CN103914831B - A kind of two-dimentional dual threshold SAR image segmentation method based on quantum telepotation - Google Patents

Abstract

The invention discloses a kind of two-dimentional dual threshold SAR image segmentation method based on quantum telepotation, implementation step is: (1) initializes population population scale M and maximum iteration time T_max, the initial position of each particle of stochastic generation；(2) calculate fitness function, be worth to optimal location and the global optimum position of current iteration of current particle according to maximum kind internal variance；(3) P is calculated_idAnd mbest_d；(4) random array is constructed；(5) arrange and define value, it is judged that its with define the relation of value, update particle position according to formula；(6) checking whether and reach termination condition, if reaching, terminating；Otherwise return step (2) to continue.(7) in store a pair optimal threshold to be found in two dimensions of the particle that global optimum position is pointed to, splits SAR image threshold value according to this.The present invention is compared with classical dividing method, and more preferable to the effect of SAR image segmentation, time complexity is relatively small.

Description

A kind of two-dimentional dual threshold SAR image segmentation method based on quantum telepotation

Technical field

The invention belongs to image data processing technology field, a kind of two dimension based on quantum telepotation Dual threshold SAR image segmentation method.

Background technology

Radar concepts was formed at for 20 beginnings of the century, was the electronic equipment utilizing electromagnetic wave detection target.It is in the second time world Great War is developed rapidly.Synthetic aperture radar (Synthetic Aperture Radar, SAR) is that one is operated in microwave The Coherent Imaging RADAR of wave band, is a kind of active microwave remote sensing sensor.Owing to it has round-the-clock, round-the-clock, long distance From the observation band with broadness, and it is prone to distinguish the ability of moving target from fixed background, is therefore widely used in geology The various fields such as exploration, urban planning, military detection, marine monitoring, vegetation growth assessment.The development of synthetic aperture radar is to state People's livelihood work, defense technological modernization, development of the national economy tool are of great significance, and Radar Technology is increasingly by the world The attention of advanced technology country, the most existing various real aperture imaging radars, and comprised information more fully pole It is combined to aperture radar (Polarimetrie Synthetie Aperture Radar, write a Chinese character in simplified form PolSAR).

SAR image segmentation and classification be SAR image process the first step, be also in image procossing the most scabrous problem it One.Therefore, seek image partition method efficient, high-precision tool to be of great significance.Image segmentation is constantly subjected to people Great attention, many scholars have paid huge effort, have had been proposed for various types of partitioning algorithm so far for this.Classical Image partition method have: split-run based on thresholding, segmentation based on edge, segmentation based on region, based on cluster Split, based on morphology and other dividing methods etc..Thresholded image dividing method is simple to operate in these methods, efficiency Height, is therefore constantly subjected to the concern of a lot of related direction researcher as a kind of common direct detection zone domain splitting method, It it is a kind of important method of image segmentation.Along with the raising of image segmentation problem complexity, the height of the method segmentation is time consuming nature Through having had a strong impact on its application.So, the efficient algorithm of searching one solves the problems referred to above and is significant.

Particle swarm optimization algorithm is succinct, be easily achieved, need the parameter adjusted less, has been used successfully in non-thread In the property optimization problem such as Filled function, Combinatorial Optimization.When the image that segmentation complexity is higher, many reference amounts are often carried out by we Information fusion, during solving many reference amounts optimal value, optimizing calculating is emphasis also difficult point, the spy of natural evolution in nature Levying and be applied in the middle of computerized algorithm, a lot of difficult problems will have been resolved.Such issues that appear as solution of particle swarm optimization algorithm Provide brand-new and efficient method.Therefore both are combined and will advance efficient, the real-time realization of image segmentation algorithm.

Summary of the invention

It is an object of the invention to provide a kind of two-dimentional dual threshold SAR image segmentation side based on quantum telepotation Method, overcomes existing classical thresholded image cutting techniques segmentation effect when SAR image is split poor and complexity sliced time High shortcoming.

For achieving the above object, the invention provides a kind of two-dimentional dual threshold SAR image based on quantum telepotation Dividing method, comprises the steps:

(1) population population scale M and maximum iteration time T are initialized_max, use random function to generate at the beginning of each particle Beginning position (x_i1,x_i2,…,x_is)。

(2) acquisition σ is calculated according to following formula_BAs fitness function, obtain current particle according to variance within clusters maximum Optimal location pbest_idWith global optimum gbest in current iteration_d。

$\begin{array}{c}tr\left({\mathrm{\σ}}_B\right)=\frac{{\mathrm{\ω}}_1{\[{\mathrm{\ω}}_0{\mathrm{\μ}}_{Ti}-{\mathrm{\μ}}_i(u_1,v_1)\]}^2+{\mathrm{\ω}}_0{\[{\mathrm{\ω}}_1{\mathrm{\μ}}_{Ti}-{\mathrm{\μ}}_i(u_2,v_2)\]}^2+{\[{\mathrm{\ω}}_1{\mathrm{\μ}}_i(u_1,v_1)-{\mathrm{\ω}}_0{\mathrm{\μ}}_i(u_2,v_2)\]}^2}{{\mathrm{\ω}}_0\×{\mathrm{\ω}}_1\×(1-{\mathrm{\ω}}_0-{\mathrm{\ω}}_1)}+\\ \frac{{\mathrm{\ω}}_1{\[{\mathrm{\ω}}_0{\mathrm{\μ}}_{Tj}-{\mathrm{\μ}}_j(u_1,v_1)\]}^2+{\mathrm{\ω}}_0{\[{\mathrm{\ω}}_1{\mathrm{\μ}}_{Tj}-{\mathrm{\μ}}_j(u_2,v_2)\]}^2+{\[{\mathrm{\ω}}_1{\mathrm{\μ}}_j(u_1,v_1)-{\mathrm{\ω}}_0{\mathrm{\μ}}_j(u_2,v_2)\]}^2}{{\mathrm{\ω}}_0\×{\mathrm{\ω}}_1\×(1-{\mathrm{\ω}}_0-{\mathrm{\ω}}_1)}\end{array}$

Wherein L is the gray level of original image, u₁,u₂For gray threshold, v₁, v₂For neighborhood gray threshold, u₁, u₂, v₁, v₂ Take [0, L) value, ω₀=ω₀(u₁,v₁), ω₁=ω₁(u₁,v₁,u₂,v₂), ω₂=ω₂(u₂,v₂) it is the general of each several part appearance Rate, μ₀、μ₁、μ₂It is the average of all kinds of gray scale condition probability of background and target, μ_i(u₁,v₁)、μ_i(u₂,v₂)、μ_j(u₁,v₁)、μ_j (u₂,v₂)、μ_i(L,L)、μ_j(L,L)、μ_Ti、μ_TjIt is background and the average of all kinds of gray scale of target；

(3) setting φ is the random number between [0,1], according to formula p_id=φ * pbest_id+(1-φ)*gbest_dCalculate P_id, According to formulaCalculate mbest_d。

(4) two position x are randomly generated_j, x_k, and x_j≠x_k≠x_i, according to formula x_id=φ * (x_k-x_j)+(1-φ) gbest_d±α*|mbest_d-x_id| * ln (1/u) constructs δ, makes δ=x_k-x_j, construct the rand random array identical with δ dimension.

(5) let R be and define value, it is judged that itself and the relation of rand value, as R ＞ rand, use formula x_id=φ * (pbest_id- gbest_d)+gbest_d±α*|mbest_d-x_id| the original particle position of location updating formula of * ln (1/u) i.e. quantum particle swarm, when R≤rand, uses formula x_id=φ * (x_k-x_j)+(1-φ)gbest_d±α*|mbest_d-x_id| * ln (1/u) comes the position of more new particle Put.

(6) check whether and reach termination condition, the most whether achieve fitness value or reach maximum iteration time；If reaching Termination condition then stops iteration, and algorithm terminates；Otherwise return step (2)；

(7)gbest_dIn store a pair optimal threshold to be found in two dimensions of the particle pointed to, right according to this SAR image is split by threshold value.

Present invention have the advantage that

A) the method is based on threshold method, but the segmentation that application simple threshold values segmentation can not meet complicated image is asked Topic, introduces two dimension dual threshold Otsu method the most herein and carrys out SAR image and split, to obtain more preferable segmentation effect.

B) in the present invention, main expense is in the complexity of object function and evaluates the aspect such as number of times, and the method one is In program is run, with the addition of judgement statement, do not increase the evaluation number of times of object function, simply in program implement With the addition of judgement operation, this is not the major cost of algorithm, is negligible.Two are the use of two dimension dual threshold Otsu calculates Method, but the time complexity of algorithm is higher, so using QPSO and improvement QPSO method to improve arithmetic speed in literary composition.

Accompanying drawing explanation

Fig. 1 is the flow chart of the present invention；

Fig. 2 is two groups of original SAR image that the present invention uses；

Fig. 3 is by the present invention and conventional two-dimensional Otsu method based on quantum telepotation and two dimension dual threshold Otsu side The simulation result figure that Fig. 2 (a) is split by method；

Fig. 4 is by the present invention and conventional two-dimensional Otsu method based on quantum telepotation and two dimension dual threshold Otsu side The simulation result figure that Fig. 2 (b) is split by method.

Detailed description of the invention

With reference to Fig. 1, the present invention to implement step as follows:

Step 1, initializes population population scale M and maximum iteration time T_max, use random function to generate each particle Initial position (x_i1,x_i2,…,x_is)。

Step 2, calculates according to following formula and obtains σ_BAs fitness function, obtain current particle according to variance within clusters maximum Optimal location pbest_idWith global optimum gbest in current iteration_d。

$\begin{array}{c}tr\left({\mathrm{\σ}}_B\right)=\frac{{\mathrm{\ω}}_1{\[{\mathrm{\ω}}_0{\mathrm{\μ}}_{Ti}-{\mathrm{\μ}}_i(u_1,v_1)\]}^2+{\mathrm{\ω}}_0{\[{\mathrm{\ω}}_1{\mathrm{\μ}}_{Ti}-{\mathrm{\μ}}_i(u_2,v_2)\]}^2+{\[{\mathrm{\ω}}_1{\mathrm{\μ}}_i(u_1,v_1)-{\mathrm{\ω}}_0{\mathrm{\μ}}_i(u_2,v_2)\]}^2}{{\mathrm{\ω}}_0\×{\mathrm{\ω}}_1\×(1-{\mathrm{\ω}}_0-{\mathrm{\ω}}_1)}+\\ \frac{{\mathrm{\ω}}_1{\[{\mathrm{\ω}}_0{\mathrm{\μ}}_{Tj}-{\mathrm{\μ}}_j(u_1,v_1)\]}^2+{\mathrm{\ω}}_0{\[{\mathrm{\ω}}_1{\mathrm{\μ}}_{Tj}-{\mathrm{\μ}}_j(u_2,v_2)\]}^2+{\[{\mathrm{\ω}}_1{\mathrm{\μ}}_j(u_1,v_1)-{\mathrm{\ω}}_0{\mathrm{\μ}}_j(u_2,v_2)\]}^2}{{\mathrm{\ω}}_0\×{\mathrm{\ω}}_1\×(1-{\mathrm{\ω}}_0-{\mathrm{\ω}}_1)}\end{array}$

Wherein L is the gray level of original image, u₁,u₂For gray threshold, v₁, v₂For neighborhood gray threshold, u₁, u₂, v₁, v₂ Take [0, L) value, ω₀=ω₀(u₁,v₁), ω₁=ω₁(u₁,v₁,u₂,v₂), ω₂=ω₂(u₂,v₂) it is the general of each several part appearance Rate, μ₀、μ₁、μ₂It is the average of all kinds of gray scale condition probability of background and target, μ_i(u₁,v₁)、μ_i(u₂,v₂)、μ_j(u₁,v₁)、μ_j (u₂,v₂)、μ_i(L,L)、μ_j(L,L)、μ_Ti、μ_TjIt is background and the average of all kinds of gray scale of target；

Step 3, if φ is the random number between [0,1], according to formula p_id=φ * pbest_id+(1-φ)gbest_dCalculate p_id, according to formulaCalculate mbest_d。

Step 4, randomly generates two position x_j, x_k, and x_j≠x_k≠x_i, according to formula x_id=φ * (x_k-x_j)+(1-φ) gbest_d±α*|mbest_d-x_id| * ln (1/u) constructs δ, makes δ=x_k-x_j, construct the rand random array identical with δ dimension.

Step 5, lets R be and defines value, it is judged that itself and the relation of rand value, as R ＞ rand, uses formula x_id=φ * (pbest_id-gbest_d)+gbest_d±α|mbest_d-x_id| the original grain of location updating formula of * ln (1/u) i.e. quantum particle swarm Sub-position, as R≤rand, uses formula x_id=φ * (x_k-x_j)+(1-φ)*gbest_d±α|mbest_d-x_id| * ln (1/u) updates The position of particle.

Step 6, checks whether and reaches termination condition (acquirement well adapts to angle value or reaches maximum iteration time).If reaching Then stop iteration to termination condition, algorithm terminates；Otherwise return step (2).

Step 7, gbest_dIn store a pair optimal threshold to be found in two dimensions of the particle pointed to.According to this SAR image is split by threshold value.

The effect of the present invention is further illustrated by following experiment simulation.

1. experiment condition and method

Experiment simulation environment is: MATLAB 7.0.4, Intel (R) Pentium (R) 1CPU 2.4GHz, Window XP Professional。

Experimental technique: have conventional two-dimensional Otsu method based on quantum telepotation and two dimension dual threshold Otsu side respectively Method and the present invention.

2. experiment content and result

Experiment content: the present invention uses two width SAR image shown in Fig. 2 to do test experiments respectively, is Fig. 2 (a) and Fig. 2 respectively (b)。

Experiment one, by existing conventional two-dimensional Otsu method based on quantum telepotation and two dimension dual threshold Otsu method Emulating Fig. 2 (a) image with the present invention, segmentation result is shown in Fig. 3, and wherein, Fig. 3 (a) is based on quantum telepotation The result figure of conventional two-dimensional Otsu method segmentation, Fig. 3 (b) is the result figure of two dimension dual threshold Otsu method segmentation, and 3 (c) is this Invention segmentation result.

Due to SAR image contain much information, background complicated, tradition Otsu algorithm can not meet the requirement of segmentation precision, Contrast above-mentioned experimental result and can be seen that background and target can preferably preferably be split by the present invention.Due to the most right The evaluation criterion that image segmentation is the most unified, the most herein two kinds of conventional evaluation indexes of selection: optimal segmentation threshold Value, time complexity evaluate the performance of three kinds of algorithms, as shown in table 1 below.By Fig. 3 and Biao 1, first we can be seen that herein Algorithm can the most reasonable by image division compared with traditional Otsu method that QPSO optimizes；Secondly the present invention and directly employing two The image of dimension dual threshold Otsu method segmentation is compared, and substantially reduces operation time, and segmentation effect is the most quite a lot of.Comprehensive above two Point considers, it is high that the present invention splits efficiency, and segmentation effect is the most more satisfactory.

Three kinds of algorithm performance evaluation results of table 1 Fig. 2 (a) image

Experiment two, with existing conventional two-dimensional Otsu method based on quantum telepotation and two dimension dual threshold Otsu side Fig. 2 (b) is emulated by method with the present invention, and classification results is shown in Fig. 4, and wherein, Fig. 4 (a) is tradition two based on quantum telepotation The result of dimension Otsu method segmentation, Fig. 4 (b) is the result figure of two dimension dual threshold Otsu method segmentation, and Fig. 4 (c) is that the present invention divides The result cut.

From fig. 4, it can be seen that the result of the present invention is significantly better than the result of existing two kinds of classical ways classification, region is drawn It is more careful and more accurate to divide, and splitting speed is very fast.

In sum, the segmentation to SAR data that the present invention proposes, can obtain more in the cutting procedure to complicated image Reasonably segmentation result, is optimized by QPSO and reduces algorithm calculating complexity, and inventive concept is fairly simple, it is readily appreciated that With application.

Exemplified as above is only the illustration to the present invention, is not intended that the restriction to protection scope of the present invention, all It is within design same or analogous with the present invention belongs to protection scope of the present invention.

The processing step that the present embodiment describes the most in detail belongs to techniques well known or conventional means, the most one by one Narration.

Claims (1)

1. a two-dimentional dual threshold SAR image segmentation method based on quantum telepotation, it is characterised in that: include walking as follows Rapid:

(1) population population scale M and maximum iteration time T are initialized_max, use random function to generate each particle initial bit Put (x_i1,x_i2,…,x_is)；

(2) acquisition σ is calculated according to following formula_BAs fitness function, obtain the optimum position of current particle according to variance within clusters maximum Put pbest_idWith global optimum gbest in current iteration_d；

$\begin{array}{c}tr\left({\mathrm{\σ}}_B\right)=\frac{{\mathrm{\ω}}_1{\[{\mathrm{\ω}}_0{\mathrm{\μ}}_{Ti}-{\mathrm{\μ}}_i\left(u_1,v_1\right)\]}^2+{\mathrm{\ω}}_0{\[{\mathrm{\ω}}_1{\mathrm{\μ}}_{Ti}-{\mathrm{\μ}}_i\left(u_2,v_2\right)\]}^2+{\[{\mathrm{\ω}}_1{\mathrm{\μ}}_i\left(u_1,v_1\right)-{\mathrm{\ω}}_0{\mathrm{\μ}}_i\left(u_2,v_2\right)\]}^2}{{\mathrm{\ω}}_0\×{\mathrm{\ω}}_1\×\left(1-{\mathrm{\ω}}_0-{\mathrm{\ω}}_1\right)}+\\ \frac{{\mathrm{\ω}}_1{\[{\mathrm{\ω}}_0{\mathrm{\μ}}_{Tj}-{\mathrm{\μ}}_j\left(u_1,v_1\right)\]}^2+{\mathrm{\ω}}_0{\[{\mathrm{\ω}}_1{\mathrm{\μ}}_{Tj}-{\mathrm{\μ}}_j\left(u_2,v_2\right)\]}^2+{\[{\mathrm{\ω}}_1{\mathrm{\μ}}_j\left(u_1,v_1\right)-{\mathrm{\ω}}_0{\mathrm{\μ}}_j\left(u_2,v_2\right)\]}^2}{{\mathrm{\ω}}_0\×{\mathrm{\ω}}_1\×\left(1-{\mathrm{\ω}}_0-{\mathrm{\ω}}_1\right)}\end{array}$

Wherein L is the gray level of original image, u₁,u₂For gray threshold, v₁, v₂For neighborhood gray threshold, u₁, u₂, v₁, v₂Take [0, L) value, ω₀=ω₀(u₁,v₁), ω₁=ω₁(u₁,v₁,u₂,v₂), ω₂=ω₂(u₂,v₂) be each several part occur probability, μ₀、 μ₁、μ₂It is the average of all kinds of gray scale condition probability of background and target, μ_i(u₁,v₁)、μ_i(u₂,v₂)、μ_j(u₁,v₁)、μ_j(u₂,v₂)、 μ_i(L,L)、μ_j(L,L)、μ_Ti、μ_TjIt is background and the average of all kinds of gray scale of target；

(3) setting φ is the random number between [0,1], according to formula:

p_id=φ * pbest_id+(1-φ)*gbest_dCalculate P_id, according to formulaCalculate mbest_d；

(4) two position x are randomly generated_j, x_k, and x_j≠x_k≠x_i, according to formula:

x_id=φ * (x_k-x_j)+(1-φ)gbest_d±α*|mbest_d-x_id| * ln (1/u) constructs δ, makes δ=x_k-x_j, structure and δ The random array of rand that dimension is identical；

(5) let R be and define value, it is judged that itself and the relation of rand value, as R ＞ rand, use formula x_id=φ * (pbest_id- gbest_d)+gbest_d±α*|mbest_d-x_id| the original particle position of location updating formula of * ln (1/u) i.e. quantum particle swarm, its Middle x_idFor particle current location, pbest_idFor the current optimal location of particle, gbest_dFor the global optimum position of particle, φ and u Being the random number between [0,1], α is expansion-contraction factor；As R≤rand, use formula x_id=φ * (x_k-x_j)+(1-φ)gbest_d ±α*|mbest_d-x_id| * ln (1/u) comes the position of more new particle；

(6) check whether and reach termination condition, the most whether achieve fitness value or reach maximum iteration time；If reaching to terminate Condition then stops iteration, and algorithm terminates；Otherwise return step (2)；

(7)gbest_dIn store a pair optimal threshold to be found in two dimensions of the particle pointed to, according to this to threshold value pair SAR image is split.