CN101266296A - High light spectrum small target detection method and apparatus - Google Patents

Abstract

The invention discloses a high spectrum small target detecting method, comprising: acquiring the pixel spectrum two-dimensional matrix of the three-dimensional high spectrum data in the detected image pixel; angle matching each pixel spectrum in the pixel spectrum two-dimensional matrix with the average spectrum of the detected image pixel to obtain the spectrum angle matching value of each pixel; obtaining the Markov distance corresponding with each pixel based on the spectrum angle matching value; comparing the Markov distance corresponding with each pixel with the preset threshold and determining the pixel point with Markov distance more than the preset threshold as small target point. The invention also provides a high spectrum small target detecting device has higher target detection accuracy and operational speed and efficiency of target detection without any prior information, using the said high spectrum small target detecting method and device.

Description

High light spectrum small target detection method and device

Technical field

The present invention relates to the high-spectrum remote-sensing Detection Techniques, relate in particular to a kind of high light spectrum small target detection method and device.

Background technology

High-spectrum remote-sensing is one of most important technological breakthrough in the earth observing system of twentieth century end, it has overcome traditional single band, the multispectral remote sensing limitation at aspects such as wave band number, wavelength band, meticulous information representations, provide sensor information with narrower wave band interval, more wave band quantity, can from spectral space, be segmented and be differentiated, obtained widespread use in fields such as resources remote sensing, environmental remote sensing, ecological Remote Sensing to atural object.High spectrum resolution remote sensing technique can be by abundant cartographic feature and spectral information, the nuance between reflection target atural object and background atural object, thus target atural object and background atural object are made a distinction.Utilize high spectrum resolution remote sensing technique, can find the terrain object that is difficult to survey with characteristics of image such as texture, edges, this detection for the little target that only covers seldom several pixels on image is highly beneficial.

Common small target detection method mainly comprises at present: RX small target detection method, projection Pursuit Method and major component decomposition method etc.

Wherein, the RX small target detection method is: suppose that background spectrum information satisfies certain multidimensional and distributes the covariance matrix of construct image; Being compressed to minority multidimensional image spectrum incoherent mutually by the method for principal component analysis (PCA) is on the space of substrate with the major component; Structure can reflect the detection operator of pixel spectrum characteristic parameter size, can give prominence to target optical spectrum information from background spectrum; Determine by the method for test of hypothesis whether little target exists again, thereby realize the detection of little target.

Projection Pursuit Method is: by choosing the objective function of representing the difference of target and background spectrum, construct one or one group of orthogonal vector as projecting direction; The method of employing optimization searching determines to make the projecting direction of objective function maximum, and spectroscopic data is projected to determined projecting direction; The multidimensional spectral information is compressed to low-dimensional, and outstanding target information, thereby detect little target by probability method.

The major component decomposition method mainly is to detect by the score of each major component is carried out exceptional value as image independently, thereby realizes the detection of little target.

In sum, existing small target detection method, complex calculation such as need decomposing the high-spectral data in the bigger zone to be measured of dimension or invert, operand is bigger, and travelling speed is slower, thereby causes detection efficiency lower.

Summary of the invention

In view of this, fundamental purpose of the present invention is to provide a kind of high light spectrum small target detection method and device, to solve the lower problem of small target detection method detection efficiency of the prior art.

For achieving the above object, technical scheme of the present invention is achieved in that

The invention provides a kind of high light spectrum small target detection method, comprising:

Obtain the pixel spectrum two-dimensional matrix of three-dimensional high-spectral data in the testing image pixel;

The averaged spectrum of each pixel spectrum in the described pixel spectrum two-dimensional matrix and described testing image pixel is carried out the angle coupling, obtain the spectrum angle matching value of each pixel;

According to described spectrum angle matching value, obtain and the corresponding mahalanobis distance of described each pixel;

The mahalanobis distance and the predetermined threshold of described each pixel correspondence are compared, and definite mahalanobis distance is little impact point greater than the pixel point of predetermined threshold.

The described pixel spectrum two-dimensional matrix that obtains three-dimensional high-spectral data in the testing image pixel specifically comprises:

Three-dimensional high-spectral data in the described testing image pixel is expressed as the pixel spectrum two-dimensional matrix of high spectrum reflection rate:

R _m×n＝[p ₁，p ₂...p _y×i+j...p _x×y]，0＜i≤x，0＜j≤y，

Or R _{M * n}=[p ₁, p ₂... p _{I+x * j}... p _{X * y}], 0＜i≤x, 0＜j≤y,

Wherein, R _{M * n}Expression pixel spectrum two-dimensional matrix, [p ₁, p ₂... p _{Y * i+j}... p _{X * y}] and [p ₁, p ₂... p _{I+x * j}... p _{X * y}] expression testing image pixel spectrum vector, m represents the wave band number, n represents the sum of testing image pixel, x represents the line number of testing image pixel, y represents the columns of testing image pixel, n=x * y.

After the described two-dimensional matrix that obtains three-dimensional high-spectral data in the testing image pixel, this method also comprises: to the pre-service that the pixel spectrum two-dimensional matrix of described testing image is proofreaied and correct spectral error, the pixel spectrum two-dimensional matrix after obtaining proofreading and correct.

Described pre-service is that continuum remove to be handled, orthonormal transformation is handled and adds at least a in handling of scatter correction.

Described averaged spectrum with each pixel spectrum in the pixel spectrum two-dimensional matrix and testing image pixel is carried out the angle coupling, obtains the spectrum angle matching value of each pixel, specifically comprises:

Obtain the averaged spectrum of described testing image pixel according to described pixel spectrum two-dimensional matrix $\stackrel{\‾}{p}=\frac{1}{n}\underset{h=1}{\overset{n}{\mathrm{\Σ}}}{p}_h,$ Wherein, p _hThe spectrum vector of representing h pixel in the described testing image;

Described each pixel spectrum and the averaged spectrum that obtains are carried out spectrum angle coupling, are calculated as follows as follows:

${\mathrm{\α}}_h={\cos }^{-1}\frac{\stackrel{\‾}{p}\·p_h}{\left|\stackrel{\‾}{p}\right|\left|p_h\right|}={\cos }^{-1}\frac{\underset{k=1}{\overset{m}{\mathrm{\Σ}}}{\stackrel{\‾}{p}}_k{p}_{\mathrm{hk}}}{\sqrt{\underset{k=1}{\overset{m}{\mathrm{\Σ}}}{\left({\stackrel{\‾}{p}}_k\right)}^2}\sqrt{\underset{k=1}{\overset{m}{\mathrm{\Σ}}}{\left(p_{\mathrm{hk}}\right)}^2}}$

Wherein, α _hThe spectrum angle matching value of h pixel in the expression testing image, p _hThe spectrum vector of h pixel in the expression testing image, p represents averaged spectrum vector, p _kExpression averaged spectrum vector is at the reflectance value of k wave band, p _HkH pixel in the expression testing image is at the reflectance value at k wave band place.

Described according to spectrum angle matching value, obtain the mahalanobis distance corresponding and be with each pixel:

${\mathrm{MD}}_h={\left[{({\mathrm{\α}}_h-\stackrel{\‾}{\mathrm{\α}})}^T{M}^{-1}({\mathrm{\α}}_h-\stackrel{\‾}{\mathrm{\α}})\right]}^{\frac{1}{2}},h=\mathrm{1,2},...,n$

Wherein, MD _hThe mahalanobis distance of h pixel in the expression testing image, α _hThe spectrum angle matching value of h pixel in the expression testing image, n represents testing image pixel sum, α represents the mean value of all pixel spectrum angle matching values, $\stackrel{\‾}{\mathrm{\α}}=\frac{1}{n}\underset{h=1}{\overset{n}{\mathrm{\Σ}}}{\mathrm{\α}}_h,$ M ^-1Be the inverse matrix of mahalanobis distance matrix M, $M=\frac{C^{T}C}{n-1},$ C represents by each α in the testing image pixel _hThe spectrum angle matching vector of forming.

Described predetermined threshold is to determine in the following way:

$\mathrm{\Δ}=u+(\mathrm{Max}-u)\sqrt{\frac{u}{\mathrm{Max}}}$

Wherein, Δ is represented predetermined threshold, $u=\frac{1}{n}\underset{h=1}{\overset{n}{\mathrm{\Σ}}}{\mathrm{MD}}_h,$ U represents the mean value of the mahalanobis distance of each pixel in the testing image, and Max is the mahalanobis distance MD of each pixel in the testing image _hIn maximal value.

The present invention also provides a kind of high light spectrum small target detection device, comprising: two-dimensional matrix acquiring unit, angle matching unit, mahalanobis distance acquiring unit and little impact point determining unit; Wherein,

Described two-dimensional matrix acquiring unit is used for obtaining the pixel spectrum two-dimensional matrix of the three-dimensional high-spectral data of testing image pixel;

Described angle matching unit is used for each the pixel spectrum of described pixel spectrum two-dimensional matrix and the averaged spectrum of described testing image pixel are carried out the angle coupling, obtains the spectrum angle matching value of each pixel;

Described mahalanobis distance acquiring unit is used for according to described spectrum angle matching value, obtains and the corresponding mahalanobis distance of described each pixel;

Described little impact point determining unit is used for the mahalanobis distance and the predetermined threshold value of described each pixel correspondence are compared, and definite mahalanobis distance is little impact point greater than the pixel point of predetermined threshold.

Described device also comprises: pretreatment unit, connect described two-dimensional matrix acquiring unit and angle matching unit, the pre-service that the pixel spectrum two-dimensional matrix that is used for that described two-dimensional matrix acquiring unit is obtained is proofreaied and correct spectral error, the pixel spectrum two-dimensional matrix after obtaining handling offers described angle matching unit.

High light spectrum small target detection method provided by the present invention and device carry out the detection of spectrum singular point from spectrum dimension angle, thereby realize the detection to unusual target, need not any prior imformation; The present invention is in testing process, and the method for employing spectrum angle coupling is compressed to one dimension with the spectroscopic data of multidimensional, and outstanding singular value target information, so operand is little, fast operation; The present invention adopts the method that adapts to threshold value in mahalanobis distance singular value testing process, can carry out small target detection on the basis of pixel spectrum statistical model the unknown, and the scope of application is more extensive, and the accuracy rate of detection is higher; In addition, The present invention be directed to the operation that spectrum dimension is carried out, promptly carry out computing after can gathering the spectrum of a pixel point, the target optical spectrum that need not to wait for whole zone is all gathered and is finished, thereby has better met the requirement of target detection on real-time.

Description of drawings

Fig. 1 is the process flow diagram of a kind of high light spectrum small target detection method of the present invention;

Fig. 2 is the synoptic diagram of the original high spectrum image of the embodiment of the invention;

Fig. 3 is the synoptic diagram of the testing image of the embodiment of the invention;

Fig. 4 is that the original spectrum of the embodiment of the invention is without pretreated spectrogram;

Fig. 5 is the spectrogram of original spectrum after continuum removal and orthogonal transformation processing of the embodiment of the invention;

Fig. 6 is the pixel averaged spectrum synoptic diagram of the testing image of the embodiment of the invention;

Fig. 7 is the angle matching value synoptic diagram of each pixel in the testing image of the embodiment of the invention;

Fig. 8 is the mahalanobis distance synoptic diagram of each pixel in the testing image of the embodiment of the invention;

Fig. 9 is the result of detection synoptic diagram of the embodiment of the invention;

Figure 10 is the composition structural representation of a kind of high light spectrum small target detection device of the present invention.

Embodiment

The technical solution of the present invention is further elaborated below in conjunction with the drawings and specific embodiments.

The process flow diagram of high light spectrum small target detection method provided by the present invention as shown in Figure 1, mainly may further comprise the steps:

Step 101 is obtained the pixel spectrum two-dimensional matrix of three-dimensional high-spectral data in the testing image pixel.

The three-dimensional high-spectral data of the testing image pixel that obtains is expressed as the pixel spectrum two-dimensional matrix of high spectrum reflection rate, as follows:

R _m×n＝[p ₁，p ₂...p _y×i+j...p _x×y]，0＜i≤x，0＜j≤y (1)

Or R _{M * n}=[p ₁, p ₂... p _{I+x * j}... p _{X * y}], 0＜i≤x, 0＜j≤y (2)

Wherein, (1) formula is the representation that the testing image pixel launches by row, and (2) formula is the representation that the testing image pixel launches by row, R _{M * n}Expression pixel spectrum two-dimensional matrix, x represents testing image pixel line number, and y represents testing image pixel columns, and m represents the wave band number, and n represents testing image pixel sum, n=x * y.[p ₁, p ₂... p _{Y * i+j}... p _{X * y}] and [p ₁, p ₂... p _{I+x * j}... p _{X * y}] the pixel spectrum vector of expression testing image, at the R that launches by row _{M * n}In, p _{Y * i+j}The capable j of i is listed as the spectrum vector of corresponding pixel in the expression testing image; In like manner, at the R that launches by row _{M * n}In, p _{I+x * j}Represent that then the capable j of i in the testing image is listed as the spectrum vector of corresponding pixel.The spectrum vector of each pixel comprises the reflectance value of this pixel at each wave band place, and for example: the spectrum vector of supposing h pixel is p _h, p then _h=[p _H1, p _H2... p _Hk... p _Hm] ^T, [p _H1, p _H2... p _Hk... p _Hm] ^TRepresentative [p _H1, p _H2... p _Hk... p _Hm] transposed matrix, p wherein _HkRepresent the reflectance value of h pixel at k wave band place.Because what imaging spectrometer obtained is the reflectance value of image picture elements point at each wave band place, so this reflectance value is a known quantity.

Step 102 is to the pre-service that the pixel spectrum two-dimensional matrix of testing image is proofreaied and correct spectral error, the pixel spectrum two-dimensional matrix after obtaining handling.

It is to proofread and correct the spectral error that causes because of atmospheric scattering etc. that two-dimensional matrix is carried out pretreated purpose, and pretreated method can be removed for continuum and handle, orthonormal transformation is handled and add at least a in handling of scatter correction.

Wherein, continuum is removed to handle and specifically comprised: determine two relative peak points in both sides, spectral absorption center as end points, joining two endpoints just constitutes the straight line of an envelope above reflectance curve, and this straight line is continuum.Relative reflectance after continuum is removed is exactly with the reflectivity of actual spectrum reflectivity divided by respective wavelength place on the continuum.After the continuum removal, the reflectivity at end points place is 1, and the reflectivity between the end points is all less than 1.

The formula that orthonormal transformation is handled is as follows:

$p_{\mathrm{hk},\mathrm{SNV}}=\frac{p_{\mathrm{hk}}-{\stackrel{\‾}{p}}_h}{\sqrt{\underset{k=1}{\overset{m}{\mathrm{\Σ}}}{(p_{\mathrm{hk}}-{\stackrel{\‾}{p}}_h)}^2}}{(m-1)}^{1/2}---\left(3\right)$

Wherein, p _{Hk, SNV}Expression is handled in the testing image of back the reflectance value of h pixel at k wave band, p through orthogonal transformation _hH pixel is at the mean value of each wave band place reflectivity in the expression testing image, and m represents the wave band number, and m-1 represents degree of freedom.

The process that additional scatter correction is handled comprises:

At first, calculate the averaged spectrum vector:

$\stackrel{\‾}{p}=\frac{1}{n}\underset{h=1}{\overset{n}{\mathrm{\Σ}}}{p}_h---\left(4\right)$

Then, each pixel spectrum is carried out linear regression:

p _h＝m _hp+b _h (5)

Add scatter correction again:

$p_{h\left(\mathrm{MSC}\right)}=\frac{(p_h-b_h)}{m_h}---\left(6\right)$

Above-mentioned (4) in (6) formula, p represents the averaged spectrum vector, Expression is to the summation of all pixel spectrum vectors, p _hThe spectrum vector of h pixel in the expression testing image, m _h, b _hRepresent h pixel spectrum vector p respectively _hWith the slope and the intercept of the linear regression of all pixel averaged spectrum vectors, p _{H (MSC)}Pixel spectrum vector behind the additional scatter correction of expression process.

After carrying out pre-service, can obtain proofreading and correct the afterwards more accurate spectral information of spectral error.It is pointed out that preprocess method of the present invention is not limited only to above-mentioned three kinds of disposal routes, other are any can proofread and correct the disposal route of drawing a spectral error that waits because of atmospheric scattering and also should belong to protection scope of the present invention; And carry out pretreated operation in the embodiment of the invention and can select wherein a kind of executable operations in above-mentioned three kinds of disposal routes, also optional majority kind is executable operations successively.But step 102 is the selection operation in the high light spectrum small target detection method of the present invention, by the pre-service of step 102, and can be so that the result of high light spectrum small target detection be more accurate.

Step 103 is carried out the angle coupling with the averaged spectrum of each pixel spectrum in the pixel spectrum two-dimensional matrix after handling and testing image pixel, obtains the spectrum angle matching value of each pixel.

The formula that calculates the spectrum angle matching value of each pixel is:

${\mathrm{\α}}_h={\cos }^{-1}\frac{\stackrel{\‾}{p}\·p_h}{\left|\stackrel{\‾}{p}\right|\left|p_h\right|}={\cos }^{-1}\frac{\underset{k=1}{\overset{m}{\mathrm{\Σ}}}{\stackrel{\‾}{p}}_k{p}_{\mathrm{hk}}}{\sqrt{\underset{k=1}{\overset{m}{\mathrm{\Σ}}}{\left({\stackrel{\‾}{p}}_k\right)}^2}\sqrt{\underset{k=1}{\overset{m}{\mathrm{\Σ}}}{\left(p_{\mathrm{hk}}\right)}^2}}---\left(7\right)$

Wherein, α _hThe spectrum angle matching value of h pixel in the expression testing image, p _hThe spectrum vector of h pixel in the expression testing image, p represents averaged spectrum vector, p _kExpression averaged spectrum vector is at the reflectance value of k wave band, p _HkH pixel in the expression testing image is at the reflectance value at k wave band place.So-called spectrum angle matching value is meant the angle between pixel spectrum to be measured and the averaged spectrum, and each the pixel correspondence in the testing image a spectrum angle matching value, and this angle matching value has reflected that each pixel departs from the level of whole testing image.P in the formula (7) is actually as a kind of reference spectra vector, realization angle coupling, and the reference spectra vector that also can adopt other in actual applications replaces the p in the formula (7) to carry out the angle coupling, be that the meadow accounts under the leading situation for example at testing image, can select the spectrum vector of grass is the reference spectra vector, replaces p to realize the computing of above-mentioned formula (7).

Step 104 according to spectrum angle matching value, is obtained the mahalanobis distance corresponding with each pixel.

At first, according to the spectrum angle matching value α of each pixel in the testing image that obtains _h, calculate the mahalanobis distance of each pixel spectrum angle matching value, the computing formula of mahalanobis distance is:

${\mathrm{MD}}_h={\left[{({\mathrm{\α}}_h-\stackrel{\‾}{\mathrm{\α}})}^T{M}^{-1}({\mathrm{\α}}_h-\stackrel{\‾}{\mathrm{\α}})\right]}^{\frac{1}{2}},h=\mathrm{1,2},...,n---\left(8\right)$

Wherein, MD _hThe mahalanobis distance of h pixel in the expression testing image, n represents testing image pixel sum, α represents the mean value of all pixel spectrum angle matching values, $\stackrel{\‾}{\mathrm{\α}}=\frac{1}{n}\underset{h=1}{\overset{n}{\mathrm{\Σ}}}{\mathrm{\α}}_h,$ M ^-1Be the inverse matrix of mahalanobis distance matrix M, M can be calculated by following formula,

$M=\frac{C^{T}C}{n-1}---\left(9\right)$

Wherein, C is by each α in the testing image pixel _hThe spectrum angle matching vector of forming.

Step 105 compares the mahalanobis distance and the predetermined threshold value of each pixel correspondence, and definite mahalanobis distance is little impact point greater than the pixel point of predetermined threshold.

Predetermined threshold delta can be determined by following formula:

$\mathrm{\Δ}=u+(\mathrm{Max}-u)\sqrt{\frac{u}{\mathrm{Max}}}---\left(10\right)$

Wherein, $u=\frac{1}{n}\underset{h=1}{\overset{n}{\mathrm{\Σ}}}{\mathrm{MD}}_h,$ U is the mean value of the mahalanobis distance of each pixel in the testing image, and Max is the mahalanobis distance MD of each pixel in the testing image _hIn maximal value, also promptly determine mahalanobis distance MD _hPixel point greater than Δ is little impact point.

Further elaborate below in conjunction with the high light spectrum small target detection method of specific embodiment the invention described above.The used high-spectrum remote sensing data of this example derives from airborne imaging spectrometer, and airborne imaging spectrometer is for adopting the imaging spectrometer of push-scanning image mode.The high spectrum image that uses in the present embodiment as shown in Figure 2, size is 614 * 512 pixels, each pixel spectrum comprises 224 wavelength, to 2506.81 nanometers, the wavelength interval is 10 nanometers to wavelength coverage from 369.85 nanometers.Concrete detection process is as follows:

A, get that the image of 43 * 43 pixels is the testing image of present embodiment in the white box shown in Fig. 2, Fig. 3 then be the synoptic diagram of this testing image, the total n=43 * 43=1849 of testing image pixel.Because the existence of bad wave band, only adopt spectral quality preferably 507.74～1324.03nm wave band as detecting band.Wherein, can see that the bigger pixel of gray-scale value in the middle of the white box is the target pixel.

B, each pixel spectrum of testing image carried out continuum is removed and orthonormal transformation is handled, the original spectrum before handling as shown in Figure 4, the spectrum after the processing is as shown in Figure 5.Horizontal ordinate among Fig. 4 and Fig. 5 is represented wavelength, and ordinate is represented reflectivity.

C, the averaged spectrum of the spectrum of each pixel in the testing image and testing image pixel is carried out spectrum angle coupling, obtain the spectrum angle matching value of each pixel in the testing image.The averaged spectrum of testing image pixel as shown in Figure 6, the horizontal ordinate among Fig. 6 is represented wavelength, ordinate is represented reflectivity.The spectrum angle matching value of each pixel that obtains as shown in Figure 7, the horizontal ordinate among Fig. 7 is represented the pixel number, ordinate is represented spectrum angle matching value.

D, according to the spectrum angle matching value of each pixel, calculate and the corresponding mahalanobis distance of each pixel, as shown in Figure 8, Fig. 8 is the mahalanobis distance synoptic diagram of each pixel in the testing image of the embodiment of the invention, horizontal ordinate among the figure is represented the pixel number, and ordinate is represented the mahalanobis distance value.

E, the pairing mahalanobis distance of each pixel spectrum is carried out singular value detect, pairing mahalanobis distance value of each pixel spectrum and predetermined threshold value are compared, and definite mahalanobis distance value is the target pixel greater than the pixel of predetermined threshold, the target pixel point at testing image center is detected accurately as can be seen from result of detection synoptic diagram shown in Figure 9, and result of detection conforms to actual conditions.

For realizing the high light spectrum small target detection method of the invention described above, the present invention also provides a kind of high light spectrum small target detection device, as shown in figure 10, this device comprises: two-dimensional matrix acquiring unit 10, pretreatment unit 20, angle matching unit 30, mahalanobis distance acquiring unit 40 and little impact point determining unit 50.Two-dimensional matrix acquiring unit 10 is used for obtaining the pixel spectrum two-dimensional matrix of the three-dimensional high-spectral data of testing image pixel.Pretreatment unit 20 connects two-dimensional matrix acquiring unit 10, the pre-service that the pixel spectrum two-dimensional matrix that is used for that two-dimensional matrix acquiring unit 10 is obtained is proofreaied and correct spectral error, and the pixel spectrum two-dimensional matrix after obtaining handling offers angle matching unit 30.Pretreated method can remove be handled for continuum, orthonormal transformation is handled and add at least a in handling of scatter correction, and pretreated purpose is to proofread and correct the spectral error that causes because of atmospheric scattering.Angle matching unit 30 connects pretreatment unit 20, is used for obtaining the spectrum angle matching value of each pixel with carrying out the angle coupling through each the pixel spectrum of pretreated pixel spectrum two-dimensional matrix and the averaged spectrum of testing image pixel.Mahalanobis distance acquiring unit 40, joint angle matching unit 30 is used for obtaining the mahalanobis distance corresponding with each pixel according to spectrum angle matching value.Little impact point determining unit 50 connects mahalanobis distance acquiring unit 40, be used for the mahalanobis distance and the predetermined threshold value of each pixel correspondence are compared, and definite mahalanobis distance is little impact point greater than the pixel point of predetermined threshold.

In sum, high light spectrum small target detection method provided by the present invention and device carry out the detection of spectrum singular point from spectrum dimension angle, thereby realize the detection to unusual target, need not any prior imformation; The present invention is in testing process, and the method for employing spectrum angle coupling is compressed to one dimension with the spectroscopic data of multidimensional, and outstanding singular value target information, so operand is little, fast operation; The present invention adopts the method that adapts to threshold value in mahalanobis distance singular value testing process, can carry out small target detection on the basis of pixel spectrum statistical model the unknown, and the scope of application is more extensive, and the accuracy rate of detection is higher; In addition, The present invention be directed to the operation that spectrum dimension is carried out, promptly carry out computing after can gathering the spectrum of a pixel point, the target optical spectrum that need not to wait for whole zone is all gathered and is finished, thereby has better met the requirement of target detection on real-time.

The above is preferred embodiment of the present invention only, is not to be used to limit protection scope of the present invention.

Claims (9)

1, a kind of high light spectrum small target detection method is characterized in that, comprising:

Obtain the pixel spectrum two-dimensional matrix of three-dimensional high-spectral data in the testing image pixel;

The averaged spectrum of each pixel spectrum in the described pixel spectrum two-dimensional matrix and described testing image pixel is carried out the angle coupling, obtain the spectrum angle matching value of each pixel;

According to described spectrum angle matching value, obtain and the corresponding mahalanobis distance of described each pixel;

The mahalanobis distance and the predetermined threshold of described each pixel correspondence are compared, and definite mahalanobis distance is little impact point greater than the pixel point of predetermined threshold.

According to the described high light spectrum small target detection method of claim 1, it is characterized in that 2, the described pixel spectrum two-dimensional matrix that obtains three-dimensional high-spectral data in the testing image pixel specifically comprises:

Three-dimensional high-spectral data in the described testing image pixel is expressed as the pixel spectrum two-dimensional matrix of high spectrum reflection rate:

R _m×n＝[p ₁，p ₂...p _y×i+j...p _x×y]，0＜i≤x，0＜j≤y，

Or R _{M * n}=[p ₁, p ₂... p _{I+x * j}... p _{X * y}], 0＜i≤x, 0＜j≤y,

Wherein, R _{M * n}Expression pixel spectrum two-dimensional matrix, [p ₁, p ₂... p _{Y * i+j}... p _{X * y}] and [p ₁, p ₂... p _{I+x * j}... p _{X * y}] expression testing image pixel spectrum vector, m represents the wave band number, n represents the sum of testing image pixel, x represents the line number of testing image pixel, y represents the columns of testing image pixel, n=x * y.

3, according to the described high light spectrum small target detection method of claim 1, it is characterized in that, after the described two-dimensional matrix that obtains three-dimensional high-spectral data in the testing image pixel, this method also comprises: to the pre-service that the pixel spectrum two-dimensional matrix of described testing image is proofreaied and correct spectral error, the pixel spectrum two-dimensional matrix after obtaining proofreading and correct.

According to the described high light spectrum small target detection method of claim 3, it is characterized in that 4, described pre-service is that continuum remove to be handled, orthonormal transformation is handled and adds at least a in handling of scatter correction.

5, according to the described high light spectrum small target detection method of claim 1, it is characterized in that, described averaged spectrum with each pixel spectrum in the pixel spectrum two-dimensional matrix and testing image pixel is carried out the angle coupling, obtains the spectrum angle matching value of each pixel, specifically comprises:

Obtain the averaged spectrum of described testing image pixel according to described pixel spectrum two-dimensional matrix $\stackrel{\‾}{p}=\frac{1}{n}\underset{h=1}{\overset{n}{\mathrm{\Σ}}}{p}_h,$ Wherein, p _hThe spectrum vector of representing h pixel in the described testing image;

Described each pixel spectrum and the averaged spectrum that obtains are carried out spectrum angle coupling, are calculated as follows as follows:

${\mathrm{\α}}_h={\cos }^{-1}\frac{\stackrel{\‾}{p}\·p_h}{\left|\stackrel{\‾}{p}\right|\left|p_h\right|}={\cos }^{-1}\frac{\underset{k=1}{\overset{m}{\mathrm{\Σ}}}{\stackrel{\‾}{p}}_k{p}_{\mathrm{hk}}}{\sqrt{\underset{k=1}{\overset{m}{\mathrm{\Σ}}}{\left({\stackrel{\‾}{p}}_k\right)}^2}\sqrt{\underset{k=1}{\overset{m}{\mathrm{\Σ}}}{\left(p_{\mathrm{hk}}\right)}^2}}$

Wherein, α _hThe spectrum angle matching value of h pixel in the expression testing image, p _hThe spectrum vector of h pixel in the expression testing image, p represents averaged spectrum vector, p _kExpression averaged spectrum vector is at the reflectance value of k wave band, p _HkH pixel in the expression testing image is at the reflectance value at k wave band place.

6, according to the described high light spectrum small target detection method of claim 1, it is characterized in that, described according to spectrum angle matching value, obtain the mahalanobis distance corresponding and be with each pixel:

${\mathrm{MD}}_h={\left[{({\mathrm{\α}}_h-\stackrel{\‾}{\mathrm{\α}})}^T{M}^{-1}({\mathrm{\α}}_h-\stackrel{\‾}{\mathrm{\α}})\right]}^{\frac{1}{2}},h=\mathrm{1,2},...,n$

Wherein, MD _hThe mahalanobis distance of h pixel in the expression testing image, α _hThe spectrum angle matching value of h pixel in the expression testing image, n represents testing image pixel sum, α represents the mean value of all pixel spectrum angle matching values, $\stackrel{\‾}{\mathrm{\α}}=\frac{1}{n}\underset{h=1}{\overset{n}{\mathrm{\Σ}}}{\mathrm{\α}}_h,$ M ^-1Be the inverse matrix of mahalanobis distance matrix M, $M=\frac{C^{T}C}{n-1},$ C represents by each α in the testing image pixel _hThe spectrum angle matching vector of forming.

According to the described high light spectrum small target detection method of claim 1, it is characterized in that 7, described predetermined threshold is to determine in the following way:

$\mathrm{\Δ}=u+(\mathrm{Max}-u)\sqrt{\frac{u}{\mathrm{Max}}}$

Wherein, Δ is represented predetermined threshold, $u=\frac{1}{n}\underset{h=1}{\overset{n}{\mathrm{\Σ}}}{\mathrm{MD}}_h,$ U represents the mean value of the mahalanobis distance of each pixel in the testing image, and Max is the mahalanobis distance MD of each pixel in the testing image _hIn maximal value.

8, a kind of high light spectrum small target detection device is characterized in that, comprising: two-dimensional matrix acquiring unit, angle matching unit, mahalanobis distance acquiring unit and little impact point determining unit; Wherein,

Described two-dimensional matrix acquiring unit is used for obtaining the pixel spectrum two-dimensional matrix of the three-dimensional high-spectral data of testing image pixel;

Described angle matching unit is used for each the pixel spectrum of described pixel spectrum two-dimensional matrix and the averaged spectrum of described testing image pixel are carried out the angle coupling, obtains the spectrum angle matching value of each pixel;

Described mahalanobis distance acquiring unit is used for according to described spectrum angle matching value, obtains and the corresponding mahalanobis distance of described each pixel;

Described little impact point determining unit is used for the mahalanobis distance and the predetermined threshold value of described each pixel correspondence are compared, and definite mahalanobis distance is little impact point greater than the pixel point of predetermined threshold.

9, described according to Claim 8 high light spectrum small target detection device, it is characterized in that, described device also comprises: pretreatment unit, connect described two-dimensional matrix acquiring unit and angle matching unit, the pre-service that the pixel spectrum two-dimensional matrix that is used for that described two-dimensional matrix acquiring unit is obtained is proofreaied and correct spectral error, the pixel spectrum two-dimensional matrix after obtaining handling offers described angle matching unit.

Images