CN106650645A - Method and device for extracting target feature of remote sensing image - Google Patents

Abstract

The invention relates to a method and a device for extracting a target feature of a remote sensing image. The method comprises the steps of dividing a target remote sensing image into a plurality of regional images, wherein each regional image comprises a plurality of pixel grids; performing interpolation on each regional image, and acquiring an interpolated array of each regional image; determining a pre-estimation array of a target feature parameter of each regional image according to the interpolated array of each regional image; determining the target feature parameter of each regional image through Gauss-Newton iterative method according to the pre-estimation array of the target feature parameter of each regional image; and determining a target feature parameter of the target remote sensing image according to the target feature parameter of each regional image. Therefore, features of a distributed target of the remote sensing image can be extracted effectively and accurately.

Description

The target's feature-extraction method and apparatus of remote sensing images

Technical field

It relates to signal processing technology field, in particular it relates to a kind of target's feature-extraction method of remote sensing images And device.

Background technology

Remote sensing images (Remote Sensing Image) refer to film (or the phase for recording various atural object electromagnetic wave sizes Piece) image, mainly include aerial image and satellite image.For example, synthetic aperture radar (Synthetic Aperture Radar, SAR) have a wide range of applications at aspects such as mapping, disaster monitorings.

The remote sensing images that SAR is generated contain amplitude, phase place and the positional information of target.Wherein phase information is for landform Mapping etc. applies extremely important, therefore, it is possible to realize that the efficient algorithm that target signature is accurately extracted is unusual necessity.Current mesh Mark feature is mainly extracted by the method for interpolation, and this method is simply effective, very high for single-point aimed at precision, but works as and apply During to distribution objectives, due to influencing each other between adjacent target, it is impossible to obtain accurate characteristic information.

The content of the invention

The purpose of the disclosure is to provide a kind of target's feature-extraction method and apparatus of simple and effective remote sensing images.

To achieve these goals, the disclosure provides a kind of target's feature-extraction method of remote sensing images.Methods described bag Include：Object in Remote Sensing is divided into into multiple area images, each area image includes multiple pixel grids；To each administrative division map As entering row interpolation, array after the interpolation of each area image is obtained；According to array after the interpolation of each area image, each is determined The pre-estimation array of the target signature parameter of area image；Counting is estimated according to the target signature parameter of each area image Group, by Gaussian-Newton method, determines the target signature parameter of each area image；According to the target of each area image Characteristic parameter determines the target signature parameter of the Object in Remote Sensing.

Alternatively, it is described that row interpolation is entered to each area image, the step of obtain array after the interpolation of each area image Including：Two dimensional discrete Fourier transform is carried out to corresponding first array of pixel grid of an area image, the second array is generated； Frequency domain zero padding is carried out to second array, the 3rd array is generated；Two-dimensional discrete Fourier inversion is carried out to the 3rd array Change, generate array after the interpolation of the area image.

Alternatively, array after the interpolation according to each area image, determines the target signature ginseng of each area image The step of several pre-estimation array, includes：According in array after the interpolation of an area image, amplitude maximum in a pixel grid The array of point, determines the pre-estimation array of the target signature parameter in the pixel grid；According to number after the interpolation of the area image In group, the pre-estimation array of the target signature parameter in each pixel grid determines the target signature parameter of the area image Pre-estimation array.

Alternatively, the pre-estimation array of the target signature parameter according to each area image, is changed by Gauss-Newton The step of Dai Fa, target signature parameter for determining each area image, includes：According to array after the interpolation of each area image and The pre-estimation array of target signature parameter, sets up the recursive iteration equation of corresponding area image；By Gauss-Newton iteration Method, calculates respectively the real part solution and imaginary part solution of the recursive iteration equation of each area image；According to the recurrence of each area image The real part solution and imaginary part solution of iterative equation, determines the target signature parameter of each area image.

Alternatively, it is described by Gaussian-Newton method, the reality of the recursive iteration equation of each area image is calculated respectively The step of portion is solved with imaginary part solution includes：By Gaussian-Newton method, the recursive iteration equation of each area image is calculated respectively Real part solution, wherein, in the area image after interpolation, the reconstruction data real part R of j-th sampled point_jFor：

R_j=Kr₁sinc(BW(nx_j-Nx₁,ny_j-Ny₁))+Kr₂sinc(BW(nx_j-Nx₂,ny_j-Ny₂))+…+Kr_isinc (BW(nx_j-Nx_i,ny_j-Ny_i))

Target function gradient matrix A r of structure_KFor：

The recursive iteration equation of structure is：

The initial estimation of target signature parameter real part is：

Wherein, BW is bandwidth, nx_jFor j-th sampled point in the area image after interpolation, R_jFor the reconstruction of j-th sampled point Data real part, Kr_iFor the real part of the point of amplitude maximum in i-th grid in an area image, Nx_i、Ny_iRespectively i-th grid The abscissa and ordinate of the maximum point of middle amplitude, k is iterations, Ar_KFor target function gradient matrix, τ r_kAdopt for j-th The difference of the number of samples factually reconstruction data real part of portion and the point, value () is the data of j-th sampled point, and Re () is number Value real part, xr^(k)For the real part of the kth time corresponding target signature parameter of iteration, δ r^(k)For real part error term, ()^-1For matrix It is inverse, ()^TFor transposed matrix, wherein, solving δ r^(k)Afterwards, xr is put^(k+1)=xr^(k)+δr^(k), k=k+1, whenWhen, stop iteration；

By Gaussian-Newton method, the imaginary part solution of the recursive iteration equation of each area image is calculated respectively, wherein, In area image after interpolation, reconstruction data imaginary part I of j-th sampled point_jFor：

I_j=Ki₁sinc(BW(nx_j-Nx₁,ny_j-Ny₁))+Ki₂sinc(BW(nx_j-Nx₂,ny_j-Ny₂))+…+Ki_isinc (BW(nx_j-Nx_i,ny_j-Ny_i))

Target function gradient matrix A i of structure_kFor：

The recursive iteration equation of structure is：

The initial estimation of target signature parameter imaginary part is：

Wherein, BW is bandwidth, nx_jFor j-th sampled point in the area image after interpolation, I_jFor the reconstruction of j-th sampled point Data imaginary part, Ki_iFor the imaginary part of the point of amplitude maximum in i-th grid in an area image, Nx_i、Ny_iRespectively i-th grid The abscissa and ordinate of the maximum point of middle amplitude, k is iterations, Ai_KFor target function gradient matrix, τ i_kAdopt for j-th The difference of the reconstruction data imaginary part of sampling point data imaginary part and the point, value () is the data of j-th sampled point, and Im () is number Value imaginary part, xi^(k)For the imaginary part of the kth time corresponding target signature parameter of iteration, δ i^(k)For imaginary part error term, ()^-1For matrix It is inverse, ()^TFor transposed matrix, wherein, solving δ i^(k)Afterwards, xi is put^(k+1)=xi^(k)+δi^(k), k=k+1, whenWhen, stop iteration.

The disclosure also provides a kind of target's feature-extraction device of remote sensing images.Described device includes：Division module, is used for Object in Remote Sensing is divided into into multiple area images, each area image includes multiple pixel grids；Interpolating module, for right Each area image enters row interpolation, obtains array after the interpolation of each area image；Pre-estimation array determining module, for basis Array after the interpolation of each area image, determines the pre-estimation array of the target signature parameter of each area image；Provincial characteristics Determining module, for according to the pre-estimation array of the target signature parameter of each area image, by Gaussian-Newton method, Determine the target signature parameter of each area image；Characteristics of image determining module, for special according to the target of each area image Levy the target signature parameter of Object in Remote Sensing described in parameter determination.

By above-mentioned technical proposal, the point in the area image that Gaussian-Newton method iterative goes out after interpolation Clarification of objective parameter such that it is able to efficiently and accurately extract the feature of the distribution objectives of remote sensing images.

Other feature and advantage of the disclosure will be described in detail in subsequent specific embodiment part.

Description of the drawings

Accompanying drawing is, for providing further understanding of the disclosure, and to constitute the part of specification, with following tool Body embodiment is used to explain the disclosure together, but does not constitute restriction of this disclosure.In the accompanying drawings：

Fig. 1 is the flow chart of the target's feature-extraction method of the remote sensing images that an exemplary embodiment is provided；

Fig. 2 is the flow chart for entering row interpolation to area image that an exemplary embodiment is provided；

Fig. 3 is the flow chart of the target signature parameter of determination each area image that an exemplary embodiment is provided；

Fig. 4 is the block diagram of the target's feature-extraction device of the remote sensing images that an exemplary embodiment is provided；

Fig. 5 is the 3-D view of the remote sensing images built according to simulation model that an exemplary embodiment is provided；

Fig. 6 is the 3-D view of the remote sensing images built according to disclosed method that an exemplary embodiment is provided.

Specific embodiment

It is described in detail below in conjunction with accompanying drawing specific embodiment of this disclosure.It should be appreciated that this place is retouched The specific embodiment stated is merely to illustrate and explains the disclosure, is not limited to the disclosure.

Fig. 1 is the flow chart of the target's feature-extraction method of the remote sensing images that an exemplary embodiment is provided.Such as Fig. 1 institutes Show, methods described may comprise steps of.

In step s 11, Object in Remote Sensing is divided into into multiple area images, each area image includes multiple pixels Grid.

It is, to accelerate solving speed, to Object in Remote Sensing subarea processing is carried out.Such as Object in Remote Sensing Pixel is MN, and each area image includes S=NaNr pixel grid, then the number for dividing back zone area image isWherein,For the symbol that rounds up.In view of the problem of amount of calculation, Na and Nr can not obtain excessive, Under normal circumstances, Na=Nr=8 can be taken.

In step s 12, row interpolation is entered to each area image, obtains array after the interpolation of each area image.

When meshes number S is less, iterative processing error is larger, it is therefore desirable to first enter row interpolation to area image, to obtain More multi-point sampling is obtained, the accuracy of result is improved.Fig. 2 is the stream for entering row interpolation to area image that an exemplary embodiment is provided Cheng Tu.As shown in Fig. 2 enter row interpolation to each area image, (step the step of obtain array after the interpolation of each area image S12) may comprise steps of.

In step S121, two-dimensional discrete Fourier change is carried out to corresponding first array of pixel grid of an area image Change, generate the second array.

For example, an area image includes S=NaNr pixel grid, and the corresponding pixel data of total-grid constitutes one Size is first array Data of NaNr_k.To the first array Data_kCarry out two dimensional discrete Fourier transform：

Generate the second array Data_f.Wherein, u-v is frequency domain coordinates system, and x-y is spatial domain coordinate system.

In step S122, frequency domain zero padding is carried out to the second array, generate the 3rd array.

Then, to the second array Data_fCarry out frequency domain zero padding.If interpolation multiple is TimesTimes (for example, Times =4).Through interpolation processing, can be by the second array Data_fExpand to the 3rd array Data that a size is NxNy_padding。 Wherein Nx=NaTimes, Ny=NrTimes.Data_paddingIn matrix, theRow is to the OK,Arrange toIt is classified as Data_f, remaining element all 0.

In step S123, two-dimensional discrete Fourier inverse transformation is carried out to the 3rd array, generate the interpolation of the area image Array afterwards.

To the 3rd array Data_paddingCarry out two-dimensional discrete Fourier inverse transformation：

Obtain array Data after the interpolation of the area image that size is NxNy.

Fig. 1 is returned, in step s 13, according to array after the interpolation of each area image, the mesh of each area image is determined The pre-estimation array of mark characteristic parameter.

It is, the region division and interpolation processing before, rough target component can be estimated first.It is theoretical On, pre-estimation array can determine according to the corresponding array of any pixel grid.In one embodiment, step S13 can be with Including step S131 and step S132.

In step S131, according in array after the interpolation of an area image, the point of amplitude maximum in a pixel grid Array, determines the pre-estimation array of the target signature parameter in the pixel grid.

For example, in an area image, altogether comprising NaNr pixel grid.Through interpolation processing, in single pixel net TimesTimes point is included in lattice.The point for being located at amplitude maximum in i-th grid is X_i, the real part is Kr_i=Re (X_i), imaginary part is Ki_i=Im (X_i), displacement abscissa value is Nx_i, displacement ordinate value is Ny_i, then target is special in the pixel grid The pre-estimation array for levying parameter is [Kr_i,Ki_i,Nx_i,Ny_i]。

In step S132, according to the target signature ginseng in array after the interpolation of the area image, in each pixel grid Several pre-estimation arrays, determines the pre-estimation array of the target signature parameter of the area image.

In step S132 on the basis of example, the area image includes S=NaNr pixel grid, then have 4S Parameter.These parametric joints are got up, the matrix X that size is S4 can be obtained, the target of the i.e. area images of matrix X is special Levy pre-estimation array X of parameter.

In step S14, according to the pre-estimation array of the target signature parameter of each area image, by Gauss-Newton Iterative method, determines the target signature parameter of each area image.

Fig. 3 is the flow chart of the target signature parameter of determination each area image that an exemplary embodiment is provided.Such as Fig. 3 It is shown, according to the pre-estimation array of the target signature parameter of each area image, by Gaussian-Newton method, determine each The step of target signature parameter of area image (step S14), may comprise steps of.

In step s 141, according to array after the interpolation of each area image and the pre-estimation array of target signature parameter, Set up the recursive iteration equation of corresponding area image.

Two groups of parameters can be read.First group be an area image interpolation after array Data；Second group is the administrative division map Pre-estimation array X of the target signature parameter of picture.The recursive iteration equation of corresponding area image is set up according to Data and X, Target signature parameter is asked for according to the recursive iteration equation set up.

In step S142, by Gaussian-Newton method, the recursive iteration equation of each area image is calculated respectively Real part solution and imaginary part solution.

, wherein it is desired to arrange sampled data and iterative initial value.Sampled data can be set as follows：

Array Data is the two-dimensional matrix of NxNy sizes after known interpolation, if abscissa sampled point nx=[1,2 ... Nx- 1, Nx], ordinate sampled point ny=[1,2 ... Ny-1, Ny].The data two-dimensional matrix Data of input is redeveloped into into one-dimensional columns Group value, size is NxNy × 1, wherein the 1st to the Ny element is the first column data in Data, Ny+1 to 2Ny unit Element is the second column data in Data, the like.

Iterative initial value can be set as follows：

4S precompensation parameter is read from pre-estimation array X, X is divided into into the columns group that 4 groups of sizes are S1 by row：K_r, K_i, N_x, N_y。K_r, K_i, N_x, N_yValue of real part array, imaginary values array, x directions displacement array, y directions displacement array are represented respectively, Using these data as iterative initial value.

In an embodiment of the disclosure, step S142 can include realistic portion solution and imaginary part solution, individually below in detail Description.

By Gaussian-Newton method, the real part solution of the recursive iteration equation of each area image is calculated respectively, wherein, In area image after interpolation, the reconstruction data real part R of j-th sampled point_jFor：

R_j=Kr₁sinc(BW(nx_j-Nx₁,ny_j-Ny₁))+Kr₂sinc(BW(nx_j-Nx₂,ny_j-Ny₂))+…+Kr_isinc (BW(nx_j-Nx_i,ny_j-Ny_i))

Target function gradient matrix A r of structure_KFor：

The recursive iteration equation of structure is：

The initial estimation of target signature parameter real part is：

Wherein, BW is bandwidth, nx_jFor j-th sampled point in the area image after interpolation, R_jFor the reconstruction of j-th sampled point Data real part, Kr_iFor the real part of the point of amplitude maximum in i-th grid in an area image, Nx_i、Ny_iRespectively i-th grid The abscissa and ordinate of the maximum point of middle amplitude, k is iterations, Ar_KFor target function gradient matrix, τ r_kAdopt for j-th The difference of the number of samples factually reconstruction data real part of portion and the point, value () is the data of j-th sampled point, and Re () is number Value real part, xr^(k)For the real part of the kth time corresponding target signature parameter of iteration, δ r^(k)For real part error term, ()^-1For matrix It is inverse, ()^TFor transposed matrix, wherein, solving δ r^(k)Afterwards, xr is put^(k+1)=xr^(k)+δr^(k), k=k+1, whenWhen, stop iteration.Now i.e. it is believed that reaching precision needed for solution target, iteration ends, xr^(k)For final result.

Similarly, the void of the recursive iteration equation of each area image can respectively be calculated by Gaussian-Newton method Portion solves, wherein, in the area image after interpolation, reconstruction data imaginary part I of j-th sampled point_jFor：

I_j=Ki₁sinc(BW(nx_j-Nx₁,ny_j-Ny₁))+Ki₂sinc(BW(nx_j-Nx₂,ny_j-Ny₂))+…+Ki_isinc (BW(nx_j-Nx_i,ny_j-Ny_i))

Target function gradient matrix A i of structure_kFor：

The recursive iteration equation of structure is：

The initial estimation of target signature parameter imaginary part is：

Wherein, BW is bandwidth, nx_jFor j-th sampled point in the area image after interpolation, I_jFor the reconstruction of j-th sampled point Data imaginary part, Ki_iFor the imaginary part of the point of amplitude maximum in i-th grid in an area image, Nx_i、Ny_iRespectively i-th grid The abscissa and ordinate of the maximum point of middle amplitude, k is iterations, Ai_KFor target function gradient matrix, τ i_kAdopt for j-th The difference of the reconstruction data imaginary part of sampling point data imaginary part and the point, value () is the data of j-th sampled point, and Im () is number Value imaginary part, xi^(k)For the imaginary part of the kth time corresponding target signature parameter of iteration, δ i^(k)For imaginary part error term, ()^-1For matrix It is inverse, ()^TFor transposed matrix, wherein, solving δ i^(k)Afterwards, xi is put^(k+1)=xi^(k)+δi^(k), k=k+1, whenWhen, stop iteration.Now i.e. it is believed that reaching precision needed for solution target, iteration ends, xi^(k)For final result.

In step S143, according to the real part solution and imaginary part solution of the recursive iteration equation of each area image, each is determined The target signature parameter of area image.

By real part solution and imaginary part solution directly in conjunction with getting up, it is possible to obtain the target signature parameter of each area image.

N after iteration_x, N_yValue is the solution of abscissa and ordinate, solves parameters.Phase place solutionAmplitude solutionSo far, the target signature of each area image can be extracted.Each The amplitude of each target, phase place, orientation exact position, distance can accurately draw to exact position in area image.

Return Fig. 1, in step S15, the target remote sensing figure according to the target signature parameter determination of each area image The target signature parameter of picture.

According to the method described above, the multiple area images for being divided all are disposed, you can obtain Object in Remote Sensing Target signature parameter.

By above-mentioned technical proposal, the point in the area image that Gaussian-Newton method iterative goes out after interpolation Clarification of objective parameter such that it is able to efficiently and accurately extract the feature of the distribution objectives of remote sensing images.

The disclosure additionally provides a kind of target's feature-extraction device of remote sensing images.Fig. 4 is that an exemplary embodiment is provided Remote sensing images target's feature-extraction device block diagram.As shown in figure 4, the target's feature-extraction device 10 of remote sensing images can be with It is true including division module 11, interpolating module 12, pre-estimation array determining module 13, provincial characteristics determining module 14 and characteristics of image Cover half block 15.

Division module 11 is used to for Object in Remote Sensing to be divided into multiple area images, and each area image includes multiple pictures Plain grid.

Interpolating module 12 is used to enter row interpolation to each area image, obtains array after the interpolation of each area image.

Pre-estimation array determining module 13 is used for according to array after the interpolation of each area image, determines each area image Target signature parameter pre-estimation array.

Provincial characteristics determining module 14 is used for the pre-estimation array of the target signature parameter according to each area image, passes through Gaussian-Newton method, determines the target signature parameter of each area image.

Characteristics of image determining module 15 is used for the target remote sensing according to the target signature parameter determination of each area image The target signature parameter of image.

Alternatively, the interpolating module 12 can include transformation submodule, frequency domain zero padding submodule and inverse transformation submodule.

Transformation submodule is used to carry out two-dimensional discrete Fourier to corresponding first array of pixel grid of an area image Conversion, generates the second array.

Frequency domain zero padding submodule is used to carry out second array frequency domain zero padding, generates the 3rd array.

Inverse transformation submodule is used to carry out two-dimensional discrete Fourier inverse transformation to the 3rd array, generates the area image Interpolation after array.

Alternatively, the pre-estimation array determining module 13 can include grid pre-estimation array determination sub-module and region Pre-estimation array determination sub-module.

Grid pre-estimation array determination sub-module is used for according in array after the interpolation of an area image, in a pixel grid The array of the point of amplitude maximum, determines the pre-estimation array of the target signature parameter in the pixel grid.

Region pre-estimation array determination sub-module is used for according in array after the interpolation of the area image, each pixel grid The pre-estimation array of interior target signature parameter, determines the pre-estimation array of the target signature parameter of the area image.

Alternatively, the provincial characteristics determining module 14 includes establishing equation submodule, real imaginary part calculating sub module and area Characteristic of field determination sub-module.

Establishing equation submodule is used for the pre-estimation according to array and target signature parameter after the interpolation of each area image Array, sets up the recursive iteration equation of corresponding area image.

Real imaginary part calculating sub module is used for by Gaussian-Newton method, and the recurrence that each area image is calculated respectively changes For the real part solution and imaginary part solution of equation.

Provincial characteristics determination sub-module is used for the real part solution and imaginary part solution of the recursive iteration equation according to each area image, Determine the target signature parameter of each area image.

Alternatively, the real imaginary part calculating sub module includes real part calculating sub module and imaginary part calculating sub module.

Real part calculating sub module is used for by Gaussian-Newton method, and the recursive iteration of each area image is calculated respectively The real part solution of equation.Wherein, in the area image after interpolation, the reconstruction data real part R of j-th sampled point_jFor：

R_j=Kr₁sinc(BW(nx_j-Nx₁,ny_j-Ny₁))+Kr₂sinc(BW(nx_j-Nx₂,ny_j-Ny₂))+…+Kr_isinc (BW(nx_j-Nx_i,ny_j-Ny_i))

Target function gradient matrix A r of structure_KFor：

The recursive iteration equation of structure is：

The initial estimation of target signature parameter real part is：

Wherein, BW is bandwidth, nx_jFor j-th sampled point in the area image after interpolation, R_jFor the reconstruction of j-th sampled point Data real part, Kr_iFor the real part of the point of amplitude maximum in i-th grid in an area image, Nx_i、Ny_iRespectively i-th grid The abscissa and ordinate of the maximum point of middle amplitude, k is iterations, Ar_KFor target function gradient matrix, τ r_kAdopt for j-th The difference of the number of samples factually reconstruction data real part of portion and the point, value () is the data of j-th sampled point, and Re () is number Value real part, xr^(k)For the real part of the kth time corresponding target signature parameter of iteration, δ r^(k)For real part error term, () -1 is matrix Inverse, ()^TFor transposed matrix, wherein, solving δ r^(k)Afterwards, xr is put^(k+1)=xr^(k)+δr^(k), k=k+1, whenWhen, stop iteration.

Imaginary part calculating sub module is used for by Gaussian-Newton method, and the recursive iteration of each area image is calculated respectively The imaginary part solution of equation.Wherein, in the area image after interpolation, reconstruction data imaginary part I of j-th sampled point_jFor：

I_j=Ki₁sinc(BW(nx_j-Nx₁,ny_j-Ny₁))+Ki₂sinc(BW(nx_j-Nx₂,ny_j-Ny₂))+…+Ki_isinc (BW(nx_j-Nx_i,ny_j-Ny_i))

Target function gradient matrix A i of structure_kFor：

The recursive iteration equation of structure is：

The initial estimation of target signature parameter imaginary part is：

Wherein, BW is bandwidth, nx_jFor j-th sampled point in the area image after interpolation, I_jFor the reconstruction of j-th sampled point Data imaginary part, Ki_iFor the imaginary part of the point of amplitude maximum in i-th grid in an area image, Nx_i、Ny_iRespectively i-th grid The abscissa and ordinate of the maximum point of middle amplitude, k is iterations, Ai_KFor target function gradient matrix, τ i_kAdopt for j-th The difference of the reconstruction data imaginary part of sampling point data imaginary part and the point, value () is the data of j-th sampled point, and Im () is number Value imaginary part, xi^(k)For the imaginary part of the kth time corresponding target signature parameter of iteration, δ i^(k)For imaginary part error term, ()^-1For matrix It is inverse, ()^TFor transposed matrix, wherein, solving δ i^(k)Afterwards, xi is put^(k+1)=xi^(k)+δi^(k), k=k+1, whenWhen, stop iteration.

With regard to the device in above-described embodiment, wherein modules perform the concrete mode of operation in relevant the method Embodiment in be described in detail, explanation will be not set forth in detail herein.

By above-mentioned technical proposal, the point in the area image that Gaussian-Newton method iterative goes out after interpolation Clarification of objective parameter such that it is able to efficiently and accurately extract the feature of the distribution objectives of remote sensing images.

In order to verify the validity of disclosed method, inventor is respectively according to simulation model and according to disclosed method Remote sensing images are processed.It is as shown in table 1 according to the target component that simulation model is obtained.

The target component that table 1 is obtained according to simulation model

Parameter	Value
		The number of point target in remote sensing images	3
The amplitude of point target	20,60,5
		Phase place (the unit of point target：°)	50,80,50
The space coordinates (x-axis) of analog image	2.4,2.4,1.8
		The space coordinates (y-axis) of analog image	3.4,3.9,2.2
Signal bandwidth	200MHz
		Orientation pulse recurrence frequency	5000Hz

It is as shown in table 2 according to the target component that disclosed method is obtained.

The target component that table 2 is obtained according to disclosed method

Can be seen that by Tables 1 and 2, the target component obtained according to simulation model and the mesh obtained according to disclosed method Mark parameter is completely the same.

Fig. 5 is the 3-D view of the remote sensing images built according to simulation model that an exemplary embodiment is provided.Fig. 6 is one The 3-D view of the remote sensing images built according to disclosed method that exemplary embodiment is provided.Can be seen that by Fig. 5 and Fig. 6, The 3-D view that obtained according to simulation model and the 3-D view obtained according to disclosed method are completely the same.Therefore, this public affairs Opening the target's feature-extraction method and apparatus of the remote sensing images of offer can efficiently and accurately extract the feature of distribution objectives.

The preferred embodiment of the disclosure is described in detail above in association with accompanying drawing, but, the disclosure is not limited to above-mentioned reality The detail in mode is applied, in the range of the technology design of the disclosure, various letters can be carried out with technical scheme of this disclosure Monotropic type, these simple variants belong to the protection domain of the disclosure.

It is further to note that each particular technique feature described in above-mentioned specific embodiment, in not lance In the case of shield, can be combined by any suitable means.In order to avoid unnecessary repetition, the disclosure to it is various can The combination of energy is no longer separately illustrated.

Additionally, can also be combined between a variety of embodiments of the disclosure, as long as it is without prejudice to this Disclosed thought, it should equally be considered as disclosure disclosure of that.

Claims (10)

1. a kind of target's feature-extraction method of remote sensing images, it is characterised in that methods described includes：

Object in Remote Sensing is divided into into multiple area images, each area image includes multiple pixel grids；

Row interpolation is entered to each area image, array after the interpolation of each area image is obtained；

According to array after the interpolation of each area image, the pre-estimation array of the target signature parameter of each area image is determined；

According to the pre-estimation array of the target signature parameter of each area image, by Gaussian-Newton method, each area is determined The target signature parameter of area image；

The target signature parameter of Object in Remote Sensing according to the target signature parameter determination of each area image.

2. method according to claim 1, it is characterised in that described to enter row interpolation to each area image, obtains each Include the step of array after the interpolation of area image：

Two dimensional discrete Fourier transform is carried out to corresponding first array of pixel grid of an area image, the second array is generated；

Frequency domain zero padding is carried out to second array, the 3rd array is generated；

Two-dimensional discrete Fourier inverse transformation is carried out to the 3rd array, array after the interpolation of the area image is generated.

3. method according to claim 1, it is characterised in that array after the interpolation according to each area image, really The step of pre-estimation array of the target signature parameter of fixed each area image, includes：

According in array after the interpolation of an area image, the array of the point of amplitude maximum, determines the pixel network in a pixel grid The pre-estimation array of the target signature parameter in lattice；

According in array after the interpolation of the area image, the pre-estimation array of the target signature parameter in each pixel grid, really The pre-estimation array of the target signature parameter of the fixed area image.

4. method according to claim 1, it is characterised in that the target signature parameter according to each area image Pre-estimation array, includes the step of by Gaussian-Newton method, the target signature parameter for determining each area image：

According to array after the interpolation of each area image and the pre-estimation array of target signature parameter, corresponding area image is set up Recursive iteration equation；

By Gaussian-Newton method, the real part solution and imaginary part solution of the recursive iteration equation of each area image are calculated respectively；

According to the real part solution and imaginary part solution of the recursive iteration equation of each area image, the target signature of each area image is determined Parameter.

5. method according to claim 4, it is characterised in that described by Gaussian-Newton method, calculates respectively each The step of real part solution and imaginary part solution of the recursive iteration equation of area image, includes：

By Gaussian-Newton method, the real part solution of the recursive iteration equation of each area image is calculated respectively, wherein, interpolation In area image afterwards, the reconstruction data real part R of j-th sampled point_jFor：

R_j=Kr₁sinc(BW(nx_j-Nx₁,ny_j-Ny₁))+Kr₂sinc(BW(nx_j-Nx₂,ny_j-Ny₂))+…+Kr_isinc(BW (nx_j-Nx_i,ny_j-Ny_i))

$\sin c\left(x\right)=\frac{sin\left(\mathrm{\π}x\right)}{\mathrm{\π}x}$

$\sin c(x,y)={\left(\frac{sin\left(\mathrm{\π}x\right)}{\mathrm{\π}x}\right)}^T\·\left(\frac{sin\left(\mathrm{\π}y\right)}{\mathrm{\π}y}\right)$

Target function gradient matrix A r of structure_KFor：

${\mathrm{Ar}}_K=\left\{\begin{array}{ccccccccc}\frac{{\mathrm{dR}}_1}{{\mathrm{dKr}}_1}& \frac{{\mathrm{dR}}_1}{{\mathrm{dKr}}_2}& \mathrm{...}& \frac{{\mathrm{dR}}_1}{{\mathrm{dNx}}_1}& \frac{{\mathrm{dR}}_1}{{\mathrm{dNx}}_2}& \mathrm{...}& \frac{{\mathrm{dR}}_1}{{\mathrm{dNy}}_1}& \frac{{\mathrm{dR}}_1}{{\mathrm{dNy}}_2}& \mathrm{...}\\ \frac{{\mathrm{dR}}_2}{{\mathrm{dKr}}_1}& \frac{{\mathrm{dR}}_2}{{\mathrm{dKr}}_2}& \mathrm{...}& \frac{{\mathrm{dR}}_2}{{\mathrm{dNx}}_1}& \frac{{\mathrm{dR}}_2}{{\mathrm{dNx}}_2}& \mathrm{...}& \frac{{\mathrm{dR}}_2}{{\mathrm{dNy}}_1}& \frac{{\mathrm{dR}}_2}{{\mathrm{dNy}}_2}& \mathrm{...}\\ .& .& .& .& .& .& .& .& .\\ .& .& .& .& .& .& .& .& .\\ .& .& .& .& .& .& .& .& .\end{array}\right\}$

${\mathrm{\τr}}_k=\left\{\begin{array}{c}{R}_1-R\left(value\right(1\left)\right)\\ R_2-R\left(value\right(2\left)\right)\\ .\\ .\\ .\end{array}\right\}$

The recursive iteration equation of structure is：

${\mathrm{xr}}^{(k+1)}={\mathrm{xr}}^{\left(k\right)}+{\mathrm{\δr}}^{\left(k\right)}={\mathrm{xr}}^{\left(k\right)}-{\left({\mathrm{Ar}}_K^T{\mathrm{Ar}}_K\right)}^{-1}{\mathrm{Ar}}_K^T{\mathrm{\τr}}_k$

The initial estimation of target signature parameter real part is：

${\mathrm{xr}}^{\left(1\right)}=\left\{\begin{array}{c}Kr\\ Nx\\ Ny\end{array}\right\}$

Wherein, BW is bandwidth, nx_jFor j-th sampled point in the area image after interpolation, R_jFor the reconstruction data of j-th sampled point Real part, Kr_iFor the real part of the point of amplitude maximum in i-th grid in an area image, Nx_i、Ny_iWidth in respectively i-th grid The abscissa and ordinate of the maximum point of degree, k is iterations, Ar_KFor target function gradient matrix, τ r_kFor j-th sampled point The difference of the reconstruction data real part of data real part and the point, value () is the data of j-th sampled point, and Re () is numerical value reality Portion, xr^(k)For the real part of the kth time corresponding target signature parameter of iteration, δ r^(k)For real part error term, ()^-1For inverse of a matrix, (·)^TFor transposed matrix, wherein, solving δ r^(k)Afterwards, xr is put^(k+1)=xr^(k)+δr^(k), k=k+1, whenWhen, stop iteration；

By Gaussian-Newton method, the imaginary part solution of the recursive iteration equation of each area image is calculated respectively, wherein, interpolation In area image afterwards, reconstruction data imaginary part I of j-th sampled point_jFor：

I_j=Ki₁sinc(BW(nx_j-Nx₁,ny_j-Ny₁))+Ki₂sinc(BW(nx_j-Nx₂,ny_j-Ny₂))+…+Ki_isinc(BW (nx_j-Nx_i,ny_j-Ny_i))

$\sin c\left(x\right)=\frac{sin\left(\mathrm{\π}x\right)}{\mathrm{\π}x}$

$\sin c(x,y)={\left(\frac{sin\left(\mathrm{\π}x\right)}{\mathrm{\π}x}\right)}^T\·\left(\frac{sin\left(\mathrm{\π}y\right)}{\mathrm{\π}y}\right)$

Target function gradient matrix A i of structure_kFor：

${\mathrm{Ai}}_K=\left\{\begin{array}{ccccccccc}\frac{{\mathrm{dI}}_1}{{\mathrm{dKi}}_1}& \frac{{\mathrm{dI}}_1}{{\mathrm{dKi}}_2}& \mathrm{...}& \frac{{\mathrm{dI}}_1}{{\mathrm{dNx}}_1}& \frac{{\mathrm{dI}}_1}{{\mathrm{dNx}}_2}& \mathrm{...}& \frac{{\mathrm{dI}}_1}{{\mathrm{dNy}}_1}& \frac{{\mathrm{dI}}_1}{{\mathrm{dNy}}_2}& \mathrm{...}\\ \frac{{\mathrm{dI}}_2}{{\mathrm{dKi}}_1}& \frac{{\mathrm{dI}}_2}{{\mathrm{dKi}}_2}& \mathrm{...}& \frac{{\mathrm{dI}}_2}{{\mathrm{dNx}}_1}& \frac{{\mathrm{dI}}_2}{{\mathrm{dNx}}_2}& \mathrm{...}& \frac{{\mathrm{dI}}_2}{{\mathrm{dNy}}_1}& \frac{{\mathrm{dI}}_2}{{\mathrm{dNy}}_2}& \mathrm{...}\\ .& .& .& .& .& .& .& .& .\\ .& .& .& .& .& .& .& .& .\\ .& .& .& .& .& .& .& .& .\end{array}\right\}$

${\mathrm{\τi}}_k=\left\{\begin{array}{c}{\mathrm{Yi}}_1-\mathrm{Im}\left(value\right(1\left)\right)\\ {\mathrm{Yi}}_2-\mathrm{Im}\left(value\right(2\left)\right)\\ .\\ .\\ .\end{array}\right\}$

The recursive iteration equation of structure is：

${\mathrm{xi}}^{(k+1)}={\mathrm{xi}}^{\left(k\right)}+{\mathrm{\δi}}^{\left(k\right)}={\mathrm{xi}}^{\left(k\right)}-{\left({\mathrm{Ai}}_K^T{\mathrm{Ai}}_K\right)}^{-1}{\mathrm{Ai}}_K^T{\mathrm{\τi}}_k$

The initial estimation of target signature parameter imaginary part is：

${\mathrm{xi}}^{\left(1\right)}=\left\{\begin{array}{c}Ki\\ Nx\\ Ny\end{array}\right\}$

Wherein, BW is bandwidth, nx_jFor j-th sampled point in the area image after interpolation, I_jFor the reconstruction data of j-th sampled point Imaginary part, Ki_iFor the imaginary part of the point of amplitude maximum in i-th grid in an area image, Nx_i、Ny_iWidth in respectively i-th grid The abscissa and ordinate of the maximum point of degree, k is iterations, Ai_KFor target function gradient matrix, τ i_kFor j-th sampled point The difference of the reconstruction data imaginary part of data imaginary part and the point, value () is the data of j-th sampled point, and Im () is that numerical value is empty Portion, xi^(k)For the imaginary part of the kth time corresponding target signature parameter of iteration, δ i^(k)For imaginary part error term, ()^-1For inverse of a matrix, (·)^TFor transposed matrix, wherein, solving δ i^(k)Afterwards, xi is put^(k+1)=xi^(k)+δi^(k), k=k+1, whenWhen, stop iteration.

6. the target's feature-extraction device of a kind of remote sensing images, it is characterised in that described device includes：

Division module, for Object in Remote Sensing to be divided into into multiple area images, each area image includes multiple pixel networks Lattice；

Interpolating module, for entering row interpolation to each area image, obtains array after the interpolation of each area image；

Pre-estimation array determining module, for array after the interpolation according to each area image, determines the mesh of each area image The pre-estimation array of mark characteristic parameter；

Provincial characteristics determining module, for according to the pre-estimation array of the target signature parameter of each area image, by Gauss- Newton iteration method, determines the target signature parameter of each area image；

Characteristics of image determining module, for Object in Remote Sensing described in the target signature parameter determination according to each area image Target signature parameter.

7. device according to claim 6, it is characterised in that the interpolating module includes：

Transformation submodule, for the first array corresponding to the pixel grid of an area image two-dimensional discrete Fourier change is carried out Change, generate the second array；

Frequency domain zero padding submodule, for carrying out frequency domain zero padding to second array, generates the 3rd array；

Inverse transformation submodule, for carrying out two-dimensional discrete Fourier inverse transformation to the 3rd array, generates the area image Array after interpolation.

8. device according to claim 6, it is characterised in that the pre-estimation array determining module includes：

Grid pre-estimation array determination sub-module, for according in array after the interpolation of an area image, width in a pixel grid The array of the maximum point of value, determines the pre-estimation array of the target signature parameter in the pixel grid；

Region pre-estimation array determination sub-module, in array after the interpolation according to the area image, in each pixel grid Target signature parameter pre-estimation array, determine the pre-estimation array of the target signature parameter of the area image.

9. device according to claim 6, it is characterised in that the provincial characteristics determining module includes：

Establishing equation submodule, for array after the interpolation according to each area image and target signature parameter counting is estimated Group, sets up the recursive iteration equation of corresponding area image；

Real imaginary part calculating sub module, for by Gaussian-Newton method, the recursive iteration side of each area image being calculated respectively The real part solution and imaginary part solution of journey；

Provincial characteristics determination sub-module, for according to the real part solution and imaginary part solution of the recursive iteration equation of each area image, really The target signature parameter of fixed each area image.

10. device according to claim 9, it is characterised in that the real imaginary part calculating sub module includes：

Real part calculating sub module, for by Gaussian-Newton method, the recursive iteration equation of each area image being calculated respectively Real part solution, wherein, in the area image after interpolation, the reconstruction data real part R of j-th sampled point_jFor：

R_j=Kr₁sinc(BW(nx_j-Nx₁,ny_j-Ny₁))+Kr₂sinc(BW(nx_j-Nx₂,ny_j-Ny₂))+…+Kr_isinc(BW (nx_j-Nx_i,ny_j-Ny_i))

$\sin c\left(x\right)=\frac{sin\left(\mathrm{\π}x\right)}{\mathrm{\π}x}$

$\sin c(x,y)={\left(\frac{sin\left(\mathrm{\π}x\right)}{\mathrm{\π}x}\right)}^T\·\left(\frac{sin\left(\mathrm{\π}y\right)}{\mathrm{\π}y}\right)$

Target function gradient matrix A r of structure_KFor：

${\mathrm{Ar}}_K=\left\{\begin{array}{ccccccccc}\frac{{\mathrm{dR}}_1}{{\mathrm{dKr}}_1}& \frac{{\mathrm{dR}}_1}{{\mathrm{dKr}}_2}& \mathrm{...}& \frac{{\mathrm{dR}}_1}{{\mathrm{dNx}}_1}& \frac{{\mathrm{dR}}_1}{{\mathrm{dNx}}_2}& \mathrm{...}& \frac{{\mathrm{dR}}_1}{{\mathrm{dNy}}_1}& \frac{{\mathrm{dR}}_1}{{\mathrm{dNy}}_2}& \mathrm{...}\\ \frac{{\mathrm{dR}}_2}{{\mathrm{dKr}}_1}& \frac{{\mathrm{dR}}_2}{{\mathrm{dKr}}_2}& \mathrm{...}& \frac{{\mathrm{dR}}_2}{{\mathrm{dNx}}_1}& \frac{{\mathrm{dR}}_2}{{\mathrm{dNx}}_2}& \mathrm{...}& \frac{{\mathrm{dR}}_2}{{\mathrm{dNy}}_1}& \frac{{\mathrm{dR}}_2}{{\mathrm{dNy}}_2}& \mathrm{...}\\ .& .& .& .& .& .& .& .& .\\ .& .& .& .& .& .& .& .& .\\ .& .& .& .& .& .& .& .& .\end{array}\right\}$

${\mathrm{\τr}}_k=\left\{\begin{array}{c}{R}_1-\mathrm{Re}\left(value\right(1\left)\right)\\ R_2-\mathrm{Re}\left(value\right(2\left)\right)\\ .\\ .\\ .\end{array}\right\}$

The recursive iteration equation of structure is：

${\mathrm{xr}}^{(k+1)}={\mathrm{xr}}^{\left(k\right)}+{\mathrm{\δr}}^{\left(k\right)}={\mathrm{xr}}^{\left(k\right)}-{\left({\mathrm{Ar}}_K^T{\mathrm{Ar}}_K\right)}^{-1}{\mathrm{Ar}}_K^T{\mathrm{\τr}}_k$

The initial estimation of target signature parameter real part is：

${\mathrm{xr}}^{\left(1\right)}=\left\{\begin{array}{c}Kr\\ Nx\\ Ny\end{array}\right\}$

Wherein, BW is bandwidth, nx_jFor j-th sampled point in the area image after interpolation, R_jFor the reconstruction data of j-th sampled point Real part, Kr_iFor the real part of the point of amplitude maximum in i-th grid in an area image, Nx_i、Ny_iWidth in respectively i-th grid The abscissa and ordinate of the maximum point of degree, k is iterations, Ar_KFor target function gradient matrix, τ r_kFor j-th sampled point The difference of the reconstruction data real part of data real part and the point, value () is the data of j-th sampled point, and Re () is numerical value reality Portion, xr^(k)For the real part of the kth time corresponding target signature parameter of iteration, δ r^(k)For real part error term, ()^-1For inverse of a matrix, (·)^TFor transposed matrix, wherein, solving δ r^(k)Afterwards, xr is put^(k+1)=xr^(k)+δr^(k), k=k+1, whenWhen, stop iteration；

Imaginary part calculating sub module, for by Gaussian-Newton method, the recursive iteration equation of each area image being calculated respectively Imaginary part solution, wherein, in the area image after interpolation, reconstruction data imaginary part I of j-th sampled point_jFor：

I_j=Ki₁sinc(BW(nx_j-Nx₁,ny_j-Ny₁))+Ki₂sinc(BW(nx_j-Nx₂,ny_j-Ny₂))+…+Ki_isinc(BW (nx_j-Nx_i,ny_j-Ny_i))

$\sin c\left(x\right)=\frac{sin\left(\mathrm{\π}x\right)}{\mathrm{\π}x}$

$\sin c(x,y)={\left(\frac{sin\left(\mathrm{\π}x\right)}{\mathrm{\π}x}\right)}^T\·\left(\frac{sin\left(\mathrm{\π}y\right)}{\mathrm{\π}y}\right)$

Target function gradient matrix A i of structure_kFor：

${\mathrm{Ai}}_K=\left\{\begin{array}{ccccccccc}\frac{{\mathrm{dI}}_1}{{\mathrm{dKi}}_1}& \frac{{\mathrm{dI}}_1}{{\mathrm{dKi}}_2}& \mathrm{...}& \frac{{\mathrm{dI}}_1}{{\mathrm{dNx}}_1}& \frac{{\mathrm{dI}}_1}{{\mathrm{dNx}}_2}& \mathrm{...}& \frac{{\mathrm{dI}}_1}{{\mathrm{dNy}}_1}& \frac{{\mathrm{dI}}_1}{{\mathrm{dNy}}_2}& \mathrm{...}\\ \frac{{\mathrm{dI}}_2}{{\mathrm{dKi}}_1}& \frac{{\mathrm{dI}}_2}{{\mathrm{dKi}}_2}& \mathrm{...}& \frac{{\mathrm{dI}}_2}{{\mathrm{dNx}}_1}& \frac{{\mathrm{dI}}_2}{{\mathrm{dNx}}_2}& \mathrm{...}& \frac{{\mathrm{dI}}_2}{{\mathrm{dNy}}_1}& \frac{{\mathrm{dI}}_2}{{\mathrm{dNy}}_2}& \mathrm{...}\\ .& .& .& .& .& .& .& .& .\\ .& .& .& .& .& .& .& .& .\\ .& .& .& .& .& .& .& .& .\end{array}\right\}$

${\mathrm{\τi}}_k=\left\{\begin{array}{c}{\mathrm{Yi}}_1-\mathrm{Im}\left(value\right(1\left)\right)\\ {\mathrm{Yi}}_2-\mathrm{Im}\left(value\right(2\left)\right)\\ .\\ .\\ .\end{array}\right\}$

The recursive iteration equation of structure is：

${\mathrm{xi}}^{(k+1)}={\mathrm{xi}}^{\left(k\right)}+{\mathrm{\δi}}^{\left(k\right)}={\mathrm{xi}}^{\left(k\right)}-{\left({\mathrm{Ai}}_K^T{\mathrm{Ai}}_K\right)}^{-1}{\mathrm{Ai}}_K^T{\mathrm{\τi}}_k$

The initial estimation of target signature parameter imaginary part is：

${\mathrm{xi}}^{\left(1\right)}=\left\{\begin{array}{c}Ki\\ Nx\\ Ny\end{array}\right\}$

Wherein, BW is bandwidth, nx_jFor j-th sampled point in the area image after interpolation, I_jFor the reconstruction data of j-th sampled point Imaginary part, Ki_iFor the imaginary part of the point of amplitude maximum in i-th grid in an area image, Nx_i、Ny_iWidth in respectively i-th grid The abscissa and ordinate of the maximum point of degree, k is iterations, Ai_KFor target function gradient matrix, τ i_kFor j-th sampled point The difference of the reconstruction data imaginary part of data imaginary part and the point, value () is the data of j-th sampled point, and Im () is that numerical value is empty Portion, xi^(k)For the imaginary part of the kth time corresponding target signature parameter of iteration, δ i^(k)For imaginary part error term, ()^-1For inverse of a matrix, (·)^TFor transposed matrix, wherein, solving δ i^(k)Afterwards, xi is put^(k+1)=xi^(k)+δi^(k), k=k+1, whenWhen, stop iteration.