Cd

Qa

b

a

ARA

KBUI

1

grtiwrndrtuVnflIc

S

h0

Optik 125 (2014) 7100–7105

Contents lists available at ScienceDirect

Optik

jo ur nal homepage: www.elsev ier .de / i j leo

haotic biogeography-based optimization approach to targetetection in UAV surveillance

ifu Zhanga,b, Haibin Duana,b,∗

State Key Laboratory of Virtual Reality Technology and Systems, Beihang University, Beijing 100191, PR ChinaScience and Technology on Aircraft Control Laboratory, Beihang University, Beijing 100191, PR China

r t i c l e i n f o

rticle history:eceived 3 December 2013ccepted 10 July 2014

eywords:iogeography-based optimization (BBO)nmanned aerial vehicle (UAV)

mage matching

a b s t r a c t

This paper describes a novel chaotic biogeography-based optimization (CBBO) algorithm for target detec-tion by means of template matching to meet the request of unmanned aerial vehicle (UAV) surveillance.Template matching has been widely applied in movement tracking and other fields and makes excel-lent performances in visual navigation. Biogeography-based optimization (BBO) algorithm emerges asa new kind of optimization method on the basis of biogeography concept. The idea of migration andmutation strategy of species in BBO contributes to solving optimization problems. Our work adds chaoticsearching strategy into BBO and applies CBBO in template matching. By utilizing chaotic strategy, thepopulation ergodicity and global searching ability are improved, thus avoiding local optimal solutions

during evolution. Applying the algorithm to resolving template matching problem overcomes the defectsof common image matching. Series of experimental results demonstrate the feasibility and effectivenessof our modified approach over other algorithms in solving template matching problems. Our modifiedBBO algorithm performs better in terms of convergence property and robustness when compared withbasic BBO.

© 2014 Elsevier GmbH. All rights reserved.

. Introduction

Unmanned aerial vehicles (UAV) are employed to execute dan-erous or dull tasks for manned aircraft in both military and civilianealms [1–3]. It is of great necessity to provide UAV with precisearget information for executing various missions, including nav-gation, autonomous landing and attack [4–6]. UAV are currently

idely used in aerial surveillance, which acquires high accuracy ofecognition. Global positioning systems (GPS) are installed on UAVavigation systems for target location generally. Advancements inigital camera technology have made significant performances inecent years with effective image processing software. Much atten-ion has been directed forward vision-based methods for theirsefulness in both commercial and military applications [7–11].ision strategy provides a better idea for safer and swifter UAVavigation [12]. A volume of works on visual control in formation

ight and aerial refueling has been conducted by many researchers.

t has been proposed to employ only natural features of scenes inomputer vision systems during navigation [13]. Vision methods

∗ Corresponding author at: State Key Laboratory of Virtual Reality Technology andystems, Beihang University, Beijing 100191, PR China. Tel.: +86 10 8231 7318.

E-mail address: hbduan@buaa.edu.cn (H. Duan).

ttp://dx.doi.org/10.1016/j.ijleo.2014.08.093030-4026/© 2014 Elsevier GmbH. All rights reserved.

allow us to regulate the aircraft system in accordance with the sur-rounding variations. Therefore, it possesses great potential to adoptvision strategy for UAV navigation.

Multiple proposed vision methods attempt to implement UAVtarget detection through feature extraction. For instance, HoughTransform [14] is applied to edge extraction of targets and makesbetter performances in runway recognition. We apply templatematching in this paper to solve the problem of target detection dur-ing surveillance. Common template matching algorithms generallyrequire searching all the parts in the searching region. These meth-ods call for tremendous calculation and end up with low efficiency.Multiple methods were proposed in succession aimed at accel-erating matching velocity [15]. Feature-based approaches shrinkthe searching regions to enhance efficiency. Optimization-basedmatching involves various optimization algorithms such as theGauss–Newton method [16], the Levenberg–Marquardt methodand multiple swarm intelligence methods. Using the genetic algo-rithm (GA) as a substitution to common ergodic searching has beenmost frequently used for image matching [17]. However, whenGA evolves to a certain extent, the accumulation of local opti-

mization turns into the next generation through intersection andmutation. These properties of GA increase the difficulty of changingthe fitness function values. In this case, the so-called “evolutionarystagnation” turns up. Swarm intelligence has absorbed the sights of

ptik 125 (2014) 7100–7105 7101

mipsorshttAspotoc

Ttcrdilaaiar

Bdst4ac

2

2

tbite

ciamtwaa

H

Tti

Mig

ratio

n ra

te

Emmigration rate

E

Immigration rateλ

μ

Q. Zhang, H. Duan / O

any researchers in recent years. Bonabeau [18] defined the swarmntelligence as “any attempt to design algorithms or distributedroblem-solving devices inspired by the collective behavior ofocial insect colonies and other animal societies.” BBO is an effectiveptimization method that was proposed by Simon in 2008 [19] foresolving engineering problems initially. It is a population-basedearching algorithm on the basis of species’ migration theory. Theabitats (or islands) of species vary adaptively through informa-ion shared between candidate solutions. A typical feature of BBO ishe modification of original population through migration concept.nother distinctive character is that BBO utilizes the fitness of eacholution to determine the immigration and emigration rates. Gooderformances have been demonstrated by BBO on various single-bjective benchmark functions [20–22]. It has also been appliedo resolving problems including sensor selection power systemptimization [23], groundwater detection [24], and satellite imagelassification [25].

Chaos exists widely in nature as a nonlinear phenomenon [26].he chaos theory was established by Lorenz [27] to simulatehe global weather system numerically. Lorenz found that subtlehanges in initial conditions lead to radical difference among finalesults from the subsequent simulation and make long-term pre-iction impossible. We can observe such sensitive dependence on

nitialization not only in complex systems, but also in the simplestogistic equations. The chaotic system can travel all the states inccordance with the objective laws of the system itself withoutny repeat within a certain range. We adopt the chaotic search-ng strategy in this paper to modify the convergence velocity andccuracy of BBO. Furthermore, the proposed CBBO is employed toesolve the template matching problem during UAV surveillance.

The present paper introduces chaotic searching strategy intoBO for solving the problem of template matching. The remain-er of this paper is organized as follows. Section 2 reviews thetandard BBO and CBBO algorithm. Section 3 describes the struc-ure of template matching approached by CBBO, while in Section, comparative experimental results are shown to elucidate thedvantages of our proposed method. The final section contains ouroncluding remarks.

. Chaotic biogeography-based optimization

.1. Standard biogeography-based optimization

BBO is a very recently proposed intelligent optimization methodhat was inspired by the theory of biogeography. The biogeography-ased model mainly concentrates on studying species distribution

n adjacent habitats. Simon [19] related the phenomenon of migra-ion and mutation during species with optimization problems inngineering and proposed BBO.

Each individual is called an island and represents a habitat thatan be separated in BBO. Islands are regarded as points in search-ng space or solutions of optimization problems. Different speciesre distributed on islands, and their values of adaptability deter-ine the scale of species. The habitat suitability index (HSI) is used

o measure the adaptability of habitats. Features that are relevantith HSI, including rainfall, temperature, diversity of vegetation

nd square of lands, can be represented with suitability index vari-bles (SIV). The relationship between HSI and SIV is shown as

SI = f (Habitat) = f (SIV1 − SIVnear) (1)

The crucial steps of BBO are migration and mutation of species.he immigration and emigration of species on islands enable habi-ats to share suitability. Random mutation among the speciesmproves the suitability of habitats furthermore. BBO seeks out the

HSI

Fig. 1. Habitat migration rates vs. habitat suitability index.

optimal solutions in the searching space by mimicking the mecha-nisms above.

2.1.1. MigrationThe geography theory describes how species migrate among

habitats. New species emerge and old ones vanish during the migra-tion. Habitats with higher HSI own more species while low-HSIhabitats contain fewer. Better solutions are similar to high-HSIhabitats, and poor solutions represent low-HSI ones. Generally, thehigh-HSI habitats share higher emigration rates and lower immi-gration rates. High-HSI habitats draw on the verge of saturation andtend to resist immigrations. On the other hand, species on high-HSIhabitats are more likely to emigrate to adjacent habitats and sharetheir properties. Low-HSI habitats share relatively higher immigra-tion rates and lower emigration rates accordingly. Immigration ofnew species improves the diversity of creatures and suitability ofhabitats.

The migration rates of islands are presented in Fig. 1. The emi-gration rate (�) and immigration rate (�) are functions of speciespopulation (S) on islands. Multiple functions can be employed todescribe the relationship, and we adopt the cosine model in thispaper, which is shown as

�k = 12

(cos(

k�

n

)+ 1)

(2)

�k = E

2

(− cos

(k�

n

)+ 1)

(3)

It is shown in Fig. 1 that the immigration rate and emigrationrate change smoothly when species population on the island getsjolly large or small. And the rates vary fast under the circumstancethat species population stays stable.

The maximum immigration and zero emigration emerge whenno species exist in the habitat. The immigration rate decreases tozero while the emigration increases to the maximum value E as thenumber of species increases. The value of S is stabilized at one pointwhen the two rates get equal.

Suppose that the species population of one habitat is S. �S and �Sare the immigration rate and emigration rate respectively, whichare defined as:

�S = I(1 − S/Smax) (4)

�S = ES/Smax (5)

Suppose that E = I for simplicity, so �S + �S = E.BBO makes decisions on whether to change SIVs according to

the values of immigration rate and emigration rate. The specific

procedure of migration is presented as follows:

Step 1: Calculate HSI of each island and sort them in the descendingorder;

7 ptik 12

A

2

svtsstr

m

wab

P

s

P

Satt

102 Q. Zhang, H. Duan / O

Step 2: Choose the islands for immigration in accordance withimmigration rates;Step 3: Choose the adjacent islands for exchange according to theemigration rates;Step 4: Choose one SIV randomly from adjacent islands forexchange;Step 5: Recalculate the HSI of each island and sort them, thusseeking out the largest HSI, corresponding to the optimal solution.Pseudo codes of the migration mechanism are shown in Algorithm1.

lgorithm 1 (Habitat migration).1: for i = 1 to Np2: Use the immigration rate �i to make decisions on whether to

immigrate to Xi

3: if rand(0,1) < �i then4: for j = 1 to Np do5: Select another habitat Xj with probability �j

6: if rand (0,1) < �j then7: Select an SIV from habitat Xj randomly. Replace a

random SIV in Xi with the SIV selected from Xj

8: end9: end10: end11: end

.1.2. MutationThe mutation mechanism contributes to increasing the diver-

ity of habitats during the evolution. The suitability index of speciesaries unexpectedly due to random events including natural disas-ers and diseases. BBO mimics the mutation mechanism to updateolutions. Mutation tends to take place on habitats with lowerpecies count probabilities (PS). Suppose that the amount of habi-ats is N, and the maximum mutation rate is mmax. The mutationate of habitats is computed as:

(S) = mmax(1 − PS/Pmax) (6)

here Pmax = arg max Pi, i = 1, 2, . . ., N, PS represents the prob-bility that one habitats contains S kind of species. The relationshipetween PS and the migration rate can be represented as:

S =

⎧⎪⎨⎪⎩

−(�S + �S)PS + �S+1PS+1 S = 0

−(�S + �S)PS + �S−1PS−1 + �S+1PS+1 1 ≤ S ≤ Smax − 1

−(�S + �S)PS + �S−1PS−1 S = Smax

(7)

Additionally, when species population in one habitat reaches atable state, the corresponding probability is

k =

⎧⎪⎪⎪⎪⎪⎨⎪⎪⎪⎪⎪⎩

P0 = 1

1 +∑n

k=1�0�1�2. . .�k−1

�1�2. . .�k

k = 0

Pk = �0�1�2. . .�k−1

�1�2. . .�k

(1 +∑n

k=1�0�1�2. . .�k−1

�1�2. . .�k

) 1 ≤ k ≤ n

(8)

The process of mutation is relatively simple in BBO. The currentIV of one habitat is displaced by another SIV generated randomlyccording to mutation rate. The mutation strategy of BBO improveshe diversity of species and modifies the solutions of low-HSI habi-ats. Meanwhile, solutions of high-HSI habitats get optimized.

The main procedure of mutation is presented as

Step 1: Calculate PS according to �S and �S;Step 2: Select the SIV for mutation in accordance with PS;Step 3: Displace the current SIV with the randomly generated SIV.

5 (2014) 7100–7105

The pseudo-code of the mutation algorithm is shown inAlgorithm 2.

Algorithm 2 (Habitat mutation).1: for j = 1 to length (SIV) do2: Use the migration rates �i and �i to figure out the probability Pi

Select a variable Xi(SIV) with probability on the basis of Pi

3: if Xi(SIV) is selected then4: Use the generated SIV to replace the current Xi(SIV)5: end6: end

According to the theories proposed in [28], the main merits ofBBO are faster velocity and higher accuracy compared with GA andparticle swarm optimization algorithm (PSO). GA involves moreparameters, and needs to design more complex coding schemesbetween solution space and chromosome space. The evolution pro-cess of GA consists of selection, intersection and mutation, whichincrease the calculation complexity. PSO involves fewer parame-ters, but the optimal solutions are achieved through the change ofvelocity indirectly. The property that similar solutions are proneto gathering together confines the searching accuracy of PSO andleads to local optimal solutions. The features of solutions in BBO canbe changed directly by the self-adapted immigration and emigra-tion, which accelerates the velocity of convergence. In addition, themutation strategy in the evolution adds the variation of solutionsand avoids local optimal solutions.

2.2. BBO based on chaotic search strategy

BBO is prone to falling into local optimal solution and presents alower convergence speed in the latter period of evolution. We adoptthe chaotic searching strategy to resolve the problem. Known as acommon nonlinear phenomenon, chaos appears to be in a mass, butfollows constant laws inside. Chaos is defined as the random motionstates obtained by constant equations. Characterized by random-ness, ergodicity and regularity, chaotic variables present chaoticstates. Ergodicity alludes that chaotic variables travel through allthe states without any repeat. Regularity implies that chaotic vari-ables are derived by constant equations. Due to the possession of allthese characters, the chaotic strategy has become an effective opti-mization technique that can be applied in various fields includingdynamic control and neural network. Additionally, it is becomingincreasingly popular to employ chaotic strategy for the modifica-tion of other algorithms. The chaotic search is implemented withchaotic mapping operators. The common Logistic map is a typicalchaotic system that is utilized in our work with the equation as

zn+1 = �zn(1 − zn); n = 0, 1, 2, . . . (9)

where � is the control parameter. The equation iterates to a fixedsequence with an arbitrary initial value z0 ∈ [0, 1] once the valueof � is determined. When � equals to 4, the system will be incomplete chaos. The ergodicity of chaotic optimization makes theevolutionary process prone to jumping out of local optimal solution.

In this paper, the chaotic search is conducted after the migrationand mutation of species. The value of control parameter is given as� = 4. We attempt to seek out better individuals in the neighbor-hood of the present global optimal solution using chaotic search.Hence, the chaotic solutions are compared with present optimalsolution and we select the better one as the final result in thepresent generation.

3. Image matching based on CBBO

3.1. Principle of template matching

Template matching is viewed as the process of seeking out thecorrect corresponding sub-image in line with the known template.

Q. Zhang, H. Duan / Optik 125 (2014) 7100–7105 7103

Sub-image

Template

IestGs

otmawc

qcoiot

f

woGmm

3

cemaa

Original-image

Fig. 2. Schematic of template matching.

t possesses higher accuracy and sensitivity for target positioningven in complicated conditions. Template matching calculates theimilarity degree between template and the candidate regions, thuso ascertain the most similar sub-image as the optimal matching.enerally, template matching is mainly conducted based on inten-ity values or features.

Template matching can be conducted by computing the degreef correlation between the two for seeking out the matching loca-ion in the search area. Fig. 2 illustrates the schematic of template

atching. Assume that the template stacks upon the original image,nd the part overrode by the template is recorded as sub-image Sx,y,here (x, y) represents the coordinates of the pixel at the upper left

orner of this sub-image.The normalized cross correlation function has been most fre-

uently employed as the fitness function of template matching byalculating the cross correlation coefficient of the template and theriginal image. However, the operation of correlation computings extremely time-consuming in the case of large image scales. Tovercome this shortage, we utilize the absolute intensity differenceo obtain the fitness value as:

(m, n) =∑x,y

|I(x + m, y + n) − T(x, y)| (10)

here I(x,y) is the intensity value of pixel (x, y) in images. The sizef original image I and template T are M × N and m × n respectively.enerally, we confirm the present candidate sub-image as the opti-al matching region when the corresponding f(m,n) reaches theinimum during iteration.

.2. Template matching based on CBBO

Large searching volume and high time complexity remains ahallenging problem in common template matching. Therefore,nhancing the searching speed and accuracy is the key issue inatching. BBO with chaotic searching strategy can be used to

chieve the optimal matching rapidly and efficiently with the stepss follows.

Step 1: Initialization. Generate the initial habitats randomly.Define and initialize the parameters including the maximumspecies population Smax, the maximum immigration rate I and themaximum emigration rate E.Step 2: Calculate the suitability, migration rates and mutationprobability of each habitat.Step 3: Conduct migration in accordance with the migration ratesachieved in the previous steps. Update SIVs and HSI of each habitat.Step 4: Conduct mutation according to the mutation rates andspecies count probabilities. Update SIVs and HSI of each habitat.Step 5: Conduct chaotic search with the logistic map. Update theSIVs and HSI of each habitat.

Step 6: Repeat steps 1–5 until the maximum iteration is reached.End the iteration and treat the searching point corresponding tothe final optimal solution as the best matching point. The flowchart of CBBO-based template matching is shown in Fig. 3.

Fig. 3. Flow chart of template matching using CBBO.

4. Experimental results and analysis

4.1. Benchmark function simulation

The main contribution of this paper is the introduction of chaoticsearch strategy in BBO for image matching in UAV surveillance. Itis needed to compare the performances of BBO with other evolu-tion algorithms (EA). Series of experiments on benchmark functionsare carried out to test the feasibility and effectiveness of BBO. Thesimulation is implemented in Matlab 2012b, and executed by thecomputer with 2.0 GHz CPU, 2 GB memory, and operation system ofWindows 7. The expressions of the benchmark functions are shownin Table 1. We select GA, PSO and ant colony optimization algorithm(ACO) as comparisons with BBO. In the algorithms, the populationsize is set to be 50, the maximum iteration is 500, and the BBO initialmutation probability is 0.05.

Fig. 4 provides a view into the comparison of BBO and otherclassical optimization methods. The comparative results of Ack-ley and Sphere functions indicate that BBO performs significantlybetter than PSO and GA. GA has not performed best due to the com-plex operations such as mutation. The convergence velocity of PSOappears significantly worse in spite of its simplicity. BBO failed toemerge as a clear winner in the case of Griewank function comparedwith ACO. However, it is more time consuming for ACO to obtainthe slightly better results than BBO. The optimal solutions achievedby the other three methods appear worse than BBO in most cases.The experimental results illustrate that BBO has superior searchingability to other methods both on convergence speed and accuracy.

4.2. Template matching

In this work, template matching is applied for UAV surveillance,which acquires high accuracy of recognition. The experimentalresults on image matching using CBBO algorithm are given inFigs. 5 and 6 respectively. Buildings on the ground (tower in Fig. 5and radars in Fig. 6) are selected as the targets to be matched. Com-parisons have been made on both of the results with the standardBBO. Both BBO and CBBO are employed to minimize the intensitydifference between template and sub-regions of the original image.The population size of standard BBO and CBBO were both 200, andthe maximum iteration is 50.

The results of Case 2 are shown asFrom Figs. 5 and 6, it is obvious that both targets can be correctly

recognized using BBO and CBBO in template matching. Especially,the pictorial representations in Fig. 6(c) present the successful

7104 Q. Zhang, H. Duan / Optik 125 (2014) 7100–7105

Table 1Test functions.

Function Equation Domain

Ackley f1(x) = 20 + e − 20e

(−0.2

√(1/D)∑D

i=1x2

i

)− e

(1/D)∑D

i=1cos(2�xi ) −32.768 ≤ xi ≤ 32.768

Griewank f2(x) =D∑

i=1

x2i

4000 −D∏

i=1

(cos

(xi√

i

))+ 1 −600 ≤ xi ≤ 600

Quartic f3(x) =

(D∑

i=1

xi

)2

−100 ≤ xi ≤ 100

Sphere f4(x) =D∑

i=1

x2i

−100 ≤ xi ≤ 100

0 50 100 150 200 250 300 350 400 450 5000

10

20

30

40

50

60

Iter

Fitn

ess

valu

e

BBOGAACOPSO

(c) Quartic

0 50 100 150 200 250 300 350 400 450 5000

20

40

60

80

100

120

Iter

Fitn

ess

valu

e

BBOGAACOPSO

(d) Sphere

0 50 100 150 200 250 300 350 400 450 5000

5

10

15

20

25

Iter

Fitn

ess

valu

e

BBOPSOGAACO

(a) Ackley

0 50 100 150 20 0 250 300 35 0 400 450 5000

100

200

300

400

500

600

Iter

Fitn

ess

valu

e

BBOPSOGAACO

(b) Griewank

Fig. 4. Comparison of BBO and other methods on benchmark functions.

Fig. 5. Experimental results of Case 1.

matching in the presence of four similar radars, which is difficult tomake distinctions using other methods. It is obvious that althoughboth the standard BBO and the proposed CBBO succeed in imple-menting successful matching and quickly converge to the optimalresults, our modified method has shown more optimal results anda higher rate of convergence. Chaotic parallelism provides strongercomputing power for template matching where simple calculationson larger set of data are required. The application of chaotic searchstrategy enables BBO to search around the optimal solution in eachgeneration randomly, which obviously accelerates the convergencevelocity. The experimental results have confirmed the effectivenessof the modified method for resolving template matching problemsunder various circumstances. Meanwhile, it is supposed to be notedthat the length of time cost for matching is lengthen in CBBO due

to the chaotic search.

Q. Zhang, H. Duan / Optik 12

5

pwontan

tafoaoa

A

B#uC2

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[method, Exploit. Appl. Softw. 27 (9) (2008) 91–94.

Fig. 6. Experimental results of Case 2.

. Conclusions

In this paper, we proposed a novel chaotic BBO approach to tem-late matching for UAV surveillance. The chaotic searching strategyas adopted for the acceleration of convergence velocity and the

ptimization of results. Series of experiments were conducted, andumerical experimental results between the proposed method andraditional methods are also given to demonstrate the feasibilitynd effectiveness of CBBO. BBO shows promising potential but stilleeds further theoretical study and relative experiments.

In the near future, we can adopt several new strategies to avoidhe local optimization of BBO. For instance, the migration mech-nism can be studied deeply for the information exchange. Ouruture work will also focus on combining template matching withther vision-based processes for UAV surveillance. In addition, BBOnd other similar population-based methods will be applied inther regions in computer vision, including image segmentationnd target recognition.

cknowledgements

This work was partially supported by National Keyasic Research Program of China (973 Project) under grant

2013CB035503, Natural Science Foundation of China (NSFC)nder grant # 61333004 and #61273054, andNational Magneticonfinement Fusion Research Program of China under grant #012GB102006.

[

[

5 (2014) 7100–7105 7105

References

[1] H.B. Duan, Q.N. Luo, G.J. Ma, Y.H. Shi, Hybrid particle swarm optimizationand genetic algorithm for multi-UAVs formation reconfiguration, IEEE Comput.Intell. Mag. 8 (3) (2013) 16–27.

[2] Q.N. Luo, H.B. Duan, An improved artificial physics approach to multipleUAVs/UGVs heterogeneous coordination, Sci. China Technol. Sci. 56 (10) (2013)2473–2479.

[3] H.B. Duan, X.X. Wei, Z.N. Dong, Multiple UCAVs cooperative air combat simula-tion platform based on PSO, ACO, and game theory, IEEE Aerosp. Electron. Syst.Mag. 28 (11) (2013) 12–19.

[4] H.B. Duan, Y.M. Deng, X.H. Wang, C.F. Xu, Small and dim target detection via lat-eral inhibition filtering and artificial bee colony based selective visual attention,PLOS ONE 8 (8) (2013) e72035.

[5] H.B. Duan, Y.M. Deng, X.H. Wang, F. Liu, Biologically eagle-eye-based visualimaging guidance simulation platform for unmanned flying vehicles, IEEEAerosp. Electron. Syst. Mag. 28 (12) (2013) 36–45.

[6] T. Wasserman, J. Lennon, E. Atkins, A modal operations paradigm for robustvision-based astronaut following, J. Aerosp. Comput. Inf. Commun. 3 (12) (2006)603–618.

[7] X.H. Wang, H.B. Duan, Predator–prey biogeography-based optimization for bio-inspired visual attention, Int. J. Comput. Intell. Syst. 6 (6) (2013) 1151–1162.

[8] B. Barber, T. McLain, B. Edwards, Vision-based landing of fixed-wing miniatureair vehicles MIT Lincoln laboratory, J. Aerosp. Comput. Inf. Commun. 6 (3) (2009)207–226.

[9] Y.M. Deng, H.B. Duan, Hybrid C2 features and spectral residual approach toobject recognition, Optik 124 (8) (2013) 3590–3595.

10] M. Campbell, M. Wheeler, Vision-based geolocation tracking system for unin-habited aerial vehicles, J. Guid. Control Dyn. 33 (2) (2010) 521–532.

11] X.H. Wang, H.B. Duan, D.L. Luo, Cauchy biogeography-based optimization basedon lateral inhibition for image matching, Optik 124 (22) (2013) 5447–5453.

12] J. Effland, B. Seanor, Y. Gu, M. Napolitano, Application of machine vision inunmanned aerial systems for autonomous target tracking, in: AIAA Guidance,Navigation and Control Conference and Exhibit, Honolulu, Hawaii, Aug. 18–21,2008.

13] E. Frew, T. McGee, Z.W. Kim, et al., Vision-based road-following using a smallautonomous aircraft, in: Proceedings of 2004 Aerospace Conference, IEEE, vol.5, 2004, pp. 3006–3015.

14] J. Shang, Z. Shi, Vision-based runway recognition for UAV autonomous landing,Int. J. Comput. Sci. Netw. Secur. 7 (3) (2007) 112–117.

15] Y. Zhang, D. Liu, J. Xin, Research of image correlation matching method basedon CPSO, J. Electron. Inf. Technol. 30 (3) (2008) 529–530.

16] P.K. Subramanian, Gauss–Newton methods for the complementarity problem,J. Optim. Theory Appl. 77 (3) (1993) 467–482.

17] P. Chalermwat, T. El-Ghazawi, J. Le Moigne, Two-phase genetic algorithm-basedimage registration on parallel clusters, J. Future Gener. Comput. Syst. 17 (3)(2001) 467–476.

18] E. Bonabeau, M. Dorigo, G. Theraulaz, Swarm intelligence: from natural to arti-ficial systems, Connect. Sci. 14 (2) (2002) 163–164.

19] H.P. Ma, X.Y. Ruan, Z.X. Pan, Handling multiple objectives with biogeography-based optimization, Int. J. Autom. Comput. 9 (1) (2012) 30–36.

20] D.W. Du, D. Simon, M. Ergezer, Biogeography-based optimization combinedwith evolutionary strategy and immigration refusal, in: Proceedings of IEEEInternational Conference on Systems, Man, and Cybernetics, IEEE, San Antonio,USA, 2009, pp. 997–1002.

21] M. Ergezer, D. Simon, D.W. Du, Oppositional biogeography-based optimization,in: Proceedings of the IEEE Conference on Systems, Man, and Cybernetics, IEEE,San Antonio, USA, 2009, pp. 1009–1014.

22] H.P. Ma, S.H. Ni, M. Sun, Equilibrium species counts and migration model trade-offs for biogeography-based optimization, in: Proceedings of the 48th IEEEConference on Decision and Control, IEEE, Shanghai, PRC, 2009, pp. 3306–3310.

23] R. Rarick, D. Simon, F.E. Villaseca, B. Vyakaranam, Biogeography-based opti-mization and the solution of the power flow problem, in: Proceedings of IEEEConference on Systems, Man, and Cybernetics, IEEE, San Antonio, USA, 2009,pp. 1003–1008.

24] H. Kundra, A. Kaur, V. Panchal, An integrated approach to biogeography basedoptimization with case based reasoning for retrieving groundwater possibil-ity, in: Proceedings of the 8th Annual Asian Conference and Exhibition onGeospatial Information, Technology and Applications, Singapore, 2009.

25] V.K. Panchal, P. Singh, N. Kaur, H. Kundra, Biogeography based satellite imageclassification, Int. J. Comput. Sci. Inf. Secur. 6 (2) (2009) 269–274.

26] H.Y. Zhang, Research on classical image matching algorithm and its improved

27] Y.X. Yang, D. Liu, J. Xin, Research of image correlation matching method basedon CPSO, J. Electron. Inf. Technol. 30 (3) (2008) 529–533.

28] D. Simon, Biogeography-based optimization, IEEE Trans. Evol. Comput. 12 (6)(2008) 702–713.