From step 2, if the position coordinates of interesting target point in known reference image, the moving displacement estimated value obtaining according to above step and formula (6) and (7), can obtain the position coordinates of this impact point any piece image Central Asia pixel scale in image sequence, and go out the current corresponding optical path difference of impact point according to this position and the corresponding position calculation of image zero optical path difference, thereby for the Fourier transform process of step 5.

Step 3: the interference strength that obtains the corresponding a certain optical path difference of interesting target point.

In step 2, obtain pushing away sweep in image interesting target point Q coordinate (x ', y '), and (x ', y ') substantially can not drop on integer pixel positions, therefore adopt interpolation method to obtain gray-scale value corresponding to target, be the interference strength value under the corresponding a certain optical path difference of impact point; Extract the gray-scale value of impact point certain piece image correspondence position in interference image sequence, can obtain impact point corresponding to the interference strength data under some optical path differences (every piece image can only obtain the interference strength of the corresponding a certain optical path difference of target).

Step 4: obtain the interferogram that interesting target point is comprised of the interference strength of different optical path differences;

The position of the M two field picture that the interesting target point Q that utilizes step 2 to obtain is corresponding, repeating step three extracts the interference strength data of the different optical path differences of its correspondences, forms the interferogram of target.

Step 5: the interferogram of interesting target point is restored to the spectrum that obtains this interesting target point by Nonuniform fast Fourier transform.

Because the variations such as carrying platform attitude is sidewindered, pitching, driftage can cause extracted interesting target point interference strength data right and wrong in optical path difference equally distributed, need to carry out the curve of spectrum that Nonuniform fast Fourier transform restores impact point.

Interferogram for target is non-equal interval sampling, utilizes Nonuniform fast Fourier transform algorithm to carry out spectrum recovering, obtains the curve of spectrum of target.

Claims (2)

1. a data processing method for imaging spectrometer, comprises following step:

Step 1: by space-time unite interferometric modulator imaging spectrometer, obtain a series of images sequence that comprises interesting target point;

The quantity that gathers image is definite according to Shannon's sampling theorem, and the quantity that gathers image is M, for interesting target point, and the corresponding optical path difference of every piece image, corresponding M the different optical path difference of M width image;

Step 2: obtain interesting target point Q in the position that pushes away continuously the interference image of sweeping by image registration;

The first two field picture in original sequence is made as to original reference image, the interference image sequence obtaining by space-time unite interferometric modulator imaging spectrometer is divided into N image sub-sequence, in each subsequence, have K _nwidth image, in each subsequence, any two field picture is reference picture, remaining image is image subject to registration;

Method based on Fourier transform is asked for the mutual displacement relation of reference picture and image subject to registration;

Given two width image f and the g that have (u, v) translation,

g(x,y)＝f(x-u,y-v)

Can obtain,

$G({\mathrm{\ω}}_x,{\mathrm{\ω}}_y)=e^{-j({\mathrm{\ω}}_{x}u+{\mathrm{\ω}}_{y}v)}F({\mathrm{\ω}}_x,{\mathrm{\ω}}_y)---\left(2\right)$

Wherein: F is the frequency spectrum of image f, G is the frequency spectrum of image g, and (x, y) is the coordinate of aerial image vegetarian refreshments, (ω _x, ω _y) be Space Angle frequency;

In subsequence, image subject to registration is as follows with respect to the acquisition methods of the motion estimated values of the reference picture of this subsequence:

A. integer level pixel displacement is estimated:

According to formula (2), can obtain computing formula (3):

$\exp \left(j({\mathrm{\ω}}_x{u}_c+{\mathrm{\ω}}_y{v}_c)\right)=\frac{F({\mathrm{\ω}}_x,{\mathrm{\ω}}_y)G^{*}({\mathrm{\ω}}_x,{\mathrm{\ω}}_y)}{\left|F({\mathrm{\ω}}_x,{\mathrm{\ω}}_y)G^{*}({\mathrm{\ω}}_x,{\mathrm{\ω}}_y)\right|}---\left(3\right)$

Wherein: F is image spectrum subject to registration, G is reference picture frequency spectrum, G ^*for the conjugation spectrum of G, F ^*for the conjugation spectrum of F, the motion estimated values of horizontal direction is u _c, the motion estimated values on vertical direction is v _c, u _cand v _cbe 1 * K _ncolumn vector; This cross-power spectrum is carried out to Fourier's inversion, obtain the cross correlation function between reference picture and image subject to registration, the corresponding numerical value of cross-correlation function value peak be the whole Pixel-level Displacement Estimation value between image subject to registration and reference picture;

B. singular value screening and correction;

Step is as follows:

1) setting threshold T, according to the whole Pixel-level Displacement Estimation of steps A gained result, compares interframe displacement and threshold value T, if be greater than T, the moving displacement parameter of image is proofreaied and correct, and enters step 2), if be less than or equal to T, enter step C;

2) sequence number of the image of extract to need proofreading and correct, take this image former frame image is reference picture, 1 in repeating step A, step B);

C. sub-pixel Displacement Estimation;

Concrete steps are as follows:

1) use Fourier transform, with reference to image and image subject to registration, transform from a spatial domain to frequency domain;

2) extract the low frequency region of reference picture and image subject to registration, calculate both phase differential ∠ (F _{n * n}/ G _{n * n}); ∠ (.) represents that two spectral phases between signal are poor, and n * n is selected low frequency region, and wherein n is peak width;

3) use least square method to phase differential ∠ (F _{n * n}/ G _{n * n}) carry out matching, calculated level direction motion estimated values u _swith vertical direction motion estimated values v _s;

D. repeating step A, B, C carry out kinematic parameter estimation to the reference picture of adjacent subimage sequence, obtain the position relationship between the reference picture of adjacent subimage sequence, make ε _ifor the motion estimated values in horizontal direction, τ _ifor the motion estimated values on vertical direction;

E. obtain image subject to registration with respect to the relative displacement of original reference image;

In n number of sub images sequence, k two field picture is as follows with respect to the relative displacement calculation expression of original reference image, n=1 wherein ..., N, k=1 ..., K _n;

$U_{n,k}=u_{c,k}+u_{s,k}+\underset{i=0}{\overset{n-1}{\mathrm{\Σ}}}{\mathrm{\ϵ}}_i---\left(4\right)$

$V_{n,k}=v_{c,k}+v_{s,k}+\underset{i=0}{\overset{n-1}{\mathrm{\Σ}}}{\mathrm{\τ}}_i---\left(5\right)$

Wherein: U _n,kand V _n,kbe respectively image subject to registration with respect to horizontal direction and the vertical direction Displacement Estimation value of original reference image, u _c,kand v _c,kbe respectively image subject to registration in n number of sub images sequence with respect to horizontal direction and the whole pixel displacement estimated value of vertical direction of the reference picture of this subimage sequence, u _s,kand v _s,kbe respectively image subject to registration in n number of sub images sequence with respect to horizontal direction and the vertical direction Displacement estimated value of the reference picture of this subimage sequence, ε _iand τ _ibe respectively the reference picture of i number of sub images sequence and the relative displacement between i+1 number of sub images sequence reference picture, ε ₀and τ ₀be 0;

F. the position of target setting on consecutive image;

According to interesting target position (x, y) on k two field picture o'clock in n number of sub images sequence, can obtain the estimating target position on the k ' two field picture (x ', y ') in the n ' number of sub images sequence:

x′＝x+U _n′,k′-U _n,k (6)

y′＝y+V _n′,k′-V _n,k (7)

Wherein, (U _n,k, V _n,k) and (U _{n ', k '}, V _{n ', k '}) be respectively the Displacement Estimation parameter of the original reference Image Acquisition with respect to complete image sequence of the k ' two field picture in k two field picture and the n ' number of sub images sequence in n number of sub images sequence;

If the position coordinates of interesting target point in known reference image, the moving displacement estimated value obtaining according to above step and formula (6) and (7), obtain the position coordinates of this impact point any piece image Central Asia pixel scale in image sequence, and according to this position and the corresponding position calculation of image zero optical path difference, go out the current corresponding optical path difference of impact point, for the Fourier transform process of step 5;

Step 3: the interference strength that obtains the corresponding a certain optical path difference of interesting target point;

In step 2, obtain pushing away and sweep in image interesting target point Q at coordinate (x ', y '), adopt interpolation method to obtain gray-scale value corresponding to target, be the interference strength value under the corresponding a certain optical path difference of impact point;

Step 4: obtain the interferogram that interesting target point is comprised of the interference strength of different optical path differences;

The position of the M two field picture that the interesting target point Q that utilizes step 2 to obtain is corresponding, repeating step three extracts the interference strength data of the different optical path differences of its correspondences, forms the interferogram of target;

Step 5: the interferogram of interesting target point is restored to the spectrum that obtains this interesting target point by Nonuniform fast Fourier transform.

2. the data processing method of a kind of imaging spectrometer according to claim 1, in step 2, setting the first two field picture in each subsequence is reference picture.

Images