CN102609717A - Cotton foreign fiber characteristic selection method based on particle swarm optimization algorithm - Google Patents

Abstract

The invention discloses a cotton foreign fiber characteristic selection method based on a particle swarm optimization algorithm and relates to the technical field of image processing. The cotton foreign fiber characteristic selection method comprises the following steps of: S1, initializing a particle swarm according to characteristic data of a characteristic training sample set obtained by characteristic extraction; S2, designing an SVM (Support Vector Machine) classifier according to the sample set; S3, classifying the sample set and calculating a fitness value of a particle; S4, optimally solving the fitness value of the current particle and comparing a globally optimal solution of a group; and updating a locally optimal solution and the globally optimal solution; S5, calculating the movement speed and new position of the particle; and S6, if a finishing condition is met, finishing and outputting an optimal characteristic set; otherwise, adding 1 into an iteration and returning to the step S2. According to the method disclosed by the invention, the optimal selection can be carried out on a cotton foreign fiber characteristic, and to the method adapts to the requirements of the SVM classifier; and furthermore, the classifying performance is further improved.

Description

Cotton foreign fiber feature selection approach based on particle swarm optimization algorithm

Technical field

The present invention relates to technical field of image processing, relate in particular to a kind of cotton foreign fiber feature selection approach based on particle swarm optimization algorithm.

Background technology

Foreign fiber in the cotton be meant cotton picking, ted and process such as purchase in sneak into and in the cotton cotton and Products Quality thereof had non-cotton fiber and the color fibre that has a strong impact on, like man-made fiber, hair, the rope made of hemp etc.Though the content of foreign fiber in gined cotton is few, and is serious to the quality influence of wollen fabrics.The foreign fiber of sneaking into surface is the cotton yarn broken end easily, reduces production efficiency; When weaving cotton cloth, influence fabric quality; During dyeing, influence outward appearance, the quality of cotton yarn and cloth cover has been caused very big harm.In China, a lot of cotton spinning of present stage processing enterprise mainly relies on and employs a large amount of workman's manual works to pick foreign fiber, has paid very big human and material resources cost for this reason.

Extremely learning algorithm based on Statistical Learning Theory is the important means in the foreign fiber characteristic special zone in the cotton.Foreign fiber is extracted the initial characteristics set that can access the description target; Guaranteeing under the prerequisite that nicety of grading does not reduce; From the initial characteristics set, select the strongest optimal feature subset of classification capacity; Can reduce the complicacy of classifier design to greatest extent and improve classification speed, be the precondition and guarantee that realizes online real-time grading.

Although at present in the achievement that has obtained a lot of excellences aspect the research of the feature selecting of cotton foreign fiber and pattern-recognition, because conditions such as region difference, the kind of foreign fiber, CF are also inequality even widely different.(Support Vector Machines, SVM) theory is a kind of machine learning algorithm based on Statistical Learning Theory to SVMs, can solve small sample, practical problems such as non-linear preferably.But existing achievement in research shows; The svm classifier device divides the time-like performance to descend to unnecessary and incoherent data set is arranged; Therefore characteristic optimization makes its requirement that can adapt to the svm classifier device, removes the data set of redundant and uncorrelated characteristic; Thereby further improve classification performance, become a research part in the machine learning.

Summary of the invention

The technical matters that (one) will solve

The technical matters that the present invention will solve is: provide a kind of and can be optimized selection to the cotton foreign fiber characteristic, and adapt to the requirement of svm classifier device, further improve the cotton foreign fiber feature selection approach based on particle swarm optimization algorithm of classification performance.

(2) technical scheme

For addressing the above problem, the invention provides a kind of cotton foreign fiber feature selection approach based on particle swarm optimization algorithm, the method comprising the steps of:

The characteristic of the features training sample set that S1. obtains according to feature extraction, the initialization population;

S2. according to said sample set design support vector machine classifier;

S3. said sample set is classified, calculate the appropriateness value of particle;

S4. the appropriateness value of current particle and the globally optimal solution of its locally optimal solution and population are compared, and upgrade the locally optimal solution of current particle and the globally optimal solution of population according to comparative result;

S5. the locally optimal solution and the globally optimal solution that are upgraded according to step S4 calculate the translational speed of particle and new position;

S6. judge whether to meet end condition, if meet, then finish and export the optimal characteristics collection, otherwise iterations adds 1, and return step S2.

Preferably, the characteristic in the said sample set comprises: color characteristic, shape facility and textural characteristics.

Preferably, in step S1, initialization particle position and initial velocity produce one group of initial value at random.

Preferably, establish sample set X=(x at step S2 _i, y _i), i=1,2 ..., N, wherein, N is a training sample quantity, x _iBe the characteristic of sample, y _iBe the classification of sample, be for each calculating particles SVMs optimal classification function model:

$f\left(x\right)=\mathrm{sign}\{\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{\mathrm{\α}}_i^{*}{y}_i(x_i\·x)+b^{*}\}$

Wherein, sign is a sign function, promptly $\mathrm{Sign}\left(x\right)=\left\{\begin{array}{c}1,x\&GreaterEqual;0\\ -1,x<0\end{array}\right.,$

Be the Lagrange coefficient, x is for treating classification samples.

Preferably, in step S3, the fitness function model that calculates said appropriateness value is:

$\mathrm{fitness}={\left[\frac{1}{N}\underset{k=1}{\overset{N}{\mathrm{\&Sigma;}}}{a_k}^2\right]}^{-\frac{1}{2}}$

Wherein, particle is a=(a ₁, a ₂..., a _k).

Preferably, step S4 further comprises step: if the appropriateness value of said particle greater than locally optimal solution, then makes the locally optimal solution of said particle equal said appropriateness value; If the appropriateness value of said particle greater than the globally optimal solution of said population, then makes the globally optimal solution of said population equal said appropriateness value.

Preferably, in step S5, the new position of said particle is:

$x^{i+1}=\left\{\begin{array}{c}0,f\left(x_a^{i+1}\right)\&le;0.5\\ 1,f\left(x_a^{i+1}\right)>0.5\end{array}\right.$

Wherein,

f (x) is the Sigmond function, i.e. f (x)=1/ (1+e-x);

The translational speed of said particle is:

v ⁱ⁺¹＝v ⁱ+c ₁r ₁(p _lB-p ⁱ)+c ₂r ₂(p _B-p ⁱ)

Wherein, p ⁱBe the appropriateness value of current particle, c ₁, c ₂Be the study factor, and, c ₁=c ₂=2, r ₁, r ₂Be the random number between [0,1].

(3) beneficial effect

Method of the present invention is to the initial characteristics set that can describe the foreign fiber target of having extracted; Guaranteeing under the prerequisite that nicety of grading does not reduce; From wherein selecting the strongest optimal feature subset of classification capacity; Can reduce the complicacy of classifier design to greatest extent, and can under the situation of not losing nicety of grading, reduce calculated amount, the raising classification speed of characteristic svm classifier.

Description of drawings

Fig. 1 is the cotton foreign fiber feature selection approach process flow diagram based on particle swarm optimization algorithm according to one embodiment of the present invention.

Embodiment

The cotton foreign fiber feature selection approach based on particle swarm optimization algorithm that the present invention proposes specifies as follows in conjunction with accompanying drawing and embodiment.

Particle group optimizing (Particle Swarm Optimization; PSO) algorithm is a kind of emerging global optimization approach; It ceaselessly moves through each individuality among the crowd and searches for optimum solution, and each particle is by current locally optimal solution and its direction of motion of globally optimal solution decision.Particle swarm optimization algorithm has that algorithm is succinct, fast convergence rate, does not have the characteristics such as adjusting of many parameters as a kind of heuristic search algorithm; Use the characteristic optimization that helps cotton foreign fiber in the present invention; Can extract the optimal characteristics collection of image; The optimal characteristics of extracting not only can shorten the time of svm classifier device classification, can also improve classification accuracy to a certain extent.Be applied to improve svm classifier device classification performance based on removing unnecessary and incoherent data set in the disaggregated model of SVM.

As shown in Figure 1, the cotton foreign fiber feature selection approach based on particle swarm optimization algorithm of accordinging to one embodiment of the present invention comprises step:

S1. extract cotton foreign fiber feature samples data, the characteristic of the features training sample set X that obtains according to feature extraction, initialization population; The position and the initial velocity that comprise population produce one group of initial value at random.

Characteristic among the sample set X mainly comprises: color characteristic, shape facility and textural characteristics.

S2. according to sample set X design SVMs svm classifier device, this svm classifier device of training on training sample set X.

S3. sample set X is classified; Calculate the appropriateness value fitness of particle; With when carrying out feature selecting, under the prerequisite of assurance even raising sorter classification performance, reduce the scale of character subset, promptly a particle can make the nicety of grading of sorter generation high more; The number of features of selecting simultaneously is few more, and its appropriateness value just should be high more.

S4. with the appropriateness value p of current particle ⁱWith its locally optimal solution p _LBAnd the globally optimal solution p of population _BCompare, and upgrade the locally optimal solution p of current particle according to comparative result _LBAnd the globally optimal solution p of population _B

After the particle initialization, particle's velocity and position are a group RANDOM SOLUTION, find optimum solution through iteration then, and in each iteration, particle upgrades oneself through two optimum solutions: one is the optimum solution that particle itself is found, i.e. locally optimal solution p _LBAnother is the optimum solution that population is found at present, is referred to as globally optimal solution p _B

S5. the locally optimal solution p that is upgraded according to step S4 _LBAnd globally optimal solution p _B, calculate the translational speed of particle and new position.

Particle is searched for optimum solution through the motion that each individuality in the population does not stop.Each particle is determined its direction of motion by two parts of globally optimal solution of own current locally optimal solution and all particles.Each particle is represented a point in the N dimension space, and its next position is determined by oneself current location and speed.

S6. judge whether to meet end condition, if meet, then finish and export the optimal characteristics collection, otherwise iterations adds 1, and return step S2.

Establish sample set X=(x at step S2 _i, y _i), i=1,2 ..., N, wherein, N is a training sample quantity, x _iBe the characteristic of sample, y _iBe the classification of sample, be for each calculating particles SVM optimal classification function model:

$f\left(x\right)=\mathrm{sign}\{\underset{i=1}{\overset{N}{\mathrm{\&Sigma;}}}{\mathrm{\&alpha;}}_i^{*}{y}_i(x_i\&CenterDot;x)+b^{*}\}$

Wherein, sign is a sign function, promptly $\mathrm{Sign}\left(x\right)=\left\{\begin{array}{c}1,x\&GreaterEqual;0\\ -1,x<0\end{array}\right.,$

Be the Lagrange coefficient, x is for treating classification samples.

In step S3, the fitness function model of calculating moderate value fitness is:

$\mathrm{fitness}={\left[\frac{1}{N}\underset{k=1}{\overset{N}{\mathrm{\&Sigma;}}}{a_k}^2\right]}^{-\frac{1}{2}}$

Wherein, particle is a=(a ₁, a ₂..., a _k).

In the method for this embodiment, step S4 further comprises step:

If the appropriateness value p of particle ⁱGreater than locally optimal solution p _LB, then make the locally optimal solution p of particle _LBEqual appropriateness value p ⁱIf the appropriateness value p of particle ⁱGlobally optimal solution p greater than population _B, then make the globally optimal solution p of population _BEqual appropriateness value p ⁱ

In addition, in step S5, the new position of particle is:

$x^{i+1}=\left\{\begin{array}{c}0,f\left(x_a^{i+1}\right)\&le;0.5\\ 1,f\left(x_a^{i+1}\right)>0.5\end{array}\right.$

Wherein,

f (x) is the Sigmond function, i.e. f (x)=1/ (1+e-x);

The translational speed of particle is:

v ⁱ⁺¹＝v ⁱ+c ₁r ₁(p _lB-p ⁱ)+c ₂r ₂(p _B-p ⁱ)

Wherein, c ₁, c ₂Be the study factor, and, c ₁=c ₂=2, r ₁, r ₂Be the random number between [0,1].

Above embodiment only is used to explain the present invention; And be not limitation of the present invention; The those of ordinary skill in relevant technologies field under the situation that does not break away from the spirit and scope of the present invention, can also be made various variations and modification; Therefore all technical schemes that are equal to also belong to category of the present invention, and scope of patent protection of the present invention should be defined by the claims.

Claims (7)

1. cotton foreign fiber feature selection approach based on particle swarm optimization algorithm is characterized in that the method comprising the steps of:

The characteristic of the features training sample set that S1. obtains according to feature extraction, the initialization population;

S2. according to said sample set design support vector machine classifier;

S3. said sample set is classified, calculate the appropriateness value of particle;

S4. the appropriateness value of current particle and the globally optimal solution of its locally optimal solution and population are compared, and upgrade the locally optimal solution of current particle and the globally optimal solution of population according to comparative result;

S5. the locally optimal solution and the globally optimal solution that are upgraded according to step S4 calculate the translational speed of particle and new position;

S6. judge whether to meet end condition, if meet, then finish and export the optimal characteristics collection, otherwise iterations adds 1, and return step S2.

2. the method for claim 1 is characterized in that, the characteristic in the said sample set comprises: color characteristic, shape facility and textural characteristics.

3. the method for claim 1 is characterized in that, in step S1, initialization particle position and initial velocity produce one group of initial value at random.

4. the method for claim 1 is characterized in that, establishes sample set X=(x at step S2 _i, y _i), i=1,2 ..., N, wherein, N is a training sample quantity, x _iBe the characteristic of sample, y _iBe the classification of sample, be for each calculating particles SVMs optimal classification function model:

$f\left(x\right)=\mathrm{sign}\{\underset{i=1}{\overset{N}{\mathrm{\&Sigma;}}}{\mathrm{\&alpha;}}_i^{*}{y}_i(x_i\&CenterDot;x)+b^{*}\}$

Wherein, sign is a sign function, promptly $\mathrm{Sign}\left(x\right)=\left\{\begin{array}{c}1,x\&GreaterEqual;0\\ -1,x<0\end{array}\right.,$

Be the Lagrange coefficient, x is for treating classification samples.

5. the method for claim 1 is characterized in that, in step S3, the fitness function model that calculates said appropriateness value is:

$\mathrm{fitness}={\left[\frac{1}{N}\underset{k=1}{\overset{N}{\mathrm{\&Sigma;}}}{a_k}^2\right]}^{-\frac{1}{2}}$

Wherein, particle is a=(a ₁, a ₂..., a _k).

6. the method for claim 1 is characterized in that, step S4 further comprises step:

If the appropriateness value of said particle greater than locally optimal solution, then makes the locally optimal solution of said particle equal said appropriateness value;

If the appropriateness value of said particle greater than the globally optimal solution of said population, then makes the globally optimal solution of said population equal said appropriateness value.

7. the method for claim 1 is characterized in that, in step S5, the new position of said particle is:

$x^{i+1}=\left\{\begin{array}{c}0,f\left(x_a^{i+1}\right)\&le;0.5\\ 1,f\left(x_a^{i+1}\right)>0.5\end{array}\right.$

Wherein,

f (x) is the Sigmond function, i.e. f (x)=1/ (1+e-x);

The translational speed of said particle is:

v ⁱ⁺¹＝v _i+c ₁r ₁(p _lB-p ⁱ)+c ₂r ₂(p _B-p ⁱ)

Wherein, p ⁱBe the appropriateness value of current particle, c ₁, c ₂Be the study factor, and, c ₁=c ₂=2, r ₁, r ₂Be the random number between [0,1].

Images