CN103578111A - Rotary kiln firing state recognition method based on flame image structure similarity - Google Patents

Abstract

The invention relates to a rotary kiln firing state recognition method based on flame picture structure similarity. According to the technical scheme, the method includes the steps that a standard flame gray level image y, a normal flame gray level image storage FN and an abnormal flame gray level image storage FA are set; a to-be-monitored flame image P is obtained, and filtering and grey level transformation are carried out on the to-be-monitored flame image P to obtain a to-be-monitored flame gray level image x; average structural similarity coefficient calculation is carried out on the to-be-monitored flame gray level image x and all standard flame gray level images to obtain a+b MSSIM coefficients (x,y); the maximum value MAXssim of the MSSIM coefficients is selected, if the standard flame gray level image y corresponding to the maximum value MAXssim of the MSSIM coefficients belongs to the normal flame gray level image storage FN, the to-be-detected flame image P belongs to a normal state, and otherwise to-be-monitored flame image P belongs to an abnormal state. The method has the advantages of being high in precision, low in computation complexity, short in processing procedure and capable of achieving online real-time monitoring of flame state changes.

Description

Rotary kiln based on flame image structural similarity burns till state identification method

Technical field

The invention belongs to the technical field that rotary kiln burns till state recognition.Be particularly related to a kind of rotary kiln based on flame image structural similarity and burn till state identification method.

Background technology

Rotary kiln is that calcining or roasting and other modes are processed Thermal Equipment used in various raw material of industry technique, for the material of input is carried out to machinery, physical or chemical treatment, in the industries such as building materials, chemical industry, metallurgy, has a wide range of applications.Be subject to the singularity of rotary structure and the impact of process complexity, gained clinker quality index is difficult to on-line measurement, the state that burns till of grog is also difficult to accurate identification, add the factors such as the multivariate strong coupling characteristic of rotary kiln process and uncertain noises, make rotary kiln operation process open loop operational phase in " manually seeing fire " still, be difficult to realize the automatic control of rotary kiln, long-time running easily causes that clinker quality index is unstable, production capacity is low, energy consumption is high and the problem such as hand labor intensity is large.

Flame image is the effecting reaction of combustion process, and monochrome information has reflected radiation temperature and the combustion efficiency in combustion process, and flame profile has reflected the shape of combustion reaction generation area.Therefore, utilize the state that burns till of rational technology judgement flame image, thereby determine running status and the combustion stability of rotary kiln, tool is of great significance.

In recent years, one of study hotspot that rotary kiln burns till status recognition technique is digital image processing techniques.After image pre-service, image are cut apart, image carried out feature extraction and chosen, the characteristics of image that extracts and choose is sent into the pattern-recognition of carrying out target in pattern classifier, thereby judging that flame burns till state.Up to the present, numerous scholars have done a large amount of and deep research to the modules of Digital Image Processing, though the recognition methods of using digital image processing techniques judgement flame image to burn till state has accuracy of identification high, the advantage that algorithm is various and ripe, but it is high that these class methods inevitably exist computation complexity, processing time is long, defect that cannot on-line real time monitoring, fundamentally do not change the pattern of " manually seeing fire ", and affect the safety and reliability of rotary kiln, cause that clinker quality index is unstable, low and the high in cost of production series of problems of production capacity.

Summary of the invention

The present invention is intended to overcome prior art defect, and the rotary kiln based on flame image structural similarity that object is to provide that a kind of precision is high, computation complexity is low, processing procedure is short and can realizes on-line real time monitoring burns till state identification method.

For completing above-mentioned task, the concrete steps of the technical solution used in the present invention are:

The first step, standard flame source image Q are demarcated by kiln operation expert; Standard flame source image Q forms standard flame source picture library L, standard flame source picture library L is divided into normal flame picture library LN and abnormal flame picture library LA, normal flame picture library LN is comprised of the standard flame source image Q of a width normal condition, and abnormal flame picture library LA is comprised of the standard flame source image Q of b width abnormality; Standard flame source image Q is carried out to filtering processing and greyscale transformation, obtain standard flame source gray level image y; All standard flame source gray level image y form standard flame source gray scale picture library F, and standard flame source gray scale picture library F is divided into normal flame gray scale picture library FN and abnormal flame gray scale picture library FA.

Second step, from the flame video of the rotary kiln clinkering zone that gathers, obtain a flame image P to be measured; Flame image P to be measured is carried out to filtering processing and greyscale transformation, obtain flame gray level image x to be measured.

The computing method of the 3rd step, employing average structure likeness coefficient, calculate the average structure likeness coefficient MSSIM (x between a flame gray level image x to be measured and each width standard flame source gray level image y, y), obtain a+b average structure likeness coefficient MSSIM (x, y).

The 4th step, a+b the average structure likeness coefficient MSSIM (x, y) that the 3rd step is obtained compare, and choose average structure likeness coefficient maximal value MAXmssim; If the corresponding standard flame source gray level image of average structure likeness coefficient maximal value MAXmssim y belongs to normal flame gray scale picture library FN, the state that burns till of judging flame image P to be measured is normal condition; If the corresponding standard flame source gray level image of average structure likeness coefficient maximal value MAXmssim y belongs to abnormal flame gray scale picture library FA, the state that burns till of judging flame image P to be measured is abnormality.

If it is normal condition that the 5th step is judged the state that burns till of flame image P to be measured, directly carry out the identification that next width flame image burns till state;

If judge, the state that burns till of flame image P to be measured is abnormality, and system alarm, improves rotary kiln flame and burn till after state, then carries out the identification that next width flame image burns till state.

The 6th step, repeat the second～five step, until finish.

Described a is 1～1000 natural number; Described b is 1～1000 natural number.

The computing method of described average structure likeness coefficient are: by flame gray level image x to be measured and standard flame source gray level image y respectively with the window of formed objects by pixel from the upper left corner, to the lower right corner, move, obtain M flame gray level image piece x to be measured _jwith M standard flame source gray level image piece y _j, calculate each flame gray level image piece x to be measured _jwith corresponding standard flame source gray level image piece y _jbetween structural similarity coefficient S SIM (x _j, y _j), obtain M structural similarity coefficient S SIM (x _j, y _j), then to M structural similarity coefficient S SIM (x _j, y _j) carry out progressive mean, obtain the average structure likeness coefficient MSSIM (x, y) of flame gray level image x to be measured and standard flame source gray level image y:

$\mathrm{MSSIM}(x,y)=\frac{1}{M}\underset{j=1}{\overset{M}{\mathrm{\Σ}}}\mathrm{SSIM}(x_j,y_j)---\left(1\right)$

In formula (1): SSIM (x _j, y _j) be flame gray level image piece x to be measured _jwith corresponding standard flame source gray level image piece y _jstructural similarity property coefficient:

$\mathrm{SSIM}(x_j,y_j)=\frac{{({2u}_{x,j}{u}_{y,j}+C_1)}^2}{{({u_{x,j}}^2+{u_{y,j}}^2+C_2)}^2}\·\frac{{2\mathrm{\σ}}_{\mathrm{xy},j}+C_2}{{{\mathrm{\σ}}_{x,j}}^2+{{\mathrm{\σ}}_{y,j}}^2+C_2}---\left(2\right)$

In formula (2): C ₁=6.5025, C ₂=58.5225;

U _{x, j}for flame gray level image piece x to be measured _javerage:

$u_{x,j}=\frac{1}{N}\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{x}_{j,i}---\left(3\right)$

U _y,jfor standard flame source gray level image piece y _javerage:

$u_{y,j}=\frac{1}{N}\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{y}_{j,i}---\left(4\right)$

σ _x,jfor flame gray level image piece x to be measured _jstandard deviation:

${\mathrm{\σ}}_{x,j}={\left(\frac{1}{N-1}\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{(x_{j,i}-u_{x,j})}^2\right)}^{1/2}---\left(5\right)$

σ _y,jfor standard flame source gray level image piece y _jstandard deviation:

${\mathrm{\σ}}_{y,j}={\left(\frac{1}{N-1}\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{(y_{j,i}-u_{y,j})}^2\right)}^{1/2}---\left(6\right)$

σ _{xy, j}for standard flame source gray level image piece y _jwith flame gray level image piece x to be measured _jbetween covariance:

${\mathrm{\σ}}_{\mathrm{xy},j}=\frac{1}{N-1}\underset{i=1}{\overset{N}{\mathrm{\Σ}}}(x_{j,i}-u_{x,j})(y_{j,i}-u_{y,j})---\left(7\right)$

In formula (3)～(7): N is flame gray level image piece x to be measured _jwith standard flame source gray level image piece y _jpixel number;

X _j,ifor flame gray level image piece x to be measured _jthe value of i pixel;

Y _j,ifor standard flame source gray level image piece y _jthe value of i pixel.

Owing to adopting technique scheme, the present invention compared with prior art tool has the following advantages:

The present invention adopts structural similarity index first in the technical field of rotary kiln flame image identification, differentiates rotary kiln burn till state from brightness, contrast and three kinds of different angles of structure, and discrimination precision is higher; In addition, the method belongs to a kind of image quality evaluating method that more meets human visual system's ultimate principle, differentiates the result subjective assessment of more fitting.

The method of discrimination that the present invention uses, without flame image is carried out to special training and study, greatly reduces computation complexity, has dwindled handling duration.

The present invention can independently differentiate the state that burns till of single width flame image, meet the condition that on-line real time monitoring rotary kiln Flame changes, can within the extremely short time, find that rotary kiln flame burns till the variation of state, and make in time corresponding adjustment, thereby increase the safety and reliability of rotary kiln, make the clinker quality of production more stable, can greatly reduce commercial production cost, for realizing the closed-loop control of monitoring in real time to whole rotary system, lay the foundation, for real realization, with " machine is seen fire " replacement, " manually see fire " basic guarantee is provided.

Therefore, the present invention has the advantages that precision is high, computation complexity is low, processing procedure is short and can realize the variation of on-line real time monitoring flame status.

Accompanying drawing explanation

Below in conjunction with drawings and the embodiments, the present invention is described in further detail:

Fig. 1 is a kind of process flow diagram of the present invention;

Fig. 2 is the flame image P to be measured in embodiment 1;

Fig. 3 is the flame gray level image x to be measured in embodiment 1;

Fig. 4 is the flame image P to be measured in embodiment 2;

Fig. 5 is the flame gray level image x to be measured in embodiment 2;

Fig. 6 is the flame image P to be measured in embodiment 3;

Fig. 7 is the flame gray level image x to be measured in embodiment 3;

Fig. 8 is the flame image P to be measured in embodiment 4;

Fig. 9 is the flame gray level image x to be measured in embodiment 4.

Embodiment

Below in conjunction with the drawings and specific embodiments, the invention will be further described, not the restriction to its protection domain:

Embodiment 1

A kind of Alumina Rotary Kiln based on flame image structural similarity burns till state identification method.The concrete steps that this rotary kiln burns till state identification method are as shown in Figure 1:

The first step, standard flame source image Q are demarcated by Alumina Rotary Kiln operation expert, and standard flame source image Q is of a size of 512 * 384 * 3.Standard flame source image Q forms standard flame source picture library L, standard flame source picture library L is divided into normal flame picture library LN and abnormal flame picture library LA, normal flame picture library LN is comprised of the standard flame source image Q of 20 width normal conditions, and abnormal flame picture library LA is comprised of the standard flame source image Q of 20 width abnormality; Standard flame source image Q is carried out to filtering and greyscale transformation, obtain standard flame source gray level image y; All standard flame source gray level image y form standard flame source gray scale picture library F, and standard flame source gray scale picture library F is divided into normal flame gray scale picture library FN and abnormal flame gray scale picture library FA.

The described wavelet decomposition that is filtered into is extracted low pass part.

Described standard flame source gray level image y is of a size of 256 * 192.

Second step, from the Alumina Rotary Kiln clinkering zone flame video gathering, obtain width colour as shown in Figure 2 flame image P to be measured, flame image P to be measured is of a size of 512 * 384 * 3; Flame image P to be measured is carried out to wavelet decomposition and extract low pass part and greyscale transformation, obtain being of a size of as shown in Figure 3 256 * 192 flame gray level image x to be measured.

The computing method of the 3rd step, employing average structure likeness coefficient; calculate the average structure likeness coefficient MSSIM (x between a flame gray level image x to be measured and each width standard flame source gray level image y; y); obtain 20 average structure likeness coefficient MSSIM (x between this flame gray level image x to be measured and the normal flame gray scale of 20 width picture library FN Plays flame gray level image y; y); obtain 20 the average structure likeness coefficient MSSIM (x, y) between this flame gray level image x to be measured and the abnormal flame gray scale of 20 width picture library FA Plays flame gray level image y simultaneously.

40 average structure likeness coefficient MSSIM (x, y) are as shown in table 1.

Table 1

The 4th step, according to shown in table 1,40 average structure likeness coefficient MSSIM (x, y) that the 3rd step is obtained compare, choosing average structure likeness coefficient maximal value MAXmssim is 0.6238.Because the corresponding standard flame source gray level image of average structure likeness coefficient maximal value MAXmssim y belongs to normal flame gray scale picture library FN, the state that burns till of judging flame image P to be measured is normal condition.

If it is normal condition that the 5th step is judged the state that burns till of flame image P to be measured, directly carry out the identification that next width flame image burns till state;

If judge, the state that burns till of flame image P to be measured is abnormality, and system alarm, improves rotary kiln flame and burn till after state, then carries out the identification that next width flame image burns till state.

Because the condition judgement that burns till of this flame image P to be measured is normal condition, therefore directly carry out the identification that next width flame image burns till state.

The 6th step, repeat the second～five step, until finish.

The computing method of the average structure likeness coefficient described in the present embodiment are: by flame gray level image x to be measured and standard flame source gray level image y respectively with 4 * 4 window by pixel from the upper left corner, to the lower right corner, move, obtain 193929 flame gray level image piece x to be measured _jwith 193929 standard flame source gray level image piece y _j, calculate each flame gray level image piece x to be measured _jwith corresponding standard flame source gray level image piece y _jbetween structural similarity coefficient S SIM (x _j, y _j), obtain 193929 structural similarity coefficient S SIM (x _j, y _j), then to 193929 structural similarity coefficient S SIM (x _j, y _j) carry out progressive mean, obtain the average structure likeness coefficient MSSIM (x, y) of flame gray level image x to be measured and standard flame source gray level image y:

$\mathrm{MSSIM}(x,y)=\frac{1}{M}\underset{j=1}{\overset{M}{\mathrm{\Σ}}}\mathrm{SSIM}(x_j,y_j)---\left(1\right)$

In formula (1): M=193929;

SSIM (x _j, y _j) be flame gray level image piece x to be measured _jwith corresponding standard flame source gray level image piece y _jstructural similarity property coefficient:

$\mathrm{SSIM}(x_j,y_j)=\frac{{({2u}_{x,j}{u}_{y,j}+C_1)}^2}{{({u_{x,j}}^2+{u_{y,j}}^2+C_2)}^2}\·\frac{{2\mathrm{\σ}}_{\mathrm{xy},j}+C_2}{{{\mathrm{\σ}}_{x,j}}^2+{{\mathrm{\σ}}_{y,j}}^2+C_2}---\left(2\right)$

In formula (2): C ₁=6.5025, C ₂=58.5225;

U _x,jfor flame gray level image piece x to be measured _javerage:

$u_{x,j}=\frac{1}{N}\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{x}_{j,i}---\left(3\right)$

U _y,jfor standard flame source gray level image piece y _javerage:

$u_{y,j}=\frac{1}{N}\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{y}_{j,i}---\left(4\right)$

σ _x,jfor flame gray level image piece x to be measured _jstandard deviation:

${\mathrm{\σ}}_{x,j}={\left(\frac{1}{N-1}\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{(x_{j,i}-u_{x,j})}^2\right)}^{1/2}---\left(5\right)$

σ _y,jfor standard flame source gray level image piece y _jstandard deviation:

${\mathrm{\σ}}_{y,j}={\left(\frac{1}{N-1}\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{(y_{j,i}-u_{y,j})}^2\right)}^{1/2}---\left(6\right)$

σ _{xy, j}for standard flame source gray level image piece y _jwith flame gray level image piece x to be measured _jbetween covariance:

${\mathrm{\σ}}_{\mathrm{xy},j}=\frac{1}{N-1}\underset{i=1}{\overset{N}{\mathrm{\Σ}}}(x_{j,i}-u_{x,j})(y_{j,i}-u_{y,j})---\left(7\right)$

In formula (3)～(7): N=16 is flame gray level image piece x to be measured _jwith standard flame source gray level image piece y _jpixel number;

X _j,ifor flame gray level image piece x to be measured _jthe value of i pixel;

Y _j,ifor standard flame source gray level image piece y _jthe value of i pixel.

Embodiment 2

A kind of Alumina Rotary Kiln based on flame image structural similarity burns till state identification method.The concrete steps that this rotary kiln burns till state identification method are as shown in Figure 1:

Described in the first step, this step, be filtered into 20 rank Butterworth LPF filtering; Described standard flame source gray level image y is of a size of 512 * 384.All the other are with embodiment 1 first step.

Second step, from the Alumina Rotary Kiln clinkering zone flame video gathering, obtain width colour as shown in Figure 4 flame image P to be measured, flame image P to be measured is of a size of 512 * 384 * 3; Flame image P to be measured with 20 rank Butterworth LPF filtering and carry out greyscale transformation, is obtained being of a size of as shown in Figure 5 512 * 384 flame gray level image x to be measured.

The 3rd step, this step except 40 average structure likeness coefficient MSSIM (x, y) as shown in table 2 outside, all the other are with the 3rd step of embodiment 1.

Table 2

The 4th step, according to shown in table 2,40 average structure likeness coefficient MSSIM (x, y) that the 3rd step is obtained compare, choosing average structure likeness coefficient maximal value MAXmssim is 0.6439.Because the corresponding standard flame source gray level image of average structure likeness coefficient maximal value MAXmssim y belongs to abnormal flame gray scale picture library FA, the state that burns till of judging flame image P to be measured is abnormality.

If it is normal condition that the 5th step is judged the state that burns till of flame image P to be measured, directly carry out the identification that next width flame image burns till state;

If judge, the state that burns till of flame image P to be measured is abnormality, and system alarm, improves rotary kiln flame and burn till after state, then carries out the identification that next width flame image burns till state.

Because the condition judgement that burns till of this flame image P to be measured is abnormality, therefore system alarm improves rotary kiln flame and burns till after state, then carry out the identification that next width flame image burns till state.

The 6th step, repeat the second～five step, until finish.

The computing method of the average structure likeness coefficient described in the present embodiment are: by flame gray level image x to be measured and standard flame source gray level image y respectively with 8 * 8 window by pixel from the upper left corner, to the lower right corner, move, obtain 190385 flame gray level image piece x to be measured _jwith 190385 standard flame source gray level image piece y _j, calculate each flame gray level image piece x to be measured _jwith corresponding standard flame source gray level image piece y _jbetween structural similarity coefficient S SIM (x _j, y _j), obtain 190385 structural similarity coefficient S SIM (x _j, y _j), then to 190385 structural similarity coefficient S SIM (x _j, y _j) carry out progressive mean, obtain the average structure likeness coefficient MSSIM (x, y) of flame gray level image x to be measured and standard flame source gray level image y:

$\mathrm{MSSIM}(x,y)=\frac{1}{M}\underset{j=1}{\overset{M}{\mathrm{\Σ}}}\mathrm{SSIM}(x_j,y_j)---\left(1\right)$

In formula (1): M=190385;

SSIM (x _j, y _j) be flame gray level image piece x to be measured _jwith corresponding standard flame source gray level image piece y _jstructural similarity property coefficient:

$\mathrm{SSIM}(x_j,y_j)=\frac{{({2u}_{x,j}{u}_{y,j}+C_1)}^2}{{({u_{x,j}}^2+{u_{y,j}}^2+C_2)}^2}\·\frac{{2\mathrm{\σ}}_{\mathrm{xy},j}+C_2}{{{\mathrm{\σ}}_{x,j}}^2+{{\mathrm{\σ}}_{y,j}}^2+C_2}---\left(2\right)$

In formula (2): C ₁=6.5025, C ₂=58.5225;

U _x,jfor flame gray level image piece x to be measured _javerage:

$u_{x,j}=\frac{1}{N}\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{x}_{j,i}---\left(3\right)$

U _y,jfor standard flame source gray level image piece y _javerage:

$u_{y,j}=\frac{1}{N}\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{y}_{j,i}---\left(4\right)$

σ _x,jfor flame gray level image piece x to be measured _jstandard deviation:

${\mathrm{\σ}}_{x,j}={\left(\frac{1}{N-1}\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{(x_{j,i}-u_{x,j})}^2\right)}^{1/2}---\left(5\right)$

σ _{y, j}for standard flame source gray level image piece y _jstandard deviation:

${\mathrm{\σ}}_{y,j}={\left(\frac{1}{N-1}\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{(y_{j,i}-u_{y,j})}^2\right)}^{1/2}---\left(6\right)$

σ _{xy, j}for standard flame source gray level image piece y _jwith flame gray level image piece x to be measured _jbetween covariance:

${\mathrm{\σ}}_{\mathrm{xy},j}=\frac{1}{N-1}\underset{i=1}{\overset{N}{\mathrm{\Σ}}}(x_{j,i}-u_{x,j})(y_{j,i}-u_{y,j})---\left(7\right)$

In formula (3)～(7): N=64 is flame gray level image piece x to be measured _jwith standard flame source gray level image piece y _jpixel number;

X _j,ifor flame gray level image piece x to be measured _jthe value of i pixel;

Y _j,ifor standard flame source gray level image piece y _jthe value of i pixel.

Embodiment 3

A kind of cement rotary kiln based on flame image structural similarity burns till state identification method.The concrete steps that this rotary kiln burns till state identification method are as shown in Figure 1:

The first step, standard flame source image Q are demarcated by cement rotary kiln operation expert, and standard flame source image Q is of a size of 352 * 288 * 3.Standard flame source image Q forms standard flame source picture library L, standard flame source picture library L is divided into normal flame picture library LN and abnormal flame picture library LA, normal flame picture library LN is comprised of the standard flame source image Q of 15 width normal conditions, and abnormal flame picture library LA is comprised of the standard flame source image Q of 10 width abnormality; Standard flame source image Q is carried out to filtering and greyscale transformation, obtain standard flame source gray level image y; All standard flame source gray level image y form standard flame source gray scale picture library F, and standard flame source gray scale picture library F is divided into normal flame gray scale picture library FN and abnormal flame gray scale picture library FA.

The described threshold filter that is filtered into.

Described standard flame source gray level image y is of a size of 352 * 288.

Second step, from the cement rotary kiln clinkering zone flame video gathering, obtain width colour as shown in Figure 6 flame image P to be measured, flame image P to be measured is of a size of 352 * 288 * 3; Flame image P to be measured is carried out to threshold filter and greyscale transformation, obtain being of a size of as shown in Figure 7 352 * 288 flame gray level image x to be measured.

The computing method of the 3rd step, employing average structure likeness coefficient; calculate the average structure likeness coefficient MSSIM (x between a flame gray level image x to be measured and each width standard flame source gray level image y; y); obtain 15 average structure likeness coefficient MSSIM (x between this flame gray level image x to be measured and the normal flame gray scale of 15 width picture library FN Plays flame gray level image y; y); obtain 10 the average structure likeness coefficient MSSIM (x, y) between this flame gray level image x to be measured and the abnormal flame gray scale of 10 width picture library FA Plays flame gray level image y simultaneously.

25 average structure likeness coefficient MSSIM (x, y) are as shown in table 3.

Table 3

The 4th step, according to shown in table 3,25 average structure likeness coefficient MSSIM (x, y) that the 3rd step is obtained compare, choosing average structure likeness coefficient maximal value MAXmssim is 0.7616.Because the corresponding standard flame source gray level image of average structure likeness coefficient maximal value MAXmssim y belongs to normal flame gray scale picture library FN, the state that burns till of judging flame image P to be measured is normal condition.

The 5th step, with the 5th step of embodiment 1.

The 6th step, with the 6th step of embodiment 1.

The computing method of described average structure likeness coefficient are: by flame gray level image x to be measured and standard flame source gray level image y respectively with 11 * 11 window by pixel from the upper left corner, to the lower right corner, move, obtain 95076 flame gray level image piece x to be measured _jwith 95076 standard flame source gray level image piece y _j, calculate each flame gray level image piece x to be measured _jwith corresponding standard flame source gray level image piece y _jbetween structural similarity coefficient S SIM (x _j, y _j), obtain 95076 structural similarity coefficient S SIM (x _j, y _j), then to 95076 structural similarity coefficient S SIM (x _j, y _j) carry out progressive mean, obtain the average structure likeness coefficient MSSIM (x, y) of flame gray level image x to be measured and standard flame source gray level image y:

$\mathrm{MSSIM}(x,y)=\frac{1}{M}\underset{j=1}{\overset{M}{\mathrm{\Σ}}}\mathrm{SSIM}(x_j,y_j)---\left(1\right)$

In formula (1): M=95076;

SSIM (x _j, y _j) be flame gray level image piece x to be measured _jwith corresponding standard flame source gray level image piece y _jstructural similarity property coefficient:

$\mathrm{SSIM}(x_j,y_j)=\frac{{({2u}_{x,j}{u}_{y,j}+C_1)}^2}{{({u_{x,j}}^2+{u_{y,j}}^2+C_2)}^2}\·\frac{{2\mathrm{\σ}}_{\mathrm{xy},j}+C_2}{{{\mathrm{\σ}}_{x,j}}^2+{{\mathrm{\σ}}_{y,j}}^2+C_2}---\left(2\right)$

In formula (2): C ₁=6.5025, C ₂=58.5225;

U _x,jfor flame gray level image piece x to be measured _javerage:

$u_{x,j}=\frac{1}{N}\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{x}_{j,i}---\left(3\right)$

U _y,jfor standard flame source gray level image piece y _javerage:

$u_{y,j}=\frac{1}{N}\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{y}_{j,i}---\left(4\right)$

σ _x,jfor flame gray level image piece x to be measured _jstandard deviation:

${\mathrm{\σ}}_{x,j}={\left(\frac{1}{N-1}\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{(x_{j,i}-u_{x,j})}^2\right)}^{1/2}---\left(5\right)$

σ _y,jfor standard flame source gray level image piece y _jstandard deviation:

${\mathrm{\σ}}_{y,j}={\left(\frac{1}{N-1}\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{(y_{j,i}-u_{y,j})}^2\right)}^{1/2}---\left(6\right)$

σ _{xy, j}for standard flame source gray level image piece y _jwith flame gray level image piece x to be measured _jbetween covariance:

${\mathrm{\σ}}_{\mathrm{xy},j}=\frac{1}{N-1}\underset{i=1}{\overset{N}{\mathrm{\Σ}}}(x_{j,i}-u_{x,j})(y_{j,i}-u_{y,j})---\left(7\right)$

In formula (3)～(7): N=121 is flame gray level image piece x to be measured _jwith standard flame source gray level image piece y _jpixel number;

X _j,ifor flame gray level image piece x to be measured _jthe value of i pixel;

Y _j,ifor standard flame source gray level image piece y _jthe value of i pixel.

Embodiment 4

A kind of cement rotary kiln based on flame image structural similarity burns till state identification method.The concrete steps that this rotary kiln burns till state identification method are as shown in Figure 1:

Described in the first step, this step, be filtered into region of interesting extraction; Described standard flame source gray level image y is of a size of 225 * 189.All the other are with embodiment 3 first steps.

Second step, from the cement rotary kiln clinkering zone flame video gathering, obtain width colour as shown in Figure 8 flame image P to be measured, flame image P to be measured is of a size of 352 * 288 * 3; Flame image P to be measured is carried out to region of interesting extraction and greyscale transformation, obtain being of a size of as shown in Figure 9 225 * 189 flame gray level image x to be measured.

The 3rd step, this step except 25 average structure likeness coefficient MSSIM (x, y) as shown in table 4 outside, all the other are with the 3rd step of embodiment 3.

Table 4

The 4th step, according to shown in table 4,25 average structure likeness coefficient MSSIM (x, y) that the 3rd step is obtained compare, choosing average structure likeness coefficient maximal value MAXmssim is 0.7720.Because the corresponding standard flame source gray level image of average structure likeness coefficient maximal value MAXmssim y belongs to abnormal flame gray scale picture library FA, the state that burns till of judging flame image P to be measured is abnormality.

The 5th step, with the 5th step of embodiment 2.

The 6th step, with the 6th step of embodiment 2.

The computing method of described average structure likeness coefficient are with the computing method of the average structure likeness coefficient of embodiment 3.

This embodiment compared with prior art tool has the following advantages:

This embodiment adopts structural similarity index first in the technical field of rotary kiln flame image identification, differentiates rotary kiln burn till state from brightness, contrast and three kinds of different angles of structure, and discrimination precision is higher; In addition, the method belongs to a kind of image quality evaluating method that more meets human visual system's ultimate principle, differentiates the result subjective assessment of more fitting.

The method of discrimination that this embodiment is used, without flame image is carried out to special training and study, greatly reduces computation complexity, has dwindled handling duration.

This embodiment can independently be differentiated the state that burns till of single width flame image, meet the condition that on-line real time monitoring rotary kiln Flame changes, can within the extremely short time, find that rotary kiln flame burns till the variation of state, and make in time corresponding adjustment, thereby increase the safety and reliability of rotary kiln, make the clinker quality of production more stable, can greatly reduce commercial production cost, for realizing the closed-loop control of monitoring in real time to whole rotary system, lay the foundation, for real realization, with " machine is seen fire " replacement, " manually see fire " basic guarantee is provided.

Therefore, this embodiment has the advantages that precision is high, computation complexity is low, processing procedure is short and can realize the variation of on-line real time monitoring flame status.

Claims (3)

1. the rotary kiln based on flame image structural similarity burns till a state identification method, it is characterized in that described rotary kiln burns till the concrete steps of state identification method:

The first step, standard flame source image Q are demarcated by kiln operation expert; Standard flame source image Q forms standard flame source picture library L, standard flame source picture library L is divided into normal flame picture library LN and abnormal flame picture library LA, normal flame picture library LN is comprised of the standard flame source image Q of a width normal condition, and abnormal flame picture library LA is comprised of the standard flame source image Q of b width abnormality; Standard flame source image Q is carried out to filtering processing and greyscale transformation, obtain standard flame source gray level image y; All standard flame source gray level image y form standard flame source gray scale picture library F, and standard flame source gray scale picture library F is divided into normal flame gray scale picture library FN and abnormal flame gray scale picture library FA;

Second step, from the flame video of the rotary kiln clinkering zone that gathers, obtain a flame image P to be measured; Flame image P to be measured is carried out to filtering processing and greyscale transformation, obtain flame gray level image x to be measured;

The computing method of the 3rd step, employing average structure likeness coefficient, calculate the average structure likeness coefficient MSSIM (x between a flame gray level image x to be measured and each width standard flame source gray level image y, y), obtain a+b average structure likeness coefficient MSSIM (x, y);

The 4th step, a+b the average structure likeness coefficient MSSIM (x, y) that the 3rd step is obtained compare, and choose average structure likeness coefficient maximal value MAXmssim; If the corresponding standard flame source gray level image of average structure likeness coefficient maximal value MAXmssim y belongs to normal flame gray scale picture library FN, the state that burns till of judging flame image P to be measured is normal condition; If the corresponding standard flame source gray level image of average structure likeness coefficient maximal value MAXmssim y belongs to abnormal flame gray scale picture library FA, the state that burns till of judging flame image P to be measured is abnormality;

If it is normal condition that the 5th step is judged the state that burns till of flame image P to be measured, directly carry out the identification that next width flame image burns till state;

If judge, the state that burns till of flame image P to be measured is abnormality, and system alarm, improves rotary kiln flame and burn till after state, then carries out the identification that next width flame image burns till state;

The 6th step, repeat the second～five step, until finish.

2. according to the rotary kiln based on flame image structural similarity claimed in claim 1, burn till state identification method, it is characterized in that described a is 1～1000 natural number; Described b is 1～1000 natural number.

3. according to the rotary kiln based on flame image structural similarity claimed in claim 1, burn till state identification method, the computing method that it is characterized in that described average structure likeness coefficient are: by flame gray level image x to be measured and standard flame source gray level image y respectively with the window of formed objects by pixel from the upper left corner, to the lower right corner, move, obtain M flame gray level image piece x to be measured _jwith M standard flame source gray level image piece y _j, calculate each flame gray level image piece x to be measured _jwith corresponding standard flame source gray level image piece y _jbetween structural similarity coefficient S SIM (x _j, y _j), obtain M structural similarity coefficient S SIM (x _j, y _j); Then to M structural similarity coefficient S SIM (x _j, y _j) carry out progressive mean, obtain the average structure likeness coefficient MSSIM (x, y) of flame gray level image x to be measured and standard flame source gray level image y:

$\mathrm{MSSIM}(x,y)=\frac{1}{M}\underset{j=1}{\overset{M}{\mathrm{\Σ}}}\mathrm{SSIM}(x_j,y_j)---\left(1\right)$

In formula (1): SSIM (x _j, y _j) be flame gray level image piece x to be measured _jwith corresponding standard flame source gray level image piece y _jstructural similarity property coefficient:

$\mathrm{SSIM}(x_j,y_j)=\frac{{({2u}_{x,j}{u}_{y,j}+C_1)}^2}{{({u_{x,j}}^2+{u_{y,j}}^2+C_2)}^2}\·\frac{{2\mathrm{\σ}}_{\mathrm{xy},j}+C_2}{{{\mathrm{\σ}}_{x,j}}^2+{{\mathrm{\σ}}_{y,j}}^2+C_2}---\left(2\right)$

In formula (2): C ₁=6.5025, C ₂=58.5225;

U _x,jfor flame gray level image piece x to be measured _javerage:

$u_{x,j}=\frac{1}{N}\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{x}_{j,i}---\left(3\right)$

U _y,jfor standard flame source gray level image piece y _javerage:

$u_{y,j}=\frac{1}{N}\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{y}_{j,i}---\left(4\right)$

σ _x,jfor flame gray level image piece x to be measured _jstandard deviation:

${\mathrm{\σ}}_{x,j}={\left(\frac{1}{N-1}\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{(x_{j,i}-u_{x,j})}^2\right)}^{1/2}---\left(5\right)$

σ _y,jfor standard flame source gray level image piece y _jstandard deviation:

${\mathrm{\σ}}_{y,j}={\left(\frac{1}{N-1}\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{(y_{j,i}-u_{y,j})}^2\right)}^{1/2}---\left(6\right)$

σ _{xy, j}for standard flame source gray level image piece y _jwith flame gray level image piece x to be measured _jbetween covariance:

${\mathrm{\σ}}_{\mathrm{xy},j}=\frac{1}{N-1}\underset{i=1}{\overset{N}{\mathrm{\Σ}}}(x_{j,i}-u_{x,j})(y_{j,i}-u_{y,j})---\left(7\right)$

In formula (3)～(7): N is flame gray level image piece x to be measured _jwith standard flame source gray level image piece y _jpixel number;

X _j,ifor flame gray level image piece x to be measured _jthe value of i pixel;

Y _j,ifor standard flame source gray level image piece y _jthe value of i pixel.

Images