CN104657949A - Method for optimizing structural elements during denoising of coal slime flotation froth image - Google Patents

Abstract

The invention relates to processing of a coal slime flotation froth image, particularly to a method for optimizing structural elements during denoising of the coal slime flotation froth image. The method for optimizing the structural elements during denoising of the coal slime flotation froth image comprises steps as follows: the coal slime flotation froth image acquired from a flotation site of a coal preparation plant is denoised by a morphological opening and closing filter, the structural elements used during denoising of the coal slime flotation froth image are optimized, and the optimized structural elements are used for image reconstruction morphological opening and closing filtering denoising processing. With the adoption of the method, selection processes are simple, convenient and time-saving, selection results are more scientific, reasonable and accurate, the blindness of selection of the structural elements is overcome, and the reconstruction morphological opening and closing filtering denoising effect is improved.

Description

The method that in the denoising of a kind of coal slime flotation froth images, structural element is optimized

Technical field

The present invention relates to the process of coal slime flotation froth images, the method that specifically in the denoising of a kind of coal slime flotation froth images, structural element is optimized.

Background technology

Coal slime flotation froth images is the actual image obtained in the Floating Production Process of coal preparation plant, because the bubble in coal slime flotation is the bubble carrying duff grain, and border fuzzy, not of uniform sizely to cause, and Image Acquisition is carried out in without the natural light closed and light environment, containing many little reflection bright spots on the bubble in image.In the acquisition process of image, also there is the noises such as ccd sensor in addition, these bring very large difficulty all to follow-up Iamge Segmentation.Therefore in order to the noise that coal slime flotation froth images exists will be removed, adopt morphology opening-closing Filter to carry out denoising to froth images and be very important.

In the denoising of Morphological Reconstruction opening and closing filtering image, the effect of structural element is very important, and its selection determines final filtering performance.But structural element choose often unfixing method and rule can be followed, how to determine that structural element shape and size become the problem perplexing people.In image filtering, select difform structural element can produce the filter effect of not same-action; In filtering, its size is also very crucial equally, if when selecting excessive structural element, at this moment the details of filtered image will be lost; Otherwise, then denoising can be made not thorough.Due to above-mentioned importance, for the theme being reasonably selected to numerous researcher discussion of structural element.When carrying out morphologic filtering; in order to obtain good filter effect; reply structural element carries out screening and optimizes; not only to consider that its geometric configuration also will optimize its size; select rational structural element not only can remove details that noise better can protect again image, guarantees the accuracy based on watershed segmentation.

Summary of the invention

Technical matters to be solved by this invention is: how when adopting morphology opening-closing Filter to carry out denoising to froth images, better removes the noise that coal slime flotation froth images exists.

The technical solution adopted in the present invention is: the method that in the denoising of a kind of coal slime flotation froth images, structural element is optimized, to the coal slime flotation froth images obtained from coal preparation plant's flotation site, morphology opening-closing Filter is adopted to carry out denoising to coal slime flotation froth images, the structural element during denoising of coal slime flotation froth images is optimized, structural element after optimization is used for the process of Image Reconstruction morphology opening and closing filtering and noise reduction, is optimized carries out according to following step the structural element during denoising of coal slime flotation froth images

Step one, choose the coal slime flotation froth images photographed aborning, its Pixel Dimensions is 256 × 256, actionradius r is respectively 2, 3, 4, the circular configuration element of 5 is tested one by one to its information capacity, find maximum information capacity and time large structural element corresponding to information capacity, then between, choose complete " 1 " matrix of a d × d again as new construction element, d is between the circular configuration element dimensions corresponding to maximum information capacity and secondary large information capacity, when Optimized Iterative, this d × d matrix by rows is launched into 1 row, form initial antibodies group thus, initial antibodies group chromosome is encode in d × d position 0/1, namely coded word hop count is L=d × d, d is natural number,

Step 2, each antibody for initial antibodies group, calculate its fitness value, namely

$f_i=C_{\mathrm{info}}={\log }_2[1+\underset{\mathrm{\ω}}{\mathrm{\Σ}}\mathrm{Norm}(G_1,G_2)]---\left(1\right)$

Wherein f _irepresent the fitness value of i-th antibody in antibody population, i is natural number, C _inforepresent the information capacity improved, Norm (G ₁, G ₂) represent based on peak value normalization two-dimensional histogram, G ₁represent the gray scale through i-th certain pixel of antibody filtered image in image, G ₂represent the gray scale of its right adjacent pixel, ω is the cumulative constraint of information capacity, can be expressed as:

$\mathrm{\ω}=\left\{\begin{array}{c}|G_1-\frac{1}{2}(G_{\max }+G_{\min })|\≤T_1\\ |G_1-G_2|\≤T_2\end{array}\right.---\left(2\right)$

T ₁, T ₂be called as nonnegativity restrictions threshold value, G _maxand G _minrepresent the maxima and minima of the gray scale in the two-dimensional histogram used by logarithm normalization respectively, in the optimization of structural element, get G _min=0, G _max=255, T ₁=128, T ₂=2;

Step 3, employing elite retention strategy, select m maximum antibody of adaptive value to be placed in elite storehouse as memory antibody to be retained as elite colony, do not participate in selection, crossover and mutation operation, directly being taken as excellent individual joins in the colony of new generation of generation, wherein m=(15% ~ 20%) N, m round numbers, N is population scale;

Step 4, take the antibody concentration of the quick calculation method antagonist group of the hamming distance based on XOR to calculate, the basis of antibody concentration calculates replication rate e _k;

Replication rate: $e_k=\frac{f_k}{{\left(C_k\right)}^{\mathrm{\β}}}---\left(3\right)$

F _kfor the fitness of antibody k, C _kfor the concentration of antibody k, β is the concentration of reflection antibody and the important parameter of fitness proportion occupied by expecting in breeding potential, gets β=2 here, the concentration C of antibody k _kthe fast algorithm based on the hamming distance of XOR is adopted to calculate, namely

$C_k=\frac{1}{N}\underset{w=1}{\overset{N}{\mathrm{\Σ}}}{a}_{\mathrm{kw}},a_{\mathrm{kw}}=\left\{\begin{array}{c}1,D\≤t\\ 0,D>t\end{array}\right.---\left(4\right)$

C _kfor the concentration of antibody k, N is population scale, a _kwbe the affinity between two antibody k and w, D is the hamming distance of two antibody, and t is hamming distance threshold value, and t=0.3*L, L are length and the coded word hop count of character string, k and w is natural number;

Step 5, for select probability p _skcalculate, select probability p can be obtained according to step 4 _sk, based on p _skvalue antagonist colony select, and carry out interlace operation

$p_{\mathrm{sk}}=\frac{e_k}{\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{e}_i}---\left(5\right)$

Wherein, e _kfor the replication rate of antibody k, N is population scale, and i is the natural number being less than or equal to N, e _ithe replication rate of any one antibody i;

Step 6, employing add adjustable factors θ, obtain new antibodies group; The computing formula adding the adaptive mutation probability of adjustable factors is:

$P_m=\left\{\begin{array}{c}\frac{k_3\left[(f_{\max }-f)\right]}{f_{\max }-f_{\mathrm{avg}}}+\mathrm{\θ},f>f_{\mathrm{avg}}\\ k_4,f\≤f_{\mathrm{avg}}\end{array}\right.---\left(6\right)$

$\mathrm{\θ}=-\frac{k}{G_{\max }-1}G+K+\frac{K}{G_{\max }-1}---\left(7\right)$

F in formula _maxrepresent the maximal value of fitness; f _avgrepresent arithmetic mean fitness value; F is ideal adaptation angle value; k ₃, k ₄refer to the regulation coefficient between 0 and 1, get k ₃=k ₄=0.1, θ is adjustable factors, and G is evolutionary generation; G _maxfor maximum evolutionary generation; The value of adjustable factors θ, here k=0.005 when k is the first generation;

Step 7, antagonist group upgrade, and call high fitness value individuality in elite colony and replace the individuality of low adaptive value in antibody population, generate antibody population of future generation;

Step 8, according to end condition to judging, if the words met terminate to optimize, export the structural element after optimizing, if do not meet, jump to step 2 and repeat, end condition is one of following: a, definition threshold epsilon=0.0001, for each antibody population, calculate the average fitness of this antibody population, the difference of the arithmetic mean fitness value of this antibody population and the arithmetic mean fitness value of previous generation antibody population is less than ε, b, continuous 15 generation antibody population the highest fitness value remain unchanged, c, reach maximum evolutionary generation.Tool of the present invention has the following advantages:

The invention has the beneficial effects as follows:

1, on the basis selecting mechanism, adopt the selection mechanism of concentration adjustment, introduce the hamming distance similarity criterion based on XOR, abandon the method for traditional information entropy, loaded down with trivial details Logarithmic calculation can be avoided, improve efficiency.

2, adopt the adaptive variation method improved, add variation regulatory factor, thus the diversity of initial stage individuality can be increased, avoid Premature Convergence.

3, outstanding genetic entities is retained, accelerate algorithm search process further.

4, in the process structural element of coal slime flotation image denoising is optimized, for coal slime flotation froth images without with reference to the singularity evaluated, adopt information capacity based on the improvement of gray level co-occurrence matrixes as the fitness function in structural element optimized algorithm in the denoising of coal slime flotation froth images.

5, initial antibodies group is generated by a kind of generation method of initial antibodies group of improvement, using circular configuration element as object, progressively expand its size, test using the information capacity of the improvement based on gray level co-occurrence matrixes as standard respectively, finally locking may obtain the structural element scope of best filter effect, in this, as chromosome coding length.

The effective ways that the present invention finds a kind of structural element of coal slime flotation froth images to choose, make process of choosing easy, save time, make the result chosen science, rationally, accurately more, overcome structural element choose in blindness, improve the denoising effect of Morphological Reconstruction opening and closing filtering.

Accompanying drawing explanation

Fig. 1 is the process flow diagram that in coal slime flotation froth images Morphological Reconstruction opening and closing denoising, structural element is optimized;

Fig. 2 is field of definition and the constraint of two-dimensional histogram;

Fig. 3 is adjustable factors change curve θ.

Embodiment:

Be described below in conjunction with accompanying drawing 1,2,3 pairs of specific embodiment of the invention:

1. initial antibodies group chromosome adopts the principle measured one by one with typical structure element to determine.Choose respectively when being reconstructed opening and closing filtering to floatation foam image radius be 2,3,4,5 circular configuration element carry out judging the dimension of initial antibodies group.Choose the coal slime flotation froth images photographed aborning, its Pixel Dimensions is 256 × 256, the circular configuration element that actionradius r is respectively 2,3,4,5 is reconstructed morphology opening and closing filtering and noise reduction to it, then the information capacity of denoising image is tested one by one, be that the information capacity of the circular configuration element gained filtering image of 2,3,4 increases successively along with the increase of radius for radius, especially when radius is the structural element of 4, the information capacity of filtered image is maximum; But when the radius of structural element is 5, after filtering, the information capacity of flotation image reduces on the contrary.In order to further test, being 3 at radius, (5 dimension) and radius are between the structural element of 4 (7 tie up), choose complete " 1 " matrix of a d × d again as new construction element, d is between the circular configuration element dimensions corresponding to maximum information capacity and secondary large information capacity, namely complete " 1 " matrix choosing 6 × 6 carries out fitness test as structural element, and the fitness recording floatation foam image after complete " 1 " matrix reconstruction morphology opening and closing filtering of 6 × 6 is greater than through radius the information capacity of the structural element filtered image being 4.Thus when various optimizing, structural element is defined as 6 × 6 matrixes, in Optimized Iterative, matrix by rows is launched into 1 row, and namely initial antibodies group chromosome is 36 0,1 coding, i.e. coded word hop count: L=36.Producing an initial population is 30, and chromosome number is the sequence set of 36, the matrix of available 30 × 36, and every a line of matrix is as an antibody, and maximum evolutionary generation is set as 50.

2. choose the froth images that a width is obtained in coal slime flotation production scene by industrial CCD camera, its Pixel Dimensions is 256 × 256.6 × 6 matrixes are arranged in order for each antibody, be the structural element corresponding to it, it can be used as the structural element in Morphological Reconstruction wave filter, denoising is carried out to froth images, image after process is adopted and carries out fitness value based on the information capacity of the improvement of gray level co-occurrence matrixes, namely

$f_i=C_{\mathrm{info}}={\log }_2[1+\underset{\mathrm{\ω}}{\mathrm{\Σ}}\mathrm{Norm}(G_1,G_2)]---\left(1\right)$

Wherein f _irepresent the fitness value of i-th antibody in antibody population, i is natural number, C _inforepresent the information capacity improved, Norm (G ₁, G ₂) represent based on peak value normalization two-dimensional histogram, G ₁represent the gray scale through i-th certain pixel of antibody filtered image in image, G ₂represent the gray scale of its right adjacent pixel, ω is the cumulative constraint of information capacity, can be expressed as:

$\mathrm{\ω}=\left\{\begin{array}{c}|G_1-\frac{1}{2}(G_{\max }+G_{\min })|\≤T_1\\ |G_1-G_2|\≤T_2\end{array}\right.---\left(2\right)$

Calculate, try to achieve the fitness value of each antibody.

3. pair fitness sorts, and chooses wherein maximum m=5 and retains as elite's antibody.

4. apply desired replication rate formula:

$e_k=\frac{f_k}{{\left(C_k\right)}^{\mathrm{\β}}}---\left(3\right)$

The desired replication rate of calculating antibody, gets β=2 in formula; Simultaneously according to formula

$C_k=\frac{1}{N}\underset{w=1}{\overset{N}{\mathrm{\Σ}}}{a}_{\mathrm{kw}},a_{\mathrm{kw}}=\left\{\begin{array}{c}1,D\≤t\\ 0,D>t\end{array}\right.---\left(4\right)$

The concentration of calculating antibody k; D is the hamming distance between two antibody, t=10, adopts the XOR algorithm improved, namely carries out XOR by turn to the binary code sequence of two antibody, the figure place being " 1 " added up in XOR result, finally cumulative and be hamming distance;

5. based on step 4, by formula namely:

$p_{\mathrm{sk}}=\frac{e_k}{\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{e}_i}---\left(5\right)$

The select probability of calculating antibody; Based on p _skvalue antagonist colony select, and carry out interlace operation.

6. carry out mutation operation based on the TSP question method antagonist improved, the aberration rate of antibody passes through formula $P_m=\left\{\begin{array}{c}\frac{k_3\left[(f_{\max }-f)\right]}{f_{\max }-f_{\mathrm{avg}}}+\mathrm{\θ},f>f_{\mathrm{avg}}\\ k_4,f\≤f_{\mathrm{avg}}\end{array}\right.---\left(6\right)$

$\mathrm{\θ}=-\frac{k}{G_{\max }-1}G+K+\frac{K}{G_{\max }-1}---\left(7\right)$

Calculate, wherein get k ₃=k ₄=0.1, k=0.005, produces new antibodies group after making a variation.

7. call high fitness value individuality in elite colony and replace the individuality of low adaptive value in antibody population, antagonist group upgrade, and generates antibody population of future generation.

8., for every generation antibody population, if meet one of following condition, terminate to optimize, export the structural element after optimizing; If do not meet, jump to 2 and repeat.1) average fitness of this antibody population is calculated; If the difference ε of the average fitness of the average fitness of this antibody population and previous generation antibody population is less than or equal to 0.0001, i.e. ε≤0.0001; 2) continuous 15 generation antibody population the highest fitness remain unchanged; 3) maximum evolutionary generation is reached.

9. after iteration terminates, select the individuality that fitness in population is maximum, it is arranged in the matrix of 6 × 6 according to order, is the structural element of final optimization pass gained.

Claims (1)

1. the method that in coal slime flotation froth images denoising, structural element is optimized, to the coal slime flotation froth images obtained from coal preparation plant's flotation site, morphology opening-closing Filter is adopted to carry out denoising to coal slime flotation froth images, it is characterized in that: the structural element during denoising of coal slime flotation froth images is optimized, structural element after optimization is used for the process of Image Reconstruction morphology opening and closing filtering and noise reduction, is optimized carries out according to following step the structural element during denoising of coal slime flotation froth images

Step one, choose the coal slime flotation froth images photographed aborning, its Pixel Dimensions is 256 × 256, actionradius r is respectively 2, 3, 4, the circular configuration element of 5 is tested one by one to its information capacity, find maximum information capacity and time large structural element corresponding to information capacity, then between, choose complete " 1 " matrix of a d × d again as new construction element, d is between the circular configuration element dimensions corresponding to maximum information capacity and secondary large information capacity, when Optimized Iterative, this d × d matrix by rows is launched into 1 row, form initial antibodies group thus, initial antibodies group chromosome is encode in d × d position 0/1, namely coded word hop count is L=d × d, d is natural number,

Step 2, each antibody for initial antibodies group, calculate its fitness value, namely

$f_i=C_{\mathrm{inf\; o}}={\log }_2[1+\underset{\mathrm{\ω}}{\mathrm{\Σ}}\mathrm{Norm}(G_1,G_2)]---\left(1\right)$

Wherein f _irepresent the fitness value of i-th antibody in antibody population, i is natural number, C _inforepresent the information capacity improved, Norm (G ₁, G ₂) represent based on peak value normalization two-dimensional histogram, G ₁represent the gray scale through i-th certain pixel of antibody filtered image in image, G ₂represent the gray scale of its right adjacent pixel, ω is the cumulative constraint of information capacity, can be expressed as:

$\mathrm{\ω}=\left\{\begin{array}{c}|G_1-\frac{1}{2}(G_{\max }+G_{\min })|\≤T_1\\ |G_1-G_2|\≤T_2\end{array}\right.---\left(2\right)$

T ₁, T ₂be called as nonnegativity restrictions threshold value, G _maxand G _minrepresent the maxima and minima of the gray scale in the two-dimensional histogram used by logarithm normalization respectively, in the optimization of structural element, get G _min=0, G _max=255, T ₁=128, T ₂=2;

Step 3, employing elite retention strategy, select m maximum antibody of adaptive value to be placed in elite storehouse as memory antibody to be retained as elite colony, do not participate in selection, crossover and mutation operation, directly being taken as excellent individual joins in the colony of new generation of generation, wherein m=(15% ~ 20%) N, m round numbers, N is population scale;

Step 4, take the antibody concentration of the quick calculation method antagonist group of the hamming distance based on XOR to calculate, the basis of antibody concentration calculates replication rate e _k;

Replication rate: $e_k=\frac{f_k}{{\left(C_k\right)}^{\mathrm{\β}}}---\left(3\right)$

F _kfor the fitness of antibody k, C _kfor the concentration of antibody k, β is the concentration of reflection antibody and the important parameter of fitness proportion occupied by expecting in breeding potential, gets β=2 here, the concentration C of antibody k _kthe fast algorithm based on the hamming distance of XOR is adopted to calculate, namely

$C_k=\frac{1}{N}\underset{w=1}{\overset{N}{\mathrm{\Σ}}}{a}_{\mathrm{kw}},a_{\mathrm{kw}}=\left\{\begin{array}{c}1,D\≤t\\ 0,D>t\end{array}\right.---\left(4\right)$

C _kfor the concentration of antibody k, N is population scale, a _kwbe the affinity between two antibody k and w, D is the hamming distance of two antibody, and t is hamming distance threshold value, and t=0.3*L, L are length and the coded word hop count of character string, k and w is natural number;

Step 5, for select probability p _skcalculate, select probability p can be obtained according to step 4 _sk, based on p _skvalue antagonist colony select, and carry out interlace operation

$p_{\mathrm{sk}}=\frac{e_k}{\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{e}_i}---\left(5\right)$

Wherein, e _kfor the replication rate of antibody k, N is population scale, and i is the natural number being less than or equal to N, e _ithe replication rate of any one antibody i;

Step 6, employing add adjustable factors θ, obtain new antibodies group; The computing formula adding the adaptive mutation probability of adjustable factors is:

$P_m=\left\{\begin{array}{c}\frac{k_3\left[(f_{\max }-f)\right]}{f_{\max }-f_{\mathrm{avg}}}+\mathrm{\θ},f>f_{\mathrm{avg}}\\ k_4,f\≤f_{\mathrm{favg}}\end{array}\right.---\left(6\right)$

$\mathrm{\θ}=-\frac{k}{G_{\max }-1}G+k+\frac{k}{G_{\max }-1}---\left(7\right)$

F in formula _maxrepresent the maximal value of fitness; f _avgrepresent arithmetic mean fitness value; F is ideal adaptation angle value; k ₃, k ₄refer to the regulation coefficient between 0 and 1, get k ₃=k ₄=0.1, θ is adjustable factors, and G is evolutionary generation; G _maxfor maximum evolutionary generation; The value of adjustable factors θ, here k=0.005 when k is the first generation;

Step 7, antagonist group upgrade, and call high fitness value individuality in elite colony and replace the individuality of low adaptive value in antibody population, generate antibody population of future generation;

Step 8, according to end condition to judging, if the words met terminate to optimize, export the structural element after optimizing, if do not meet, jump to step 2 and repeat, end condition is one of following: a, definition threshold epsilon=0.0001, for each antibody population, calculate the average fitness of this antibody population, the difference of the arithmetic mean fitness value of this antibody population and the arithmetic mean fitness value of previous generation antibody population is less than ε, b, continuous 15 generation antibody population the highest fitness value remain unchanged, c, reach maximum evolutionary generation.