CN105005795A - Space-hierarchical-matching-based hyper spectral classification method and apparatus - Google Patents

Abstract

The invention discloses a space-hierarchical-matching-based hyper spectral classification method and apparatus. The method comprises: normalization processing is carried out on a to-be-matched spectrum and all spectrums in a spectrum base and a hierarchical intersected histogram and a distance vector are obtained respectively; according to the distance vector, a similarity value of the to-be-matched spectrum and each spectrum in the spectrum base is calculated; one spectrum having the highest sampling similarity value with the to-be-matched spectrum is selected from the spectrum base as a matching classification object. According to the invention, the histogram information of the hyper spectrum is analyzed; quantization and space delamination are carried out on the spectrums after normalization to obtain a spectrum having a dimension far less than that of the original spectrum, thereby completing spectrum dimension reduction, obviously reducing the operand of subsequent matching, and realizing rapid and real-time classification of the hyper spectrum. Compared with the traditional method, the operand of the matching classification can be reduced obviously; the real-time performance is good; the anti-noise property and robustness are good; the matching precision is high.

Description

A kind of ultraphotic profile classification method based on space delamination coupling and device

Technical field

The present invention relates to ultraphotic spectrum dimensionality reduction sorting technique field, specifically, the present invention relates to a kind of ultraphotic profile classification method based on space delamination coupling and device.

Background technology

In the last few years, two large class methods are roughly had to classify for spectral fingerprint.First class methods are the methods based on coding, and spectrum finger-print codes is become code word by it, then adopt spectrum similarity measure such as Euclidean distance and Hamming distance etc. to carry out analysis of spectrum.One of them is exactly typically binary coding (BC), and it adopts a thresholding to compare to each wave band, then the result compared is become binary number.The complexity of BC is very low, but there is very large error due to quantizing process, may cause the loss of some important spectral informations.In addition, due to the homogeney of the spectrum fingerprint of similar spectrum, cause it bad to similar spectral classification effect.In order to solve the problem, other some algorithms have also put forward, such as spectral modeling analysis (SPAM), the binary coding (SFBC) made earnest efforts based on spectrum and spectrum Gradient Features coding (SDFC) etc.These methods adopt additional bit to encode to spectral signature, and as spectrum gradient, adjacent band difference etc., more spectral information can be taken into account by they.Second largest class methods are estimated based on spectral signature, and as estimated, spectral profile is as spectral signature.Then general criterion of least squares is adopted to carry out analysis of spectrum, as based on the method such as small echo and Kalman filter.Above-mentioned two class methods are verified has good effect in multispectral or even EO-1 hyperion.But ultraphotic is composed, owing to having higher spectral resolution and larger data dimension, and between wave band, there is redundancy, the relevant information between wave band is not considered when classic method adopts the mode by wave band to be applied to ultraphotic spectrum, cause consuming time huge, make traditional method can not meet the requirement of real-time.Therefore, need to carry out dimensionality reduction to spectrum, greatly can reduce time loss like this to reach the requirement of real-time.

In order to reach the object of dimensionality reduction, wherein a kind of typical method for spectrum dimensionality reduction is exactly band selection.Such as, based on the mutual information of the method calculating observation spectrum of mutual information (MI) and the reference spectra of library of spectra, the wave band of relatively high MI is then selected.Adaptive band selection (ABS) the method choice information as much as possible wave band that correlativity is little as far as possible simultaneously.Although these methods can reduce time loss in the process of classification, for each matching process, they need a large amount of observation spectrum, cause them still not reach the requirement of real-time like this.Recently, the people such as side propose a kind of matching process based on passing through feature, and it can meet the requirement of real-time, has very high nicety of grading simultaneously.But its adopts ruler as feature, and adopts Euclidean distance to mate, due to noise-sensitive, this will cause nicety of grading to decline.In addition, the method for some characteristic matching in computer vision also can promote the use of in the middle of Spectral matching, but they have individual condition precedent to be exactly that spectral signature must extract in advance.

For high HYPERSPECTRAL SENSOR, signal to noise ratio (S/N ratio) (SNR) is directly proportional with the square root of t sweep time, spectral resolution △ v and radiant flux E, that is, when scanning constant time t and radiant flux E time, along with the raising of spectral resolution △ v, SNR can be caused like this to reduce.When low signal-to-noise ratio, the Euclidean distance of detecting light spectrum and real spectrum can become large, can reduce like this based on ED the performance of sorting algorithm of HD.Meanwhile, the photon effect of sensor and correction error etc. can inevitably give high hyperspectral data introduce noise, the reduction of classification performance can be caused like this.Therefore, be badly in need of improving the real-time of sorting algorithm and the robustness to noise.

Summary of the invention

For overcoming prior art defect, the present invention proposes a kind of ultraphotic profile classification method based on space delamination coupling and device, by quantizing and space delamination the spectrum after normalization, thus obtain dimension much smaller than original spectrum, complete the dimensionality reduction of spectrum, significantly reduce operand during subsequent match, achieve and quick real-time grading is carried out to ultraphotic spectrum, relative classic method significantly can reduce the operand of coupling classification, real-time is good, there is good noise immunity and robustness, the advantage that matching precision is high.

For reaching this object, the present invention by the following technical solutions:

Based on a ultraphotic profile classification method for space delamination coupling, comprise the following steps:

Step 1, spectral normalization; Be normalized respectively all spectrum in classification spectrum to be matched and library of spectra, normalization formula is as follows:

x＝(z-z _min)/(z _max-z _min),

Wherein, x is the spectral radiance value after normalization, and z is the radiation value of original spectrum, z _maxfor the maximal value of radiation value in this original spectrum, z _minfor the minimum value of radiation value in this original spectrum;

Step 2, calculating distance vector; X and y represents a spectrum in the classification spectrum to be matched that the N after normalization ties up and library of spectra respectively, wherein N represents the dimension of spectrum x and y, spectral value after normalization is quantized into the individual discrete grade of M, wherein M represents quantification gradation number, then at spatial resolution levels l=0 ..., during L, construct a series of grid, wherein L is maximum space level of resolution, L<=log ₂n, makes, when spatial resolution is l, spectrum to be divided into 2 along wavenumber axes by spectrum simultaneously ^lindividual cell; Calculate the histogram that two each cells of spectrum are corresponding again, therefore, it is as follows that spectrum x and y each sub-block space delamination when quantification gradation M and spatial resolution levels are l intersects histogrammic account form:

$P_M(H_x^l,H_y^l)=\underset{i=1}{\overset{D}{\&Sigma;}}min\left(H_x^l\right(i),H_y^l(i\left)\right),$

Wherein, D (D=2 ^l) represent the sum of cell in grid, with represent the histogram of spectrum x and y i-th cell when quantification gradation M and spatial resolution levels l respectively, min (a, b) represents the smaller value getting a and b;

Therefore, the distance vector of spectrum x and y when quantification gradation M and maximum space level of resolution are L is represented account form as follows:

$\begin{array}{c}{q}_M^L(x,y)=P_M(H_x^L,H_y^L)+\underset{l=0}{\overset{L-1}{\mathrm{\&Sigma;}}}\frac{1}{2^{L-1}}\left(P_M\right(H_x^l,H_y^l)-P_M(H_x^{l+1},H_y^{l+1}\left)\right)\\ =\frac{1}{2^L}{P}_M(H_x^0,H_y^0)+\underset{l=1}{\overset{L}{\mathrm{\&Sigma;}}}\frac{1}{2^{L-l+1}}{P}_M(H_x^l,H_y^l)\end{array},$

Wherein, with represent crossing histogram when quantification gradation M and spatial resolution levels 0 and l respectively, especially, represent directly to the crossing histogram of original normalization spectrum x and y when quantification gradation M;

Step 3, calculate space delamination similarity; To adjust the distance vector m component summation of vector obtains space delamination similarity, mathematically, and space delamination similarity account form as follows:

$Q_M^L(x,y)=sum\left(q_M^L\right(x,y\left)\right),$

Wherein, sum (a) expression is sued for peace to each component of vectorial a, be worth larger, the similarity of two spectrum x and y is higher;

Step 4, Spectral matching are classified; Be included in library of spectra and choose a highest spectrum of similarity of sampling with spectrum to be matched as mating object of classification.

Based on a ultraphotic spectrum sorter for space delamination coupling, comprising:

Spectral normalization unit, for spectral normalization; Be normalized respectively all spectrum in classification spectrum to be matched and library of spectra, normalization formula is as follows:

x＝(z-z _min)/(z _max-z _min),

Wherein, x is the spectral radiance value after normalization, and z is the radiation value of original spectrum, z _maxfor the maximal value of radiation value in this original spectrum, z _minfor the minimum value of radiation value in this original spectrum;

Distance vector computing unit, for calculating distance vector; X and y represents a spectrum in the classification spectrum to be matched that the N after normalization ties up and library of spectra respectively, wherein N represents the dimension of spectrum x and y, spectral value after normalization is quantized into the individual discrete grade of M, wherein M represents quantification gradation number, then at spatial resolution levels l=0 ..., during L, construct a series of grid, wherein L is maximum space level of resolution, L<=log ₂n, makes, when spatial resolution is l, spectrum to be divided into 2 along wavenumber axes by spectrum simultaneously ^lindividual cell; Calculate the histogram that two each cells of spectrum are corresponding again, therefore, it is as follows that spectrum x and y each sub-block space delamination when quantification gradation M and spatial resolution levels are l intersects histogrammic account form:

$P_M(H_x^l,H_y^l)=\underset{i=1}{\overset{D}{\&Sigma;}}min\left(H_x^l\right(i),H_y^l(i\left)\right),$

Wherein, D (D=2 ^l) represent the sum of cell in grid, with represent the histogram of spectrum x and y i-th cell when quantification gradation M and spatial resolution levels l respectively, min (a, b) represents the smaller value getting a and b;

Therefore, spectrum x and y distance vector when quantification gradation M and maximum space level of resolution are L is represented account form as follows:

$\begin{array}{c}{q}_M^L(x,y)=P_M(H_x^L,H_y^L)+\underset{l=0}{\overset{L-1}{\mathrm{\&Sigma;}}}\frac{1}{2^{L-1}}\left(P_M\right(H_x^l,H_y^l)-P_M(H_x^{l+1},H_y^{l+1}\left)\right)\\ =\frac{1}{2^L}{P}_M(H_x^0,H_y^0)+\underset{l=1}{\overset{L}{\mathrm{\&Sigma;}}}\frac{1}{2^{L-l+1}}{P}_M(H_x^l,H_y^l)\end{array},$

Wherein, with represent crossing histogram when quantification gradation M and spatial resolution levels 0 and l respectively, especially, represent directly to the crossing histogram of original normalization spectrum x and y when quantification gradation M;

Space delamination similarity calculated, for calculating space delamination similarity; To adjust the distance vector m component summation of vector obtains space delamination similarity, mathematically, and space delamination similarity account form as follows:

$Q_M^L(x,y)=sum\left(q_M^L\right(x,y\left)\right),$

Wherein, sum (a) expression is sued for peace to each component of vectorial a, be worth larger, the similarity of two spectrum x and y is higher;

Spectral matching taxon, classifies for Spectral matching; Be included in library of spectra and choose a highest spectrum of similarity of sampling with spectrum to be matched as mating object of classification.

Beneficial effect:

A kind of ultraphotic profile classification method based on space delamination coupling of the present invention and device, described method is by quantizing and space delamination the spectrum after normalization, thus obtain dimension much smaller than original spectrum, complete the dimensionality reduction of spectrum, significantly reduce operand during subsequent match, achieve and quick real-time grading is carried out to ultraphotic spectrum, relative classic method significantly can reduce the operand of coupling classification, real-time is good, has good noise immunity and robustness, the advantage that matching precision is high.

Accompanying drawing explanation

Fig. 1 is the process flow diagram of a kind of ultraphotic profile classification method based on space delamination coupling that the embodiment of the present invention provides.

Fig. 2 is the spectrogram of the spectrum x of the embodiment of the present invention.

Fig. 3 is the spectrogram of the spectrum y of the embodiment of the present invention.

Fig. 4 is the quantification spectrum histogram of sub-block 1 correspondence of the spectrum x of the embodiment of the present invention.

Fig. 5 is the quantification spectrum histogram of sub-block 2 correspondence of the spectrum x of the embodiment of the present invention.

Fig. 6 is the quantification spectrum histogram of sub-block 1 correspondence of the spectrum y of the embodiment of the present invention.

Fig. 7 is the quantification spectrum histogram of sub-block 2 correspondence of the spectrum y of the embodiment of the present invention.

Fig. 8 is that the quantification spectrum of Fig. 4 and Fig. 6 of the embodiment of the present invention intersects histogram.

Fig. 9 is that the quantification spectrum of Fig. 5 and Fig. 7 of the embodiment of the present invention intersects histogram.

Figure 10 is the histogram summation of Fig. 8 and Fig. 9 of the embodiment of the present invention.

Figure 11 is the structural representation of a kind of ultraphotic spectrum sorter based on space delamination coupling that the embodiment of the present invention provides.

Embodiment

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

Embodiment 1:

Fig. 1 is the process flow diagram of a kind of ultraphotic profile classification method based on space delamination coupling that the embodiment of the present invention provides.As shown in Figure 1, the present invention mainly comprises 4 steps: spectral normalization, calculates distance vector, calculates space delamination similarity, and spectrum corresponding when getting maximum Similarity value is to complete coupling classification.The present embodiment chooses the library of spectra containing 1432 kinds of substance spectra, and often kind of material only has a spectroscopic data in library of spectra.Spectral resolution is Δ σ=0.1cm ^-1, wavelength coverage is 2-14 μm, and corresponding wave-number range is 5000-714cm ^-1.Total N=42861 sampled point in this spectral range, namely original spectrum is 42861 dimensions.

During concrete enforcement, technical solution of the present invention can adopt computer software technology to realize automatic operational scheme.It is as follows that embodiment performs step:

Step 1, spectral normalization, comprise and being normalized respectively all spectrum in classification spectrum to be matched and library of spectra, normalization formula is as follows,

x＝(z-z _min)/(z _max-z _min),(1)

Wherein, x is the spectral radiance value after normalization, and z is the radiation value of original spectrum, z _maxfor the maximal value of radiation value in this original spectrum, z _minfor the minimum value of radiation value in this original spectrum.

Step 2, calculating distance vector, in embodiment, to arbitrary spectrum in the spectrum to be matched after normalization and the library of spectra after normalization, according to the spectrogram after corresponding normalization, perform respectively and comprise following operation:

(2.1) x and y represents the spectrum of the dimension of the N after normalization respectively, and wherein one is spectrum to be matched, and other one is a spectrum in library of spectra, and wherein N represents the dimension of spectrum x and y.Spectral value after normalization is quantized into the individual discrete grade of M, wherein M represents quantification gradation number, then at spatial resolution levels l=0 ..., construct a series of grid during L, wherein L is maximum space level of resolution, L<=log ₂n, makes, when spatial resolution is l, spectrum to be divided into 2 along wavenumber axes by spectrum simultaneously ^lindividual cell.Calculate the histogram that two each cells of spectrum are corresponding again, therefore, it is as follows that spectrum x and y each sub-block space delamination when quantification gradation M and spatial resolution levels are l intersects histogrammic account form:

$P_M(H_x^l,H_y^l)=\underset{i=1}{\overset{D}{\&Sigma;}}min\left(H_x^l\right(i),H_y^l(i\left)\right),---\left(2\right)$

Wherein, D (D=2 ^l) represent the sum of cell in grid, with represent the histogram of spectrum x and y i-th cell when quantification gradation M and spatial resolution levels l respectively, min (a, b) represents the smaller value getting a and b.

Here, M is represented that quantification gradation number is described: N is the sampling number of wavelength direction, simultaneously, each sampled point also has corresponding size, here discrete level is conceptive is similar to analog-to-digital quantification gradation, exactly the size of the spectral value obtained of sampling is quantified as the grade of M equidistant from distance.

N=42861 in embodiment, the quantification gradation of spectrum is set to M=30 in this example, and during spatial resolution levels l=1, D=2, so be divided into 2 sub-blocks by spectrum.In embodiment, the spectrum of spectrum x and y respectively as shown in Figures 2 and 3; Then ask quantification histogram to 2 sub-blocks of spectrum x and y respectively, the quantification histogram of its correspondence respectively as shown in figs. 4-7; Then, utilize two corresponding to spectrum x and y respectively again sub-blocks of formula (2) to ask and quantize to intersect histogram, (it should be noted that, quantification mentioned here is intersected each sub-block space delamination that histogram calculated by formula (2) exactly and is intersected histogram.) quantification of its correspondence intersects histogram respectively as shown in FIG. 8 and 9; Finally obtain space delamination histogram to the frequency summation that two quantize to intersect each reflectivity quantized interval of histogram 8 and 9 corresponding, the result obtained as shown in Figure 10.

(2.2) histogram is intersected in the layering of spatial resolution levels l time space corresponding weight is set to it is inversely proportional with the width of the cell under corresponding spatial resolution level.Therefore, spectrum x and y distance vector when quantification gradation M and maximum space level of resolution are L is represented account form as follows:

$\begin{array}{c}{q}_M^L(x,y)=P_M(H_x^L,H_y^L)+\underset{l=0}{\overset{L-1}{\mathrm{\&Sigma;}}}\frac{1}{2^{L-1}}\left(P_M\right(H_x^l,H_y^l)-P_M(H_x^{l+1},H_y^{l+1}\left)\right)\\ =\frac{1}{2^L}{P}_M(H_x^0,H_y^0)+\underset{l=1}{\overset{L}{\mathrm{\&Sigma;}}}\frac{1}{2^{L-l+1}}{P}_M(H_x^l,H_y^l)\end{array},$

Wherein, with represent crossing histogram when quantification gradation M and spatial resolution levels 0 and l respectively, especially, represent directly to the crossing histogram of original normalization spectrum x and y when quantification gradation M;

L=3 in embodiment.

Step 3, calculate space delamination similarity, spectrum to be matched is calculated space delamination similarity one by one with each spectrum in library of spectra respectively, then sees which value is maximum, this spectrum so in library of spectra is exactly one that mates most with spectrum to be matched.To adjust the distance vector m component summation of vector obtains space delamination similarity, mathematically, and space delamination similarity account form as follows:

$Q_M^L(x,y)=sum\left(q_M^L\right(x,y\left)\right),$

Wherein, sum (a) expression is sued for peace to each component of vectorial a. be worth larger, the similarity of two spectrum x and y is higher.

Step 4, Spectral matching are classified, and are included in library of spectra and choose a highest spectrum of similarity of sampling with spectrum to be matched as mating object of classification.

There are 1432 spectrum in the library of spectra of embodiment, then calculate every bar spectrum in spectrum to be measured and library of spectra respectively value, chooses be worth spectrum in the library of spectra of maximum correspondence as coupling classification spectrum.

Embodiment 2:

The present embodiment is device embodiment, and the embodiment of the method for device embodiment of the present invention and above-described embodiment 1 belongs to same technical conceive, and the content of not detailed description in device embodiment, refers to embodiment of the method.

Figure 11 is the structural representation of a kind of ultraphotic spectrum sorter based on space delamination coupling that the embodiment of the present invention provides.As shown in Figure 10, a kind of ultraphotic spectrum sorter based on space delamination coupling of the present invention, comprising:

Spectral normalization unit, for spectral normalization; Be normalized respectively all spectrum in classification spectrum to be matched and library of spectra, normalization formula is as follows:

x＝(z-z _min)/(z _max-z _min),

Wherein, x is the spectral radiance value after normalization, and z is the radiation value of original spectrum, z _maxfor the maximal value of radiation value in this original spectrum, z _minfor the minimum value of radiation value in this original spectrum;

Distance vector unit, for calculating distance vector; X and y represents a spectrum in the classification spectrum to be matched that the N after normalization ties up and library of spectra respectively, wherein N represents the dimension of spectrum x and y, spectral value after normalization is quantized into the individual discrete grade of M, wherein M represents quantification gradation number, then at spatial resolution levels l=0 ..., during L, construct a series of grid, wherein L is maximum space level of resolution, L<=log ₂n, makes, when spatial resolution is l, spectrum to be divided into 2 along wavenumber axes by spectrum simultaneously ^lindividual cell; Calculate the histogram that two each cells of spectrum are corresponding again, therefore, it is as follows that spectrum x and y each sub-block space delamination when quantification gradation M and spatial resolution levels are l intersects histogrammic account form:

$P_M(H_x^l,H_y^l)=\underset{i=1}{\overset{D}{\&Sigma;}}min\left(H_x^l\right(i),H_y^l(i\left)\right),$

Wherein, D (D=2 ^l) represent the sum of cell in grid, with represent the histogram of spectrum x and y i-th cell when quantification gradation M and spatial resolution levels l respectively, min (a, b) represents the smaller value getting a and b;

Histogram is intersected in the layering of spatial resolution levels l time space corresponding weight is set to it is inversely proportional with the width of the cell under corresponding spatial resolution level, therefore, represents the distance vector of spectrum x and y when quantification gradation M and maximum space level of resolution are L account form as follows:

$\begin{array}{c}{q}_M^L(x,y)=P_M(H_x^L,H_y^L)+\underset{l=0}{\overset{L-1}{\mathrm{\&Sigma;}}}\frac{1}{2^{L-1}}\left(P_M\right(H_x^l,H_y^l)-P_M(H_x^{l+1},H_y^{l+1}\left)\right)\\ =\frac{1}{2^L}{P}_M(H_x^0,H_y^0)+\underset{l=1}{\overset{L}{\mathrm{\&Sigma;}}}\frac{1}{2^{L-l+1}}{P}_M(H_x^l,H_y^l)\end{array},$

Wherein, with represent crossing histogram when quantification gradation M and spatial resolution levels 0 and l respectively, especially, represent directly to the crossing histogram of original normalization spectrum x and y when quantification gradation M;

Space delamination similarity calculated, for calculating space delamination similarity; To adjust the distance vector m component summation of vector obtains space delamination similarity, mathematically, and space delamination similarity account form as follows:

$Q_M^L(x,y)=sum\left(q_M^L\right(x,y\left)\right),$

Wherein, sum (a) expression is sued for peace to each component of vectorial a, be worth larger, the similarity of two spectrum x and y is higher;

Spectral matching taxon, classifies for Spectral matching; Be included in library of spectra and choose a highest spectrum of similarity of sampling with spectrum to be matched as mating object of classification.

On the whole, ultraphotic profile classification method based on space delamination coupling proposed by the invention, by quantizing and space delamination the spectrum after normalization, thus obtain dimension much smaller than original spectrum, complete the dimensionality reduction of spectrum, significantly reduce operand during subsequent match, achieve and quick real-time grading is carried out to ultraphotic spectrum.Relative classic method significantly can reduce the operand of coupling classification, and real-time is good, has good noise immunity and robustness, the advantage that matching precision is high.

Should be understood that, the part that this instructions does not elaborate all belongs to prior art.

Should be understood that; the above-mentioned description for embodiment is comparatively detailed; therefore the restriction to scope of patent protection of the present invention can not be thought; those of ordinary skill in the art is under enlightenment of the present invention; do not departing under the ambit that the claims in the present invention protect; can also make and replacing or distortion, all fall within protection scope of the present invention, request protection domain of the present invention should be as the criterion with claims.

Claims (2)

1., based on a ultraphotic profile classification method for space delamination coupling, it is characterized in that, comprise the following steps:

Step 1, spectral normalization; Be normalized respectively all spectrum in classification spectrum to be matched and library of spectra, normalization formula is as follows:

x＝(z-z _min)/(z _max-z _min),

Wherein, x is the spectral radiance value after normalization, and z is the radiation value of original spectrum, z _maxfor the maximal value of radiation value in this original spectrum, z _minfor the minimum value of radiation value in this original spectrum;

Step 2, calculating distance vector; X and y represents a spectrum in the classification spectrum to be matched that the N after normalization ties up and library of spectra respectively, wherein N represents the dimension of spectrum x and y, spectral value after normalization is quantized into the individual discrete grade of M, wherein M represents quantification gradation number, then at spatial resolution levels l=0 ..., during L, construct a series of grid, wherein L is maximum space level of resolution, L<=log ₂n, makes, when spatial resolution is l, spectrum to be divided into 2 along wavenumber axes by spectrum simultaneously ^lindividual cell; Calculate the histogram that two each cells of spectrum are corresponding again, therefore, it is as follows that spectrum x and y each sub-block space delamination when quantification gradation M and spatial resolution levels are l intersects histogrammic account form:

$P_M(H_x^l,H_y^l)=\underset{i=1}{\overset{D}{\&Sigma;}}min\left(H_x^l\right(i),H_y^l(i\left)\right),$

Wherein, D (D=2 ^l) represent the sum of cell in grid, with represent the histogram of spectrum x and y i-th cell when quantification gradation M and spatial resolution levels l respectively, min (a, b) represents the smaller value getting a and b;

Therefore, the distance vector of spectrum x and y when quantification gradation M and maximum space level of resolution are L is represented account form as follows:

$\begin{array}{c}{q}_M^L(x,y)=P_M(H_x^L,H_y^L)+\underset{l=0}{\overset{L-1}{\mathrm{\&Sigma;}}}\frac{1}{2^{L-1}}\left(P_M\right(H_x^l,H_y^l)-P_M(H_x^{l+1},H_y^{l+1}\left)\right)\\ =\frac{1}{2^L}{P}_M(H_x^0,H_y^0)+\underset{l=1}{\overset{L}{\mathrm{\&Sigma;}}}\frac{1}{2^{L-l+1}}{P}_M(H_x^l,H_y^l)\end{array},$

Wherein, with represent crossing histogram when quantification gradation M and spatial resolution levels 0 and l respectively, especially, represent directly to the crossing histogram of original normalization spectrum x and y when quantification gradation M;

Step 3, calculate space delamination similarity; To adjust the distance vector m component summation of vector obtains space delamination similarity, mathematically, and space delamination similarity account form as follows:

$Q_M^L(x,y)=sum\left(q_M^L\right(x,y\left)\right),$

Wherein, sum (a) expression is sued for peace to each component of vectorial a, be worth larger, the similarity of two spectrum x and y is higher;

Step 4, Spectral matching are classified; Be included in library of spectra and choose a highest spectrum of similarity of sampling with spectrum to be matched as mating object of classification.

2., based on a ultraphotic spectrum sorter for space delamination coupling, it is characterized in that, comprising:

Spectral normalization unit, for spectral normalization; Be normalized respectively all spectrum in classification spectrum to be matched and library of spectra, normalization formula is as follows:

x＝(z-z _min)/(z _max-z _min),

Wherein, x is the spectral radiance value after normalization, and z is the radiation value of original spectrum, z _maxfor the maximal value of radiation value in this original spectrum, z _minfor the minimum value of radiation value in this original spectrum;

Distance vector computing unit, for calculating distance vector; X and y represents a spectrum in the classification spectrum to be matched that the N after normalization ties up and library of spectra respectively, wherein N represents the dimension of spectrum x and y, spectral value after normalization is quantized into the individual discrete grade of M, wherein M represents quantification gradation number, then at spatial resolution levels l=0 ..., during L, construct a series of grid, wherein L is maximum space level of resolution, L<=log ₂n, makes, when spatial resolution is l, spectrum to be divided into 2 along wavenumber axes by spectrum simultaneously ^lindividual cell; Calculate the histogram that two each cells of spectrum are corresponding again, therefore, it is as follows that spectrum x and y each sub-block space delamination when quantification gradation M and spatial resolution levels are l intersects histogrammic account form:

$P_M(H_x^l,H_y^l)=\underset{i=1}{\overset{D}{\&Sigma;}}min\left(H_x^l\right(i),H_y^l(i\left)\right),$

Wherein, D (D=2 ^l) represent the sum of cell in grid, with represent the histogram of spectrum x and y i-th cell when quantification gradation M and spatial resolution levels l respectively, min (a, b) represents the smaller value getting a and b;

Therefore, the distance vector of spectrum x and y when quantification gradation M and maximum space level of resolution are L is represented account form as follows:

$\begin{array}{c}{q}_M^L(x,y)=P_M(H_x^L,H_y^L)+\underset{l=0}{\overset{L-1}{\mathrm{\&Sigma;}}}\frac{1}{2^{L-1}}\left(P_M\right(H_x^l,H_y^l)-P_M(H_x^{l+1},H_y^{l+1}\left)\right)\\ =\frac{1}{2^L}{P}_M(H_x^0,H_y^0)+\underset{l=1}{\overset{L}{\mathrm{\&Sigma;}}}\frac{1}{2^{L-l+1}}{P}_M(H_x^l,H_y^l)\end{array},$

Wherein, with represent crossing histogram when quantification gradation M and spatial resolution levels 0 and l respectively, especially, represent directly to the crossing histogram of original normalization spectrum x and y when quantification gradation M;

Space delamination similarity calculated, for calculating space delamination similarity; To adjust the distance vector m component summation of vector obtains space delamination similarity, mathematically, and space delamination similarity account form as follows:

$Q_M^L(x,y)=sum\left(q_M^L\right(x,y\left)\right),$

Wherein, sum (a) expression is sued for peace to each component of vectorial a, be worth larger, the similarity of two spectrum x and y is higher;

Spectral matching taxon, classifies for Spectral matching; Be included in library of spectra and choose a highest spectrum of similarity of sampling with spectrum to be matched as mating object of classification.