CN105303184A - Method for accurately identifying ground features in satellite remote-sensing image - Google Patents

Abstract

The invention discloses a method for accurately identifying ground features in a satellite remote-sensing image. The method comprises steps of: extracting vegetation and water areas according to a vegetation index and a water index of multispectral image data registered with a satellite remote-sensing panchromatic image, and determining a specific category according to the geometrical morphological characteristics of the vegetation and water areas; according to the gray characteristics of the ground features, segmenting the extracted satellite remote-sensing panchromatic image of the vegetation and water areas into different areas by using gray consistency technology, and extracting factory areas according to the geometrical characteristics of the areas; determining that the rest areas are man-made ground feature areas after the factory areas and the vegetation and water areas are extracted, segmenting the man-made ground feature areas by using a textural feature extracting method, and performing a SVM classification identification process according to the textural feature of each segmented man-made ground feature area in order to obtain the specific category of each man-made ground feature area. The method may accurately detect and identify the ground features in the satellite remote-sensing image.

Description

Atural object precise recognition method in a kind of satellite remote-sensing image

Technical field

The present invention relates to satellite remote-sensing image identification field, particularly relate to atural object precise recognition method in a kind of satellite remote-sensing image.

Background technology

Satellite remote-sensing image Objects recognition, as the basis of earth's surface variation monitoring, has important effect.By the detecting and identifying of ground object target, analyze the change of earth environment and resource, detection and early warning disaster, collect a large amount of natures and Information of Ancient Human Activity, can be widely used in the field such as national economy and military affairs.Existing remote sensing images feature changes detection technique is owing to lacking suitable evaluation criterion and theoretical foundation, major part is all by simple Pixel Analysis, do not form the geometric properties information in automatic change detecting method system and target future, the distortion of outer bound pair remotely-sensed data cannot be eliminated.

The identification of remote sensing image target is generally carried out for man-made features, not only according to its spectral signature, also to a great extent according to target shape, Spatial Semantics relation etc., the target classification ownership of its foothold small scale often.Existing remote sensing Objects recognition technology, can be divided into substantially based on pixel classification and object-oriented image analysing computer two kinds of methods.During the former is applicable to, low spatial resolution image information extract, the latter's suitable treatment high spatial resolution image certificate.

OO image analysis methods shows obvious advantage in many applications.But also comparatively accurate, imperfect in high-resolution remote sensing image object oriented analysis scheme at present, this seriously constrains the application of high-resolution remote sensing image in various fields.Therefore, systematically study high spatial resolution remote sense image information extraction technology, exploitation remote sensing image intelligence interpretation system seems and to be even more important and urgently.

Summary of the invention

The object of this invention is to provide atural object precise recognition method in a kind of satellite remote-sensing image, identification can be detected accurately to the atural object in satellite remote-sensing image.

The object of the invention is to be achieved through the following technical solutions:

Atural object precise recognition method in a kind of satellite remote-sensing image, comprising:

According to satellite remote sensing panchromatic image registration after the vegetation index of multispectral image data and water body index, extract vegetation and region, waters, and determine concrete kind according to the geometric shape feature in vegetation and region, waters;

According to the gray scale feature of atural object, adopt gray consistency technology that the satellite remote sensing panchromatic image extracting vegetation and region, waters is divided into different regions, then go out factory area according to the Extraction of Geometrical Features in region;

Extract factory area, the remaining area in vegetation and region, waters is man-made features region, the extracting method of textural characteristics is adopted to carry out dividing processing to man-made features region, and the Classification and Identification process of SVM is carried out according to each man-made features regional texture feature after segmentation, obtain the concrete kind in each man-made features region.

The vegetation index of the multispectral image data after described basis and satellite remote sensing panchromatic image registration and water body index, extract vegetation and region, waters, and determine that concrete kind comprises according to the geometric shape feature in vegetation and region, waters:

Use air impedance vegetation index, extract forest and part factory, region that area is less than setting value obtains wood land to use chain code technology to remove;

Use the additive method except air impedance vegetation index to extract whole vegetation area, after removing wood land, obtain field and city part vegetation, remove after area is less than the region of setting value and obtain region, field;

Use water body index water lift territory, and determine the concrete kind in waters according to the geometric properties in waters.

The satellite remote sensing panchromatic image extracting vegetation and region, waters is divided into different regions and comprises by described employing gray consistency technology:

Two-sided filter is adopted to carry out denoising to the satellite remote sensing panchromatic image extracting vegetation and region, waters;

Adopt gray consistency growth method to split the image after denoising, by region segmentation identical for gray-scale value out, obtain the region that some gray-scale values are identical.

Described employing gray consistency growth method is split the image after denoising, region segmentation identical for gray-scale value is out comprised:

Initial point set A is a point in the image after denoising, and average gray Avg by current collection A the mean value of a gray-scale value;

Point alternatively point around some set A, if the difference of the gray-scale value of candidate point and Avg is less than the threshold value of setting, then in this addition point set A, and upgrades Avg value;

Repeat this step, until do not have new point to add, now set A is the region with grey similarity.

The extracting method of described employing textural characteristics carries out dividing processing to man-made features region and comprises:

Gabor filter is adopted to carry out filtering process to man-made features region;

Texture feature extraction is carried out to filtered image, obtains texture image;

Texture image is carried out to the classification of SVM, thus complete the precise classification to segmentation rear region.

As seen from the above technical solution provided by the invention, twice segmentation is carried out to satellite remote-sensing image, by the respective continuum of different atural object can be obtained after first time rough segmentation, by secondary fine segmentation, precise classification can be carried out to atural object, avoid the interference of different classes of adjacent; Finally, to the extraction carrying out exponential sum geometric properties of cutting object, determine the concrete kind of different cutting object, thus realize identifying the accurate detection of atural object.

Accompanying drawing explanation

In order to be illustrated more clearly in the technical scheme of the embodiment of the present invention, below the accompanying drawing used required in describing embodiment is briefly described, apparently, accompanying drawing in the following describes is only some embodiments of the present invention, for those of ordinary skill in the art, under the prerequisite not paying creative work, other accompanying drawings can also be obtained according to these accompanying drawings.

The schematic diagram of the accurate identifying of atural object in a kind of satellite remote-sensing image that Fig. 1 provides for the embodiment of the present invention;

The texture feature extraction process flow diagram of the satellite remote-sensing image that Fig. 2 provides for the embodiment of the present invention;

The Gabor texture blending village of the satellite remote-sensing image that Fig. 3 provides for the embodiment of the present invention, building, factory building schematic diagram.

Embodiment

Below in conjunction with the accompanying drawing in the embodiment of the present invention, be clearly and completely described the technical scheme in the embodiment of the present invention, obviously, described embodiment is only the present invention's part embodiment, instead of whole embodiments.Based on embodiments of the invention, those of ordinary skill in the art, not making the every other embodiment obtained under creative work prerequisite, belong to protection scope of the present invention.

As shown in Figure 1, for the embodiment of the present invention provides atural object precise recognition method schematic diagram in a kind of satellite remote-sensing image, it mainly comprises the steps:

Step 1, according to satellite remote sensing panchromatic image registration after the vegetation index of multispectral image data and water body index, extract vegetation and region, waters, and determine concrete kind according to the geometric shape feature in vegetation and region, waters.

The multispectral data of remote sensing image contains the absorption reflection potential of variety classes atural object to different spectrum, in the embodiment of the present invention, satellite remote sensing panchromatic image in advance with multispectral data registration, thus according to the vegetation index of multispectral image and water body index, the vegetation in image and region, waters can be extracted, then get out of the wood according to the geometric shape feature extraction in region, ground object target that field, lake, river, pool etc. are concrete.

For example, in remote sensing multispectral data, visible infrared ripple is easily by Chlorophyll absorption, and infrared waves has stronger transmittance to chlorophyll.Utilize chlorophyll to the selective calculation vegetation index of light wave, generally various combinatorial operation is done to these two kinds of light waves.

(1) formula expression of normalized differential vegetation index NDVI (NormalizedDifferenceVegetationIndex) normalized differential vegetation index is:

$NDVI=\frac{NIR-R}{NIR+R}$

Wherein, NIR is near-infrared reflection value, and R is visible red wave reflection value.NDVI is that RVI obtains after non-linear normalizing, and scope is between [-1,1].NDVI eliminates the impact of the factors such as the sun elevation angle, landform, cloud layer, moonscope angle and atmospheric conditions to a certain extent.

Normalized differential vegetation index describes the upgrowth situation of plant, and coverage rate positive correlation.Research shows, leaf area index (LAT), vegetation coverage, photosynthesis, carbon stay admittedly, green bio amount, surface temperature, Net primary productivity (NPP), integral shroud impedance, water evaporation etc. can have close associating with normalized differential vegetation index, and the seasonal variety of quantity of precipitation, carbon dioxide also can be reflected to normalized differential vegetation index in addition.

(2) difference vegetation index DVI (DifferenceVegetationIndex)

The change of this exponent pair Soil Background is very sensitive, is also environmental vegetation index, and its computing formula is:

DVI＝NIR-R

Wherein NIR is near-infrared reflection value, and R is visible red wave reflection value.

It is fewer that difference vegetation index is applied, and due to the susceptibility to Soil Background, monitoring capability declines to some extent when vegetation coverage is higher.So be applicable to the early metaphase of vegetation growth, in vegetation ecological environmental monitoring, there is certain advantage.

(3) soil adjustment vegetation index SAVI (SoilAdjustedVegetationIndex)

Soil regulates vegetation index to be intended to reduce the impact of Soil Background, and formula is:

$SAVI=\left(\frac{NIR-R}{NIR+R+L}\right)(1+L)$

Wherein NIR represents near-infrared band, and R represents visible red wave band, and L represents that soil regulates index, is also Vegetation canopy regulatory factor, is used for controlling different soils change of reflection to the impact of vegetation index.L is according to practical application value, and when L value is 0, SAVI exponential sum NDVI index is of equal value, represents that earth's surface does not have vegetative coverage, when L etc. with 1 time, represent that earth's surface is all by vegetative coverage.The factor (1+L) ensures that the span of SAVI is for [-1,1].In order to reduce the interference of soil noise to index, propose modified soil and regulate vegetation index (MSAVI), formula is:

$MSAVI=\frac{(2NIR+1)-\sqrt{{(2NIR+1)}^2-8(NIR-R)}}{2}$

Replace Vegetation canopy regulatory factor with automatic regulatory factor in formula, more can distinguish the boundary of vegetation and non-vegetation, reduce further the image of different soils change of reflection to vegetation index.

(4) air impedance vegetation index ARVI (AtmosphericallyResistantVegetationIndex)

Air impedance vegetation index introduces the impact that blue light reduces air, and expression formula is

$ARVI=\frac{NIR-\[R-\mathrm{\γ}(B-R)\]}{NIR+\[R-\mathrm{\γ}(B-R)\]}$

Wherein NIR represents near infrared wave spectrum, and B represents visible blue wave spectrum, and R represents the red wave spectrum of visible ray, and γ represents blue factor of influence, usually desirable 1.ARVI index is the improvement of NDVI index, and scope is [-1,1], with blue wave spectrum correction atmospheric scattering on the impact of vegetation, is mainly used to the error eliminating gasoloid generation.So it is applicable to the higher area of atmospheric aerosol concentration, the vegetation index that can be used under haze weather extracts.

Forest: use air impedance vegetation index to extract highlight area, extract forest and part factory, because factory's area is relatively little, the region that chain code technology can be used to remove area be less than setting value, obtains wood land.

Field: use the additive method except air impedance vegetation index to extract whole vegetation area, field and city part vegetation is obtained after removing forest part, because city tree and grass coverage is relatively little, can remove after area is less than the region of setting value and obtain region, field.

Waters has special reflecting and absorption characteristic to different wave length, and namely visible ray weakens gradually to middle-infrared band reflection potential, near infrared and middle infrared wavelength, receptivity is the strongest.

Normalization difference water body index NDWI (NormalizedDifferenceWaterIndex)

The normalization difference water body index be made up of the contrast of visible ray green band and near-infrared band, its formula is:

$NDWI=\frac{G-NIR}{G+NIR}---\left(20\right)$

Wherein, G represents the green wave spectrum of visible ray, and NIR represents near infrared wave spectrum.The span of normalization difference water body index, in [-1,1], can strengthen Water-Body Information, simultaneously the interference of Background suppression terrestrial object information.

After using water body index water lift territory, by extracting the geometric properties attribute in these waters, as area, girth, length breadth ratio etc., analyze the classification in waters.

(1) area S

Directly use the area of pixel sum as region in region.

(2) girth C

$C=\underset{i=0}{\Σ}{a}_i$

Coefficient a in formula _iaccording to the walking direction between neighbor, the coefficient in horizontal and vertical direction is 1, and other directions are

(3) length breadth ratio R

$R=\frac{W_{MER}}{L_{MER}}$

Rotary process is adopted to ask the minimum enclosed rectangle of profile, W _mERwide for minimum enclosed rectangle, L _mERfor the length of minimum enclosed rectangle.Length breadth ratio reflects the fat or thin shape in region.

(4) rectangular degree F

$F=\frac{S}{S_{MER}}$

Wherein S _mERfor the area of minimum enclosed rectangle.Rectangular degree is the ratio of area with its minimum enclosed rectangle of contour area, its ratio more close to 1, declare area and regular rectangular shape more close.

(5) circularity G

$G=\frac{S}{S_{MEC}}$

Wherein S _mECfor the area of minimum circumscribed circle.Circularity is the ratio of area with its minimum circumscribed circle of contour area, and its ratio is more close to 1, and declare area and rule are justified more close.

The geometric properties constraint in waters

Lake: use water body index to extract, length and width are smaller, circularity the greater is lake.

River: use water body index to extract, length breadth ratio is comparatively large, circularity smaller is river.

Step 2, gray scale feature according to atural object, adopt gray consistency technology that the satellite remote sensing panchromatic image extracting vegetation and region, waters is divided into different regions, then go out factory area according to the Extraction of Geometrical Features in region.

In the embodiment of the present invention, because abovementioned steps 1 has extracted vegetation and region, waters, therefore, vegetation and region, waters can be removed from satellite remote sensing panchromatic image, subsequent treatment does not relate to vegetation and region, waters yet, and the satellite remote sensing panchromatic image in this step is the satellite remote sensing panchromatic image removing vegetation and region, waters.

This step carries out rough segmentation to satellite remote sensing panchromatic image, by carrying out Image Segmentation to satellite remote sensing panchromatic image according to gray consistency, be the different regional of gray scale by satellite image division of teaching contents, thus the process of image is risen to object rank from pixel level.It mainly comprises: images filter and atural object split two parts.Specific as follows:

1) two-sided filter is adopted to carry out denoising to satellite remote sensing panchromatic image.

Adopt two-sided filter to the pixel (x, y) of image h (x, y), at neighborhood of pixel points S _x,yinside carry out noise-removed filtering, adopt local weighted averaging method synthetic image

$\hat{f}(x,y)=\frac{\underset{(i,j)\∈S_{x,y}}{\Σ}w(i,j)h(i,j)}{\underset{(i,j)\∈S_{x,y}}{\Σ}w(i,j)}$

To the neighborhood territory pixel of each pixel h (i, j), its weighting coefficient w (i, j) is by spatial neighbor factor w _d(i, j) and brightness similar factors w _rthe product composition of (i, j): w (i, j)=w _d(i, j) w _r(i, j)

Wherein:

$w_d(i,j)=e^{-\frac{|i-x{|}^2+|j-y{|}^2}{2{\mathrm{\σ}}_d^2}}$

$w_r(i,j)=e^{-\frac{|g(i,j)-g(x,y){|}^2}{2{\mathrm{\σ}}_r^2}}$

σ _dfor the standard deviation of spatial domain Gaussian function, σ _rfor the standard deviation of codomain Gaussian function, spatial neighbor factor w _d(i, j) and brightness similar factors w _rthe nonlinear combination of (i, j) constitutes the weighting coefficient of two-sided filter.

2) adopt gray consistency growth method to split the image after denoising, by region segmentation identical for gray-scale value out, obtain the region that some gray-scale values are identical.

The steps include: that initial point set A is a point in image after denoising, average gray Avg by current collection A the mean value of a gray-scale value; Point alternatively point around some set A, if the difference of the gray-scale value of candidate point and Avg is less than the threshold value of setting, then in this addition point set A, and upgrades Avg value; Repeat this step, until do not have new point to add, now set A is the region with grey similarity.

Above-mentioned steps can adopt following manner to realize:

(1) { A}, { A}, then { the mean value Avg of A} is the gray-scale value of this pixel, and { count in A} Count=1 for this current pixel point p (x, y) to be put into point set in existing some set.

(2) by certain neighborhood territory pixel p (x of pixel p (x, y) _i, y _i) gray-scale value G (x _i, y _i) poor with Avg, be denoted as t.If t < is T ₀, then tagging f=1; Otherwise, put f=0, (wherein T ₀for threshold value).Repeat this operation, until all neighborhood territory pixel marks terminate.

(3) for the pixel p (x, y) being labeled as 1, leave in A}, cumulative Count, and upgrade Avg according to following formula.

${\mathrm{Avg}}_{new}=\frac{{\mathrm{Avg}}_{old}+G(x_1,y_1)+\mathrm{...}+G(x_i,y_i)}{Count+i}$

(4) repeat 2 ~ 3 steps, until Count no longer changes, now { A} is the required region with grey similarity, describes this region by geometric shape method.

(5) reselect new pixel, repeat 1 ~ 4 operation, until pixels all in image is all divided in similarity region.

By analysis, factory is made up of very large factory building usually, according to the feature of factory, adopts grey similarity partition means, filters out the target of rectangle.Road remains the highest region segmentation of priority, if having a large amount of rectangular targets in the region that surrounds of road, then thinks that this region may be factory.

Exemplary, geometric properties constraint is as follows:

Step 3, extract factory area, the remaining area in vegetation and region, waters is man-made features region, the extracting method of textural characteristics is adopted to carry out dividing processing to man-made features region, and the Classification and Identification process of SVM is carried out according to each man-made features regional texture feature after segmentation, obtain the concrete kind in each man-made features region.

In abovementioned steps 2, atural object on a large scale in satellite remote-sensing image only can effectively be classified by gray consistency Iamge Segmentation, relative to most region, man-made features areal extent is less, and component part more complicated, is difficult to use grey similarity to carry out the Accurate Segmentation of man-made features kind.

In the embodiment of the present invention, because abovementioned steps 1 ~ step 2 has extracted vegetation and region, waters and factory area, therefore, remaining area is man-made features region, by the extraction of textural characteristics, can obtain the accurate segmentation in man-made features region.

For example, the reason that Gabor transformation can be good at extracting of texture important is, Gabor base has Gaussian characteristics, it also inherits the feature of small echo simultaneously, on different frequency domain yardstick, different directions, different textural characteristics is converted to different numerical value by Gabor filter, thus completes the extraction of textural characteristics.

The step that Gabor filter extracts texture is: first build Gabor filter group, and carry out Gabor filtering to the image in man-made features region on multi-direction and multiple dimensioned; Then texture feature extraction is carried out to the filtering image of each yardstick in each direction, obtain texture image; Finally cluster and classification are carried out to texture image, thus complete the Accurate Segmentation to segmentation rear region.Its treatment scheme as shown in Figure 2.

1) build Gabor filter group, carry out Gabor filtering.

Adopt the Fourier formalism that two-dimensional Gabor basis function is corresponding with it, formula is respectively:

$g(x,y)=\left(\frac{1}{2{\mathrm{\&pi;\&sigma;}}_x{\mathrm{\&sigma;}}_y}\right)\exp \&lsqb;-\frac{1}{2}(\frac{x^2}{{\mathrm{\&sigma;}}_x^2}+\frac{y^2}{{\mathrm{\&sigma;}}_y^2})+2\mathrm{\&pi;}jWx\&rsqb;$

$G(u,v)=\exp \{-\frac{1}{2}\&lsqb;\frac{(u-W)}{{\mathrm{\&sigma;}}_u^2}+\frac{v^2}{{\mathrm{\&sigma;}}_v^2}\&rsqb;\}$

In formula, σ _xfor the effective frequency belt width in spatial domain x direction, σ _yfor the effective frequency belt width in spatial domain y direction, σ _uand σ _vfor effective frequency belt width corresponding in frequency domain.σ _u=1/2 π σ _x, σ _v=1/2 π σ _y, j is imaginary unit, and x, y are spatial domain variable, and u, v are frequency domain variable, and W is the frequency of sinusoidal normal direction, is also the multiple modulation frequency of Gaussian function, shows the position of Gabor filter in frequency domain.Gabor function forms a complete Non-orthogonal basis set, describes localized frequency on this basis by spread signal, and Gabor wavelet is called as a kind of function of self similarity.Suppose that g (x, y) is Gabor morther wavelet, the bank of filters of so this self similarity obtains by the rotary extension of g (x, y):

g _mn(x,y)＝a ^-mg(x′,y′)

x′＝a ^-m(xcosθ+ysinθ)y′＝a ^-m(-xsinθ+ycosθ)

Wherein, a > 1, m, n is integer, θ=n π/K, and K is the direction number of Gabor filter group, m=0,1 ... S-1, S are yardstick numbers.In order to reduce the redundant information of filtering image, if U _land U _hinterested lowest frequency and most high frequency, U _kfor the frequency of a kth area-of-interest, so this strategy must be guaranteed under frequency spectrum, support that half peak value of filter response contacts with each other.

Computing formula is as follows:

$\begin{array}{cc}a={(U_h/U_l)}^{\frac{1}{S-1}}& {\mathrm{\&sigma;}}_u=\frac{(a-1)U_h}{(a+1)\sqrt{21n2}}\end{array}$

${\mathrm{\&sigma;}}_v=tan\left(\frac{\mathrm{\&pi;}}{2k}\right)\&lsqb;U_h-2ln\left(\frac{2{\mathrm{\&sigma;}}_u^2}{U_k}\right)\&rsqb;{\&lsqb;2ln2-\frac{{\left(2ln2\right)}^2{\mathrm{\&sigma;}}_u^2}{U_h^2}\&rsqb;}^{-\frac{1}{2}}$

2) textural characteristics describes.

For picture I (x, y), Gabor filter is defined as follows:

W _mn(x,y)＝∫I(x ₁,y ₁)g _mn*(x-x ₁,y-y ₁)dx ₁dy ₁

Wherein, (x ₁, y ₁) be the pixel in image, * represents complex conjugate, and its hypothesis local grain region is spatially uniform, and the average μ of transform coefficient magnitude _mnand meansquaredeviationσ _mnfor representing the attribute description in this region, namely represent a texture description vector by the average in this region and mean square deviation:

μ _mn＝∫∫|W _mn(xy)|dxdy

${\mathrm{\&sigma;}}_{mn}=\sqrt{\&Integral;\&Integral;{\left(\right|W_{mn}(x,y)|-{\mathrm{\&mu;}}_{mn})}^{2}dxdy}$

So, the textural characteristics in this region is by average μ corresponding to all Gabor filter groups _mnand meansquaredeviationσ _mncombination.Suppose the Gabor filter group using 4 yardstick S=4,6 direction K=6 in testing, so this texture description comprises 48 components:

$\stackrel{\&OverBar;}{f}=\&lsqb;{\mathrm{\&mu;}}_{00}{\mathrm{\&sigma;}}_{00}{\mathrm{\&mu;}}_{01}{\mathrm{\&sigma;}}_{01}\mathrm{...}{\mathrm{\&mu;}}_{35}{\mathrm{\&sigma;}}_{35}\&rsqb;$

Concrete steps are:

A () carries out Fourier transform to former figure, obtain real part and the imaginary part frequency domain data of image;

B () carries out Fourier transform to Gabor base, obtain real part and the imaginary part frequency domain data of bank of filters;

C the real part of () former figure is multiplied with imaginary part with the real part of imaginary part frequency domain data with all Gabor filter, obtain real part and the imaginary data of a series of new filtering image;

D () carries out Fourier inversion to the real part of new image data and imaginary data, obtain filtering image.

E () asks average and the variance of each secondary filtering image, synthesize a textural characteristics and describe.

Under frequency domain, carry out the calculation of filtered data that are multiplied with image by Gabor base, it requires that the image of Gabor base is consistent with the size of texture image, if so Gabor base is too little, be consistent as size and texture image by expanded view.Following formula describes texture blending method:

$f(x_1,x_2,\&CenterDot;\&CenterDot;\&CenterDot;x_n)=T(I\&CircleTimes;G)$

Wherein, certain computing of T representing matrix, such as average, variance, maximal value, minimum value etc., I represents texture image, and G represents Gabor filter group, and it contains n Gabor base, and final texture represents that the number of f with Gabor base is the same. computing illustrates Gabor filtering method, is multiplied by Gabor base with texture image under frequency domain.

After carrying out Accurate Segmentation to man-made features region, then SVM can be utilized to carry out Classification and Identification, the concrete kind in each man-made features region.

As shown in Figure 3, Gabor texture can village in identification satellite remote sensing image, building, factory building; Wherein, being that node represents village with rectangle, take circle as the expression building of node, take triangle as the expression factory building of node.Gabor filter has very strong stability, is applicable to the texture blending doing remote sensing image, but the texture that it generates is more.SVM can adapt to the texture of Gabor just, has carried out sample training and image classification, have good classifying quality to village, building, factory building.

In sum, by the step of above-described embodiment, then can carry out accurate and effective Ground Split to remote sensing image, and determine the classification of cut zone, relatively, feature changes situation can be obtained.

Through the above description of the embodiments, those skilled in the art can be well understood to above-described embodiment can by software simulating, and the mode that also can add necessary general hardware platform by software realizes.Based on such understanding, the technical scheme of above-described embodiment can embody with the form of software product, it (can be CD-ROM that this software product can be stored in a non-volatile memory medium, USB flash disk, portable hard drive etc.) in, comprise some instructions and perform method described in each embodiment of the present invention in order to make a computer equipment (can be personal computer, server, or the network equipment etc.).

The above; be only the present invention's preferably embodiment, but protection scope of the present invention is not limited thereto, is anyly familiar with those skilled in the art in the technical scope that the present invention discloses; the change that can expect easily or replacement, all should be encompassed within protection scope of the present invention.Therefore, protection scope of the present invention should be as the criterion with the protection domain of claims.

Claims (5)

1. an atural object precise recognition method in satellite remote-sensing image, is characterized in that, comprising:

According to satellite remote sensing panchromatic image registration after the vegetation index of multispectral image data and water body index, extract vegetation and region, waters, and determine concrete kind according to the geometric shape feature in vegetation and region, waters;

According to the gray scale feature of atural object, adopt gray consistency technology that the satellite remote sensing panchromatic image extracting vegetation and region, waters is divided into different regions, then go out factory area according to the Extraction of Geometrical Features in region;

Extract factory area, the remaining area in vegetation and region, waters is man-made features region, the extracting method of textural characteristics is adopted to carry out dividing processing to man-made features region, and the Classification and Identification process of SVM is carried out according to each man-made features regional texture feature after segmentation, obtain the concrete kind in each man-made features region.

2. method according to claim 1, it is characterized in that, the vegetation index of the multispectral image data after described basis and satellite remote sensing panchromatic image registration and water body index, extract vegetation and region, waters, and determine that concrete kind comprises according to the geometric shape feature in vegetation and region, waters:

Use air impedance vegetation index, extract forest and part factory, region that area is less than setting value obtains wood land to use chain code technology to remove;

Use the additive method except air impedance vegetation index to extract whole vegetation area, after removing wood land, obtain field and city part vegetation, remove after area is less than the region of setting value and obtain region, field;

Use water body index water lift territory, and determine the concrete kind in waters according to the geometric properties in waters.

3. method according to claim 1, is characterized in that, the satellite remote sensing panchromatic image extracting vegetation and region, waters is divided into different regions and comprises by described employing gray consistency technology:

Two-sided filter is adopted to carry out denoising to the satellite remote sensing panchromatic image extracting vegetation and region, waters;

Adopt gray consistency growth method to split the image after denoising, by region segmentation identical for gray-scale value out, obtain the region that some gray-scale values are identical.

4. method according to claim 3, is characterized in that, described employing gray consistency growth method is split the image after denoising, region segmentation identical for gray-scale value is out comprised:

Initial point set A is a point in the image after denoising, and average gray Avg by current collection A the mean value of a gray-scale value;

Point alternatively point around some set A, if the difference of the gray-scale value of candidate point and Avg is less than the threshold value of setting, then in this addition point set A, and upgrades Avg value;

Repeat this step, until do not have new point to add, now set A is the region with grey similarity.

5. method according to claim 1, is characterized in that, the extracting method of described employing textural characteristics carries out dividing processing to man-made features region and comprises:

Gabor filter is adopted to carry out filtering process to man-made features region;

Texture feature extraction is carried out to filtered image, obtains texture image;

Texture image is carried out to the classification of SVM, thus complete the precise classification to segmentation rear region.