CN105976357A - Hyper-spectral image end member extraction method and hyper-spectral image end member extraction system - Google Patents

Abstract

The invention discloses a hyper-spectral image end member extraction method and a hyper-spectral image end member extraction system. Optimized variables in the problem of hyper-spectral image end member extraction correspond to nectar source locations in an artificial bee colony algorithm, and optimization of each nectar source is determined by a fitness function. Nectar collecting bees search for an optimal nectar source, and sends nectar source information to follower bees after search; the follower bees search for an optimal nectar source the second time, and use a selected optimal nectar source to update a previously obtained optimal solution; and when there is a nectar collecting bee not updating the nectar source in a preset period of time, the nectar collecting bee is transformed into an investigation bee, and the investigation bee compares the fitness value of a randomly selected nectar source with the fitting value of a nectar source used as an optimal solution and calibrates the nectar source used as an optimal solution. According to the invention, the end member extraction problem is converted into a combination optimization problem solving process based on the artificial bee colony algorithm, the advantages of the artificial bee colony algorithm are given full play, and the nectar source used as an optimal solution is calibrated by the investigation bee to reduce the risk that the optimization process is trapped into a local optimal solution.

Description

A kind of high optical spectrum image end member extraction method and system

Technical field

The present invention relates to hyper-spectral image technique field, in particular, relate to a kind of high optical spectrum image end member extraction side Method and system.

Background technology

High-spectrum seems a kind of spectral resolution spectrum picture in the range of 101 orders of magnitude.Due to imaging spectrometer The restriction of spatial resolution so that mixed pixel problem is widely present in high spectrum image.Solve the side of mixed pixel problem Method is referred to as Decomposition of Mixed Pixels, its objective is to analyze in mixed pixel comprise which material (referred to as end member) and they shared by Ratio (referred to as abundance).

High optical spectrum image end member extraction method is mainly based upon the high optical spectrum image end member extraction method of ant group algorithm at present (ACOEE), high optical spectrum image end member extraction method based on discrete particle (DPSOEE) and its deriving method, these methods are all Be the root-mean-square error (RootMean Square Error, RMSE) with original image and back mixing image as object function, logical Cross optimization objective function to seek optimal solution.

Although, current high optical spectrum image end member extraction method can obtain more satisfied result under certain condition, But they there is also some problems.Due to the usual more than one of the purest pixel of same class atural object in high spectrum image, a lot The spectral reflectivity closing on pixel is identical, and this results in the phenomenon repeating to identify end member of the same race and frequently occurs, the identification knot of end member Fruit differs greatly with the metadata using Hysime algorithm (a kind of conventional end member quantity survey algorithm) to calculate, and makes Gao Guang The result that spectrum image end member extracts is easily trapped into locally optimal solution.

Summary of the invention

In view of this, the open a kind of high optical spectrum image end member extraction method of the present invention and system, to solve high spectrum image The problem that the result of Endmember extraction is easily trapped into locally optimal solution.

A kind of high optical spectrum image end member extraction method, including:

Obtain the candidate's end member in high spectrum image, and determine fitness function；

Initiation parameter, including, adopt bee numbers m, follow honeybee quantity m, maximum iteration time maxIter；

Randomly generating m feasible solution in solution space, each described feasible solution is corresponding as a gathering honey honeybee Nectar source；

Gathering honey honeybee randomly chooses a nectar source as current first nectar source, according to searching function in described current first nectar source Neighborhood in search the first new nectar source, utilize the fitness value in described fitness function calculating the described first new nectar source, and select Described current first nectar source of replacement that in described current first nectar source and the described first new nectar source, fitness value is big；

Gathering honey honeybee shares nectar source information to following honeybee；

Follow honeybee according to following probability selection gathering honey honeybee, using current first nectar source up-to-date for this gathering honey honeybee as current second Nectar source, searches the second new nectar source according to described search function secondary in the neighborhood in described current second nectar source, utilizes described suitable Response function calculates the fitness value in the described second new nectar source, and selects in described current second nectar source and the described second new nectar source Described current second nectar source is replaced in the nectar source that fitness value is big；

Utilize and follow the optimal solution obtained before the nectar source renewal that honeybee is chosen；

Whether judge in m gathering honey honeybee in preset time period either with or without the gathering honey honeybee updating nectar source；

If there is not updating the gathering honey honeybee in nectar source, then will not update the gathering honey honeybee in nectar source in described preset time period Be converted to investigate honeybee；

Investigation honeybee randomly selects a nectar source as current 3rd nectar source in search volume, exists according to described search function Search the 3rd new nectar source in the neighborhood in described current 3rd nectar source, utilize described fitness function to calculate the described 3rd new nectar source Fitness value；

Judge whether the fitness value in the described 3rd new nectar source is less than the fitness value in described 3rd nectar source；

If less than the fitness value in described 3rd nectar source, then will follow nectar source that honeybee chooses as current optimal solution, and Judge whether current iteration number of times reaches described maximum iteration time maxIter；

If not less than the fitness value in described 3rd nectar source, then this investigation honeybee being converted to gathering honey honeybee, continue to search for new Nectar source；

If there is no not updating the gathering honey honeybee in nectar source, then judge whether current iteration number of times reaches described greatest iteration Number of times maxIter；

If reaching described maximum iteration time maxIter, then output Endmember extraction result；

Without reaching described maximum iteration time maxIter, then gathering honey honeybee continues to search for new nectar source.

Preferably, the expression formula of described fitness function is as follows:

$fit\left(X_i\right)=\frac{1}{f\left(X_i\right)};$

In formula, fit (X_i) represent i-th nectar source fitness function, X_iRepresent i-th nectar source, X_i= (x_i1x_i2......x_iM) ', M is the end member quantity in described high spectrum image, f (X_i) it is the object function of optimal problem.

Preferably, the expression formula of the object function of described optimal problem is as follows:

$f\left(X_i\right)=RMSE({\left\{x_i\right\}}_{i=1}^N,C^{-1}(X_i\left)\right)+u\×g\left(d_i\right);$

In formula, u represents that penalty coefficient, peer-to-peer two ends play the effect of adjustment；

g(d_i) represent distance terms,IfRepresent a set, d_iFor end member in image and end member Between Euclidean distance,e_pRepresent pth end member, e_qRepresent q-th End member；

C^-1ForInverse function, x_iRepresent pixel,The limit of pixel Condition processed ise_kRepresent end member, p_ikFor the ratio that end member is shared in pixel, ε_iFor error, M For the end member quantity in described high spectrum image, N is the quantity of pixel in described high spectrum image, i=1 ... .., N, C_{M, N}For Search volume；

RMSE represents root-mean-square error,L represents described Gao Guang The wave band number of spectrogram picture, x_iRepresent the pixel in described high spectrum image,Represent in the back mixing image of described high spectrum image Pixel.

Preferably, the expression formula following probability described in is as follows:

$p_j=\frac{fit\left(X_j\right)}{{\mathrm{\Σ}}_{i=1}^{Num}fit\left(X_i\right)};$

In formula, p_jRepresent that following honeybee selects the probability of following in jth nectar source, fit (X_i) represent i-th nectar source fitness Function, fit (X_j) represent jth nectar source fitness function, Num represents nectar source quantity.

Preferably, the expression formula of described search function is as follows:

φ_ij=x_ij+Δ(x_ij-x_sj)(4)；

In formula, φ_ijRepresent and search function, x_ijRepresent i-th nectar source jth gathering honey honeybee/investigation honeybee position, x_sjTable Showing the s nectar source jth gathering honey honeybee/investigation honeybee position, Δ is the random number between [-1,1], s ≠ i and s ∈ 1, 2 ... M}, j ∈ 1,2 ... M}, M are the end member quantity in described high spectrum image.

A kind of high optical spectrum image end member extraction system, including:

Acquiring unit, for obtaining the candidate's end member in high spectrum image, and determines fitness function；

Initialization unit, for initiation parameter, including, adopt bee numbers m, follow honeybee quantity m, maximum iteration time maxIter；

Feasible solution generation unit, for randomly generating m feasible solution in solution space, each described feasible solution is made It it is the nectar source that a gathering honey honeybee is corresponding；

First searches unit, randomly chooses a nectar source as current first nectar source for gathering honey honeybee, according to searching function In the neighborhood in described current first nectar source, search the first new nectar source, utilize described fitness function to calculate the described first new nectar source Fitness value, and select the replacement described current that in described current first nectar source and the described first new nectar source, fitness value is big One nectar source；

Share unit, share nectar source information to following honeybee for gathering honey honeybee；

Second searches unit, is used for following honeybee according to following probability selection gathering honey honeybee, by up-to-date for this gathering honey honeybee current the One nectar source, as current second nectar source, searches second according to described search function secondary in the neighborhood in described current second nectar source New nectar source, utilizes described fitness function to calculate the fitness value in the described second new nectar source, and selects described current second nectar source Described current second nectar source is replaced in the nectar source big with fitness value in the described second new nectar source；

Updating block, follows, for utilizing, the optimal solution obtained before the nectar source renewal that honeybee is chosen；

First judging unit, for judging in m gathering honey honeybee in preset time period whether adopting either with or without renewal nectar source Apis, if it is, perform the first converting unit, otherwise, performs the 4th judging unit；

Described first converting unit, for being converted to investigation by the gathering honey honeybee not updating nectar source in described preset time period Honeybee；

Computing unit, is used for investigating honeybee and randomly selects a nectar source in search volume as current 3rd nectar source, according to Described search function searches the 3rd new nectar source in the neighborhood in described current 3rd nectar source, utilizes described fitness function to calculate institute State the fitness value in the 3rd new nectar source；

Second judging unit, for judging whether the fitness value in the described 3rd new nectar source is less than the suitable of described 3rd nectar source Answer angle value, if it is, perform the 3rd judging unit, if it is not, then perform the second converting unit；

Described 3rd judging unit, for following nectar source that honeybee chooses as current optimal solution, and performs the described 4th Judging unit；

Described second converting unit, in the case of described second judging unit is judged as NO, then by this investigation honeybee Be converted to gathering honey honeybee, and return the described first search unit of execution, continue to search for new nectar source；

Described 4th judging unit, is used for judging whether current iteration number of times reaches described maximum iteration time maxIter, If it is, execution output unit, otherwise, return and perform described first search unit, continue to search for new nectar source；

Described output unit, is used for exporting Endmember extraction result.

Preferably, the expression formula of described fitness function is as follows:

$fit\left(X_i\right)=\frac{1}{f\left(X_i\right)};$

In formula, fit (X_i) represent i-th nectar source fitness function, X_iRepresent i-th nectar source, X_i= (x_i1x_i2......x_iM) ', M is the end member quantity in described high spectrum image, f (X_i) it is the object function of optimal problem.

Preferably, the expression formula of the object function of described optimal problem is as follows:

$f\left(X_i\right)=RMSE({\left\{x_i\right\}}_{i=1}^N,C^{-1}(X_i\left)\right)+u\×g\left(d_i\right);$

In formula, u represents that penalty coefficient, peer-to-peer two ends play the effect of adjustment；

g(d_i) represent distance terms,d_iFor Euclidean distance between end member and end member in image, ifRepresent a set,e_pRepresent pth end member, e_qQ end Unit；

C^-1ForInverse function, x_iRepresent pixel,The limit of pixel Condition processed ise_kRepresent end member, p_ikFor the ratio that end member is shared in pixel, ε_iFor error, M For the end member quantity in described high spectrum image, N is the quantity of pixel in described high spectrum image, i=1 ... .., N, C_{M, N}For Search volume；

RMSE represents root-mean-square error,L represents described Gao Guang The wave band number of spectrogram picture, x_iRepresent the pixel in described high spectrum image,Represent in the back mixing image of described high spectrum image Pixel.

Preferably, the expression formula following probability described in is as follows:

$p_j=\frac{fit\left(X_j\right)}{{\mathrm{\Σ}}_{i=1}^{Num}fit\left(X_i\right)};$

In formula, p_jRepresent that following honeybee selects the probability of following in jth nectar source, fit (X_i) represent i-th nectar source fitness Function, fit (X_j) represent jth nectar source fitness function, Num represents nectar source quantity.

Preferably, the expression formula of described search function is as follows:

φ_ij=x_ij+Δ(x_ij-x_sj)；

In formula, φ_ijRepresent and search function, x_ijRepresent i-th nectar source jth gathering honey honeybee/investigation honeybee position, x_sjTable Showing the s nectar source jth gathering honey honeybee/investigation honeybee position, Δ is the random number between [-1,1], s ≠ i and s ∈ 1, 2 ... M}, j ∈ 1,2 ... M}, M are the end member quantity in described high spectrum image.

From above-mentioned technical scheme it can be seen that the invention discloses a kind of high optical spectrum image end member extraction method and be System, by the position, nectar source in the optimized variable correspondence artificial bee colony algorithm in high optical spectrum image end member extraction problem, each nectar source Optimization determined by fitness function, the number of gathering honey honeybee is consistent with the number of feasible solution.Gathering honey honeybee was to corresponding nectar source before this Neighborhood is once searched for, and selects the current nectar source of replacement that in current nectar source and the new nectar source that searches, fitness value is big；Adopt After Apis completes search, nectar source information is followed honeybee, by following honeybee, the neighborhood in the selected current nectar source of gathering honey honeybee is carried out the Binary search, and select the current nectar source of replacement that in current nectar source and the new nectar source that searches, fitness value is big, then utilize with The optimal solution that the nectar source chosen with honeybee obtains before updating, when there is the gathering honey honeybee not updating nectar source in preset time period, This gathering honey honeybee is converted to investigate honeybee, and investigation honeybee is by the fitness value that will randomly select nectar source and the nectar source as optimal solution Fitness value compares, and verifies the nectar source as optimal solution, if randomly selecting the fitness value in nectar source less than making For the fitness value in the nectar source of optimal solution, then showing to follow the nectar source that honeybee finally chooses is optimal solution, then repeats said process, Until iterations reaches maximum iteration time, obtaining Endmember extraction result, otherwise gathering honey honeybee continues to search for new nectar source.Can see Going out, Endmember extraction problem, based on artificial bee colony algorithm, is converted into the solution procedure of combinatorial optimization problem, with this by the present invention Reduce the end member extraction method undue dependence for the quality of data, the advantage giving full play to artificial bee colony algorithm, and pass through to detect Look into honeybee the nectar source as optimal solution is verified so that optimization process is absorbed in the risk of locally optimal solution and is substantially reduced.

Accompanying drawing explanation

In order to be illustrated more clearly that the embodiment of the present invention or technical scheme of the prior art, below will be to embodiment or existing In having technology to describe, the required accompanying drawing used is briefly described, it should be apparent that, the accompanying drawing in describing below is only this Inventive embodiment, for those of ordinary skill in the art, on the premise of not paying creative work, it is also possible to according to Disclosed accompanying drawing obtains other accompanying drawing.

Fig. 1 is the method flow diagram of a kind of high optical spectrum image end member extraction method disclosed in the embodiment of the present invention；

Fig. 2 is the structural representation of a kind of high optical spectrum image end member extraction system disclosed in the embodiment of the present invention.

Detailed description of the invention

Below in conjunction with the accompanying drawing in the embodiment of the present invention, the technical scheme in the embodiment of the present invention is carried out clear, complete Describe, it is clear that described embodiment is only a part of embodiment of the present invention rather than whole embodiments wholely.Based on Embodiment in the present invention, it is every other that those of ordinary skill in the art are obtained under not making creative work premise Embodiment, broadly falls into the scope of protection of the invention.

The embodiment of the invention discloses a kind of high optical spectrum image end member extraction method and system, to solve high spectrum image end The problem that the result that unit extracts is easily trapped into locally optimal solution.

Artificial bee colony algorithm (Artificial Bee Colony, ABC) is by the foraging behavior of bee colony in simulation nature Realize solving of optimization problem.For optimization problem to be solved, the search volume of solution space correspondence bee colony, i.e. one The corresponding nectar source (Nectar Food) of feasible solution.The quantity comprising nectar in nectar source is referred to as fitness (Fitness), adapts to Spend with the feasible solution in nectar source produced by target function value relevant, preferable feasible solution can produce higher fitness value, can inhale Draw more Apis and come this nectar source, then this nectar source is that the probability in optimal nectar source is the most maximum.Apis is divided into by ant colony algorithm Three classes: gathering honey honeybee, follow honeybee, investigation honeybee, three class Apiss scan for according to respective rule respectively and identify.

The present invention is by the position, nectar source in the optimized variable correspondence artificial bee colony algorithm in high optical spectrum image end member extraction problem Putting, the optimization in each nectar source is determined by fitness function, and the number of gathering honey honeybee is consistent with the number of feasible solution.High spectrum image end Unit's extraction process is specific as follows:

See Fig. 1, the method flow diagram of a kind of high optical spectrum image end member extraction method disclosed in the embodiment of the present invention, including Step:

Step S11, the candidate's end member obtained in high spectrum image, and determine fitness function；

Step S12, initiation parameter, including, adopt bee numbers m, follow honeybee quantity m, maximum iteration time maxIter；

It should be noted that depending on the concrete foundation of the numerical value of initiation parameter is actually needed, the present invention does not limits at this.

Step S13, randomly generating m feasible solution in solution space, each described feasible solution is as a gathering honey The nectar source that honeybee is corresponding；

Step S14, gathering honey honeybee are by searching neighborhood nectar source, it is thus achieved that more excellent nectar source；

Concrete, gathering honey honeybee randomly chooses a nectar source as current first nectar source, according to search function described currently Search the first new nectar source in the neighborhood in the first nectar source, utilize described fitness function to calculate the fitness in the described first new nectar source Value, and select described current first nectar source of replacement that in described current first nectar source and the described first new nectar source, fitness value is big.

It is to say, every corresponding nectar source (and fitness) of gathering honey honeybee, gathering honey honeybee can be at the neighborhood in this nectar source Inside carry out Local Search and find new nectar source, if the fitness value in new nectar source is higher than the fitness value in green molasses source, then updating nectar source Position, will be updated to position, new nectar source, otherwise, abandon new nectar source and continue the neighbour in green molasses source again in position, green molasses source Search in territory.

If X_i=(x_i1x_i2......x_iM) ' represent i-th nectar source, the position at namely jth gathering honey honeybee place, then exist It is formula (1) that current position, nectar source carries out the neighborhood Local Search expression formula searching formula used, and formula (1) is specific as follows:

φ_ij=x_ij+Δ(x_ij-x_sj)(1)；

In formula, φ_ijRepresent and search function, x_ijRepresent i-th nectar source jth gathering honey honeybee/investigation honeybee position, x_sjTable Showing the s nectar source jth gathering honey honeybee/investigation honeybee position, Δ is the random number between [-1,1], s ≠ i and s ∈ 1, 2 ... M}, j ∈ 1,2 ... M}, M are the end member quantity in described high spectrum image.

Step S15, gathering honey honeybee share nectar source information to following honeybee；

Nectar source Information Sharing is generally followed honeybee by modes such as swings by gathering honey honeybee.

Nectar source information refers to the up-to-date nectar source that gathering honey honeybee is chosen.

Step S16, follow honeybee by second time search neighborhood nectar source, it is thus achieved that more excellent nectar source；

Concrete, follow honeybee according to following probability selection gathering honey honeybee, using current first nectar source up-to-date for this gathering honey honeybee as Current second nectar source, searches the second new nectar source, profit according to described search function secondary in the neighborhood in described current second nectar source Calculate the fitness value in the described second new nectar source with described fitness function, and select described current second nectar source and described second Described current second nectar source is replaced in the nectar source that in new nectar source, fitness value is big.

It should be noted that every gathering honey honeybee provides a fitness value, following honeybee can provide according to each gathering honey honeybee Fitness value selects to follow which gathering honey honeybee, if the fitness value that gathering honey honeybee provides is relatively big, then follows honeybee and selects to follow it Probability just big.Follow after honeybee selectes gathering honey honeybee, perform the work identical with gathering honey honeybee.

That follows that honeybee selects jth nectar source " follows Probability p_j" obtained by formula (2), formula (2) is specific as follows:

$p_j=\frac{fit\left(X_j\right)}{{\mathrm{\Σ}}_{i=1}^{Num}fit\left(X_i\right)}---\left(2\right);$

In formula, p_jRepresent that following honeybee selects the probability of following in jth nectar source, fit (X_i) represent i-th nectar source fitness Function, fit (X_j) represent jth nectar source fitness function, Num represents nectar source quantity.

Step S17, utilization follow the optimal solution obtained before the nectar source renewal that honeybee is chosen；

Step S18, judge in m gathering honey honeybee in preset time period whether either with or without updating the gathering honey honeybee in nectar source, if It is then to perform step S19, otherwise, perform step S24；

Wherein, depending on the concrete numerical basis of preset time period is actually needed.

Step S19, the gathering honey honeybee not updating nectar source in described preset time period is converted to investigate honeybee；

It should be noted that when the long-term search in the nectar source neighborhood of its correspondence of gathering honey honeybee is more excellent less than more current nectar source Xie Shi, can abandon this nectar source and be converted to investigate honeybee.

Step S20, investigation honeybee obtain the fitness value in random nectar source；

Concrete, investigation honeybee randomly selects a nectar source as current 3rd nectar source in search volume, searches according to described Seek function in the neighborhood in described current 3rd nectar source, search the 3rd new nectar source, utilize described fitness function to calculate the described 3rd The fitness value in new nectar source.

Step S21, judge the fitness value in the described 3rd new nectar source whether less than the fitness value in described 3rd nectar source, as Fruit is, then perform step S22, otherwise perform step S23；

Step S22, nectar source that honeybee chooses will be followed as current optimal solution, and perform step S24；

Step S23, this investigation honeybee is converted to gathering honey honeybee, returns and perform step S14, continue to search for new nectar source；

Step S24, judge whether current iteration number of times reaches described maximum iteration time maxIter, if it is, perform Step S25, otherwise, returns and again performs step S14；

Step S25, output Endmember extraction result.

It should be noted that in an iterative process, optimal this of the fitness value in the nectar source of search will be labeled every time, if Nectar source is not updated, then it is believed that this nectar source is optimum nectar source.

Searching optimal path is regarded as a kind of combinatorial optimization problem by artificial bee colony algorithm, finds the mistake of optimal solution with this Journey, this algorithm has gradually formed new study hotspot.The presenter Karaboga of artificial bee colony algorithm finds under study for action, uses Artificial bee colony algorithm solve function optimization problem in terms of solving non-limiting numerical optimization function than other heuritic approaches more Tool potentiality, compared with other swarm intelligence algorithms, artificial bee colony algorithm finds the probability of globally optimal solution bigger.

In summary it can be seen, high optical spectrum image end member extraction method disclosed by the invention, high optical spectrum image end member is extracted The position, nectar source in optimized variable correspondence artificial bee colony algorithm in problem, the optimization in each nectar source is determined by fitness function, The number of gathering honey honeybee is consistent with the number of feasible solution.The neighborhood in corresponding nectar source was once searched for by gathering honey honeybee before this, and selected The current nectar source of replacement that current nectar source is big with fitness value in the new nectar source searched；After gathering honey honeybee completes search, nectar source is believed Breath is followed honeybee, is carried out the neighborhood in the current nectar source that gathering honey honeybee is selected searching for for the second time by following honeybee, and selects current nectar source The replacement current nectar source big with fitness value in the new nectar source searched, then utilizes and obtains before following the nectar source renewal that honeybee is chosen The optimal solution obtained, when there is the gathering honey honeybee not updating nectar source in preset time period, being converted to this gathering honey honeybee investigate honeybee, detecing Look into honeybee by being compared by the fitness value of the fitness value randomly selecting nectar source with the nectar source as optimal solution, to as The nectar source of excellent solution verifies, if randomly selecting the fitness value fitness value less than the nectar source as optimal solution in nectar source, Then showing to follow the nectar source that honeybee finally chooses is optimal solution, then repeats said process, until iterations reaches greatest iteration Number of times, obtains Endmember extraction result, and otherwise gathering honey honeybee continues to search for new nectar source.It can be seen that the present invention is with artificial bee colony algorithm Based on, Endmember extraction problem is converted into the solution procedure of combinatorial optimization problem, with this reduce end member extraction method for The undue dependence of the quality of data, the advantage giving full play to artificial bee colony algorithm, and by investigation honeybee to the nectar source as optimal solution Verify so that optimization process is absorbed in the risk of locally optimal solution and is substantially reduced.

It should be noted that the present invention uses Areca trees mould when obtaining the candidate's end member in high spectrum image Type (Linear Spectral Mixture Model, LSMM).

Assuming that mixed pixel is to have end member to form with end member linear hybrid, mixed pixel is distributed in the inside of monomorphous, end Unit is distributed in the summit of monomorphous in feature space.The pure spectra signal that can represent atural object classification in high spectrum image is referred to as End member, certain end member ratio shared by mixed pixel is referred to as abundance.

The expression formula of linear spectral mixture model is shown in formula (3):

$\begin{array}{cc}{x}_i=\underset{k=1}{\overset{M}{\Σ}}{p}_{ik}{e}_k+{\mathrm{\ϵ}}_i& i=1,...,N\end{array}---\left(3\right);$

In formula, x_iRepresent pixel, e_kRepresent end member_,p_ikFor the ratio that end member is shared in pixel_,ε_iFor error, M is described End member quantity in high spectrum image, N is the quantity of pixel in described high spectrum image.

Wherein, the restrictive condition of formula (3) is shown in formula (4):

$\left\{\begin{array}{cc}{p}_{\mathrm{ik}}\≥0& \∀k=1,...,M\\ \underset{k=1}{\overset{M}{\mathrm{\Σ}}}{p}_{\mathrm{ik}}=1& \end{array}\right.---\left(4\right);$

The abundance inversion method meeting (4) formula constraints is referred to as staff cultivation method of least square (Full Constraints Least Square, FCLS).If known end member e_kAnd high spectrum imageCan obtain representing that end member accounts for Gao Guang according to FCLS The abundance matrix of spectrum image scaled, utilizes abundance matrix to be multiplied by end member vector and obtains high spectrum imageBack mixing image

In above-described embodiment, fitness function is obtained by formula (5), and formula (5) is specific as follows:

$fit\left(X_i\right)=\frac{1}{f\left(X_i\right)}---\left(5\right);$

In formula, fit (X_i) represent i-th nectar source fitness function, X_iRepresent i-th nectar source, X_i= (x_i1x_i2......x_iM) ', M is the end member quantity in described high spectrum image, f (X_i) it is the object function of optimal problem.

Wherein, the object function of the optimization problem i.e. optimizing model of Endmember extraction.

It should be noted that owing to original ant colony algorithm realizes under the conditions of solution space continuous print, to For solving combinatorial optimization problem, need solution space is carried out discretization, for ease of computer identification, can represent with 0 Background, 1 represents end member.

If Set-search spacec_iCorresponding i-th end Unit, if X_i∈C_{N, M}, then X_iRepresenting one 0 and the n figure place word string of 1 composition, M is the end member quantity in high spectrum image, and N is Gao Guang The quantity of pixel in spectrogram picture, i=1 ... the numerical value on .., N, M position is 1, and the numerical value of other position is 0.Definition Mapping relations, shown in these mapping relations such as formula (6):

$\begin{array}{c}H:Z({\left\{x_i\right\}}_{i=1}^N,M)\→C_{N,M}\\ E\→(c_1,c_2,\mathrm{...},c_N)\end{array}---\left(6\right);$

For high spectrum image, if certain pixel x in high spectrum image_iEnd member, then x is selected after being considered_iCorresponding c_i Peek value be 1, otherwise peek value is 0 (i.e. end member is 1, and background is 0).If end member quantity M it is known that, can use C_M,NIt is used as 0 With the search volume of 1 composition, search problem is i.e. to find out 0,1 optimal sequence in search volume.

Evaluate the end member that extracts comprise original high spectrum image quantity of information number time, current universally recognized method is Contrast the back mixing image of original high spectrum image and root-mean-square error (the Root Mean Square of original high spectrum image Error, RMSE), root-mean-square error the least explanation difference degree is the lowest, and the result obtained is the best.

The present invention introduces distance terms g (d in object function_i), when the two end member locus identified are too near, can enter Enter to re-search for, it is to avoid repeat to extract same end member.

The expression formula of the object function of optimal problem, is specifically shown in formula (7):

$f\left(X_i\right)=RMSE({\left\{x_i\right\}}_{i=1}^N,C^{-1}(X_i\left)\right)+u\×g\left(d_i\right)---\left(7\right);$

In formula, u represents that penalty coefficient, peer-to-peer two ends play the effect of adjustment；

g(d_i) represent distance terms,d_iFor Euclidean distance between end member and end member in image, ifRepresent a set,e_pRepresent pth end member, e_qQ end Unit；

C^-1For the inverse function of formula (4), x_iRepresent pixel,The restrictive condition of pixel ise_kRepresent end member_,p_ikFor the ratio that end member is shared in pixel_,ε_iFor error, M is described height End member quantity in spectrum picture, N is the quantity of pixel in described high spectrum image, i=1 ... .., N, C_{M, N}For search volume；

RMSE represents root-mean-square error,L represents described Gao Guang The wave band number of spectrogram picture, x_iRepresent the pixel in high spectrum image,Represent the pixel in the back mixing image of high spectrum image.

Formula (7) is brought into the final expression formula that formula (5) can obtain the object function of optimal problem, is specifically shown in formula (8):

$fit\left(X_i\right)=\frac{1}{RMSE({\left\{x_i\right\}}_{i=1}^N,C^{-1}(X_i))+u\×g\left(d_i\right)}---\left(8\right);$

Thus, it will be seen that the optimized essence of artificial bee colony is to obtain the maximum of object function, it is specifically shown in formula (9):

$\begin{array}{cc}\max & fit\left(X_i\right)\\ s.t.& E\∈Z({\left\{x_i\right\}}_{i=1}^N,M)\end{array}---\left(9\right).$

In summary it can be seen, the present invention is based on artificial bee colony algorithm, after primary colony algorithm is carried out discretization, Endmember extraction problem is converted into and solves combinatorial optimization problem, construct object function by root-mean-square error and distance terms, pass through Obtain the maximum of object function, it is achieved the extraction to high optical spectrum image end member.The present invention gives full play to original artificial bee colony and calculates The advantage of method, the risk being absorbed in locally optimal solution during making optimization is substantially reduced, and simultaneously by the constraint of " distance terms ", effectively keeps away Exempted from because the situation repeating to identify same end member that is near and that cause is crossed in the end member position identified in prior art, so from Extract correct result in the solution space dissipated, and the tram at end member place can be marked.

Corresponding with said method embodiment, the invention also discloses a kind of high optical spectrum image end member extraction system.

See Fig. 2, the structural representation of a kind of high optical spectrum image end member extraction system disclosed in the embodiment of the present invention, bag Include:

Acquiring unit 21, for obtaining the candidate's end member in high spectrum image, and determines fitness function；

Initialization unit 22, for initiation parameter, including, adopt bee numbers m, follow honeybee quantity m, maximum iteration time maxIter；

It should be noted that depending on the concrete foundation of the numerical value of initiation parameter is actually needed, the present invention does not limits at this.

Feasible solution generation unit 23, for randomly generating m feasible solution in solution space, each described feasible solution As the nectar source that a gathering honey honeybee is corresponding；

First searches unit 24, randomly chooses a nectar source as current first nectar source for gathering honey honeybee, according to searching letter Number searches the first new nectar source in the neighborhood in described current first nectar source, utilizes described fitness function to calculate the described first new honey The fitness value in source, and select the replacement that in described current first nectar source and the described first new nectar source, fitness value is big described currently First nectar source；

It is to say, every corresponding nectar source (and fitness) of gathering honey honeybee, gathering honey honeybee can be at the neighborhood in this nectar source Inside carry out Local Search and find new nectar source, if the fitness value in new nectar source is higher than the fitness value in green molasses source, then updating nectar source Position, will be updated to position, new nectar source, otherwise, abandon new nectar source and continue the neighbour in green molasses source again in position, green molasses source Search in territory.

If X_i=(x_i1x_i2......x_iM) ' represent i-th nectar source, the position at namely jth gathering honey honeybee place, then exist It is formula (1) that current position, nectar source carries out the neighborhood Local Search expression formula searching formula used, and formula (1) is specific as follows:

φ_ij=x_ij+Δ(x_ij-x_sj)(1)；

In formula, φ_ijRepresent and search function, x_ijRepresent i-th nectar source jth gathering honey honeybee/investigation honeybee position, x_sjTable Showing the s nectar source jth gathering honey honeybee/investigation honeybee position, Δ is the random number between [-1,1], s ≠ i and s ∈ 1, 2 ... M}, j ∈ 1,2 ... M}, M are the end member quantity in described high spectrum image.

Share unit 25, share nectar source information to following honeybee for gathering honey honeybee；

Nectar source Information Sharing is generally followed honeybee by modes such as swings by gathering honey honeybee.

Nectar source information refers to the up-to-date nectar source that gathering honey honeybee is chosen.

Second searches unit 26, is used for following honeybee according to following probability selection gathering honey honeybee, by up-to-date for this gathering honey honeybee current First nectar source as current second nectar source, searches the according to described search function secondary in the neighborhood in described current second nectar source Two new nectar sources, utilize described fitness function to calculate the fitness value in the described second new nectar source, and select described current second honey Described current second nectar source is replaced in the nectar source that in source and the described second new nectar source, fitness value is big；

It should be noted that every gathering honey honeybee provides a fitness value, following honeybee can provide according to each gathering honey honeybee Fitness value selects to follow which gathering honey honeybee, if the fitness value that gathering honey honeybee provides is relatively big, then follows honeybee and selects to follow it Probability just big.Follow after honeybee selectes gathering honey honeybee, perform the work identical with gathering honey honeybee.

That follows that honeybee selects jth nectar source " follows Probability p_j" obtained by formula (2), formula (2) is specific as follows:

$p_j=\frac{fit\left(X_j\right)}{{\mathrm{\Σ}}_{i=1}^{Num}fit\left(X_i\right)}---\left(2\right);$

In formula, p_jRepresent that following honeybee selects the probability of following in jth nectar source, fit (X_i) represent i-th nectar source fitness Function, fit (X_j) represent jth nectar source fitness function, Num represents nectar source quantity.

Updating block 27, follows, for utilizing, the optimal solution obtained before the nectar source renewal that honeybee is chosen；

Whether the first judging unit 28, for judging in m gathering honey honeybee in preset time period either with or without renewal nectar source Gathering honey honeybee, if it is, perform the first converting unit 29, otherwise, performs the 4th judging unit 34；

Wherein, depending on the concrete numerical basis of preset time period is actually needed.

First converting unit 29, for being converted to investigation by the gathering honey honeybee not updating nectar source in described preset time period Honeybee；

Computing unit 30, is used for investigating honeybee and randomly selects a nectar source in search volume as current 3rd nectar source, root In the neighborhood in described current 3rd nectar source, search the 3rd new nectar source according to described search function, utilize described fitness function to calculate The fitness value in the described 3rd new nectar source；

Second judging unit 31, for judging that whether the fitness value in the described 3rd new nectar source is less than described 3rd nectar source Fitness value, if it is, perform the 3rd judging unit 32, if it is not, then perform the second converting unit 33；

3rd judging unit 32, for following nectar source that honeybee chooses as current optimal solution, and it is single to perform the 4th judgement Unit 34；

Second converting unit 33, for this investigation honeybee is converted to gathering honey honeybee, and returns execution the first search unit 24, continues The new nectar source of continuous search；

4th judging unit 34, is used for judging whether current iteration number of times reaches described maximum iteration time maxIter, as Fruit is, then perform output unit 35, otherwise, return and perform the first search unit 24, continue to search for new nectar source；

Output unit 35, is used for exporting Endmember extraction result.

It should be noted that in an iterative process, optimal this of the fitness value in the nectar source of search will be labeled every time, if Nectar source is not updated, then it is believed that this nectar source is optimum nectar source.

Searching optimal path is regarded as a kind of combinatorial optimization problem by artificial bee colony algorithm, finds the mistake of optimal solution with this Journey, this algorithm has gradually formed new study hotspot.The presenter Karaboga of artificial bee colony algorithm finds under study for action, uses Artificial bee colony algorithm solve function optimization problem in terms of solving non-limiting numerical optimization function than other heuritic approaches more Tool potentiality, compared with other swarm intelligence algorithms, artificial bee colony algorithm finds the probability of globally optimal solution bigger.

In summary it can be seen, high optical spectrum image end member extraction system disclosed by the invention, high optical spectrum image end member is extracted The position, nectar source in optimized variable correspondence artificial bee colony algorithm in problem, the optimization in each nectar source is determined by fitness function, The number of gathering honey honeybee is consistent with the number of feasible solution.The neighborhood in corresponding nectar source was once searched for by gathering honey honeybee before this, and selected The current nectar source of replacement that current nectar source is big with fitness value in the new nectar source searched；After gathering honey honeybee completes search, nectar source is believed Breath is followed honeybee, is carried out the neighborhood in the current nectar source that gathering honey honeybee is selected searching for for the second time by following honeybee, and selects current nectar source The replacement current nectar source big with fitness value in the new nectar source searched, then utilizes and obtains before following the nectar source renewal that honeybee is chosen The optimal solution obtained, when there is the gathering honey honeybee not updating nectar source in preset time period, being converted to this gathering honey honeybee investigate honeybee, detecing Look into honeybee by being compared by the fitness value of the fitness value randomly selecting nectar source with the nectar source as optimal solution, to as The nectar source of excellent solution verifies, if randomly selecting the fitness value fitness value less than the nectar source as optimal solution in nectar source, Then showing to follow the nectar source that honeybee finally chooses is optimal solution, then repeats said process, until iterations reaches greatest iteration Number of times, obtains Endmember extraction result, and otherwise gathering honey honeybee continues to search for new nectar source.It can be seen that the present invention is with artificial bee colony algorithm Based on, Endmember extraction problem is converted into the solution procedure of combinatorial optimization problem, with this reduce end member extraction method for The undue dependence of the quality of data, the advantage giving full play to artificial bee colony algorithm, and by investigation honeybee to the nectar source as optimal solution Verify so that optimization process is absorbed in the risk of locally optimal solution and is substantially reduced.

It should be noted that the present invention uses Areca trees mould when obtaining the candidate's end member in high spectrum image Type (Linear Spectral Mixture Model, LSMM).

Assuming that mixed pixel is to have end member to form with end member linear hybrid, mixed pixel is distributed in the inside of monomorphous, end Unit is distributed in the summit of monomorphous in feature space.The pure spectra signal that can represent atural object classification in high spectrum image is referred to as End member, certain end member ratio shared by mixed pixel is referred to as abundance.

The expression formula of linear spectral mixture model is shown in formula (3):

$\begin{array}{cc}{x}_i=\underset{k=1}{\overset{M}{\Σ}}{p}_{ik}{e}_k+{\mathrm{\ϵ}}_i& i=1,\mathrm{...},N\end{array}---\left(3\right);$

In formula, x_iRepresent pixel, e_kRepresent end member_,p_ikFor the ratio that end member is shared in pixel_,ε_iFor error, M is described End member quantity in high spectrum image, N is the quantity of pixel in described high spectrum image.

Wherein, the restrictive condition of formula (3) is shown in formula (4):

The abundance inversion method meeting (4) formula constraints is referred to as staff cultivation method of least square (Full Constraints Least Square, FCLS).If known end member e_kAnd high spectrum imageCan obtain representing that end member accounts for Gao Guang according to FCLS The abundance matrix of spectrum image scaled, utilizes abundance matrix to be multiplied by end member vector and obtains high spectrum imageBack mixing image

In above-described embodiment, fitness function is obtained by formula (5), and formula (5) is specific as follows:

$fit\left(X_i\right)=\frac{1}{f\left(X_i\right)}---\left(5\right);$

In formula, fit_i(X_i) represent i-th nectar source fitness function, X_iRepresent i-th nectar source, X_i= (x_i1x_i2......x_iM) ', M is the end member quantity in described high spectrum image, f (X_i) it is the object function of optimal problem.

Wherein, the object function of the optimization problem i.e. optimizing model of Endmember extraction.

It should be noted that owing to original ant colony algorithm realizes under the conditions of solution space continuous print, to For solving combinatorial optimization problem, need solution space is carried out discretization, for ease of computer identification, can represent with 0 Background, 1 represents end member.

If Set-search spacec_iRepresent i-th end member, If X_i∈C_{N, M}, then X_iRepresenting one 0 and the n figure place word string of 1 composition, M is the end member quantity in high spectrum image, and N is EO-1 hyperion The quantity of pixel in image, i=1 ... the numerical value on .., N, M position is 1, and the numerical value of other position is 0.Definition is reflected Penetrate relation, shown in these mapping relations such as formula (6):

$\begin{array}{c}H:Z({\left\{x_i\right\}}_{i=1}^N,M)\→C_{N,M}\\ E\→(c_1,c_2,\mathrm{...},c_N)\end{array}---\left(6\right);$

For high spectrum image, if certain pixel x in high spectrum image_iEnd member, then x is selected after being considered_iCorresponding c_i Peek value be 1, otherwise peek value is 0 (i.e. end member is 1, and background is 0).If end member quantity M it is known that, can use C_M,NIt is used as 0 With the search volume of 1 composition, search problem is i.e. to find out 0,1 optimal sequence in search volume.

Evaluate the end member that extracts comprise original high spectrum image quantity of information number time, current universally recognized method is Contrast the back mixing image of original high spectrum image and root-mean-square error (the Root Mean Square of original high spectrum image Error, RMSE), root-mean-square error the least explanation difference degree is the lowest, and the result obtained is the best.

The present invention introduces distance terms g (d in object function_i), when the two end member locus identified are too near, can enter Enter to re-search for, it is to avoid repeat to extract same end member.

The expression formula of the object function of optimal problem, is specifically shown in formula (7):

$f\left(X_i\right)=RMSE({\left\{x_i\right\}}_{i=1}^N,C^{-1}(X_i\left)\right)+u\×g\left(d_i\right)---\left(7\right);$

In formula, u represents that penalty coefficient, peer-to-peer two ends play the effect of adjustment；

g(d_i) represent distance terms,d_iFor Euclidean distance between end member and end member in image, ifRepresent a set,e_pRepresent pth end member, e_qQ end Unit；

C^-1For the inverse function of formula (4), x_iRepresent pixel,The restrictive condition of pixel ise_kRepresent end member, p_ikFor the ratio that end member is shared in pixel, ε_iFor error, M is described height End member quantity in spectrum picture, N is the quantity of pixel in described high spectrum image, i=1 ... .., N, C_{M, N}For search volume；

RMSE represents root-mean-square error,L represents described Gao Guang The wave band number of spectrogram picture, x_iRepresent the pixel in high spectrum image,Represent the pixel in the back mixing image of high spectrum image.

Formula (7) is brought into the final expression formula that formula (5) can obtain the object function of optimal problem, is specifically shown in formula (8):

$fit\left(X_i\right)=\frac{1}{RMSE({\left\{x_i\right\}}_{i=1}^N,C^{-1}(X_i))+u\×g\left(d_i\right)}---\left(8\right);$

Thus, it will be seen that the optimized essence of artificial bee colony is to obtain the maximum of object function, it is specifically shown in formula (9):

$\begin{array}{cc}\max & fit\left(X_i\right)\\ s.t.& E\∈Z({\left\{x_i\right\}}_{i=1}^N,M)\end{array}---\left(9\right).$

In summary it can be seen, the present invention is based on artificial bee colony algorithm, after primary colony algorithm is carried out discretization, Endmember extraction problem is converted into and solves combinatorial optimization problem, construct object function by root-mean-square error and distance terms, pass through Obtain the maximum of object function, it is achieved the extraction to high optical spectrum image end member.The present invention gives full play to original artificial bee colony and calculates The advantage of method, the risk being absorbed in locally optimal solution during making optimization is substantially reduced, and simultaneously by the constraint of " distance terms ", effectively keeps away Exempted from because the situation repeating to identify same end member that is near and that cause is crossed in the end member position identified in prior art, so from Extract correct result in the solution space dissipated, and the tram at end member place can be marked.

It should be noted that the operation principle of each ingredient refers to embodiment of the method correspondence portion in system embodiment Point, here is omitted.

Finally, in addition it is also necessary to explanation, in this article, the relational terms of such as first and second or the like be used merely to by One entity or operation separate with another entity or operating space, and not necessarily require or imply these entities or operation Between exist any this reality relation or order.And, term " includes ", " comprising " or its any other variant meaning Containing comprising of nonexcludability, so that include that the process of a series of key element, method, article or equipment not only include that A little key elements, but also include other key elements being not expressly set out, or also include for this process, method, article or The key element that equipment is intrinsic.In the case of there is no more restriction, statement " including ... " key element limited, do not arrange Except there is also other identical element in including the process of described key element, method, article or equipment.

In this specification, each embodiment uses the mode gone forward one by one to describe, and what each embodiment stressed is and other The difference of embodiment, between each embodiment, identical similar portion sees mutually.

Described above to the disclosed embodiments, makes professional and technical personnel in the field be capable of or uses the present invention. Multiple amendment to these embodiments will be apparent from for those skilled in the art, as defined herein General Principle can realize without departing from the spirit or scope of the present invention in other embodiments.Therefore, the present invention It is not intended to be limited to the embodiments shown herein, and is to fit to and principles disclosed herein and features of novelty phase one The widest scope caused.

Claims (10)

1. a high optical spectrum image end member extraction method, it is characterised in that including:

Obtain the candidate's end member in high spectrum image, and determine fitness function；

Initiation parameter, including, adopt bee numbers m, follow honeybee quantity m, maximum iteration time maxIter；

Randomly generating m feasible solution in solution space, each described feasible solution is as nectar source corresponding to a gathering honey honeybee；

Gathering honey honeybee randomly chooses a nectar source as current first nectar source, according to searching the function neighbour in described current first nectar source Search the first new nectar source in territory, utilize described fitness function to calculate the fitness value in the described first new nectar source, and select described Described current first nectar source of replacement that in current first nectar source and the described first new nectar source, fitness value is big；

Gathering honey honeybee shares nectar source information to following honeybee；

Follow honeybee according to following probability selection gathering honey honeybee, using current first nectar source up-to-date for this gathering honey honeybee as current second honey Source, searches the second new nectar source according to described search function secondary in the neighborhood in described current second nectar source, utilizes described adaptation Degree function calculates the fitness value in the described second new nectar source, and selects in described current second nectar source and the described second new nectar source suitable The nectar source that angle value is big is answered to replace described current second nectar source；

Utilize and follow the optimal solution obtained before the nectar source renewal that honeybee is chosen；

Whether judge in m gathering honey honeybee in preset time period either with or without the gathering honey honeybee updating nectar source；

If there is not updating the gathering honey honeybee in nectar source, then will not update the gathering honey honeybee conversion in nectar source in described preset time period For investigation honeybee；

Investigation honeybee randomly selects a nectar source as current 3rd nectar source, according to described search function described in search volume Search the 3rd new nectar source in the neighborhood in current 3rd nectar source, utilize described fitness function to calculate the adaptation in the described 3rd new nectar source Angle value；

Judge whether the fitness value in the described 3rd new nectar source is less than the fitness value in described 3rd nectar source；

If less than the fitness value in described 3rd nectar source, then will follow nectar source that honeybee chooses as current optimal solution, and judge Whether current iteration number of times reaches described maximum iteration time maxIter；

If not less than the fitness value in described 3rd nectar source, then this investigation honeybee being converted to gathering honey honeybee, continuing to search for new nectar source；

If there is no not updating the gathering honey honeybee in nectar source, then judge whether current iteration number of times reaches described maximum iteration time maxIter；

If reaching described maximum iteration time maxIter, then output Endmember extraction result；

Without reaching described maximum iteration time maxIter, then gathering honey honeybee continues to search for new nectar source.

High optical spectrum image end member extraction method the most according to claim 1, it is characterised in that the table of described fitness function Reach formula as follows:

$fit\left(X_i\right)=\frac{1}{f\left(X_i\right)};$

In formula, fit (X_i) represent i-th nectar source fitness function, X_iRepresent i-th nectar source, X_i= (x_i1x_i2......x_iM) ', M is the end member quantity in described high spectrum image, f (X_i) it is the object function of optimal problem.

High optical spectrum image end member extraction method the most according to claim 2, it is characterised in that the target of described optimal problem The expression formula of function is as follows:

$f\left(X_i\right)=RMSE({\left\{x_i\right\}}_{i=1}^N,C^{-1}(X_i\left)\right)+u\×g\left(d_i\right);$

In formula, u represents that penalty coefficient, peer-to-peer two ends play the effect of adjustment；

g(d_i) represent distance terms,d_iFor Euclidean distance between end member and end member in image, ifRepresent One set,e_p∈Q,e_q∈ Q, e_pRepresent pth end member, e_qRepresent q-th end member；

C^-1ForInverse function, x_iRepresent pixel,The restrictive condition of pixel Fore_kRepresent end member, p_ikFor the ratio that end member is shared in pixel_,ε_iFor error_,M is described End member quantity in high spectrum image, N is the quantity of pixel in described high spectrum image, i=1 ... .., N, C_{M, N}For search sky Between；

RMSE represents root-mean-square error,L represents described high-spectrum The wave band number of picture, x_iRepresent the pixel in described high spectrum image,Represent the picture in the back mixing image of described high spectrum image Unit.

High optical spectrum image end member extraction method the most according to claim 1, it is characterised in that described in follow the expression of probability Formula is as follows:

$p_j=\frac{fit\left(X_j\right)}{{\mathrm{\Σ}}_{i=1}^{Num}fit\left(X_i\right)};$

In formula, p_jRepresent that following honeybee selects the probability of following in jth nectar source, fit (X_i) represent i-th nectar source fitness function, fit(X_j) represent jth nectar source fitness function, Num represents nectar source quantity.

High optical spectrum image end member extraction method the most according to claim 1, it is characterised in that the expression of described search function Formula is as follows:

φ_ij=x_ij+Δ(x_ij-x_sj) (4)；

In formula, φ_ijRepresent and search function, x_ijRepresent i-th nectar source jth gathering honey honeybee/investigation honeybee position, x_sjRepresent s Individual nectar source jth gathering honey honeybee/investigation honeybee position, △ is the random number between [-1,1], s ≠ i and s ∈ 1,2 ... M}, J ∈ 1,2 ... M}, M are the end member quantity in described high spectrum image.

6. a high optical spectrum image end member extraction system, it is characterised in that including:

Acquiring unit, for obtaining the candidate's end member in high spectrum image, and determines fitness function；

Initialization unit, for initiation parameter, including, adopt bee numbers m, follow honeybee quantity m, maximum iteration time maxIter；

Feasible solution generation unit, for randomly generating m feasible solution in solution space, each described feasible solution is as one The nectar source that gathering honey honeybee is corresponding；

First searches unit, randomly chooses a nectar source as current first nectar source for gathering honey honeybee, according to searching function in institute Search the first new nectar source in stating the neighborhood in current first nectar source, utilize described fitness function to calculate the suitable of the described first new nectar source Answer angle value, and select described current first honey of replacement that in described current first nectar source and the described first new nectar source, fitness value is big Source；

Share unit, share nectar source information to following honeybee for gathering honey honeybee；

Second searches unit, is used for following honeybee according to following probability selection gathering honey honeybee, by current first honey up-to-date for this gathering honey honeybee The second new honey, as current second nectar source, is searched according to described search function secondary in the neighborhood in described current second nectar source in source Source, utilizes described fitness function to calculate the fitness value in the described second new nectar source, and selects described current second nectar source and institute State the nectar source that in the second new nectar source, fitness value is big and replace described current second nectar source；

Updating block, follows, for utilizing, the optimal solution obtained before the nectar source renewal that honeybee is chosen；

First judging unit, for judging in m gathering honey honeybee in preset time period whether gathering honey honeybee either with or without renewal nectar source, If it is, perform the first converting unit, otherwise, perform the 4th judging unit；

Described first converting unit, for being converted to investigation honeybee by the gathering honey honeybee not updating nectar source in described preset time period；

Computing unit, is used for investigating honeybee and randomly selects a nectar source in search volume as current 3rd nectar source, according to described Search function in the neighborhood in described current 3rd nectar source, searches the 3rd new nectar source, utilize described fitness function calculating described the The fitness value in three new nectar sources；

Second judging unit, for judging whether the fitness value in the described 3rd new nectar source is less than the fitness in described 3rd nectar source Value, if it is, perform the 3rd judging unit, if it is not, then perform the second converting unit；

Described 3rd judging unit, for following nectar source that honeybee chooses as current optimal solution, and performs described 4th judgement Unit；

Described second converting unit, in the case of described second judging unit is judged as NO, then changes this investigation honeybee For gathering honey honeybee, and return execution described first and search unit, continue to search for new nectar source；

Described 4th judging unit, is used for judging whether current iteration number of times reaches described maximum iteration time maxIter, if It is then to perform output unit, otherwise, return and perform described first search unit, continue to search for new nectar source；

Described output unit, is used for exporting Endmember extraction result.

High optical spectrum image end member extraction system the most according to claim 6, it is characterised in that the table of described fitness function Reach formula as follows:

$fit\left(X_i\right)=\frac{1}{f\left(X_i\right)};$

In formula, fit (X_i) represent i-th nectar source fitness function, X_iRepresent i-th nectar source, X_i= (x_i1x_i2......x_iM) ', M is the end member quantity in described high spectrum image, f (X_i) it is the object function of optimal problem.

High optical spectrum image end member extraction system the most according to claim 7, it is characterised in that the target of described optimal problem The expression formula of function is as follows:

$f\left(X_i\right)=RMSE({\left\{x_i\right\}}_{i=1}^N,C^{-1}(X_i\left)\right)+u\×g\left(d_i\right);$

In formula, u represents that penalty coefficient, peer-to-peer two ends play the effect of adjustment；

g(d_i) represent distance terms,IfRepresent a set, d_iFor in image between end member and end member Euclidean distance,e_p∈Q,e_q∈ Q, e_pRepresent pth end member, e_qRepresent q-th end member；

C^-1ForInverse function, x_iRepresent pixel,The restrictive condition of pixel Fore_kRepresent end member, p_ikFor the ratio that end member is shared in pixel_,ε_iFor error_,M is described End member quantity in high spectrum image, N is the quantity of pixel in described high spectrum image, i=1 ... .., N, C_{M, N}For search sky Between；

RMSE represents root-mean-square error,L represents described high-spectrum The wave band number of picture, x_iRepresent the pixel in described high spectrum image,Represent the picture in the back mixing image of described high spectrum image Unit.

High optical spectrum image end member extraction system the most according to claim 6, it is characterised in that described in follow the expression of probability Formula is as follows:

$p_j=\frac{fit\left(X_j\right)}{{\mathrm{\Σ}}_{i=1}^{Num}fit\left(X_i\right)};$

In formula, p_jRepresent that following honeybee selects the probability of following in jth nectar source, fit (X_i) represent i-th nectar source fitness function, fit(X_j) represent jth nectar source fitness function, Num represents nectar source quantity.

High optical spectrum image end member extraction system the most according to claim 6, it is characterised in that the table of described search function Reach formula as follows:

φ_ij=x_ij+Δ(x_ij-x_sj)；

In formula, φ_ijRepresent and search function, x_ijRepresent i-th nectar source jth gathering honey honeybee/investigation honeybee position, x_sjRepresent s Individual nectar source jth gathering honey honeybee/investigation honeybee position, △ is the random number between [-1,1], s ≠ i and s ∈ 1,2 ... M}, J ∈ 1,2 ... M}, M are the end member quantity in described high spectrum image.