CN105760813A - Unmanned aerial vehicle target detection method based on plant branch and root evolution behaviors - Google Patents

Abstract

The invention discloses an unmanned aerial vehicle target detection method based on plant branch and root evolution behaviors. The method comprises the following steps: 1, an aerial image is acquired; 2, the edge potential field of the image is calculated; 3, parameters of a branch and root search optimization method are initialized; 4, a cost function is designed; 5, a stolon search operator and a root search operator are used for optimization; and 6, the optimization result is stored and verified. On the premise of the given target image, the method of the invention can be used for finding the position of the target image in an acquired image. The method has high accuracy and robustness, and a basis is provided for realizing situation assessment and automatic decision making on the unmanned aerial vehicle.

Description

A kind of unmanned plane target detection method based on plant ramose root evolved behavior

[technical field]

The present invention is a kind of autonomous object detection method of unmanned plane based on plant ramose root evolved behavior, belongs to UAV Intelligent perception and fields of measurement.

[background technology]

Unmanned plane (UnmannedAerialVehicle, UAV) it is a kind of unpiloted aviation aircraft, also referred to as air-robot, nowadays it goes on the arena of history by last 100 years, mainly has a characteristic that flexible operation, good concealment, survival ability are strong etc..Since the nineties in last century, the development of unmanned plane is day by day flourishing, especially in Military Application.Start most unmanned plane and be only used as target drone, later along with the needs of technical progress and war, scientific and technical personnel begin at and are mounted with various equipment on unmanned plane so that it is possessed the ability performing the tasks such as intelligence reconnaissance, electronic countermeasure, radar decoy, situation of battlefield monitoring and target attack.Military unmanned air vehicle advanced in the world at present mainly includes the global hawk of the U.S., X-47B unmanned plane etc..In civilian, unmanned plane application in meteorological observation, geophysical surveying, resource exploration, traffic patrolling and forest fire protection etc. is also more and more extensive.

At present, unmanned plane just develops towards microminiaturized, modularity and function diversification, intelligentized direction.In order to realize unmanned plane path planning, the function such as barrier is evaded, safe landing, target identification, search, and finally realize intellectuality, make unmanned plane have the ability of autonomous location and independent navigation, it appears particularly important.This is accomplished by unmanned plane when the task of execution, can the status information of constantly perception self, and the relative information of surrounding, and assisted by multi information and process, effectively obtain self position and attitude information in real time, thus carrying out intelligent decision according to himself posture information.

At present, vision sensor is widely used on unmanned plane.Unmanned plane utilizes optical pickocff to obtain image, then passes through and image is processed, it is possible to carry out target detection, identification, in order to unmanned plane carries out Situation Assessment.Can be obtained the Position and orientation parameters of unmanned plane by optical pickocff, thus providing guidance information for unmanned plane, this mode is a kind of feasible unmanned plane independent navigation solution, and has become as the focus of research both at home and abroad.Compared to other pose measuring methods, camera or video camera that vision measurement uses have the advantages that price is low, volume is little, power consumption is weak, convenient and practical, it can fully obtain scene information around, simultaneously because the service band of vision measurement system is special, it is made to be not easily susceptible to the impact of electrical measurement interference, there is extraordinary on-the-spot application, obtain research widely and application in recent years.Computer vision methods is used for the target detection of unmanned plane by present patent application.

In image detecting method, representative and coupling based on shape are very important branching methods, and are widely used in target identification technology.Generally, Shape Matching Technique generally includes two steps: feature extraction and similarity detection.Having had multiple method for detecting the similarity of two-dimensional image at present, main method has the coupling based on geometric moment, based on the coupling etc. of Hausdorff distance, and the matching process of distinguished point based.But geometric moment existing characteristics expressive faculty is more weak and the shortcoming such as noise-sensitive, when picture noise big of low quality time, the object detection method effect based on geometric moment is poor, and the matching process calculation cost based on Hausdorff distance is high, speed is slow, and the matching process rotational invariance of distinguished point based is poor.

So that object detection method has stronger robustness and real-time, the edge potential function method that the present invention adopts is a kind of new similarity detection method.This method is inspired by charging charge generation electromotive force to obtain, and a series of charging charge can produce electromotive force in homogenous medium, and the size of electromotive force and density depend on distance and the electric field dielectric constant of tested point and field source.In image-detection process, the template to be matched being searched can be equivalent to the test object attracted by a series of charged particles.Therefore, if the object in image is more high with template image similarity, the attraction energy of fringing field suffered by it is also more big.Owing to remaining such as the key character such as marginal position, tension force in the calculating process of edge potential field, therefore it can as the similarity detection method of form fit problem.Simultaneously as edge potential field is a Solving Multimodal Function, the present invention adopts ramose root chess game optimization method to carry out edge potential field function optimization by the mode of iteration optimizing.

Ramose root chess game optimization method (Runner-RootAlgorithm, RRA) it is a kind of novel heuristic colony intelligence optimized algorithm proposed by F.Merrikh-Bayat in 2015, this algorithm is subject to the inspiration of plant propagation, namely plant is in the process of breeding, successively finds water resource and mineral according to the breeding of stolon and root.According to this characteristic, ramose root chess game optimization algorithm proposes operator two kinds corresponding, and respectively stolon searching operators and root searching operators, carry out this characteristic of simulating plant breeding, and both operators are combined solution optimization problem.

[summary of the invention]

1, goal of the invention:

The present invention proposes a kind of autonomous object detection method of unmanned plane based on plant ramose root evolved behavior, its objective is to provide a kind of object detection method being applied on unmanned aerial vehicle platform, for unmanned plane Situation Assessment and offer basis of making decisions on one's own.

The method writes corresponding program by Matlab, and the specific objective in Aerial Images carries out accurately detection and location.

2, technical scheme:

The present invention utilizes the features such as colony intelligence optimization method ability of searching optimum is strong, application is wide, develops a kind of autonomous object detection method of unmanned plane developed based on plant ramose root, and the step of the method is as follows:

Step one: obtain Aerial Images

By UAV flight's industrial camera, carry out Real-time Collection, obtain Aerial Images.

Step 2: calculate image border potential field

Read the image that industrial camera collects, image be first converted into gray-scale map and carry out medium filtering, recycle sobel operator extraction image border, finally calculate the edge potential field of image, be shown below:

$EPF(x,y)=\frac{Q_{eq}(x_i,y_i)}{4{\mathrm{\π\ϵ}}_{eq}}\underset{(x_i,y_i)\∈W}{\Σ}\frac{1}{\sqrt{{(x-x_i)}^2+{(y-y_i)}^2}}---\left(1\right)$

ε in above formula_eqValue and picture background environmental correclation, if image edge relatively horn of plenty or signal noise ratio (snr) of image are low, then by ε_eqTaking less value, usual value is 1.If image scene is simple and signal to noise ratio is high, then by ε_eqTaking bigger value, usual value is 10.IfFor the equivalent charge number of each marginal point, x_iAnd y_iRepresent each marginal point abscissa in the picture and vertical coordinate respectively.(x is y) (x, y) the edge potential function value at place for original image coordinate to EPF.Therefore in image, the edge potential field of each pixel can be obtained by (1) formula by the edge graph of image.

Step 3: initialize ramose root chess game optimization method parameter

(1) parameters optimization dimension D is initialized

This method finds target by ramose root chess game optimization method in two dimensional image, the rotation process of utilize matlab program that target is carried out zoom operations that yardstick is 0.8-1.2 times and 0-360 degree, again the top left corner pixel point of target image is updated in edge potential field function, find the place that edge potential field functional value is maximum, i.e. target location in the picture.In the process, it is necessary to utilize the ramose root chess game optimization method scale factor to target, the anglec of rotation, and top left corner pixel point abscissa and these 4 parameters of vertical coordinate to be optimized, so D is 4.

(2) population quantity N is initialized_pop

Population quantity N_popThe effect of optimization of colony intelligence optimized algorithm is affected very big.When population quantity is larger than 100, faster but calculation cost is higher for ramose root chess game optimization method convergence rate；When population quantity less less than 100 time ramose root chess game optimization method be easily absorbed in local convergence.Therefore, the setting of population quantity should select according to practical problem, by arranging different population quantities, comparative test result, selects suitable population quantity, takes into account rapidity and the accuracy of ramose root optimization method.

(3) population position is initialized

In the method, it is necessary to random initializtion population position in solution space.If X_lFor the lower limit set of solution space, X_uFor the upper limit set of solution space, then each individuality in population initializes by following formula:

X_i=X_l+rand·(X_u-X_l)(2)

In formula, rand is the random number between 0 to 1.

(4) ramose root chess game optimization method preset parameter is set

Ramose root chess game optimization method is made up of stolon searching operators and root searching operators.In stolon searching operators, plant carries out flourish by growing the stolon made new advances, in D dimension space, and the positional information X of i-th plant_iEvery generation updates once, and concrete replacement criteria is shown below:

X_daughter(t)=X_mother(t)+d_runner×r₁(3)

d_runner=x_u-x_l(4)

In formula, X_motherT () represents i-th plant position in D dimension space in after the t time iterative process, d_runnerRepresent the ultimate range of maternal plant and sub-strain, be normally taken from the transformation range of variable, r₁Represent the random number of 0 to 1, X_daughterT offspring individual that () generates through stolon reproduction for plant, x_uFor the upper dividing value of independent variable, x_lFloor value for independent variable.In order to reduce the calculation cost of algorithm, when t is much better than t-1 for the fitness function of sub-strain for the fitness function of sub-strain, it is possible to root Local Search need not be carried out, for function minimum optimization problem, it is judged that shown in criterion such as formula (3):

$\left|\frac{\min f\left(x_{daughter}\right(t))-\min f\left(x_{daughter}\right(t-1))}{\min f\left(x_{daughter}\right(t-1))}\right|\≥tol---\left(5\right)$

In formula, tol is the threshold value that ramose root chess game optimization method sets in advance, is set to a number less than 1 generally according to experience.

minf(x_daughter(t)) it is the sub-strain of the best of the t time iteration, minf (x_daughter(t-1)) it is the sub-strain of the best of the t-1 time iteration, if the formula of being unsatisfactory for (5), carries out stolon Local Search.

x_perturbed,k=diag (1,1 ..., 1+d_runnern_k,1,...,1)*x_{daughter,best}(t)(6)

By (6) formula it can be seen that stolon Local Search mode is antithetical phrase strain one-dimensional carry out disturbing after can obtain the sub-strain of disturbance, wherein n_kBeing 0 for average, variance is the random number of 1.x_perturbed,kFor disturbing sub-strain, x_{daughter,best}T () is optimum sub-strain value after t iteration, if disturbance sub-strain fitness function value is better than the fitness function value of optimum sub-strain, then optimum sub-strain is the sub-strain of disturbance.

By (3) formula it can be seen that the ultimate range d of maternal plant and sub-strain_runnerSize directly control the search amplitude of stolon searching operators, generally by d_runnerIt is taken as the scope of independent variable.

After plant growing goes out sub-strain, current region can be carried out Local Search by the root of every sub-strain, finds the position that water resource and mineral are the abundantest, and root search formula is as follows:

x_perturbed,k=diag (1,1 ..., 1+d_rootn_k,1,...,1)*x_{daughter,best}(t)(7)

In above formula, d_rootFor root hunting zone, it is typically set at a less value, it is possible to be set to the number less than 1.Similar with stolon Local Search, if the fitness function value of the disturbance individuality after Local Search is better than the fitness function value of best sub-strain, then best sub-strain is replaced with the sub-strain of disturbance.

After having carried out the basic operator of above-mentioned two, ramose root chess game optimization method is using sub-for the best of this generation strain as first maternal plant of future generation, and all the other maternal plants in next round iterative process are generated by roulette mode by current sub-strain.By being constantly iterated renewal, until meeting the condition stopping iteration, ramose root chess game optimization method obtains target optimum position in gathering image.The overall flow figure of ramose root chess game optimization method is as shown in Figure 1.

(5) ramose root chess game optimization method iterations is set

The effect of optimization of swarm intelligence algorithm is affected very big by ramose root chess game optimization method iterations.When ramose root chess game optimization method iterations is very few, ramose root chess game optimization method can be made to be not reaching to optimum and then to stop；When iterations is excessive, ramose root chess game optimization method is likely to restrain already.For practical problem, should first analyze its average rate of convergence, the iterations of ramose root searching method is empirically set.

Step 4: design cost function

Cost function is the core of intelligent optimization method, the effectiveness of determining method and robustness.First target is rotated and change of scale by this method, calculates the potential function value that target image edge produces in gathering the edge potential field that image produces, and computing formula is as follows:

$fitness\left(X_i\right)=\frac{1}{N}\underset{i=1}{\overset{N}{\Σ}}\left\{EPF(x_i,y_i)\right\}---\left(8\right)$

In formula, N is the sum of all pixels that object edge overlaps with collection image border, can obtain trying to achieve cost function value by (1) formula and (8) formula.

Step 5: utilize stolon searching operators and root searching operators optimizing

Utilize initialized colony and the position of history colony, according to formula (3) and formula (4), current group is carried out stolon searching operators and carry out optimizing operation.In order to reduce the calculation cost of ramose root chess game optimization method, when t meets formula (5) for the cost function of sub-strain, then do not carry out stolon Local Search, otherwise then carry out stolon Local Search according to formula (6) and obtain the sub-strain of disturbance, if disturbance sub-strain cost function value is better than the cost function value of optimum sub-strain, then optimum sub-strain is replaced with the sub-strain of disturbance.

Similar with stolon Local Search, according to formula (7) to carrying out root Local Search, if the cost function value of the disturbance individuality after Local Search is better than the cost function value of best sub-strain, then best sub-strain is replaced with the sub-strain of disturbance.

After having carried out stolon searching operators and root searching operators optimizing, ramose root chess game optimization method is using sub-for the best of this generation strain as first maternal plant of future generation, and the maternal plant in next round iterative process is generated by roulette mode by current sub-strain.Whole algorithm is constantly iterated updating, and when the maximum iteration time that number of run sets more than algorithm, stops searching process.

Step 6: store optimum results and verify

After ramose root chess game optimization method stops optimizing, select the individuality that global optimum is corresponding, this result is preserved, obtain final object detection results, and preserve the iterativecurve of lower searching process.

3, advantage and effect:

The present invention proposes a kind of unmanned plane target detection method based on plant ramose root evolved behavior, its objective is to provide a kind of object detection method.Under the premise of given target image, it is possible to use this method finds out target image position in gathering image.Originally there is stronger accuracy and robustness, to realizing the Situation Assessment of unmanned plane and offer basis of making decisions on one's own.

[accompanying drawing explanation]

Fig. 1 is ramose root chess game optimization method overall flow figure.

Fig. 2 independently repeats experiment iterativecurve figure for 20 times.

Fig. 3 is best iterativecurve figure.

Fig. 4 is average evolution curve chart.

Number in the figure and symbol description are as follows:

Nc optimized algorithm iterations；N is unsatisfactory for condition (no)；Y satisfies condition (YES)

[detailed description of the invention]

The effectiveness of object detection method proposed by the invention is verified below by a concrete unmanned plane target detection example.This experimental calculation machine is configured to i7-5600U processor, 2.60Ghz dominant frequency, 8G internal memory, and software is MATLAB2015a version.

This example to implement step as follows:

Step one: obtain Aerial Images

By UAV flight's industrial camera and airborne processor, run camera continuous acquisition program, the shooting of unmanned plane during flying to certain altitude is obtained Aerial Images

Step 2: calculate image border potential field

Read the image that collected by camera arrives, it is first converted into gray-scale map and carries out medium filtering, recycle sobel operator extraction image border, calculate the edge potential field of image according to formula (6).In the present invention, ε_eqTake 1, Q_eq(x_i,y_i) it is 20, therefore in image, the edge potential field of each pixel can be obtained by the edge graph of image.

Step 3: initialize ramose root chess game optimization method parameter

(1) parameters optimization dimension D is initialized

This method finds target by ramose root chess game optimization method in two dimensional image, it is necessary to target is zoomed in and out and rotation process, find target position in the picture, so D is 4.

(2) population quantity N is initialized_pop

Population quantity N_popThe effect of optimization of colony intelligence optimization method is affected very big.When population quantity is bigger, faster but calculation cost is higher for algorithm the convergence speed；When population quantity is less, algorithm is easily absorbed in local convergence.Therefore, the setting of population quantity should select according to practical problem, takes into account rapidity and the accuracy of algorithm.In the present invention, population quantity N_popIt is set to 200.

(3) population position is initialized

In the method, it is necessary to random initializtion population position in solution space.If X_lFor the lower limit set of solution space, X_uUpper limit set for solution space.X in the present invention_l=[1,1,0,10], X_u=[256,256,360,14], individual by formula (7) initialization in first time iterative process.

(4) algorithm preset parameter is set

For colony intelligence optimized algorithm, algorithm performance can be subject to the impact of some important parameters.In this method, maternal plant and sub-strain distance range d_runnerSize directly control the search amplitude of stolon searching operators, root hunting zone d_rootThe Local Search amplitude of control algolithm.For target detection problems, d_runnerIt is set to 5, d_rootBeing 1, the tol in formula (5) is set to 0.01.

(5) ramose root chess game optimization method iterations N is set_c

Ramose root chess game optimization method iterations is very big on effect of optimization impact.When iterations is very few, algorithm can be made to be not reaching to optimum and then to stop；When iterations is excessive, algorithm is likely to restrain already.For practical problem, should first analyze its average rate of convergence, then the iterations of algorithm is set.In the present invention, maximum iteration time is set to 100.

Step 4: design cost function

Cost function is the core of intelligent optimization algorithm, determines the quality of algorithm performance.First target is rotated and change of scale by this method, calculates the potential function value that target image edge produces in gathering the edge potential field that image produces, shown in computing formula such as formula (8).

Step 5: utilize stolon searching operators and root searching operators optimizing

Utilize initialized colony and the position of history colony, according to formula (3) and formula (4), current group is carried out stolon searching operators and carry out optimizing operation.In order to reduce the calculation cost of algorithm, when t meets formula (5) for the cost function of sub-strain, then do not carry out stolon Local Search, otherwise then carry out stolon Local Search according to formula (6) and obtain the sub-strain of disturbance, if disturbance sub-strain cost function value is better than the cost function value of optimum sub-strain, then optimum sub-strain is replaced with the sub-strain of disturbance.

Similar with stolon Local Search, according to formula (7) to carrying out root Local Search, if the cost function value of the disturbance individuality after Local Search is better than the cost function value of best sub-strain, then best sub-strain is replaced with the sub-strain of disturbance.

After having carried out the basic operator of above-mentioned two, ramose root optimized algorithm is using sub-for the best of this generation strain as first maternal plant of future generation, and the maternal plant in next round iterative process is generated by roulette mode by current sub-strain.Whole algorithm is constantly iterated updating, and when the maximum iteration time that number of run sets more than algorithm, stops searching process.

Step 6: store optimum results and verify

After algorithm stops optimizing, selecting the individuality that global optimum is corresponding, this result preserved, obtain final object detection results, target area broken box goes out, and preserves the iterativecurve of lower searching process.In order to verify robustness and the accuracy of this method, carried out altogether 20 times and independently repeated experiment, each iterativecurve as in figure 2 it is shown, best iterativecurve as it is shown on figure 3, average evolution curve as shown in Figure 4.

Be can be seen that by the simulation result in example, target can be carried out detecting in many experiments process and position by goal approach that the present invention proposes exactly, the algorithm of target detection that the visible present invention proposes has significantly high accuracy and robustness, can be widely applied to UAV Intelligent decision-making and Situation Assessment field.

Claims (8)

1., based on the autonomous object detection method of unmanned plane of plant ramose root evolved behavior, the step of the method is as follows:

Step 1: obtain Aerial Images

By UAV flight's industrial camera, carry out Real-time Collection, obtain Aerial Images；

Step 2: calculate image border potential field

Read the image that industrial camera collects, image be first converted into gray-scale map and carry out medium filtering, recycle sobel operator extraction image border, finally calculate the edge potential field of image, as shown in formula (1):

$EPF(x,y)=\frac{Q_{eq}(x_i,y_i)}{4{\mathrm{\π\ϵ}}_{eq}}\underset{(x_i,y_i)\∈W}{\Σ}\frac{1}{\sqrt{{(x-x_i)}^2+{(y-y_i)}^2}}---\left(1\right)$

ε in formula (1)_eqValue and picture background environmental correclation, Q_eq(x_i,y_i) for the equivalent charge number of each marginal point, x_iAnd y_iRepresent each marginal point abscissa in the picture and vertical coordinate respectively；(x is y) (x, y) the edge potential function value at place for original image coordinate to EPF；Therefore, in image, the edge potential field of each pixel is obtained by formula (1) by the edge graph of image；

Step 3: initialize ramose root chess game optimization method parameter

Step 3.1 initializes parameters optimization dimension D

In two dimensional image, target is found by ramose root chess game optimization method, the rotation process of utilize matlab program that target is carried out zoom operations that yardstick is 0.8-1.2 times and 0-360 degree, again the top left corner pixel point of target image is updated in edge potential field function, find the place that edge potential field functional value is maximum, i.e. target location in the picture；

Step 3.2 initializes population quantity N_pop

Population quantity N_popThe effect of optimization of colony intelligence optimized algorithm there is impact；When population quantity is more than 100, ramose root chess game optimization method fast convergence rate but calculation cost is high；When population quantity is less than 100, ramose root chess game optimization method is absorbed in local convergence；By arranging different population quantities, comparative test result, select suitable population quantity；

Step 3.3 initializes population position

Need random initializtion population position in solution space；If X_lFor the lower limit set of solution space, X_uFor the upper limit set of solution space, then each individuality in population initializes by following formula:

X_i=X_l+rand·(X_u-X_l)(2)

In formula, rand is the random number between 0 to 1；

Step 3.4 arranges ramose root chess game optimization method preset parameter

Ramose root chess game optimization method is made up of stolon searching operators and root searching operators；In stolon searching operators, plant carries out flourish by growing the stolon made new advances, in D dimension space, and the positional information X of i-th plant_iEvery generation updates once, and concrete replacement criteria is shown below:

X_daughter(t)=X_mother(t)+d_runner×r₁(3)

d_runner=x_u-x_l(4)

In formula, X_motherT () represents i-th plant position in D dimension space in after the t time iterative process, d_runnerRepresent the ultimate range of maternal plant and sub-strain, take from the transformation range of variable, r₁Represent the random number of 0 to 1, X_daughterT offspring individual that () generates through stolon reproduction for plant, x_uFor the upper dividing value of independent variable, x_lFloor value for independent variable；In order to reduce the calculation cost of algorithm, when t is better than t-1 for the fitness function of sub-strain for the fitness function of sub-strain, root Local Search need not be carried out, for function minimum optimization problem, it is judged that shown in criterion such as formula (3):

$\left|\frac{\min f\left(x_{daughter}\right(t\left)\right)-\min f\left(x_{daughter}\right(t-1\left)\right)}{\min f\left(x_{daughter}\right(t-1\left)\right)}\right|\≥tol---\left(5\right)$

In formula, tol is the threshold value that ramose root chess game optimization method sets in advance, is set to a number less than 1；minf(x_daughter(t)) it is the sub-strain of the best of the t time iteration, minf (x_daughter(t-1)) it is the sub-strain of the best of the t-1 time iteration, if the formula of being unsatisfactory for (5), carries out stolon Local Search；

x_perturbed,k=diag (1,1 ..., 1+d_runnern_k,1,...,1)*x_{daughter,best}(t)(6)

Being found out by formula (6), stolon Local Search mode is obtain the sub-strain of disturbance, wherein n after antithetical phrase strain one-dimensional carries out disturbing_kBeing 0 for average, variance is the random number of 1；x_perturbed,kFor disturbing sub-strain, x_{daughter,best}T () is optimum sub-strain value after t iteration, if disturbance sub-strain fitness function value is better than the fitness function value of optimum sub-strain, then optimum sub-strain is the sub-strain of disturbance；

Found out by formula (3), the ultimate range d of maternal plant and sub-strain_runnerSize directly control the search amplitude of stolon searching operators, by d_runnerIt is taken as the scope of independent variable；

After plant growing goes out sub-strain, current region can be carried out Local Search by the root of every sub-strain, finds the position that water resource and mineral are the abundantest, and root search formula is as shown in (7):

x_perturbed,k=diag (1,1 ..., 1+d_rootn_k,1,...,1)*x_{daughter,best}(t)(7)

In formula (7), d_rootFor root hunting zone, it is set to the number less than 1, if the individual fitness function value of the disturbance after Local Search is better than the fitness function value of best sub-strain, then best sub-strain is replaced with the sub-strain of disturbance；

After having carried out the basic operator of above-mentioned two, ramose root chess game optimization method is using sub-for the best of this generation strain as first maternal plant of future generation, and all the other maternal plants in next round iterative process are generated by roulette mode by current sub-strain；By being constantly iterated renewal, until meeting the condition stopping iteration, ramose root chess game optimization method obtains target optimum position in gathering image；

Step 3.5 arranges ramose root chess game optimization method iterations

When ramose root chess game optimization method iterations is few, ramose root chess game optimization method can be made to be not reaching to optimum and then to stop；When iterations is big, ramose root chess game optimization method restrains already；So its average rate of convergence should first be analyzed, then the iterations of ramose root searching method is set；

Step 4: design cost function

Calculate the potential function value that target image edge produces in gathering the edge potential field that image produces, shown in computing formula such as following formula (8):

$fitness\left(X_i\right)=\frac{1}{N}\underset{i=1}{\overset{N}{\Σ}}\left\{EPF(x_i,y_i)\right\}---\left(8\right)$

In formula, N is that object edge tries to achieve cost function value with gathering the sum of all pixels that image border overlaps, through type (1) and formula (8)；

Step 5: utilize stolon searching operators and root searching operators optimizing

Utilize initialized colony and the position of history colony, according to formula (3) and formula (4), current group is carried out stolon searching operators and carry out optimizing operation；In order to reduce the calculation cost of ramose root chess game optimization method, when t meets formula (5) for the cost function of sub-strain, then do not carry out stolon Local Search, otherwise then carry out stolon Local Search according to formula (6) and obtain the sub-strain of disturbance, if disturbance sub-strain cost function value is better than the cost function value of optimum sub-strain, then optimum sub-strain is replaced with the sub-strain of disturbance；

According to formula (7) to carrying out root Local Search, if the cost function value of the disturbance individuality after Local Search is better than the cost function value of best sub-strain, then best sub-strain is replaced with the sub-strain of disturbance；

After having carried out stolon searching operators and root searching operators optimizing, ramose root chess game optimization method is using sub-for the best of this generation strain as first maternal plant of future generation, and the maternal plant in next round iterative process is generated by roulette mode by current sub-strain；Whole algorithm is constantly iterated updating, and when the maximum iteration time that number of run sets more than algorithm, stops searching process；

Step 6: store optimum results and verify

After ramose root chess game optimization method stops optimizing, select the individuality that global optimum is corresponding, this result is preserved, obtain final object detection results, and preserve the iterativecurve of lower searching process.

2. a kind of autonomous object detection method of unmanned plane based on plant ramose root evolved behavior according to claim 1, it is characterised in that: in step 2, if image edge relatively horn of plenty or signal noise ratio (snr) of image are low, then by ε_eqTaking less value, usual value is 1；If image scene is simple and signal to noise ratio is high, then by ε_eqTaking bigger value, usual value is 10.

3. a kind of autonomous object detection method of unmanned plane based on plant ramose root evolved behavior according to claim 1, it is characterized in that: in step 3.1, need the scale factor utilizing ramose root chess game optimization method to target, the anglec of rotation, and top left corner pixel point abscissa and these 4 parameters of vertical coordinate are optimized, so D is 4.

4. a kind of autonomous object detection method of unmanned plane based on plant ramose root evolved behavior according to claim 1, it is characterised in that: in step 2, ε_eqTake 1, Q_eq(x_i,y_i) it is 20.

5. a kind of autonomous object detection method of unmanned plane based on plant ramose root evolved behavior according to claim 1, it is characterised in that: in step 3.2, population quantity N_popIt is set to 200.

6. a kind of autonomous object detection method of unmanned plane based on plant ramose root evolved behavior according to claim 1, it is characterised in that: in step 3.3, X_l=[1,1,0,10], X_u=[256,256,360,14].

7. a kind of autonomous object detection method of unmanned plane based on plant ramose root evolved behavior according to claim 1, it is characterised in that: in step 3.4, d_runnerIt is set to 5, d_rootBeing 1, the tol in formula (5) is set to 0.01.

8. a kind of autonomous object detection method of unmanned plane based on plant ramose root evolved behavior according to claim 1, it is characterised in that: in step 3.5, maximum iteration time is set to 100.