CN102521624A - Classification method for land use types and system - Google Patents

Abstract

An embodiment of the invention provides a classification method for land use types and a system. The method includes collecting historical land use vector data of areas to be classified, and building a plurality of land use type three-dimensional characteristic space models according to the historical land use vector data; acquiring satellite remote-sensing image grid data of the areas to be classified; realizing overlaying processing for the historical land use vector data and the satellite remote-sensing image grid data, sequentially judging whether each pixel is positioned within the land use type three-dimensional characteristic space model which belongs to the same land use type with the pixel or not, acquiring pixels positioned outside the land use type three-dimensional characteristic space models, and marking the pixels, which are positioned outside the land use type three-dimensional characteristic space models, of the satellite remote-sensing image grid data as a first type of varying pixels; and converting current land use types of the first type of varying pixels into actual land use types to which the first type of varying pixels belong, and realizing classification of the land use types.

Description

A kind of method and system of land use pattern classification

Technical field

The present invention relates to the land classification technical field, particularly a kind of method and system of land use pattern classification.

Background technology

Along with going deep into of global change research due, soil utilization/cover change research has become the core realm content of global environmental change research.Decipher and the classification that Base of Remote Sensing Techniques combining geographic information system (Geographic Information System, be called for short GIS), computer technology and traditional investigation method carry out land use pattern become the important means that current acquisition large scale, high-timeliness soil are utilized space-time data.

In the research of remote sensing technology, differentiating all types of target through remote sensing image is an important ring of development of remote sensing, and the foundation in remotely-sensed data storehouse concerns importances such as specialized information extraction, dynamic change prediction and Thematic Cartography.

The soil utilizes the remote sensing classification to be actually the process of the automatic Classification and Identification of remote sensing images, just uses computing machine simulating human consciousness, accomplishes the process of remote Sensing Image Analysis and understanding.The soil utilize the key problem of remote sensing classification promptly be one to the extraction of remote sensing images signature analysis, image segmentation and cluster, carry out the process of Classification and Identification.It is to be classified as the kind in the land use pattern categorizing system to each pixel in the remote sensing images or zone that the soil utilizes the detailed process of remote sensing classification; Just select characteristic parameter through spectral signature analysis to all kinds of atural objects; Mark off feature space, the pixel of remote sensing images is divided in the feature space.In the prior art, soil commonly used utilizes the remote sensing sorting technique to have: visual interpretation method, supervised classification and unsupervised classification method.

The visual interpretation method mainly is through remote sensing image processing software remote sensing images to be amplified arbitrarily, dwindle and image enhancement processing, reaching best visual effect, and the interpretation personnel type boundary line, ground of directly sketching out along the image feature edge.The shortcoming of this method is that the main mode of artificial decipher that relies on is classified, the length that not only expends time in, and also different personnel's decipher result is different, causes classification results there are differences, and can't realize automatic classification.

Supervised classification; Claim the training area classification again, basic characteristics are before classification, to pass through sample survey on the spot, cooperate the artificial visual interpretation; Realization is known in advance to the atural object category attribute in some sampling zone on the remote sensing images; Promptly choose training area, make computing machine remove to draw up decision function again, thereby realize classification land use pattern according to the known type characteristic.Defective of the method and visual interpretation method are similar, and the process of in the method, choosing training area is to lean on artificial judgment equally, so workload is big, and length consuming time, and also the result there are differences, and can't realize robotization.

The unsupervised classification method promptly is only to classify with the regularity of distribution of the spectral signature of remote sensing images, different types realized distinguishing, and be a kind of mode of classifying automatically.The defective of the method is, according to this method classification, can only distinguish dissimilar soils, can't confirm its land use pattern, and degree of accuracy is lower, can't satisfy the needs of practical application.

Summary of the invention

In view of this, the object of the present invention is to provide a kind of method and system of land use pattern classification, to realize the land use pattern classification of high-level efficiency, robotization and pinpoint accuracy.

For realizing above-mentioned purpose, one embodiment of the present of invention provide a kind of method and system of land use pattern classification, and concrete technical scheme is following:

A kind of method of land use pattern classification, this method may further comprise the steps:

Collection wait the to classify historical soil in area utilizes vector data; Utilize vector data to set up some land use pattern three-dimensional feature spatial models according to said historical soil; Wherein, Said historical soil utilizes vector data to comprise patch and patch land use pattern information, and each land use pattern three-dimensional feature spatial model is corresponding with a kind of land use pattern;

The acquisition satellite remote-sensing image raster data in area of waiting to classify, said satellite remote-sensing image raster data comprises some pixels, each pixel be the satellite remote-sensing image with the area of waiting to classify utilize grid division and subelement;

Utilize vector data and satellite remote-sensing image raster data to carry out overlap-add procedure in said historical soil, obtain the land use pattern information of each pixel in the satellite remote-sensing image raster data;

Judge successively whether each pixel is belonging to it within land use pattern three-dimensional feature spatial model of same land use pattern; Acquisition drops on the pixel outside the land use pattern three-dimensional feature spatial model, in the satellite remote-sensing image raster data, the said pixel that drops on outside the land use pattern three-dimensional feature spatial model is labeled as the first kind and changes pixel;

The first kind is changed the current land use pattern of pixel convert its actual affiliated land use pattern into, realize the land use pattern classification.

Said collection wait the to classify historical soil in area utilizes vector data, and the step of utilizing vector data to set up some land use pattern three-dimensional feature spatial models according to said historical soil comprises:

Collection wait the to classify historical soil in area utilizes vector data; Said historical soil utilizes vector data to comprise patch and patch land use pattern information; For every kind of land use pattern is chosen a plurality of sample patches; And obtain the corresponding land use pattern information of sample patch, wherein, comprise that the vector data of sample patch and sample patch land use pattern information is the sample vector data;

Said sample vector data is carried out the training area analysis, obtain the training area vector data;

Through with training area vector data and satellite remote-sensing image raster data overlap-add procedure, for each land use pattern is set up a land use pattern three-dimensional feature spatial model.

Saidly be specially for a plurality of sample patches of choosing of every kind of land use pattern:

To same land use pattern; To belong to all patches of this land use pattern sorts according to area from big to small; And the area of patch is added up successively according to this ordering; When the ratio of accumulation result and all patch total areas reaches predefined first threshold scope, then getting a plurality of patches that add up is the sample patch;

For every kind of land use pattern, repeat said process, obtain the respective sample patch.

Said said sample vector data is carried out the training area analysis, obtains the training area vector data and be specially:

Take a sample sample patch in this vector data, said sample patch comprises patch shape information and plaque area information;

Keep sample patch shape information constant, area is dwindled, make the ratio of sample patch and the area of former sample patch after dwindling reach second threshold range that is provided with in advance;

Getting the patch that ratio that area dwindles back and the area of former vector data reaches second threshold range that is provided with in advance is training area; The land use pattern information of training area is consistent with sample patch land use pattern information, comprises that the vector data of training area and training area land use pattern information is called the training area vector data.

Said training area vector data and satellite remote-sensing image raster data overlap-add procedure are specially:

To a kind of land use pattern, will belong to the training area of this land use pattern and the multispectral or high spectrum satellite remote-sensing image raster data of process standardization and carry out overlay analysis, obtain containing the sample spectroscopic data of a plurality of wave bands; Said sample spectroscopic data is represented with the form of matrix, a plurality of characteristics that pixel comprised in each row representative sample spectroscopic data of said matrix, and the characteristic quantity that each pixel comprised equates with the wave band quantity of sample spectroscopic data;

Said sample spectroscopic data is carried out principal component analysis (PCA), obtain the major component of spectroscopic data, the quantity of said major component equates with wave band quantity, and chooses and comprise three maximum major components of characteristic; Wherein, said principal component analysis (PCA) is specially quadrature decomposition conversion;

To the training area of all land use patterns, repeat said process, obtain three major components of every kind of land use pattern training area.

Saidly set up a land use pattern three-dimensional feature spatial model for each land use pattern and be specially:

To a kind of land use pattern, utilize three major components of this land use pattern training area to set up land use pattern three-dimensional feature spatial model;

Land use pattern three-dimensional feature spatial model is expressed as

$\underset{j=1}{\overset{3}{\mathrm{\&Sigma;}}}\frac{{(P_j-M{P}_j)}^2}{{\mathrm{\&sigma;}}_{\mathrm{<figure-callout\; id="2"\; label="j"\; filenames="HDA0000122751920000011.png"\; state="\{\{state\}\}">j</figure-callout>}}^2}<{\mathrm{<figure-callout\; id="2"\; label="c"\; filenames="HDA0000122751920000011.png"\; state="\{\{state\}\}">c</figure-callout>}}^2$

P wherein in the formula _jRepresent the value of the j major component of substitution model pixel, MP _jRepresent in this land use pattern, the average of j major component, Represent the standard deviation of j major component in this land use pattern, c ²Be constant, represent the feature space index;

Utilize three major components of above-mentioned formula and every kind of land use pattern training area, for every kind of land use pattern is set up a land use pattern three-dimensional feature spatial model.

Said said historical soil is utilized vector data and satellite remote-sensing image raster data overlap-add procedure, is specially:

Utilize vector data and satellite remote-sensing image raster data stack to carry out overlay analysis in historical soil, obtain containing the common spectroscopic data of a plurality of wave bands through standardization; Said common spectroscopic data is represented with the form of matrix, a plurality of characteristics that pixel comprised in the common spectroscopic data of each row representative of said matrix, and the characteristic quantity that each pixel comprised equates with the wave band quantity of common spectroscopic data;

Said common spectroscopic data is carried out principal component analysis (PCA), obtain the major component of spectroscopic data, the quantity of said major component equates with wave band quantity, and chooses and comprise three maximum major components of characteristic; Wherein, said principal component analysis (PCA) is specially quadrature decomposition conversion.

Saidly the first kind is changed the current land use pattern of pixel convert its land use pattern under actual into and be specially:

Three major components that the first kind changed pixel are updated to all land use pattern three-dimensional feature spatial models; Calculate the distance that this first kind changes pixel and each land use pattern three-dimensional feature spatial model, and find and a nearest land use pattern three-dimensional feature spatial model of this first kind variation pixel;

The land use pattern that the first kind is changed pixel converts into and the nearest corresponding land use pattern of land use pattern three-dimensional feature spatial model of this first kind variation pixel.

The distance that the said calculating first kind changes pixel and land use pattern three-dimensional feature spatial model is specially:

According to formula $\mathrm{Di}=\left(\underset{j=1}{\overset{3}{\mathrm{\&Sigma;}}}\frac{(P_j-M{P}_{\mathrm{Ji}})}{{\mathrm{\&sigma;}}_{\mathrm{Ji}}^2}\right)/{\mathrm{<figure-callout\; id="2"\; label="c\; i"\; filenames="HDA0000122751920000011.png"\; state="\{\{state\}\}">c</figure-callout>}}_i^{}$

Wherein di represents the distance of the pixel and the i land use pattern three-dimensional feature spatial model of substitution; P _jRepresent the value of substitution pixel j major component, MP _JiRepresent the average of j major component in the i land use pattern,

Represent the standard deviation of j major component in the i land use pattern,

Represent the feature space index in the i land use pattern three-dimensional feature spatial model.

Said method further comprises:

Before the conversion first kind changes pixel land use pattern; Further be provided with one and shift the resistance matrix; Comprised the inclusive resistance coefficient of conversion needs between any two kinds of land use patterns in the said transfer resistance matrix, said resistance coefficient is stipulated according to the probability size that two kinds of land use patterns change; The distance that the first kind is changed pixel and each land use pattern three-dimensional feature spatial model further multiply by corresponding resistance coefficient; Obtain changing the distance of pixel and each land use pattern three-dimensional feature spatial model based on the first kind under the transformation rule; Acquisition changes the current land use pattern of pixel with the first kind and converts into based on changing the nearest pairing land use pattern of land use pattern three-dimensional feature spatial model of pixel with the first kind under the transformation rule based on changing the nearest land use pattern three-dimensional feature spatial model of pixel with the first kind under the transformation rule.

Saidly in the satellite remote-sensing image raster data, the said pixel that drops on outside the land use pattern three-dimensional feature spatial model is labeled as after the first kind changes pixel; Said method also comprises, mark is crossed satellite remote-sensing image raster data that the first kind changes pixel as first kind raster data;

After the first kind changed the current land use pattern of pixel and convert its actual land use pattern into, said method also comprised, the first kind that will pass through the land use pattern conversion changes pixel and is labeled as second type and changes pixel; Said mark is crossed second type of satellite remote-sensing image raster data that changes pixel as second type of raster data

Said method also comprises:

Contrast first kind raster data and second type of raster data carry out the correction of classified information according to the corrosion algorithm to the discrete pixel that distributes after the conversion land use pattern.

A kind of system of land use pattern classification, said system comprises:

Collector unit, modelling unit, computing unit and converting unit;

Collector unit is used to collect wait the to classify historical soil in area and utilizes vector data and satellite remote-sensing image raster data;

The modelling unit connects collector unit, is used for utilizing vector data to set up some land use pattern three-dimensional feature spatial models according to said historical soil;

Computing unit, link model is set up the unit, is used for said historical soil is utilized vector data and satellite remote-sensing image raster data overlap-add procedure, obtains the land use pattern information of each pixel in the satellite remote-sensing image raster data; Judge successively whether each pixel is belonging to it within land use pattern three-dimensional feature spatial model of same land use pattern; Acquisition drops on the pixel outside the land use pattern three-dimensional feature spatial model, in the satellite remote-sensing image raster data, the said pixel that drops on outside the land use pattern three-dimensional feature spatial model is labeled as the first kind and changes pixel;

Converting unit connects computing unit, is used for that the first kind is changed the current land use pattern of pixel and converts its actual affiliated land use pattern into.

Said modelling unit comprises:

The sample collection unit, training area analytic unit and construction unit;

The sample collection unit is used to every kind of land use pattern and chooses a plurality of sample patches, and obtains the corresponding land use pattern information of sample patch, wherein, comprises that the vector data of sample patch and sample patch land use pattern information is the sample vector data;

The training area analytic unit is used for said sample vector data is carried out the training area analysis, obtains the training area vector data;

Construction unit is used for training area vector data and satellite remote-sensing image raster data overlap-add procedure, for each land use pattern is set up a land use pattern three-dimensional feature spatial model.

Said converting unit comprises:

Metrics calculation unit and type information change unit;

Metrics calculation unit is used to calculate the distance that the first kind changes pixel and all land use pattern three-dimensional feature spatial models, and finds and the nearest land use pattern three-dimensional feature spatial model of this first kind variation pixel;

Type information change unit, the land use pattern that is used for the first kind is changed pixel converts the land use pattern three-dimensional feature spatial model corresponding land use pattern nearest with this variation pixel into.

Said converting unit further comprises:

Shift moment of resistance array element; Be used to be provided with transfer resistance matrix; Comprised between any two kinds of land use patterns the inclusive resistance coefficient of conversion needs in the said transfer resistance matrix, the probability that said resistance coefficient changes based on two kinds of land use patterns is big or small and stipulate; Shift the moment of resistance array element distance that the first kind changes pixel and each land use pattern three-dimensional feature spatial model further multiply by corresponding resistance coefficient; Obtain changing the distance of pixel and each land use pattern three-dimensional feature spatial model, obtain based on changing the nearest land use pattern three-dimensional feature spatial model of pixel with the first kind under the transformation rule based on the first kind under the transformation rule;

The land use pattern that said type information change unit changes pixel with the first kind convert into rule-based under the land use pattern three-dimensional feature spatial model corresponding land use pattern nearest with this variation pixel.

Said system further comprises:

After said converting unit also was used for the first kind changed the current land use pattern of pixel and convert its actual land use pattern into, the first kind that in the satellite remote-sensing image raster data, will pass through the land use pattern conversion changed pixel and is labeled as second type and changes pixel;

Said system also comprises the corrosion algorithm unit, connects computing unit and converting unit, is used to extract the computing unit mark and crosses satellite remote-sensing image raster data that the first kind changes pixel as first kind raster data; Extract mark and cross second type of satellite remote-sensing image raster data that changes pixel as second type of raster data; Contrast first kind raster data and second type of raster data carry out the correction of classified information according to the corrosion algorithm to the discrete pixel that distributes after the conversion land use pattern.

Can know based on above technical scheme; There is following beneficial effect in the present invention: through choosing the sample vector data; Train to analyze and set up land use pattern three-dimensional feature spatial model again; Thereby utilize land use pattern three-dimensional feature spatial model to utilize the vector data that comprises in the patch to analyze, obtain first kind variation pixel and change the said classification of pixel revising for the first kind to historical soil; Thus the method for the invention realize land use pattern accurately, the classification of unified standard, and realize classification automatically, raise the efficiency greatly.

Description of drawings

In order to be illustrated more clearly in the embodiment of the invention or technical scheme of the prior art; To do to introduce simply to the accompanying drawing of required use in embodiment or the description of the Prior Art below; Obviously, the accompanying drawing in describing below is some embodiments of the present invention, for those of ordinary skills; Under the prerequisite of not paying creative work, can also obtain other accompanying drawing according to these accompanying drawings.

Fig. 1 is the method for the invention process flow diagram;

Fig. 2 is a system architecture synoptic diagram according to the invention.

Embodiment

For the purpose, technical scheme and the advantage that make the embodiment of the invention clearer; To combine the accompanying drawing in the embodiment of the invention below; Technical scheme in the embodiment of the invention is carried out clear, intactly description; Obviously, described embodiment is the present invention's part embodiment, rather than whole embodiment.Based on the embodiment among the present invention, those of ordinary skills are not making the every other embodiment that is obtained under the creative work prerequisite, all belong to the scope of the present invention's protection.

The inventive method mainly comprises:

Collection wait the to classify historical soil in area utilizes vector data; Utilize vector data to set up some land use pattern three-dimensional feature spatial models according to said historical soil; Wherein, Said historical soil utilizes vector data to comprise patch and patch land use pattern information, and each land use pattern three-dimensional feature spatial model is corresponding with a kind of land use pattern;

The acquisition satellite remote-sensing image raster data in area of waiting to classify, said satellite remote-sensing image raster data comprises some pixels, each pixel be the satellite remote-sensing image with the area of waiting to classify utilize grid division and subelement;

Utilize vector data and satellite remote-sensing image raster data to carry out overlap-add procedure in said historical soil, obtain the land use pattern information of each pixel in the satellite remote-sensing image raster data;

Judge successively whether each pixel is belonging to it within land use pattern three-dimensional feature spatial model of same land use pattern; Acquisition drops on the pixel outside the land use pattern three-dimensional feature spatial model, in the satellite remote-sensing image raster data, the said pixel that drops on outside the land use pattern three-dimensional feature spatial model is labeled as the first kind and changes pixel;

The first kind is changed the current land use pattern of pixel convert its actual affiliated land use pattern into, realize the land use pattern classification.

With reference to shown in Figure 1, the method for the invention, concrete steps are following:

S1, the historical soil of collecting the area of waiting to classify utilize vector data, and said historical soil utilizes vector data to comprise patch and patch land use pattern information, the land use pattern of said each patch correspondence of patch land use pattern information record; Be a plurality of sample patches of choosing of every kind of land use pattern, and obtain sample patch corresponding sample patch land use pattern information, comprise that the vector data of model patch and sample patch land use pattern information is the sample vector data;

S2, said sample vector data is carried out the training area analysis, obtain the training area vector data; Said training area is an image-region pure in the patch, single type, known land use pattern;

S3, through with training area vector data and satellite remote-sensing image raster data overlap-add procedure, for each land use pattern is set up a land use pattern three-dimensional feature spatial model; Said satellite remote-sensing image raster data is meant that it is the raster data that a plurality of subelements obtain that the satellite remote-sensing image with the area of waiting to classify utilizes grid division, and said subelement is called as pixel;

S4, said historical soil is utilized vector data and satellite remote-sensing image raster data overlap-add procedure; Stack back pixel drops on historical soil and utilizes on the patch in the vector data; The land use pattern information of this patch is the land use pattern information of all pixels that drop in this patch; Judge that again all pixels are whether within the land use pattern three-dimensional feature spatial model of its corresponding types; Find the pixel that drops on outside the land use pattern three-dimensional feature spatial model, to drop on the pixel outside the land use pattern three-dimensional feature spatial model be that the first kind changes pixel to mark in the satellite remote-sensing image raster data; The raster data that the said mark first kind is changed pixel is as first kind raster data;

S5, the first kind is changed the current land use pattern of pixel convert its actual land use pattern into; It is second type of variation pixel that the first kind of conversion back mark conversion land use pattern in the satellite remote-sensing image raster data changes pixel; Mark is crossed second type of satellite remote-sensing image raster data that changes pixel as second type of raster data, change pixel land use pattern through the conversion first kind thus and realize the land use pattern classification.

In the middle of step s1; The said sample patch of choosing every kind of land use pattern is specially in specific embodiment, to a kind of land use pattern; To belong to all patches of this land use pattern sorts according to area from big to small; Comprise vector data and land use pattern information in the said patch, vector data is represented the shape and the area of this patch, according to formula:

$\mathrm{Pa}=\frac{\underset{j=0}{\overset{x}{\mathrm{\&Sigma;}}}A{c}_j}{\mathrm{As}};$

Wherein Pa representes first threshold, or is referred to as to choose sample threshold;

representative is added to i patch sum in order according to the patch after the descending ordering of area;

As represents the total area of these all patches of land use pattern.

First threshold Pa value is set at 55%-65% among the embodiment; That is to say; When bigger i plaque area sum accounts for all patch total area 55%-65%, then choose this bigger i patch as sample, the vector data that comprises in the sample patch is the sample vector data; Repeat aforesaid operations to each land use pattern, obtain the sample patch and the sample vector data of each land use pattern.

In step s2, said sample vector data is carried out the training area analysis, obtain the training area vector data and in specific embodiment do, the sample patch in this vector data of taking a sample, said sample patch comprises patch shape information and plaque area information; Keep sample patch shape information constant, area is dwindled, make the ratio of sample patch and the area of former sample patch after dwindling reach second threshold range that is provided with in advance; Getting the patch that ratio that area dwindles back and the area of former vector data reaches second threshold range that is provided with in advance is training area; The land use pattern information of training area is consistent with sample patch land use pattern information, training area and training area land use pattern information composing training district vector data.

With reference to formula:

$P_{\mathrm{buffer}}=\frac{A_d}{A};$

A wherein _dBe the area after the sample vector data is scaled, A is the original area of sample vector data, when the ratio of the two, i.e. and patch training area assay value P _BufferReach the scope that preestablishes second threshold value, then think A _dBe training area through obtaining after the training area analysis; Said P _BufferCan think setting in an embodiment as required; According to the method described above each sample vector data is carried out the training area analysis, obtain corresponding training area.

Described in the step s3; Setting up land use pattern three-dimensional feature spatial model concrete grammar in an embodiment for each land use pattern is; To a kind of land use pattern; The training area of this land use pattern and the multispectral or high spectrum satellite remote-sensing image raster data of process standardization are carried out overlap-add procedure, obtain containing the sample spectroscopic data of a plurality of wave bands; Adopt No. 1 moonlet remote sensing image of environment to superpose in the present embodiment, the sample spectroscopic data of said a plurality of wave bands specifically comprises visible light 1, visible light 2, visible light 3, near infrared ray totally 4 wave bands, shown in the following matrix of expression mode:

$\left[\begin{array}{cccc}{D}_{11}& D_{12}& D_{13}& D_{14}\\ D_{21}& D_{22}& D_{23}& D_{24}\\ ...& ...& ...& ...\\ D_{n1}& D_{n2}& D_{n3}& D_{n4}\end{array}\right]$

N represents in this land use pattern in this matrix, and No. 1 moonlet remote sensing image of environment comprises the quantity of pixel, and each row is represented the spectral value of 1,2,3,4 four wave bands of a pixel in the matrix;

Again above-mentioned matrix is carried out principal component analysis (PCA); Principal component analysis (PCA) is to manage original numerous spectral values with a plurality of wave bands of certain correlativity; Again be configured to the major component of one group of new irrelevant mutually comprehensive index value as the sample spectroscopic data; Do not comprise repeated characteristic between each major component, and the characteristic that each major component comprises reduces successively.

The mathematical method of principal component analysis (PCA) described in the present embodiment is that quadrature decomposes conversion, is specially the spectral value that utilizes four wave bands and makes up four mutual uncorrelated new variables, and said new variables is called as major component.The method that said quadrature decomposes conversion is the widely used mathematical measure in present this area, and this does not do and gives unnecessary details again.

Calculate the variance of each major component; The size of the said major component variance numerical value that calculates is the characteristic of this major component representative; The characteristic of this major component representative of the big more expression of variance data is many more, and just this major component explains that the ability of said sample spectroscopic data is strong more.Therefore choose the maximum major component of variance as first principal component,, then need further to choose a plurality of major components again and reflect original characteristic if first principal component is not enough to represent the included characteristic of original a plurality of spectral value.

Obtaining major component matrix note does:

$\left[\begin{array}{cccc}{P}_{11}& P_{12}& P_{13}& P_{14}\\ P_{21}& P_{22}& P_{23}& P_{24}\\ ...& ...& ...& ...\\ P_{n1}& P_{n2}& P_{n3}& P_{n4}\end{array}\right]$

Choose in the present embodiment and comprise three maximum major components of characteristic, i.e. first three major component;

After the principal component analysis (PCA) of sample spectroscopic data process; Can think that first three major component represented most characteristic in the sample spectroscopic data; So getting first three major component carries out modeling and has promptly reduced dimension; Promptly realize the foundation of three feature spaces, in the model of setting up, comprised most characteristic in the sample spectroscopic data again;

Said land use pattern three-dimensional feature spatial model expression formula is following:

$\underset{j=1}{\overset{3}{\mathrm{\&Sigma;}}}\frac{{(P_j-M{P}_j)}^2}{{\mathrm{\&sigma;}}_{\mathrm{<figure-callout\; id="2"\; label="j"\; filenames="HDA0000122751920000011.png"\; state="\{\{state\}\}">j</figure-callout>}}^2}<{\mathrm{<figure-callout\; id="2"\; label="c"\; filenames="HDA0000122751920000011.png"\; state="\{\{state\}\}">c</figure-callout>}}^2$

P wherein _jRepresent the value of the j major component of substitution model pixel, MP _jRepresent in this land use pattern, the average of j major component, Represent the standard deviation of j major component in this land use pattern, c ²Be constant, represent the feature space index, set up on their own according to concrete needs in an embodiment;

According to said method, every kind of land use pattern is set up corresponding land use pattern three-dimensional feature spatial model.

According to the said method of step s3 in the present embodiment; In step s4; Said historical soil is utilized vector data and GMS remote sensing image overlap-add procedure, obtain a common spectroscopic data, stack back pixel drops in the patch that the soil utilizes vector data; Promptly all drop on pixel land use pattern information in this patch to this patch land use pattern information; Method according to above-mentioned principal component analysis (PCA) is carried out principal component analysis (PCA) to the common spectroscopic data that obtains again, obtains first three major component of common each pixel of spectroscopic data, and the land use pattern three-dimensional feature spatial model formula that this pixel of first three major component substitution is corresponding of pixel one by one; Judge that whether pixel drops within the land use pattern three-dimensional feature spatial model, finds the pixel that drops on outside the land use pattern three-dimensional feature spatial model.

For the pixel that drops within the land use pattern three-dimensional feature spatial model, it is constant pixel that note is done, and thinks that the said land use pattern of this pixel is original type, does not change; For the pixel that drops on outside the land use pattern three-dimensional feature spatial model, in the satellite remote-sensing image raster data, be labeled as the first kind and change pixel.

In addition, need to prove herein that the land use pattern three-dimensional feature spatial model of formulating according to formula described in the step s3 is shaped as an elliposoidal space, the size in said elliposoidal space is to confirm through the feature space index.In concrete the application, it promptly is the pixel that drops on outside the elliposoidal space in essence that the first kind changes pixel.Think that the quantity and the first kind that the control first kind changes pixel change pixel at all pixel proportions; When formulating a certain land use pattern three-dimensional feature spatial model; The feature space index should in the light of actual conditions be set, cause result of calculation within zone of reasonableness.

In the present embodiment; According to said process; Then step s5 is specially: collect all first kind that obtain among the step s4 and change pixel; Three major components that each first kind changed pixel are updated to all types of land use pattern three-dimensional feature spatial models once more, calculate the distance that this first kind changes pixel and each land use pattern three-dimensional feature spatial model, and find and a nearest land use pattern three-dimensional feature spatial model of this first kind variation pixel;

Said computed range formula is following:

$\mathrm{di}=\left(\underset{j=1}{\overset{3}{\mathrm{\&Sigma;}}}\frac{(P_j-M{P}_{\mathrm{ji}})}{{\mathrm{\&sigma;}}_{\mathrm{ji}}^2}\right)/{\mathrm{<figure-callout\; id="2"\; label="c\; i"\; filenames="HDA0000122751920000011.png"\; state="\{\{state\}\}">c</figure-callout>}}_i^{}$

min(di)

Wherein di represents the distance of the pixel and the i land use pattern three-dimensional feature spatial model of substitution; P _jRepresent the value of substitution pixel j major component, MP _JiRepresent the average of j major component in the i land use pattern,

Represent the standard deviation of j major component in the i land use pattern,

Represent the feature space index in the i land use pattern three-dimensional feature spatial model; On behalf of this first kind, min (di) change the distance of minimum in the distance of pixel and all land use pattern three-dimensional feature spatial models.

According to aforementioned calculation; The land use pattern that all first kind is changed pixel converts into and the nearest corresponding land use pattern of land use pattern three-dimensional feature spatial model of each first kind variation pixel; The pixel of mark conversion land use pattern is second type of variation pixel in the satellite remote-sensing image raster data, and mark is crossed second type of satellite remote-sensing image raster data that changes pixel as second type of raster data.

In order to make classification more accurate; In the present embodiment; Before the conversion first kind changes pixel land use pattern; Further be provided with one and shift the resistance matrix, comprised the inclusive resistance coefficient of conversion needs between any two kinds of land use patterns in the said transfer resistance matrix, said resistance coefficient is stipulated according to the probability size that two kinds of land use patterns change; According to following formula, the distance that the first kind that obtains among the step s5 is changed pixel and each land use pattern three-dimensional feature spatial model further multiply by corresponding resistance coefficient:

dLi＝ρ _i*di

min(dLi)

ρ wherein _iRepresent the resistance coefficient of the original land use pattern of substitution pixel to the conversion of i class land use pattern;

DLi then represents the distance based on substitution pixel under the transformation rule and i land use pattern three-dimensional feature spatial model;

Min (dLi) then represents based under the transformation rule, minimum distance in the distance of this first kind variation pixel and all land use pattern three-dimensional feature spatial models.

According to the method described above; Described in the step s5 first kind being changed the current land use pattern of pixel converts its actual land use pattern into and in specific embodiment, can be specially; The land use pattern that all first kind are changed pixels converts into and changes the nearest corresponding land use pattern of land use pattern three-dimensional feature spatial model of pixel with each first kind under rule-based; The pixel of the rule-based conversion land use pattern of mark is the 3rd a type of variation pixel in the satellite remote-sensing image raster data, and said mark is crossed the 3rd type of satellite remote-sensing image raster data that changes pixel as the 3rd type of raster data.

Be the classification levels of precision of further raising the method for the invention, said method also comprises the steps:

First kind raster data and second type of raster data are carried out the overlay analysis contrast, and less to area in the different land use type according to the corrosion algorithm, the discrete patch that distributes is revised.

Or first kind raster data and the 3rd type of raster data carried out the overlay analysis contrast, less to area in the different land use type according to the corrosion algorithm, the discrete patch that distributes is revised.

Corresponding said method also discloses a kind of land use classification system of realizing said method among the present invention, with reference to shown in Figure 2.

Said system comprises:

Collector unit, modelling unit, computing unit and converting unit;

Collector unit is used to collect wait the to classify historical soil in area and utilizes vector data and satellite remote-sensing image raster data;

The modelling unit connects collector unit, is used for utilizing vector data to set up some land use pattern three-dimensional feature spatial models according to said historical soil;

Computing unit, link model is set up the unit, is used for said historical soil is utilized vector data and satellite remote-sensing image raster data overlap-add procedure, obtains the land use pattern information of each pixel in the satellite remote-sensing image raster data; Judge successively whether each pixel is belonging to it within land use pattern three-dimensional feature spatial model of same land use pattern; Acquisition drops on the pixel outside the land use pattern three-dimensional feature spatial model, in the satellite remote-sensing image raster data, the said pixel that drops on outside the land use pattern three-dimensional feature spatial model is labeled as the first kind and changes pixel;

Converting unit connects computing unit, is used for that the first kind is changed the current land use pattern of pixel and converts its actual affiliated land use pattern into.

Said modelling unit comprises:

The sample collection unit, training area analytic unit and construction unit;

The sample collection unit is used to every kind of land use pattern and chooses a plurality of sample patches, and obtains the corresponding land use pattern information of sample patch, wherein, comprises that the vector data of sample patch and sample patch land use pattern information is the sample vector data;

The training area analytic unit is used for said sample vector data is carried out the training area analysis, obtains the training area vector data;

Construction unit is used for training area vector data and satellite remote-sensing image raster data overlap-add procedure, for each land use pattern is set up a land use pattern three-dimensional feature spatial model.

Said converting unit comprises:

Metrics calculation unit and type information change unit;

Metrics calculation unit is used to calculate the distance that the first kind changes pixel and all land use pattern three-dimensional feature spatial models, and finds and the nearest land use pattern three-dimensional feature spatial model of this first kind variation pixel;

Type information change unit, the land use pattern that is used for the first kind is changed pixel converts the land use pattern three-dimensional feature spatial model corresponding land use pattern nearest with this variation pixel into.

Said converting unit further comprises:

Shift moment of resistance array element; Be used to be provided with transfer resistance matrix; Comprised between any two kinds of land use patterns the inclusive resistance coefficient of conversion needs in the said transfer resistance matrix, the probability that said resistance coefficient changes based on two kinds of land use patterns is big or small and stipulate; Shift the moment of resistance array element distance that the first kind changes pixel and each land use pattern three-dimensional feature spatial model further multiply by corresponding resistance coefficient; Obtain changing the distance of pixel and each land use pattern three-dimensional feature spatial model, obtain based on changing the nearest land use pattern three-dimensional feature spatial model of pixel with the first kind under the transformation rule based on the first kind under the transformation rule;

The land use pattern that said type information change unit changes pixel with the first kind convert into rule-based under the land use pattern three-dimensional feature spatial model corresponding land use pattern nearest with this variation pixel.

Said system further comprises:

After said converting unit also was used for the first kind changed the current land use pattern of pixel and convert its actual land use pattern into, the first kind that in the satellite remote-sensing image raster data, will pass through the land use pattern conversion changed pixel and is labeled as second type and changes pixel;

Said system also comprises the corrosion algorithm unit, connects computing unit and converting unit, is used to extract the computing unit mark and crosses satellite remote-sensing image raster data that the first kind changes pixel as first kind raster data; Extract mark and cross second type of satellite remote-sensing image raster data that changes pixel as second type of raster data; Contrast first kind raster data and second type of raster data carry out the correction of classified information according to the corrosion algorithm to the discrete pixel that distributes after the conversion land use pattern.

The above only is a preferred implementation of the present invention; Should be pointed out that for those skilled in the art, under the prerequisite that does not break away from the principle of the invention; Can also make some improvement and retouching, these improvement and retouching also should be regarded as protection scope of the present invention.

Claims (16)

1. the method for land use pattern classification is characterized in that this method may further comprise the steps:

Collection wait the to classify historical soil in area utilizes vector data; Utilize vector data to set up some land use pattern three-dimensional feature spatial models according to said historical soil; Wherein, Said historical soil utilizes vector data to comprise patch and patch land use pattern information, and each land use pattern three-dimensional feature spatial model is corresponding with a kind of land use pattern;

The acquisition satellite remote-sensing image raster data in area of waiting to classify, said satellite remote-sensing image raster data comprises some pixels, each pixel be the satellite remote-sensing image with the area of waiting to classify utilize grid division and subelement;

Utilize vector data and satellite remote-sensing image raster data to carry out overlap-add procedure in said historical soil, obtain the land use pattern information of each pixel in the satellite remote-sensing image raster data;

Judge successively whether each pixel is belonging to it within land use pattern three-dimensional feature spatial model of same land use pattern; Acquisition drops on the pixel outside the land use pattern three-dimensional feature spatial model, in the satellite remote-sensing image raster data, the said pixel that drops on outside the land use pattern three-dimensional feature spatial model is labeled as the first kind and changes pixel;

The first kind is changed the current land use pattern of pixel convert its actual affiliated land use pattern into, realize the land use pattern classification.

2. according to the said method of claim 1, it is characterized in that said collection wait the to classify historical soil in area utilizes vector data, the step of utilizing vector data to set up some land use pattern three-dimensional feature spatial models according to said historical soil comprises:

Collection wait the to classify historical soil in area utilizes vector data; Said historical soil utilizes vector data to comprise patch and patch land use pattern information; For every kind of land use pattern is chosen a plurality of sample patches; And obtain the corresponding land use pattern information of sample patch, wherein, comprise that the vector data of sample patch and sample patch land use pattern information is the sample vector data;

Said sample vector data is carried out the training area analysis, obtain the training area vector data;

Through with training area vector data and satellite remote-sensing image raster data overlap-add procedure, for each land use pattern is set up a land use pattern three-dimensional feature spatial model.

3. according to the said method of claim 2, it is characterized in that, saidly be specially for a plurality of sample patches of choosing of every kind of land use pattern:

To same land use pattern; To belong to all patches of this land use pattern sorts according to area from big to small; And the area of patch is added up successively according to this ordering; When the ratio of accumulation result and all patch total areas reaches predefined first threshold scope, then getting a plurality of patches that add up is the sample patch;

For every kind of land use pattern, repeat said process, obtain the respective sample patch.

4. according to the said method of claim 2, it is characterized in that, said said sample vector data carried out the training area analysis, obtain the training area vector data and be specially:

Take a sample sample patch in this vector data, said sample patch comprises patch shape information and plaque area information;

Keep sample patch shape information constant, area is dwindled, make the ratio of sample patch and the area of former sample patch after dwindling reach second threshold range that is provided with in advance;

Getting the patch that ratio that area dwindles back and the area of former vector data reaches second threshold range that is provided with in advance is training area; The land use pattern information of training area is consistent with sample patch land use pattern information, comprises that the vector data of training area and training area land use pattern information is called the training area vector data.

5. according to the said method of claim 2, it is characterized in that, said training area vector data and satellite remote-sensing image raster data overlap-add procedure be specially:

To a kind of land use pattern, will belong to the training area of this land use pattern and the multispectral or high spectrum satellite remote-sensing image raster data of process standardization and carry out overlay analysis, obtain containing the sample spectroscopic data of a plurality of wave bands; Said sample spectroscopic data is represented with the form of matrix, a plurality of characteristics that pixel comprised in each row representative sample spectroscopic data of said matrix, and the characteristic quantity that each pixel comprised equates with the wave band quantity of sample spectroscopic data;

Said sample spectroscopic data is carried out principal component analysis (PCA), obtain the major component of spectroscopic data, the quantity of said major component equates with wave band quantity, and chooses and comprise three maximum major components of characteristic; Wherein, said principal component analysis (PCA) is specially quadrature decomposition conversion;

To the training area of all land use patterns, repeat said process, obtain three major components of every kind of land use pattern training area.

6. according to the said method of claim 5, it is characterized in that, saidly set up a land use pattern three-dimensional feature spatial model for each land use pattern and be specially:

To a kind of land use pattern, utilize three major components of this land use pattern training area to set up land use pattern three-dimensional feature spatial model;

Land use pattern three-dimensional feature spatial model is expressed as

$\underset{j=1}{\overset{3}{\mathrm{\&Sigma;}}}\frac{{(P_j-M{P}_j)}^2}{{\mathrm{\&sigma;}}_j^2}\mathrm{\&Phi;}{c}^2$

P wherein in the formula _jRepresent the value of the j major component of substitution model pixel, MP _jRepresent in this land use pattern, the average of j major component, Represent the standard deviation of j major component in this land use pattern, c ²Be constant, represent the feature space index;

Utilize three major components of above-mentioned formula and every kind of land use pattern training area, for every kind of land use pattern is set up a land use pattern three-dimensional feature spatial model.

7. according to the said method of claim 1, it is characterized in that, said said historical soil utilized vector data and satellite remote-sensing image raster data overlap-add procedure, be specially:

Utilize vector data and satellite remote-sensing image raster data stack to carry out overlay analysis in historical soil, obtain containing the common spectroscopic data of a plurality of wave bands through standardization; Said common spectroscopic data is represented with the form of matrix, a plurality of characteristics that pixel comprised in the common spectroscopic data of each row representative of said matrix, and the characteristic quantity that each pixel comprised equates with the wave band quantity of common spectroscopic data;

Said common spectroscopic data is carried out principal component analysis (PCA), obtain the major component of spectroscopic data, the quantity of said major component equates with wave band quantity, and chooses and comprise three maximum major components of characteristic; Wherein, said principal component analysis (PCA) is specially quadrature decomposition conversion.

8. according to the said method of claim 6, it is characterized in that, saidly the first kind is changed the current land use pattern of pixel convert its land use pattern under actual into and be specially:

Three major components that the first kind changed pixel are updated to all land use pattern three-dimensional feature spatial models; Calculate the distance that this first kind changes pixel and each land use pattern three-dimensional feature spatial model, and find and a nearest land use pattern three-dimensional feature spatial model of this first kind variation pixel;

The land use pattern that the first kind is changed pixel converts into and the nearest corresponding land use pattern of land use pattern three-dimensional feature spatial model of this first kind variation pixel.

9. said according to Claim 8 system is characterized in that, the distance that the said calculating first kind changes pixel and land use pattern three-dimensional feature spatial model is specially:

According to formula $\mathrm{Di}=\left(\underset{j=1}{\overset{3}{\mathrm{\&Sigma;}}}\frac{(P_j-M{P}_{\mathrm{Ji}})}{{\mathrm{\&sigma;}}_{\mathrm{Ji}}^2}\right)/c_i^2$

Wherein di represents the distance of the pixel and the i land use pattern three-dimensional feature spatial model of substitution; P _jRepresent the value of substitution pixel j major component, MP _JiRepresent the average of j major component in the i land use pattern, Represent the standard deviation of j major component in the i land use pattern,

Represent the feature space index in the i land use pattern three-dimensional feature spatial model.

10. said according to Claim 8 method is characterized in that said method further comprises:

Before the conversion first kind changes pixel land use pattern; Further be provided with one and shift the resistance matrix; Comprised the inclusive resistance coefficient of conversion needs between any two kinds of land use patterns in the said transfer resistance matrix, said resistance coefficient is stipulated according to the probability size that two kinds of land use patterns change; The distance that the first kind is changed pixel and each land use pattern three-dimensional feature spatial model further multiply by corresponding resistance coefficient; Obtain changing the distance of pixel and each land use pattern three-dimensional feature spatial model based on the first kind under the transformation rule; Acquisition changes the current land use pattern of pixel with the first kind and converts into based on changing the nearest pairing land use pattern of land use pattern three-dimensional feature spatial model of pixel with the first kind under the transformation rule based on changing the nearest land use pattern three-dimensional feature spatial model of pixel with the first kind under the transformation rule.

11., it is characterized in that according to the said method of claim 1:

Saidly in the satellite remote-sensing image raster data, the said pixel that drops on outside the land use pattern three-dimensional feature spatial model is labeled as after the first kind changes pixel; Said method also comprises, mark is crossed satellite remote-sensing image raster data that the first kind changes pixel as first kind raster data;

After the first kind changed the current land use pattern of pixel and convert its actual land use pattern into, said method also comprised, the first kind that will pass through the land use pattern conversion changes pixel and is labeled as second type and changes pixel; Said mark is crossed second type of satellite remote-sensing image raster data that changes pixel as second type of raster data

Said method also comprises:

Contrast first kind raster data and second type of raster data carry out the correction of classified information according to the corrosion algorithm to the discrete pixel that distributes after the conversion land use pattern.

12. the system of a land use pattern classification is characterized in that said system comprises:

Collector unit, modelling unit, computing unit and converting unit;

Collector unit is used to collect wait the to classify historical soil in area and utilizes vector data and satellite remote-sensing image raster data;

The modelling unit connects collector unit, is used for utilizing vector data to set up some land use pattern three-dimensional feature spatial models according to said historical soil;

Computing unit, link model is set up the unit, is used for said historical soil is utilized vector data and satellite remote-sensing image raster data overlap-add procedure, obtains the land use pattern information of each pixel in the satellite remote-sensing image raster data; Judge successively whether each pixel is belonging to it within land use pattern three-dimensional feature spatial model of same land use pattern; Acquisition drops on the pixel outside the land use pattern three-dimensional feature spatial model, in the satellite remote-sensing image raster data, the said pixel that drops on outside the land use pattern three-dimensional feature spatial model is labeled as the first kind and changes pixel;

Converting unit connects computing unit, is used for that the first kind is changed the current land use pattern of pixel and converts its actual affiliated land use pattern into.

13., it is characterized in that said modelling unit comprises according to the said system of claim 12:

The sample collection unit, training area analytic unit and construction unit;

The sample collection unit is used to every kind of land use pattern and chooses a plurality of sample patches, and obtains the corresponding land use pattern information of sample patch, wherein, comprises that the vector data of sample patch and sample patch land use pattern information is the sample vector data;

The training area analytic unit is used for said sample vector data is carried out the training area analysis, obtains the training area vector data;

Construction unit is used for training area vector data and satellite remote-sensing image raster data overlap-add procedure, for each land use pattern is set up a land use pattern three-dimensional feature spatial model.

14., it is characterized in that said converting unit comprises according to the said system of claim 12:

Metrics calculation unit and type information change unit;

Metrics calculation unit is used to calculate the distance that the first kind changes pixel and all land use pattern three-dimensional feature spatial models, and finds and the nearest land use pattern three-dimensional feature spatial model of this first kind variation pixel;

Type information change unit, the land use pattern that is used for the first kind is changed pixel converts the land use pattern three-dimensional feature spatial model corresponding land use pattern nearest with this variation pixel into.

15., it is characterized in that said converting unit further comprises according to the said system of claim 14:

Shift moment of resistance array element; Be used to be provided with transfer resistance matrix; Comprised between any two kinds of land use patterns the inclusive resistance coefficient of conversion needs in the said transfer resistance matrix, the probability that said resistance coefficient changes based on two kinds of land use patterns is big or small and stipulate; Shift the moment of resistance array element distance that the first kind changes pixel and each land use pattern three-dimensional feature spatial model further multiply by corresponding resistance coefficient; Obtain changing the distance of pixel and each land use pattern three-dimensional feature spatial model, obtain based on changing the nearest land use pattern three-dimensional feature spatial model of pixel with the first kind under the transformation rule based on the first kind under the transformation rule;

The land use pattern that said type information change unit changes pixel with the first kind convert into rule-based under the land use pattern three-dimensional feature spatial model corresponding land use pattern nearest with this variation pixel.

16., it is characterized in that said system further comprises according to the said system of claim 12:

After said converting unit also was used for the first kind changed the current land use pattern of pixel and convert its actual land use pattern into, the first kind that in the satellite remote-sensing image raster data, will pass through the land use pattern conversion changed pixel and is labeled as second type and changes pixel;

Said system also comprises the corrosion algorithm unit, connects computing unit and converting unit, is used to extract the computing unit mark and crosses satellite remote-sensing image raster data that the first kind changes pixel as first kind raster data; Extract mark and cross second type of satellite remote-sensing image raster data that changes pixel as second type of raster data; Contrast first kind raster data and second type of raster data carry out the correction of classified information according to the corrosion algorithm to the discrete pixel that distributes after the conversion land use pattern.

Images