CN103927456B - A kind of remote sensing image rational function model structural optimization method based on ant colony algorithm - Google Patents

Abstract

A kind of remote sensing image rational function model optimization method based on ant colony algorithm, relates to remotely sensed image geometric model and optimizes field.Crossing, because of parameter item, the low problem of model accuracy that redundancy causes for remotely sensed image geometric model based on rational function model, be optimized this model with ant colony algorithm, step is as follows: one, construct binary set x to be solved and income degree function f (x)；Two, ant colony algorithm parameter is initialized；Three, bee colony initializes；Four, starting iteration, E honeybee optimizes；Five, calculate E honeybee and recruit probability；Six, O honeybee optimizes；Seven, globally optimal solution is updated；Eight, judging stopping criterion for iteration, if meeting, optimization completes；Otherwise carry out step 9；Nine, judging whether each solution meets S honeybee entry condition, if being unsatisfactory for, returning step 4；If meeting, carry out step 10；Ten, start S honeybee, produce a new explanation and replace former solution, return step 4.The method can simplify remotely sensed image geometric model structure so that it is approaches imaging geometry more accurately.

Description

A kind of remote sensing image rational function model structural optimization method based on ant colony algorithm

Technical field

The present invention relates to remote sensing image rational function model structural optimization method, belong to remotely sensed image geometric model and optimize skill Art field.

Background technology

Remote sensing image imaging geometry model is generally divided into rigorous geometric model and general geometric model two kinds.Right in order to reduce The demand of user's professional standards, protects design of satellites parameter and the technology secret of sensitivity, including accurate sensor geometry simultaneously Parameter and satellite almanac data etc..Thus under normal circumstances, user is only capable of obtaining a kind of general geometric model rational function Model (Rational Function Model, RFM), substitutes the rigorous geometric model based on collinear condition by this.So And, rational function model is highly dependent on model structure, is heavily dependent on again ground control point (Ground Control Points, GCP) quantity and precision.The generally degree of polynomial of each fraction of RFM model may be selected to be 1, and 2,3 times, number of times is the highest, The fine intensity of variation of the geometrical relationship that RFM can describe is the most accurate, simultaneously need to GCP quantity the highest.Generally, Each polynomial whole items in each fraction are wanted to both participate in composition model.Find through Recent study, the mould of the whole item of this selection Type framework mode, not only increases necessary GCP quantity, it is also possible to cause model accuracy to decline, and this is the knot that parameter item crosses redundancy Really.Its reason is, the coordinate relation described by a lot of parameter items is not appropriate for the truth of atural object captured by current image.Cause And, each polynomial parameter item being required for participating in constituting RFM is in optimized selection, it is achieved the optimum structure of RFM model, i.e. Target is to reduce parameter item in order to set up as far as possible on the premise of ensureing model accuracy.But, due to RFM each parameter item also There is no definite physical interpretation, then need to be difficult to framework function clear and definite, direct between object function and the decision variable realized Relation, it is difficult to be optimized by traditional method.Just have more than 30 parameter items for 2 RFM models of single width image and have to be selected Select, then can form the solution of thousands of kinds of various combination modes, thus general trial and error method to be optimized be unpractiaca, these Problem makes the RFM model of acquisition optimum structure become technical barrier.

For many years for the method for RFM parameter item optimized choice with such as the swarm intelligence algorithm such as genetic algorithm, particle cluster algorithm Result relative ideal, but still suffer from the convergence of bigger local extremum and the problem such as solving precision is low, thus exist certain Room for improvement.

Summary of the invention

The present invention is to solve that existing remote sensing RFM model is crossed redundancy when describing imaging geometry because of parameter item and caused The low problem of model accuracy, it is proposed that a kind of remote sensing image rational function model structural optimization method based on ant colony algorithm.

The present invention solves that above-mentioned technical problem adopts the technical scheme that:

Step one: construct a binary set x={x to be solved_i},(i=1,2,…,D,x_i=0,1}), its length D is Rational function model RFM (Rational Function Model, RFM) the parameter item coefficient of remote sensing image to be optimized Number；According to the actual coordinate of known ground control point (Ground Control Points, GCP), will be according to RFM Models computed Going out the image coordinate of corresponding image point, the error between image coordinate corresponding with actual GCP, as income degree function f (x), then Obviously f (x) is the least, then the income degree optimized is the best；And then solution x making f (x) minimize value is found by following steps；

Step 2: initializing ant colony algorithm parameter, described parameter includes Apis sum PN, is divided three classes by Apis, S honeybee PN/2, E honeybee 0, O honeybee PN/2, set S honeybee and start threshold value Limit, solution space dimension D；Set stopping criterion for iteration: Big iterations (maximum iteration number, MCN) and income degree threshold value (Fitness Threshold, FT)；

Step 3: bee colony initializes, makes corresponding one of each Apis solve x, is first found at the beginning of PN/2 by PN/2 S honeybee Begin to solve x_i(i=1,2 ..., PN/2), one S honeybee is set for each solution and starts enumerator Failure (i), and by its zero setting, calculate Each income degree f (x solved_i), store current globally optimal solution x_best, S honeybee all becomes E honeybee, the most now PN/2 E honeybee pair subsequently Answer PN/2 initially to dissolve, subsequently enter the E honeybee optimizing phase；

Step 4: iteration starts: be first the E honeybee optimizing phase: each E honeybee correction its homographic solution x_i(i=1,2,…,PN/ 2) the income degree f (x' corresponding to revised solution, is calculated_i), compare f (xi) and f (x'_i), if f (xi) >=f (x'_i), then retain Current solution xi corresponding to E honeybee is constant, Failure (i)=Failure (i)+1 simultaneously；If f (xi) < f (x'_i) then use x'_iReplace Current solution x corresponding to E honeybee_i, after having replaced, x is still used in unification_iRepresent；

Step 5: calculate each E honeybee and recruit probability $p\left(i\right)=1-\frac{f\left(x_i\right)}{\underset{i=1}{\overset{\mathrm{PN}/2}{\mathrm{\&Sigma;}}}f\left(x_i\right)}(i=\mathrm{1,2},\&CenterDot;\&CenterDot;\&CenterDot;,\mathrm{PN}/2);$

Step 6: O honeybee optimizing phase: PN/2 the solution x obtained after PN/2 E honeybee is optimized_i(i=1,2,…,PN/ 2), each O honeybee, according to roulette criterion, firstly generates the random number p between [0,1]_o, gradually with each E honeybee corresponding to Recruitment Probability p (i) compare, if p_o< p (i), then this E honeybee is modified by current O honeybee, and each O honeybee only revises one E honeybee.With this principle, can guarantee that and recruit the E honeybee that probability is big, i.e. preferably solve the correction number of times obtaining bigger probability；Calculate each By O honeybee revised solution x'_iCorresponding income degree f (x'_i), similar with step 4, compare f (x'_i) and the current income solved Degree f (x_i), if f (xi) >=f (x'_i), then retain and solve x_iConstant, Failure (i)=Failure (i)+1 simultaneously；If f is (x_i)<f (x'_i) then use x'_iReplace and currently solve x_i, after having replaced, x is still used in unification_iRepresent；

Step 7: calculate and compare the income degree f (x of PN/2 solution after O honeybee optimizes_i) (i=1,2 ..., PN/2), Select solution x of corresponding minimum f (x) value_iter, and then compare f (x_iter) with income degree f corresponding to current globally optimal solution (x_best), if f is (x_iter)≥f(x_best), then retain former optimal solution x_bestConstant；If f is (x_iter)<f(x_best), then use x_iterReplace x_bestBecoming new globally optimal solution, after having replaced, x is still used in unification_bestRepresent；

Step 8: judge stopping criterion for iteration situation, if iterations has arrived maximum iteration time MCN, or currently Globally optimal solution correspondence income degree f (x_best) reached or less than income degree threshold value FT, then optimized, by right for the solution of gained Answering former RFM model parameter item, retaining numerical value is the item of 1, gives up the item that numerical value is 0, is the RFM model after optimization；If it is discontented Foot, then carry out step 9；

Step 9: judge that each S honeybee solving correspondence starts whether enumerator Failure (i) value exceedes threshold value Limit, if not Exceed, then return step 4；If exceeding, then carry out step 10；

Step 10: start S honeybee, now set an extra S honeybee, produce new solution x, replace corresponding Failure I () is beyond solution x of Limit_i, become a new E honeybee, and return step 4, carry out next round iteration, until step 8 meets Stopping criterion for iteration.

The invention has the beneficial effects as follows:

The present invention is directed to remotely sensed image model based on rational function model and cross redundancy (over-due to parameter item Parameterization) the low problem of model accuracy caused, utilizes ant colony algorithm to be optimized for RFM model.Utilize The method can efficiently solve traditional method and cross redundancy issue at RFM model parameter item, can obtain so that obtained rational function Model approaches remotely sensed image geometrical relationship more accurately.

This method is used for RFM model topology optimization, and each parameter item of RFM model is corresponded to a binary set to solve by it Certainly object function framework problem, the most corresponding a kind of parameter item assembled scheme of each solution vector, and then take full advantage of ant colony algorithm Colony intelligence characteristic, E honeybee extensively searches for wide area solution, and O is to the preferable excelsior improvement of structure optimization scheme, and S honeybee helps to jump Go out local extremum.Utilize the method can effectively select parameter item crucial in the RFM model structure of remote sensing image, remove superfluous Remainder, thus while simplifying RFM model structure, improve again model accuracy.

Accompanying drawing explanation

Fig. 1 is the flow chart of the inventive method；Whole parameter items are corresponded to the schematic diagram of a binary set by Fig. 2；Fig. 3 For have selected 30GCP from the IKONOS satellite remote-sensing image that ground resolution is 1m/pixel；Fig. 4 is for from ground resolution to be The WorldView2 satellite remote-sensing image of 0.5m/pixel have selected 30 GCP.

Detailed description of the invention

Detailed description of the invention one: combining Fig. 1 and present embodiment is described, the step of present embodiment is as follows:

Step one: construct a binary set x={x to be solved_i},(i=1,2,…,D,x_i=0,1}), its length D is Rational function model RFM (Rational Function Model, RFM) the parameter item coefficient of remote sensing image to be optimized Number；According to the actual coordinate of known ground control point (Ground Control Points, GCP), will be according to RFM Models computed Go out the image coordinate of corresponding image point, error between image coordinate corresponding with actual GCP, as income degree function f (x), then show So f (x) is the least, then the income degree optimized is the best；And then solution x making f (x) minimize value is found by following steps；

Step 2: initializing ant colony algorithm parameter, described parameter includes Apis sum PN, is divided three classes by Apis, S honeybee PN/2, E honeybee 0, O honeybee PN/2, set S honeybee and start threshold value Limit, solution space dimension D；Set stopping criterion for iteration: Big iterations (maximum iteration number, MCN) and income degree threshold value (Fitness Threshold, FT)；

Step 3: bee colony initializes, makes corresponding one of each Apis solve x, is first found at the beginning of PN/2 by PN/2 S honeybee Begin to solve x_i(i=1,2 ..., PN/2), one S honeybee is set for each solution and starts enumerator Failure (i), and by its zero setting, calculate Each income degree f (x solved_i), store current globally optimal solution x_best, S honeybee all becomes E honeybee, the most now PN/2 E honeybee pair subsequently Answer PN/2 initially to dissolve, subsequently enter the E honeybee optimizing phase；

Step 4: iteration starts: be first the E honeybee optimizing phase: each E honeybee correction its homographic solution x_i(i=1,2,…,PN/ 2) the income degree f (x' corresponding to revised solution, is calculated_i), compare f (xi) and f (x'_i), if f (xi) >=f (x'_i), then retain Current solution x corresponding to E honeybee_iConstant, Failure (i)=Failure (i)+1 simultaneously；If f is (x_i)<f(x'_i) then use x'_iReplace Current solution x corresponding to E honeybee_i, after having replaced, x is still used in unification_iRepresent；

Step 5: calculate each E honeybee and recruit probability $p\left(i\right)=1-\frac{f\left(x_i\right)}{\underset{i=1}{\overset{\mathrm{PN}/2}{\mathrm{\&Sigma;}}}f\left(x_i\right)}(i=\mathrm{1,2},\&CenterDot;\&CenterDot;\&CenterDot;,\mathrm{PN}/2);$

Step 6: O honeybee optimizing phase: PN/2 the solution x obtained after PN/2 E honeybee is optimized_i(i=1,2,…,PN/ 2), each O honeybee, according to roulette criterion, firstly generates the random number p between [0,1]_o, gradually with each E honeybee corresponding to Recruitment Probability p (i) compare, if p_o< p (i), then this E honeybee is modified by current O honeybee, and each O honeybee only revises one E honeybee.With this principle, can guarantee that and recruit the E honeybee that probability is big, i.e. preferably solve the correction number of times obtaining bigger probability；Calculate each By O honeybee revised solution x'_iCorresponding income degree f (x'_i), similar with step 4, compare f (x'_i) and the current income solved Degree f (x_i), if f is (x_i)≥f(x'_i), then retain and solve x_iConstant, Failure (i)=Failure (i)+1 simultaneously；If f is (x_i)<f (x'_i) then use x'_iReplace and currently solve x_i, after having replaced, x is still used in unification_iRepresent；

Step 7: calculate and compare the income degree f (x of PN/2 solution after O honeybee optimizes_i) (i=1,2 ..., PN/2), Select solution x of corresponding minimum f (x) value_iter, and then compare f (x_iter) with income degree f corresponding to current globally optimal solution (x_best), if f is (x_iter)≥f(x_best), then retain former optimal solution x_bestConstant；If f is (x_iter)<f(x_best), then use x_iterReplace x_bestBecoming new globally optimal solution, after having replaced, x is still used in unification_bestRepresent；

Step 8: judge stopping criterion for iteration situation, if iterations has arrived maximum iteration time MCN, or currently Globally optimal solution correspondence income degree f (x_best) reached or less than income degree threshold value FT, then optimized, by right for the solution of gained Answering former RFM model parameter item, retaining numerical value is the item of 1, gives up the item that numerical value is 0, is the RFM model after optimization；If it is discontented Foot, then carry out step 9；

Step 9: judge that each S honeybee solving correspondence starts whether enumerator Failure (i) value exceedes threshold value Limit, if not Exceed, then return step 4；If exceeding, then carry out step 10；

Step 10: start S honeybee, now set an extra S honeybee, produce new solution x, replace corresponding Failure I () is beyond solution x of Limit_i, become a new E honeybee, and return step 4, carry out next round iteration, until step 8 meets Stopping criterion for iteration.

Detailed description of the invention two: the RFM model that present embodiment and detailed description of the invention one difference are in step one Such as formula (1):

$r=\frac{P_1(X,Y,Z)}{P_3(X,Y,Z)},c=\frac{P_2(X,Y,Z)}{P_4(X,Y,Z)}---\left(1\right)$

Wherein r, c are remote sensing image coordinate, and X, Y, Z are true geographical coordinate.P_i(X, Y, Z) composition form such as formula (2):

$\begin{array}{cc}{P}_i(X,Y,Z)=a_{i0}+a_{i1}X+a_{i2}Y+a_{i3}Z+a_{i4}\mathrm{XY}& \\ +a_{i5}\mathrm{XZ}+a_{i6}\mathrm{YZ}+a_{i7}{X}^2+a_{i8}{Y}^2& i=\mathrm{1,2,3,4}\\ +a_{i9}{Z}^2+a_{i10}\mathrm{XYZ}+...& \end{array}---\left(2\right)$

The multinomial of the most each fraction high reps takes 2, and makes P₃(X,Y,Z)=P₄(X, Y, Z), such as Fig. 2, by RFM model In the coefficient ai of all parameter items_jCombine, the corresponding two-value number of each coefficient, and form the vector x of an a length of D^j (j=1,2 ... D), if x_jTake 0, represent and give up respective items；If x_jTake 1, represent and retain respective items.

Other step is identical with detailed description of the invention one.

Detailed description of the invention three: the suitability degree letter that present embodiment and detailed description of the invention one difference are in step one Number f (x), computing formula such as formula (3):

$\begin{array}{c}f\left(x\right)=\sqrt{\frac{1}{N}\underset{n=1}{\overset{N}{\mathrm{\&Sigma;}}}({(r_n\&prime;-r_n)}^2+{(c_n\&prime;-c_n)}^2)}\\ =\sqrt{\frac{1}{N}\underset{n=1}{\overset{N}{\mathrm{\&Sigma;}}}({(\frac{P_1\&prime;(X_n,Y_n,Z_n)}{P_3\&prime;(X_n,Y_n,Z_n)}-r_n)}^2+{(\frac{P_2\&prime;(X_n,Y_n,Z_n)}{P_4\&prime;(X_n,Y_n,Z_n)}-c_n)}^2)}\end{array}---\left(3\right)$

N number of GCP is substituted into the RFM model after optimizing, the corresponding image ranks coordinate r solved_n' and c_n'；Formula in (3) is P₁'(X_n,Y_n,Z_n),P₂'(X_n,Y_n,Z_n),P₃'(X_n,Y_n,Z_n),P₄'(X_n,Y_n,Z_n) for the fraction item of RFM model after optimizing, as Form in formula (1).X_n,Y_n,Z_n(n=1,2 ..., N) it is the actual three-dimensional coordinate of N number of known GCP, r_n, c_nIt is respectively (X_n,Y_n,Z_n) Remote sensing image row, column coordinate corresponding to Dian；This income degree functional value is the least, then corresponding Xie Yuehao.

Other step is identical with detailed description of the invention one or two.

Detailed description of the invention four: present embodiment and detailed description of the invention one difference are in step 3 to form initial honeybee The method of group can use randomized or empirical method:

Randomized can realize PN/2 initial solution vector x of stochastic generation according to formula (4)_i ^(j)(i=1,2 ... PN/2, j= 1,2 ... D), in formulaFor meeting equally distributed random number between [0,1]；

Empirical method according to demand or part or all of in PN/2 initial solution of expertise can set one group and fix 's

Other step and detailed description of the invention one, two or three are identical.

Detailed description of the invention five: embodiment and detailed description of the invention one difference are in step 4, for i-th E Solution vector x of honeybee_iCorrection carry out according to formula (5):

$\begin{array}{ccc}{x}_i^{\&prime;\left(j\right)}=[x_i^{\left(j\right)}+{\mathrm{\&lambda;}}_i^{\left(j\right)}]& {\mathrm{ifx}}_i^{\&prime;\left(j\right)}<0& x_i^{\&prime;\left(j\right)}=0\\ & {\mathrm{ifx}}_i^{\&prime;\left(j\right)}& x_i^{\&prime;\left(j\right)}=1\end{array}---\left(5\right)$

J is the decision variable to be revised randomly choosed；Revised solution vectorFor meeting between [-1,1] Equally distributed random number.

Other step and detailed description of the invention one, two, three or four are identical.

Detailed description of the invention six: embodiment and detailed description of the invention one difference are in step 10, when S starts, its The method producing new explanation is to utilize formula (4), and wherein i value starts the right of threshold value Limit for meeting Failure (i) more than S honeybee Should be worth.

Other step and detailed description of the invention one, two, three, four or five are identical.

Present invention is not limited only to the content of the respective embodiments described above, such as formula (4) or formula (5), also can use to On the strategy that rounds or round downwards, have an opportunity to obtain same or more preferable RFM for different remote sensing images and optimize structure.

Application effect:

Apply the inventive method to the RFM model optimization of actual remote sensing image, for Fig. 3, respectively from IKONOS in Fig. 4 Test with 30 GCP selected in WorldView2 satellite remote-sensing image, carry out according to the picture point mean error of formula (3) Evaluate.Before utilizing institute of the present invention extracting method to optimize rational function model, each GCP correspondence picture point mean error in IKONOS image Being 1.71, optimizing post-evaluation error is 0.60；In WorldView2 image, each GCP correspondence picture point mean error is 0.87, optimizes Post-evaluation error is 0.62.

Claims (5)

1. a remote sensing image rational function model structural optimization method based on ant colony algorithm, it is characterised in that its step is such as Under:

Step one: construct a binary set x={x to be solved_i, wherein, i=1,2 ..., D, x_i=0,1}, its length D Number for the rational function model RFM parameter item coefficient of remote sensing image to be optimized；Reality according to known ground control point GCP Border coordinate, will calculate the image coordinate of corresponding image point, the mistake between image coordinate corresponding with actual GCP according to RFM model solution Difference, as income degree function f (x), then obviously f (x) is the least, then the income degree optimized is the best；And then found by following steps F (x) is made to minimize solution x of value；

RFM model such as formula (1):

(r, c)=RFM (X, Y, Z) are wherein

Wherein r, c are remote sensing image coordinate, and X, Y, Z are true geographical three-dimensional coordinate；P_i(X, Y, Z) composition form such as formula (2):

$\begin{array}{c}{P}_i(X,Y,Z)=a_{i0}+a_{i1}X+a_{i2}Y+a_{i3}Z+a_{i4}XY\\ +a_{i5}XZ+a_{i6}YZ+a_{i7}{X}^2+a_{i8}{Y}^2\\ +a_{i9}{Z}^2\end{array},i=1,2,3,4---\left(2\right)$

In formula, the multinomial of each fraction high reps takes 2, and makes P₃(X, Y, Z)=P₄(X, Y, Z), by parameters all in RFM model The coefficient a of item_ijCombine, the corresponding two-value number of each coefficient, and form the vector x of an a length of D^j, wherein, j= 1,2 ... D, if x_jTake 0, represent and give up respective items；If x_jTake 1, represent and retain respective items；

Step 2: initializing ant colony algorithm parameter, described parameter includes Apis sum PN, is divided three classes by Apis, S honeybee PN/2 Individual, E honeybee 0, O honeybee PN/2, set S honeybee and start threshold value Limit, solution space dimension D；Set stopping criterion for iteration: maximum is repeatedly Generation number and income degree threshold value；

Step 3: bee colony initializes, makes corresponding one of each Apis solve x, is first found PN/2 initial solution by PN/2 S honeybee x_i, wherein, i=1,2 ..., PN/2, one S honeybee is set for each solution and starts enumerator Failure (i), and by its zero setting, meter Calculate the income degree f (x of each solution_i), store current globally optimal solution x_best, S honeybee all becomes E honeybee, the most now PN/2 E honeybee subsequently Initially dissolve, subsequently enter the E honeybee optimizing phase for corresponding PN/2；

Step 4: iteration starts: be first the E honeybee optimizing phase: each E honeybee correction its homographic solution x_i, wherein, i=1,2 ..., PN/ 2, calculate the income degree f (x' corresponding to revised solution_i), compare f (x_i) and f (x'_i), if f is (x_i)≥f(x'_i), then retain and work as Solution x corresponding to front E honeybee_iConstant, Failure (i)=Failure (i)+1 simultaneously；If f is (x_i)<f(x'_i) then use x'_iReplace and work as Solution x corresponding to front E honeybee_i, after having replaced, x is still used in unification_iRepresent；

Step 5: calculate each E honeybee and recruit probability

Wherein, i=1,2 ..., PN/2；

Step 6: O honeybee optimizing phase: PN/2 the solution x obtained after PN/2 E honeybee is optimized_i, wherein, i=1,2 ..., PN/ 2, each O honeybee, according to roulette criterion, firstly generates the random number p between [0,1]_o, gradually with each E honeybee corresponding to Recruit Probability p (i) to compare, if p_o< p (i), then this E honeybee is modified by current O honeybee, and each O honeybee only revises an E Honeybee；With this principle, can guarantee that and recruit the E honeybee that probability is big, i.e. preferably solve the correction number of times obtaining bigger probability；Calculate each quilt O honeybee revised solution x'_iCorresponding income degree f (x'_i), similar with step 4, compare f (x'_i) and the current income degree solved f(x_i), if f is (x_i)≥f(x'_i), then retain and solve x_iConstant, Failure (i)=Failure (i)+1 simultaneously；If f is (x_i)<f (x'_i) then use x'_iReplace and currently solve x_i, after having replaced, x is still used in unification_iRepresent；

Step 7: calculate and compare the income degree f (x of PN/2 solution after O honeybee optimizes_i), wherein, i=1,2 ..., PN/2, Select solution x of corresponding minimum f (x) value_iter, and then compare f (x_iter) with income degree f corresponding to current globally optimal solution (x_best), if f is (x_iter)≥f(x_best), then retain former optimal solution x_bestConstant；If f is (x_iter)<f(x_best), then use x_iterReplace x_bestBecoming new globally optimal solution, after having replaced, x is still used in unification_bestRepresent；

Step 8: judge stopping criterion for iteration situation, if iterations has arrived maximum iteration time MCN, or the currently overall situation Optimal solution correspondence income degree f (x_best) reached or less than income degree threshold value FT, then optimized, by former for the solution correspondence of gained RFM model parameter item, retaining numerical value is the item of 1, gives up the item that numerical value is 0, is the RFM model after optimization；If being unsatisfactory for, then Carry out step 9；

Step 9: judge that each S honeybee solving correspondence starts whether enumerator Failure (i) value exceedes threshold value Limit, if not less than, Then return step 4；If exceeding, then carry out step 10；

Step 10: start S honeybee, now set an extra S honeybee, produce new solution x, replaces corresponding Failure (i) and surpasses Go out solution x of Limit_i, become a new E honeybee, and return step 4, carry out next round iteration, until step 8 meets iteration End condition.

Remote sensing image rational function model structural optimization method based on ant colony algorithm the most according to claim 1, it is special Levy the computing formula such as formula (3) of income degree function f (x) being in step one:

$\begin{array}{c}f\left(x\right)=\sqrt{\frac{1}{N}\underset{n=1}{\overset{N}{\&Sigma;}}({\left({r_n}^{\&prime;}-r_n\right)}^2+{\left({c_n}^{\&prime;}-c_n\right)}^2)}\\ =\sqrt{\frac{1}{N}\underset{n=1}{\overset{N}{\&Sigma;}}({\left(\frac{{P_1}^{\&prime;}(X_n,Y_n,Z_n)}{{P_3}^{\&prime;}(X_n,Y_n,Z_n)}-r_n\right)}^2+{\left(\frac{{P_2}^{\&prime;}(X_n,Y_n,Z_n)}{{P_4}^{\&prime;}(X_n,Y_n,Z_n)}-c_n\right)}^2)}\end{array}---\left(3\right)$

N number of GCP is substituted into the RFM model after optimizing, the corresponding image ranks coordinate r solved_n' and c_n'；Formula (3) is P₁'(X_n, Y_n,Z_n),P₂'(X_n,Y_n,Z_n),P₃'(X_n,Y_n,Z_n),P₄'(X_n,Y_n,Z_n) for the fraction item of RFM model after optimizing, in formula (1) Form；X_n,Y_n,Z_nFor the actual three-dimensional coordinate of N number of known GCP, wherein, n=1,2 ..., N, r_n, c_nIt is respectively (X_n,Y_n,Z_n) point Corresponding remote sensing image row, column coordinate；This income degree functional value is the least, then corresponding Xie Yuehao.

Remote sensing image rational function model structural optimization method based on ant colony algorithm the most according to claim 2, it is special Levy and be bee colony initialization employing randomized or empirical method in step 3:

Randomized can realize PN/2 initial solution vector of stochastic generation according to formula (4)Wherein, i=1,2 ... PN/2, j= 1,2 ... D, in formulaFor meeting equally distributed random number between [0,1]；

Empirical method can according to demand or part or all of in PN/2 initial solution of expertise, set one group fixing at the beginning of Begin to solve

Remote sensing image rational function model structural optimization method based on ant colony algorithm the most according to claim 3, it is special Levy and be in step 4, for solution vector x of i-th E honeybee_iCorrection carry out according to formula (5):

$\begin{array}{cc}{x}_i^{\&prime;\left(j\right)}=\&lsqb;x_i^{\left(j\right)}+{\mathrm{\&lambda;}}_i^{\left(j\right)}\&rsqb;& \begin{array}{cc}{\mathrm{ifx}}_i^{\&prime;\left(j\right)}<0& x_i^{\&prime;\left(j\right)}=0\\ {\mathrm{ifx}}_i^{\&prime;\left(j\right)}>0& x_i^{\&prime;\left(j\right)}=1\end{array}\end{array}---\left(5\right)$

J is the decision variable to be revised randomly choosed；For revised solution vector,For meeting uniformly between [-1,1] The random number of distribution.

5. according to the remote sensing image rational function model structural optimization method based on ant colony algorithm described in claim 3 or 4, its Being characterised by step 10, when S starts, its method producing new explanation is to utilize formula (4), and wherein i value is for meeting Failure I () starts the respective value of threshold value Limit more than S honeybee.