CN103295030A - Classification method and device based on hyperspectral remote sensing images - Google Patents

Abstract

An embodiment of the invention provides a classification method and device based on hyperspectral remote sensing images. Spatial neighhood information of pixels in the hyperspectral remote sensing images serves as a constraint condition, and an optimal clustering center of the hyperspectral remote sensing images is determined through an artificial bee colony algorithm according to a clustering center assembly obtained by calculation and is utilized to classify targets in the hyperspectral remote sensing images. Accordingly, in the process of classifying the targets in the images, not only do spectrums of the pixels serve as a basis, but also the spatial neighhood information of the pixels serves as a basis, and according to the principle that the closer spatially in geography, the higher the possibility of belonging to an identical class, the problem of classification excess can be solved since the possibility of classifying the pixels large in spectrum difference and spatially close as one class is greatly increased according to the spatial neighhood information of the pixels.

Description

A kind of sorting technique and device based on high-spectrum remote sensing

Technical field

The present invention relates to the remote sensing field, relate in particular to a kind of sorting technique based on high-spectrum remote sensing and device.

Background technology

High-spectrum remote sensing is a kind of remote sensing images that use high spectrum sensor to obtain, and the target in the high-spectrum remote sensing is classified, and can understand the situation that ground covers.Existing sorting technique at high-spectrum remote sensing, the common classification of identifying different target according to the spectral information of different target in the high spectrum image.

Comprise coverage rate comparatively widely in the remote sensing images, the user wishes the primary categories that the normally ground understood covers by high-spectrum remote sensing, but not details, and the foundation of the sorting technique of existing high spectrum image differentiation different target is the spectral signature of different target, therefore, cause the excessively problem of classification easily.

For example, high-spectrum remote sensing at the urban area, the user wishes to recognize by it road layout in city, and so, only the road that need identify in these remote sensing images gets final product, and the well lid on the road does not need to distinguish with road, but the spectrum of well lid and road is completely different, according to existing sorting technique to high-spectrum remote sensing, well lid and road will inevitably be divided into two classes, thus the result who causes continuous road to be divided.

Summary of the invention

In view of this, the embodiment of the invention provides a kind of sorting technique and device of high-spectrum remote sensing, and purpose is to solve the existing excessive problem of classification that causes at the sorting technique of high-spectrum remote sensing.

To achieve these goals, the embodiment of the invention provides following technical scheme:

A kind of sorting technique of high-spectrum remote sensing comprises:

Calculate the cluster centre set of the predetermined number of high-spectrum remote sensing;

The spatial neighborhood information of pixel in the described high-spectrum remote sensing as constraint condition, according to the cluster centre set of described predetermined number, is determined the optimum cluster centre set of described high-spectrum remote sensing by artificial ant colony algorithm;

According to the optimum cluster centre set of described high-spectrum remote sensing, the target in the described high-spectrum remote sensing is classified.

Preferably, described spatial neighborhood information with the pixel in the described high-spectrum remote sensing according to the cluster centre set of described predetermined number, determines that the optimum cluster centre of described high-spectrum remote sensing comprises as constraint condition by artificial ant colony algorithm:

With the food source of the cluster centre of described predetermined number set as gathering honey honeybee in the artificial ant colony algorithm;

Carry out following steps successively, until satisfying default first condition:

Described gathering honey honeybee is determined the food source of the fitness maximum of described gathering honey honeybee correspondence according to the spatial neighborhood information of pixel in described food source and the described high-spectrum remote sensing;

Follow honeybee according to the spatial neighborhood information of pixel in described food source and the described high-spectrum remote sensing, determine described food source of following the fitness maximum of honeybee correspondence;

The food source of fitness maximum in the food source of the food source of the fitness maximum of described gathering honey honeybee correspondence and described fitness maximum of following the honeybee correspondence is defined as optimum cluster centre set;

Search bee is determined food source at random from described high-spectrum remote sensing.

Preferably, described gathering honey honeybee determines that according to the spatial neighborhood information of pixel in described food source and the described high-spectrum remote sensing food source of the fitness maximum of described gathering honey honeybee correspondence comprises:

Calculate the fitness of described food source;

Gathering honey honeybee that will be corresponding with described food source is gathered as the cluster centre that upgrades according to the new food source that described food source searches;

According to the cluster centre set of described renewal, determine the cluster result of the target in the described high-spectrum remote sensing;

By local iteration's condition model, calculate the objective function of the cluster centre set of described renewal according to described cluster result;

According to described objective function, calculate the fitness of the cluster centre set of described renewal;

The fitness of the fitness by more described food source and the set of the cluster centre of described renewal is with the food source of fitness the greater as the fitness maximum of described gathering honey honeybee correspondence.

Preferably, described spatial neighborhood information of following pixel in the honeybee described food source of foundation and the described high-spectrum remote sensing, determine that described food source of following the fitness maximum of honeybee correspondence comprises:

Follow honeybee according to default probability, from the food source of described gathering honey honeybee, select one as food source;

Calculate the fitness of described food source;

The honeybee of following that will be corresponding with described food source is gathered as the cluster centre that upgrades according to the new food source that described food source searches;

According to the cluster centre set of described renewal, determine the cluster result of the target in the described high-spectrum remote sensing;

By local iteration's condition model, calculate the objective function of the cluster centre set of described renewal according to described cluster result;

According to described objective function, calculate the fitness of the cluster centre set of described renewal;

The fitness of the fitness by more described food source and the set of the cluster centre of described renewal, with fitness the greater as described food source of following the fitness maximum of honeybee correspondence.

Preferably, described search bee is determined food source at random from described high-spectrum remote sensing before, also comprise:

When described gathering honey honeybee is satisfied default second condition, described gathering honey honeybee is converted to search bee.

Preferably, gather according to the optimum cluster centre of described high-spectrum remote sensing, the target in the described high-spectrum remote sensing is classified to be comprised:

By local iteration's condition model, described cluster result is carried out subseries again;

Cluster centre set corresponding stored with sorting result and described renewal again;

Obtain the classification results corresponding with described food source;

With the classification results of the food source correspondence of the described fitness maximum classification results as the target in the described high-spectrum remote sensing.

Preferably, the set of the cluster centre of described calculating high-spectrum remote sensing comprises:

Use the K mean algorithm to calculate the cluster centre set of the predetermined number of high-spectrum remote sensing.

A kind of sorter of high-spectrum remote sensing comprises:

Cluster centre set computing module is used for the cluster centre set of the predetermined number of calculating high-spectrum remote sensing;

Optimum cluster centre set determination module, be used for spatial neighborhood information with described high-spectrum remote sensing pixel as constraint condition, according to the cluster centre set of described predetermined number, determine the optimum cluster centre set of described high-spectrum remote sensing by artificial ant colony algorithm;

Sort module is used for the optimum cluster centre set according to described high-spectrum remote sensing, and the target in the described high-spectrum remote sensing is classified.

Preferably, described optimum cluster centre set determination module comprises:

Food source is chosen the unit, for the food source of the cluster centre of described predetermined number being gathered as artificial ant colony algorithm gathering honey honeybee;

Iteration unit is used for carrying out successively following steps, until satisfying default first condition: described gathering honey honeybee according to described food source and described high-spectrum remote sensing in the spatial neighborhood information of pixel, determine the food source of the fitness maximum of described gathering honey honeybee correspondence; Follow honeybee according to the spatial neighborhood information of pixel in described food source and the described high-spectrum remote sensing, determine described food source of following the fitness maximum of honeybee correspondence; The food source of fitness maximum in the food source of the food source of the fitness maximum of described gathering honey honeybee correspondence and described fitness maximum of following the honeybee correspondence is defined as optimum cluster centre set; Search bee is determined food source at random from described high-spectrum remote sensing.

Preferably, described iteration unit comprises:

Gathering honey honeybee search subelement is for the fitness that calculates described food source; Gathering honey honeybee that will be corresponding with described food source is gathered as the cluster centre that upgrades according to the new food source that described food source searches; According to the cluster centre set of described renewal, determine the cluster result of the target in the described high-spectrum remote sensing; By local iteration's condition model, calculate the objective function of the cluster centre set of described renewal according to described cluster result; According to described objective function, calculate the fitness of the cluster centre set of described renewal; The fitness of the fitness by more described food source and the set of the cluster centre of described renewal is with the food source of fitness the greater as the fitness maximum of described gathering honey honeybee correspondence.

Preferably, described iteration module comprises:

Follow honeybee search subelement, be used for following honeybee according to default probability, from the food source of described gathering honey honeybee, select one as food source; Calculate the fitness of described food source; The honeybee of following that will be corresponding with described food source is gathered as the cluster centre that upgrades according to the new food source that described food source searches; According to the cluster centre set of described renewal, determine the cluster result of the target in the described high-spectrum remote sensing; By local iteration's condition model, calculate the objective function of the cluster centre set of described renewal according to described cluster result; According to described objective function, calculate the fitness of the cluster centre set of described renewal; The fitness of the fitness by more described food source and the set of the cluster centre of described renewal, with fitness the greater as described food source of following the fitness maximum of honeybee correspondence.

Preferably, described sort module comprises:

Taxon is used for by local iteration's condition model described cluster result being carried out subseries again;

Storage unit, the cluster centre that is used for again sorting result and described renewal is gathered corresponding stored;

Acquiring unit is used for obtaining the classification results corresponding with described food source;

Determining unit is used for the classification results of the food source correspondence of the described fitness maximum classification results as the target of described high-spectrum remote sensing.

The sorting technique of the high-spectrum remote sensing that the embodiment of the invention provides and device, with the spatial neighborhood information of pixel in the described high-spectrum remote sensing as constraint condition, cluster centre set according to the high-spectrum remote sensing that calculates, determine the optimum cluster centre of described high-spectrum remote sensing by artificial ant colony algorithm, and use optimum cluster centre that the target in the high-spectrum remote sensing is classified, that is to say, end user worker bee group algorithm is further optimized cluster centre, and in the process of optimizing with the spatial neighborhood information of pixel as constraint condition, therefore, in the process to the target classification in the image, not only the spectrum with pixel is foundation, and be foundation with the spatial neighborhood information of pixel, according to more closing on the space in the geography, then belong to the more high principle of possibility of same classification, after considering the spatial neighborhood information of pixel, SPECTRAL DIVERSITY greatly but the pixel that spatially closes on is classified as the possibility of a class increases greatly, therefore, can avoid the excessive problem of classifying.

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, apparently, accompanying drawing in describing below only 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 process flow diagram of the sorting technique of the disclosed a kind of high-spectrum remote sensing of the embodiment of the invention;

Fig. 2 is the process flow diagram of the sorting technique of disclosed another high-spectrum remote sensing of the embodiment of the invention;

Fig. 3 is the process flow diagram of the sorting technique of disclosed another high-spectrum remote sensing of the embodiment of the invention;

Fig. 4 is the sorter structural representation of the disclosed a kind of high-spectrum remote sensing of the embodiment of the invention.

Embodiment

The embodiment of the invention discloses a kind of sorting technique and device of high-spectrum remote sensing, its core inventive point is, with the spatial neighborhood information of pixel in the high-spectrum remote sensing as constraint condition, end user worker bee group algorithm is optimized the cluster centre set of high-spectrum remote sensing, obtain optimum cluster centre, use optimum cluster centre that high-spectrum remote sensing is classified.

Below in conjunction with the accompanying drawing in the embodiment of the invention, the technical scheme in the embodiment of the invention is clearly and completely described, obviously, described embodiment only is the present invention's part embodiment, rather than whole embodiment.Based on the embodiment among the present invention, those of ordinary skills belong to the scope of protection of the invention not making the every other embodiment that obtains under the creative work prerequisite.

The sorting technique of the disclosed a kind of high-spectrum remote sensing of the embodiment of the invention as shown in Figure 1, comprising:

S101: the cluster centre set of calculating the predetermined number of high-spectrum remote sensing;

Usually, use clustering algorithm to calculate the cluster centre set of high-spectrum remote sensing, repeatedly use clustering algorithm, can access the set of many group cluster centres, according to default quantity, repeatedly use clustering algorithm can obtain the cluster centre set of predetermined number.

S102: the spatial neighborhood information of pixel in the described high-spectrum remote sensing as constraint condition, according to described cluster centre set, is determined the optimum cluster centre of described high-spectrum remote sensing by artificial ant colony algorithm;

Spatial neighborhood information refers to the relation of pixel on the locus in the high-spectrum remote sensing, and the pixel on a non-border has eight neighborhood pixels usually, in geography, it has been generally acknowledged that the pixel that spatially more closes on, and the possibility that belongs to same classification is more high.

S103: according to the optimum cluster centre of described high-spectrum remote sensing, the target in the described high-spectrum remote sensing is classified.

In the prior art, difference according to spectrum is classified to high-spectrum remote sensing, as long as the big target of spectrum difference, all be divided into different classes ofly, and the geographic range that remote sensing images comprise is wide, target is numerous, the user wishes the upper information in normally some zones of obtaining by remote sensing images, but not the next information, this also is to use the original intention of remote sensing images observation, therefore, only be that The classification basis can make that undoubtedly remote sensing images are excessively classified with spectrum, and can not satisfy user's demand.

The described sorting technique of the embodiment of the invention, in the process that cluster centre is optimized, with the spatial neighborhood information of pixel in the high-spectrum remote sensing as constraint condition, in the process that different target is classified, follow spectrum and spatial neighborhood information, the target of different spectrum is divided into the possibility height of different classifications, and the target that approaches on the space is divided into the possibility height of same classification, therefore, can revise being divided into different classes of target by the usage space neighborhood information, obtain optimum cluster centre, so, carry out sorting result according to optimum cluster centre and can avoid the excessively problem of classification, thereby satisfy the user to the demand of the granularity of remote sensing image classification.

The sorting technique of another high-spectrum remote sensing disclosed by the invention as shown in Figure 2, comprising:

S201: use the K mean algorithm to calculate the N of high-spectrum remote sensing _eIndividual cluster centre set;

Clustering algorithm is a kind of sorting technique of non-supervision, and the K mean algorithm is a kind of clustering algorithm of classics, and the principle of using the K mean algorithm to classify to high-spectrum remote sensing is:

To comprise n(n is positive integer) L(L of individual pixel is positive integer) wave band high-spectral data collection $R=\{r_i=(r_{i1},r_{i2},\·\·\·,r_{\mathrm{iL}}{)}^T{\}}_{i=1}^n$ Be divided into m subclass

Satisfy

If for pixel r _iWith subclass R _j, definition

ω _i=j if r _i∈ R _j(1) then set

A classification results can representing high-spectral data collection X, ω _iBe called pixel r _iThe classification mark, wherein, r _IjRepresent the reflectivity (or spoke brightness) on j the wave band of i pixel.

If with c _j=(c _J1, c _J2..., c _JL) ^TRepresent the cluster centre of j class and be criterion to cluster centre apart from the sum minimum with pixel that then the clustering problem of high spectrum image can be described as following optimization problem:

$\min f\left(C\right)=\underset{i=1}{\overset{n}{\mathrm{\Σ}}}d(r_i,c_{{\mathrm{\ω}}_i})---\left(2\right)$

s.t.ω _i=arg min _j{d(r _i,c _j)}

Wherein,

M * L dimensional vector that expression joins end to end m cluster centre to obtain.D () is distance function, and remotely-sensed data adopts the Euclidean distance between the pixel spectrum usually, namely

d _ij=d(r _i,c _j)=||r _i-c _j|| ₂ （4）

For optimization problem (2), in each iterative process, can calculate each pixel to distances of clustering centers d according to formula (4) _Ij, the principle minimum according to distance is attributed to certain cluster centre with each pixel, obtains cluster result, calculates new cluster centre according to cluster result, and algorithm does not stop when cluster centre does not change before and after iteration, namely finishes classification.The method of upgrading cluster centre is to calculate the average of all kinds of pixels, namely

$c_j=\frac{1}{n_j}\underset{i=1}{\overset{n_j}{\mathrm{\Σ}}}{r}_i,{\mathrm{\ω}}_i=j---\left(5\right)$

N wherein _jThe quantity of pixel among the expression classification j.

Use the k-means algorithm idiographic flow can be shown in following false code:

=====k-means algorithm false code=====

1: input high-spectrum remote sensing R and categorical measure m;

2: select m cluster centre at random

3:REPEAT

4: calculate d according to formula (4) _Ij, obtain classification results W and target function value f according to formula (2)

Each classification j{ of 5:FOR

Upgrade cluster centre according to formula (5) }

The 6:UNTIL cluster centre no longer changes;

7: output category is W, cluster centre set as a result

With target function value f.

By above-mentioned false code as can be seen, high-spectrum remote sensing being carried out cluster can obtain high-spectrum remote sensing and carry out sorting result, cluster centre set and objective function according to spectrum.In the present embodiment, N _eInferior use K mean algorithm can obtain N _eIndividual cluster centre set wherein comprises m cluster centre in each cluster centre set.

S202: with N _eIndividual cluster centre set is as the food source of gathering honey honeybee in the artificial ant colony algorithm;

(Artificial Bee Colony Algorithm ABC) realizes finding the solution of optimization problem by the foraging behavior of bee colony in simulating nature circle to artificial ant colony algorithm.For optimization problem to be found the solution, the search volume of the corresponding bee colony in feasible solution space, a feasible solution is called a food source (food scource), the quantity that comprises nectar in the food source is called fitness (fitness), relevant with the target function value that the corresponding feasible solution of this food source produces, feasible solution can produce higher fitness preferably, also will attract more honeybee to come this food source gathering honey.

Whole honeybees in the artificial ant colony algorithm are divided into three classes: gathering honey honeybee (employed bee), follow honeybee (onlooker bee) and search bee (scout bee).Three class honeybees respectively according to strategy separately search for, judgement and type conversion.

In the present embodiment, the quantity of gathering honey honeybee is consistent with the quantity of food source, is N _eIndividual, the quantity of following honeybee can preestablish.

S203: described gathering honey honeybee is determined the food source of the fitness maximum of described gathering honey honeybee correspondence according to the spatial neighborhood information of pixel in described food source and the described high-spectrum remote sensing;

In artificial ant colony algorithm, the corresponding food source (and fitness) of each gathering honey honeybee, the gathering honey honeybee can carry out Local Search and find new food source in the neighborhood of this food source, if the fitness of new food source is better than former food source, then its corresponding relation is updated to new food source (and fitness), otherwise abandons the new food source and continue in the neighborhood of former food source, searching for.

In the present embodiment, particularly, if

Represent i food source, the position of i gathering honey honeybee just, then the neighborhood Local Search can be expressed as

x _ij′=x _ij+φ(x _ij-x _kj) （6）

Wherein, k is another food source that is different from i of selecting at random, j be from 1,2 ..., an integer of selecting at random among m * L}, φ ∈ (0,1) is random number.

Food source x _iFitness be

${\mathrm{fit}}_i=\frac{1}{1+f\left(x_i\right)}---\left(7\right)$

In the present embodiment, f (x) is the objective function of determining according to the spatial neighborhood information of pixel in the described high-spectrum remote sensing.

S204: follow honeybee according to the spatial neighborhood information of pixel in described food source and the described high-spectrum remote sensing, determine described food source of following the fitness maximum of honeybee correspondence;

Each follows the fitness of all food sources that honeybee can obtain according to the gathering honey honeybee, select a food source (being equal to the gathering honey honeybee) and in the neighborhood of this food source, carry out Local Search according to certain probability, if the fitness in the new food source of finding is better than former food source, then that gathering honey honeybee (rather than following honeybee) is corresponding food source is updated to new food source, otherwise abandons the new food source.

Follow " following probability " p that honeybee selects j food source _jFor

$p_j=\frac{{\mathrm{fit}}_j}{{\mathrm{\Σ}}_{i=1}^S\mathrm{fi}{t}_i}---\left(8\right)$

Wherein, fit _iThe fitness of representing i food source, S are represented the total quantity of food source, just the total quantity of gathering honey honeybee.If the objective function of optimization problem is non-negative, then food source x _iFitness

${\mathrm{fit}}_i=\frac{1}{1+f\left(x_i\right)}---\left(9\right)$

In the present embodiment, f (x) is the objective function of determining according to the spatial neighborhood information of pixel in the described high-spectrum remote sensing.

S205: the food source of fitness maximum in the food source of the food source of the fitness maximum of described gathering honey honeybee correspondence and described fitness maximum of following the honeybee correspondence is defined as optimum cluster centre set;

S206: when described gathering honey honeybee is satisfied default second condition, described gathering honey honeybee is converted to search bee;

Default second condition is that gathering honey honeybee continuous T in its corresponding food source neighborhood is searched for less than the solution more excellent than the current foodstuff source for 3 times, namely can abandon this food source and be converted to search bee less than than the more excellent solution in current foodstuff source the time when the continuous T 3 times search in its corresponding food source neighborhood of gathering honey honeybee.

S207: search bee is determined food source at random from described high-spectrum remote sensing;

S208: judge whether to satisfy default first condition, if, carry out S209, if not, carry out S203;

In the present embodiment, the iterations T1 that default first condition can equal to preset for S203 to S207 number of iterations, perhaps, optimum cluster centre set continuous T does not change for 2 times, and T2 is default convergence threshold.

S209: according to the optimum cluster centre set of described high-spectrum remote sensing, the target in the described high-spectrum remote sensing is classified.Usually, clustering algorithm converges to locally optimal solution easily and can't jump out this locally optimal solution, the sorting technique of the described high spectrum image of present embodiment, set is optimized end user worker bee group algorithm to cluster centre, because the characteristic of artificial ant colony algorithm globally optimal solution, can make the cluster centre set jump out locally optimal solution, again because when end user worker bee group algorithm, serve as the fitness that food source is determined in constraint with spatial neighborhood information, so, can avoid the problem that remote sensing images are excessively classified.

The sorting technique of disclosed another high-spectrum remote sensing of the embodiment of the invention as shown in Figure 3, comprising:

S301: the N that calculates high-spectrum remote sensing _eIndividual cluster centre set;

S302: with N _eIndividual cluster centre set is as the food source of gathering honey honeybee in the artificial ant colony algorithm;

S303: the fitness that calculates described food source;

S304: gathering honey honeybee that will be corresponding with described food source is gathered as the cluster centre that upgrades according to the new food source that described food source searches;

S305: according to new cluster centre set, determine the cluster result of the target in the described high-spectrum remote sensing;

In the present embodiment, can determine cluster result according to the K mean algorithm.

S306: (Iteration Conditional Model ICM), calculates the objective function of the cluster centre set of described renewal according to described cluster result by local iteration's condition model;

Usually, Markov random field (Markov Random Field, MRF) model is a kind of effective integration spectral information that is widely used and the sorter of spatial information, mainly classify by the posterior probability that spatial coherence between pixel is added spectral signature, can effectively describe the local spatial feature of pixel and its neighborhood.Discriminant classification function based on Markov random field can simply be expressed as:

$U_{\mathrm{ij}}={\left|\right|r_i-c_j\left|\right|}_2+\mathrm{\β}\underset{\∂i}{\mathrm{\Σ}}(1-\mathrm{\δ}(j,{\mathrm{\ω}}_{\∂i}))---\left(10\right)$

Wherein,

Expression r _iPixel in the neighborhood,

Expression r _iClassification in the neighborhood under certain pixel; δ () is the Kronecker function.

According to above-mentioned discriminant classification function, high spectrum image unsupervised classification problem can be converted into following optimization problem:

$\begin{array}{cc}\min & U=\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{U}_{{\mathrm{i\ω}}_i}\end{array}$

s.t.ω _i=arg min _j{U _ij} （11）

In the following formula, U is objective function.

In this enforcement, the false code of objective function of calculating the cluster centre set of described renewal according to described cluster result is:

=======ICM algorithm false code======

1: input high spectrum image R and cluster result W;

2: calculate cluster centre according to formula (5)

3: calculate d according to formula (5) _Ij

4:REPEAT

Each pixel x of 5:FOR _i{

Each classification j{ of 6:FOR

Calculate U according to formula (10) _Ij

7: upgrade classification results W and target function value U according to formula (11);

The 8:UNTIL classification results no longer changes;

9: output category is W and target function value U as a result.

ICM is the approximate data commonly used of finding the solution MRF, and its cardinal rule is iteration in the solution of closing on, and objective function is progressively optimized, till stable.This model iteration finishes, and the classification mark of all pixels can make objective function reach local minimum on the image.ICM algorithm flexible and convenient, and can provide approximate solution preferably.

" neighborhood of pixel " is included in 4 pixels that common edge is arranged with this pixel in the image in the present embodiment.

S307: according to described objective function, calculate the fitness of the cluster centre set of described renewal;

In the present embodiment, the calculating of fitness can be carried out according to following formula (7), is about to the objective function f (x) in the above-mentioned objective function U substitution formula (7) that obtains.

S308: the fitness of the fitness by more described food source and the set of the cluster centre of described renewal, with the food source of fitness the greater as the fitness maximum of described gathering honey honeybee correspondence.

S309: follow honeybee according to default probability, from the food source of described gathering honey honeybee, select one as food source;

S310: follow honeybee according to the spatial neighborhood information of pixel in described food source and the described high-spectrum remote sensing, determine described food source of following the fitness maximum of honeybee correspondence;

After following the selected food source of honeybee, according to the step identical with the gathering honey honeybee, determine the food source of corresponding fitness maximum, concrete steps are as follows:

Calculate the fitness of described food source;

The honeybee of following that will be corresponding with described food source is gathered as the cluster centre that upgrades according to the new food source that described food source searches;

According to the cluster centre set of described renewal, determine the cluster result of the target in the described high-spectrum remote sensing;

By local iteration's condition model, calculate the objective function of the cluster centre set of described renewal according to described cluster result;

According to described objective function, calculate the fitness of the cluster centre of described renewal;

The fitness of the fitness by more described food source and the set of the cluster centre of described renewal, with fitness the greater as described food source of following the fitness maximum of honeybee correspondence.

S311: the food source of fitness maximum in the food source of the food source of the fitness maximum of described gathering honey honeybee correspondence and described fitness maximum of following the honeybee correspondence is defined as optimum cluster centre set;

S312: search bee is determined food source at random from described high-spectrum remote sensing;

S313: judge whether to satisfy default first condition, if, carry out S314, if not, return S302;

In the present embodiment, the iterations T1 that default first condition can equal to preset for S203 to S207 number of iterations, perhaps, optimum cluster centre set continuous T does not change for 2 times, and T2 is default convergence threshold.

S314: by local iteration's condition model, described cluster result is carried out subseries again;

S315: the cluster centre of sorting result and described renewal is gathered corresponding stored again;

S316: obtain the classification results corresponding with described food source;

S317: with the classification results of the food source correspondence of the described fitness maximum classification results as the target in the described high-spectrum remote sensing.

The described high-spectrum remote sensing sorting technique of present embodiment, the ICM model is introduced in the artificial ant colony algorithm, be used for determining objective function, be further used for determining optimum cluster centre set, and use optimum cluster centre set that high-spectrum remote sensing is classified, therefore, can not cause the problem of excessive classification, for example, for the well lid on the road in the high-spectrum remote sensing, because itself and road spatially close on, therefore, according to spatial neighborhood information, itself and road can be divided into different classifications.

With said method embodiment accordingly, the embodiment of the invention also discloses a kind of sorter of high-spectrum remote sensing, as shown in Figure 4, comprising:

Cluster centre set computing module 401 is used for the cluster centre set of the predetermined number of calculating high-spectrum remote sensing;

Optimum cluster centre set determination module 402, be used for spatial neighborhood information with described high-spectrum remote sensing pixel as constraint condition, according to the cluster centre set of described predetermined number, determine the optimum cluster centre set of described high-spectrum remote sensing by artificial ant colony algorithm;

Sort module 403 is used for the optimum cluster centre set according to described high-spectrum remote sensing, and the target in the described high-spectrum remote sensing is classified.

Further, described optimum cluster centre set determination module can specifically comprise:

Food source is chosen the unit, for the food source of the cluster centre of described predetermined number being gathered as artificial ant colony algorithm gathering honey honeybee;

Iteration unit is used for carrying out successively following steps, until satisfying default first condition: described gathering honey honeybee according to described food source and described high-spectrum remote sensing in the spatial neighborhood information of pixel, determine the food source of the fitness maximum of described gathering honey honeybee correspondence; Follow honeybee according to the spatial neighborhood information of pixel in described food source and the described high-spectrum remote sensing, determine described food source of following the fitness maximum of honeybee correspondence; The food source of fitness maximum in the food source of the food source of the fitness maximum of described gathering honey honeybee correspondence and described fitness maximum of following the honeybee correspondence is defined as optimum cluster centre set; Search bee is determined food source at random from described high-spectrum remote sensing.

Wherein, iteration unit can comprise:

Gathering honey honeybee search subelement is for the fitness that calculates described food source; Gathering honey honeybee that will be corresponding with described food source is gathered as the cluster centre that upgrades according to the new food source that described food source searches; According to the cluster centre set of described renewal, determine the cluster result of the target in the described high-spectrum remote sensing; By local iteration's condition model, calculate the objective function of the cluster centre set of described renewal according to described cluster result; According to described objective function, calculate the fitness of the cluster centre set of described renewal; The fitness of the fitness by more described food source and the set of the cluster centre of described renewal is with the food source of fitness the greater as the fitness maximum of described gathering honey honeybee correspondence;

And, follow honeybee search subelement, be used for following honeybee according to default probability, from the food source of described gathering honey honeybee, select one as food source; Calculate the fitness of described food source; The honeybee of following that will be corresponding with described food source is gathered as the cluster centre that upgrades according to the new food source that described food source searches; According to the cluster centre set of described renewal, determine the cluster result of the target in the described high-spectrum remote sensing; By local iteration's condition model, calculate the objective function of the cluster centre set of described renewal according to described cluster result; According to described objective function, calculate the fitness of the cluster centre set of described renewal; The fitness of the fitness by more described food source and the set of the cluster centre of described renewal, with fitness the greater as described food source of following the fitness maximum of honeybee correspondence.

Further, described sort module can comprise:

Taxon is used for by local iteration's condition model described cluster result being carried out subseries again;

Storage unit, the cluster centre that is used for again sorting result and described renewal is gathered corresponding stored;

Acquiring unit is used for obtaining the classification results corresponding with described food source;

Determining unit is used for the classification results of the food source correspondence of the described fitness maximum classification results as the target of described high-spectrum remote sensing.

The assorting process of the described sorter of present embodiment is identical with said method embodiment, here repeat no more, thereby, can revise being divided into different classes of target by the usage space neighborhood information, obtain optimum cluster centre, so, carry out sorting result according to optimum cluster centre and can avoid the excessively problem of classification, thereby satisfy the user to the demand of the granularity of remote sensing image classification.

If the described function of present embodiment method realizes with the form of SFU software functional unit and during as independently production marketing or use, can be stored in the computing equipment read/write memory medium.Based on such understanding, the part that the embodiment of the invention contributes to prior art or the part of this technical scheme can embody with the form of software product, this software product is stored in the storage medium, comprise that some instructions are with so that a computing equipment (can be personal computer, server, mobile computing device or the network equipment etc.) carry out all or part of step of the described method of each embodiment of the present invention.And aforesaid storage medium comprises: various media that can be program code stored such as USB flash disk, portable hard drive, ROM (read-only memory) (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), magnetic disc or CD.

Each embodiment adopts the mode of going forward one by one to describe in this instructions, and what each embodiment stressed is and the difference of other embodiment that same or similar part is mutually referring to getting final product between each embodiment.

To the above-mentioned explanation of the disclosed embodiments, make this area professional and technical personnel can realize or use the present invention.Multiple modification to these embodiment will be apparent concerning those skilled in the art, and defined General Principle can realize under the situation that does not break away from the spirit or scope of the present invention in other embodiments herein.Therefore, the present invention will can not be restricted to these embodiment shown in this article, but will meet the wideest scope consistent with principle disclosed herein and features of novelty.

Claims (12)

1. the sorting technique of a high-spectrum remote sensing is characterized in that, comprising:

Calculate the cluster centre set of the predetermined number of high-spectrum remote sensing;

The spatial neighborhood information of pixel in the described high-spectrum remote sensing as constraint condition, according to the cluster centre set of described predetermined number, is determined the optimum cluster centre set of described high-spectrum remote sensing by artificial ant colony algorithm;

According to the optimum cluster centre set of described high-spectrum remote sensing, the target in the described high-spectrum remote sensing is classified.

2. method according to claim 1, it is characterized in that, described spatial neighborhood information with the pixel in the described high-spectrum remote sensing is as constraint condition, according to the cluster centre set of described predetermined number, determine that by artificial ant colony algorithm the optimum cluster centre of described high-spectrum remote sensing comprises:

With the food source of the cluster centre of described predetermined number set as gathering honey honeybee in the artificial ant colony algorithm;

Carry out following steps successively, until satisfying default first condition:

Described gathering honey honeybee is determined the food source of the fitness maximum of described gathering honey honeybee correspondence according to the spatial neighborhood information of pixel in described food source and the described high-spectrum remote sensing;

Follow honeybee according to the spatial neighborhood information of pixel in described food source and the described high-spectrum remote sensing, determine described food source of following the fitness maximum of honeybee correspondence;

The food source of fitness maximum in the food source of the food source of the fitness maximum of described gathering honey honeybee correspondence and described fitness maximum of following the honeybee correspondence is defined as optimum cluster centre set;

Search bee is determined food source at random from described high-spectrum remote sensing.

3. method according to claim 2 is characterized in that, described gathering honey honeybee determines that according to the spatial neighborhood information of pixel in described food source and the described high-spectrum remote sensing food source of the fitness maximum of described gathering honey honeybee correspondence comprises:

Calculate the fitness of described food source;

Gathering honey honeybee that will be corresponding with described food source is gathered as the cluster centre that upgrades according to the new food source that described food source searches;

According to the cluster centre set of described renewal, determine the cluster result of the target in the described high-spectrum remote sensing;

By local iteration's condition model, calculate the objective function of the cluster centre set of described renewal according to described cluster result;

According to described objective function, calculate the fitness of the cluster centre set of described renewal;

The fitness of the fitness by more described food source and the set of the cluster centre of described renewal is with the food source of fitness the greater as the fitness maximum of described gathering honey honeybee correspondence.

4. method according to claim 2 is characterized in that, described spatial neighborhood information of following pixel in the honeybee described food source of foundation and the described high-spectrum remote sensing determines that described food source of following the fitness maximum of honeybee correspondence comprises:

Follow honeybee according to default probability, from the food source of described gathering honey honeybee, select one as food source;

Calculate the fitness of described food source;

The honeybee of following that will be corresponding with described food source is gathered as the cluster centre that upgrades according to the new food source that described food source searches;

According to the cluster centre set of described renewal, determine the cluster result of the target in the described high-spectrum remote sensing;

By local iteration's condition model, calculate the objective function of the cluster centre set of described renewal according to described cluster result;

According to described objective function, calculate the fitness of the cluster centre set of described renewal;

The fitness of the fitness by more described food source and the set of the cluster centre of described renewal, with fitness the greater as described food source of following the fitness maximum of honeybee correspondence.

5. according to claim 2,3 or 4 described methods, it is characterized in that, described search bee is determined food source at random from described high-spectrum remote sensing before, also comprise:

When described gathering honey honeybee is satisfied default second condition, described gathering honey honeybee is converted to search bee.

6. according to claim 3 or 4 described methods, it is characterized in that gather according to the optimum cluster centre of described high-spectrum remote sensing, the target in the described high-spectrum remote sensing is classified to be comprised:

By local iteration's condition model, described cluster result is carried out subseries again;

Cluster centre set corresponding stored with sorting result and described renewal again;

Obtain the classification results corresponding with described food source;

With the classification results of the food source correspondence of the described fitness maximum classification results as the target in the described high-spectrum remote sensing.

7. method according to claim 1 is characterized in that, the cluster centre set of described calculating high-spectrum remote sensing comprises:

Use the K mean algorithm to calculate the cluster centre set of the predetermined number of high-spectrum remote sensing.

8. the sorter of a high-spectrum remote sensing is characterized in that, comprising:

Cluster centre set computing module is used for the cluster centre set of the predetermined number of calculating high-spectrum remote sensing;

Optimum cluster centre set determination module, be used for spatial neighborhood information with described high-spectrum remote sensing pixel as constraint condition, according to the cluster centre set of described predetermined number, determine the optimum cluster centre set of described high-spectrum remote sensing by artificial ant colony algorithm;

Sort module is used for the optimum cluster centre set according to described high-spectrum remote sensing, and the target in the described high-spectrum remote sensing is classified.

9. device according to claim 8 is characterized in that, described optimum cluster centre set determination module comprises:

Food source is chosen the unit, for the food source of the cluster centre of described predetermined number being gathered as artificial ant colony algorithm gathering honey honeybee;

Iteration unit is used for carrying out successively following steps, until satisfying default first condition: described gathering honey honeybee according to described food source and described high-spectrum remote sensing in the spatial neighborhood information of pixel, determine the food source of the fitness maximum of described gathering honey honeybee correspondence; Follow honeybee according to the spatial neighborhood information of pixel in described food source and the described high-spectrum remote sensing, determine described food source of following the fitness maximum of honeybee correspondence; The food source of fitness maximum in the food source of the food source of the fitness maximum of described gathering honey honeybee correspondence and described fitness maximum of following the honeybee correspondence is defined as optimum cluster centre set; Search bee is determined food source at random from described high-spectrum remote sensing.

10. device according to claim 9 is characterized in that, described iteration unit comprises:

Gathering honey honeybee search subelement is for the fitness that calculates described food source; Gathering honey honeybee that will be corresponding with described food source is gathered as the cluster centre that upgrades according to the new food source that described food source searches; According to the cluster centre set of described renewal, determine the cluster result of the target in the described high-spectrum remote sensing; By local iteration's condition model, calculate the objective function of the cluster centre set of described renewal according to described cluster result; According to described objective function, calculate the fitness of the cluster centre set of described renewal; The fitness of the fitness by more described food source and the set of the cluster centre of described renewal is with the food source of fitness the greater as the fitness maximum of described gathering honey honeybee correspondence.

11. device according to claim 9 is characterized in that, described iteration module comprises:

Follow honeybee search subelement, be used for following honeybee according to default probability, from the food source of described gathering honey honeybee, select one as food source; Calculate the fitness of described food source; The honeybee of following that will be corresponding with described food source is gathered as the cluster centre that upgrades according to the new food source that described food source searches; According to the cluster centre set of described renewal, determine the cluster result of the target in the described high-spectrum remote sensing; By local iteration's condition model, calculate the objective function of the cluster centre set of described renewal according to described cluster result; According to described objective function, calculate the fitness of the cluster centre set of described renewal; The fitness of the fitness by more described food source and the set of the cluster centre of described renewal, with fitness the greater as described food source of following the fitness maximum of honeybee correspondence.

12. device according to claim 8 is characterized in that, described sort module comprises:

Taxon is used for by local iteration's condition model described cluster result being carried out subseries again;

Storage unit, the cluster centre that is used for again sorting result and described renewal is gathered corresponding stored;

Acquiring unit is used for obtaining the classification results corresponding with described food source;

Determining unit is used for the classification results of the food source correspondence of the described fitness maximum classification results as the target of described high-spectrum remote sensing.

Images