CN101457267B - Intelligent extraction method of blast burden temperature field isotherm - Google Patents

Abstract

The invention provides an intelligent extraction method of an isotherm of a blast furnace charge level temperature field. In the method, collected infrared images of the blast furnace charge level are selected, gray scale isotherms are extracted, each gray scale isotherm is corrected and temperature values of the isotherms are calibrated respectively by adopting the optimum image selection strategy of an image gray scale distribution variance comprehensive quality index, as well as a gray scale isotherm extraction technology, a gray scale isotherm correction algorithm and an isotherm temperature calibration method which are based on statistics and a variable segmentation threshold according to digital image processing techniques. As the intelligent extraction is performed on the blast furnace charge level temperature field isotherm based on the infrared image by adopting the method, the extracted isotherm can characterize principal characteristic information of the blast furnace charge level temperature field and is characterized by simple and convenient computation. The invention provides a reliable basis for knowing the condition of the blast furnace charge level temperature field, and further provides objective and quantized reference information for judging the distribution of gas current and the charge level in the blast furnace ironmaking process; and the method has relatively strong practicability.

Description

Intelligent extraction method of blast burden temperature field isotherm

Technical field

The present invention relates to process monitoring and control technology field that Ferrous Metallurgy is produced, a kind of intelligent extraction method of blast burden temperature field isotherm particularly is provided.

Background technology

Iron and steel is as the important foundation raw material and the strategic materials of Chinese national economy and defence and military development, be widely used in industry-by-industries such as machinery, electronics, building materials, traffic, space flight, aviation, defence and military, in the national economic development, had crucial status.

Blast furnace is the key equipment of Iron and Steel Production, the internal environment complexity, and many states are difficult to direct detection.Blast furnace charge level temperature field directly influences heat energy distribution, chemical reaction, coal gas distributions and the furnace pressure in block district in the blast furnace, thereby understand the variation of blast furnace charge level temperature field in real time, to the state of development of prediction gas fluid in blast furnace, optimize blast furnace operating, guarantee that blast furnace is steady very crucial along operation.

The blast furnace infrared image is the hot state that can directly show blast furnace charge level temperature field, but because the complicacy of blast furnace internal-response, the charge level infrared image that causes thermal camera to take is subjected to multiple factor and disturbs, and causes being difficult to accurately obtain the situation of charge level temperature field.At the blast furnace characteristics of IR images, studying new method and technology handles original infrared image, extract the important information that isotherm in the image characterizes blast furnace charge level temperature field accurately and efficiently, to judging in the stove that coal gas distributes and the control blast furnace quality of production has extremely important meaning.

Summary of the invention

The problem that is difficult to accurately obtain for the characteristic information that effectively solves charge level temperature field in the large blast furnace production, the invention provides a kind of intelligent extraction method of blast burden temperature field isotherm based on infrared image, adopt this method can extract the isothermal distribution situation of charge level temperature field effectively, reflect the key character information of blast furnace charge level temperature field more exactly.

The technical scheme that the present invention solves the problems of the technologies described above may further comprise the steps:

(1) gathers multiple image, determine best pending image, optimized image is carried out mean filter, first filtering overriding noise signal according to every width of cloth gradation of image distribution variance quality comprehensive index;

(2) adopt based on statistical method and the dynamic feature extracting method that changes the segmentation threshold scope, to image F after the filtering ₁Carrying out the gray scale level line advances to extract the image F that obtains _2iAs preliminary isothermal map picture;

(3) a few width of cloth images in the last step are superposeed, then the image after the stack is carried out Filtering Processing based on square error and minimum estimate, obtain image F ₃

(4) outline line in the isothermal line image is revised,, made temperature line have more continuity as processing such as deburring, benefit breakaway poings;

(5) according to temperature and the gray scale corresponding relation between the thermopair recently in the cross temperature of the point on each bar gray scale level line, demarcate temperature value respectively, then temperature value is revised, obtain to have the blast furnace charge level isothermal map of temperature mark at last as F _Final

What the present invention proposed comes best infrared image is selected according to gradation of image distribution variance quality comprehensive index, selects to have more reliability and dirigibility with the image local value by rule of thumb than tradition.Compare with fixing traditional threshold segmentation method such as segmentation threshold, bimodal split plot design, based on statistics and dynamically variation cut the method for threshold value, more can adapt to cutting apart to the blast furnace charge level infrared image of dynamic change under the multi-state.Image is carried out cutting apart of different threshold values and is divided into several sub-pictures effectively preserving important information than the normally used Threshold Segmentation that manys on a sub-picture.

Description of drawings

Fig. 1 isotherm extracting method of the present invention process flow diagram;

Fig. 23 * 3 mean filter window figure of the present invention;

Image scanning mode synoptic diagram in the step 2 among Fig. 3 the present invention;

Fig. 4 is five width of cloth images of different threshold values after to same width of cloth image segmentation

Several cut apart the superimposed image of figure Fig. 5;

The isothermal map that Fig. 6 is rough;

Isothermal map after Fig. 7 repairs.

Embodiment

Technical scheme for a better understanding of the present invention is further described embodiments of the present invention below in conjunction with accompanying drawing in the instructions, and Fig. 1 is the realization block diagram of method.

At first, choose optimized image.Because infrared band has intrinsic resolving power and be subjected to the influence of Atmospheric Absorption and scattering process in transmission course, causes infrared image quality very strong to the dependence of shooting environmental.And high furnace interior is a high temperature, many floating dust, intensity of light source complex environment pockety, make the blast furnace infrared image usually affected by environment, too much or the under-exposed phenomenon of partial exposure appears, cause charge level information to be difficult to reflection, and according to operating experience, charge level temperature field changes inviolent, even within a short period of time can regard as stable, therefore, gather multiple image at short notice, and, choose image wherein affected by environment by comparing intensity profile information, can effectively prevent the detection misalignment that the blast furnace environmental factor causes.

This optimized image selection strategy has more reliability and dirigibility than selecting image with experience directly perceived and being worth size to be judged according to the image local pixel grey scale.

The infrared image F that the blast furnace camera acquisition is arrived _{T-4 τ}, F _{T-3 τ}, F _{T-2 τ}, F _T-τ, F _t, (F wherein _{T-i τ}Be t-i τ images acquired constantly, τ=500ms) carry out based on the screening of gradation of image distribution variance quality index, its basic procedure is as follows.

Step 1: to F _{T-4 τ}, F _{T-3 τ}, F _{T-2 τ}, F _T-τ, F _tCarry out gray-scale pixels statistics respectively,, the number of pixels of each each gray level of width of cloth image is added up, obtain because the infrared image gray scale has 0～255 totally 256 grades

Step 2: to each width of cloth gray distribution of image by formula (1) and (2) ask for variance respectively:

${\mathrm{\μ}}_{f_{t-\mathrm{n\τ}}}=\frac{1}{256}n---\left(1\right)$

${\mathrm{\σ}}_{f_{t-\mathrm{n\τ}}}^2=\frac{1}{256}\underset{m=0}{\overset{255}{\mathrm{\Σ}}}{(g_{f_{t-\mathrm{n\τ}}m}-{\mathrm{\μ}}_{f_{t-\mathrm{n\τ}}m})}^2---\left(2\right)$

Obtain every width of cloth gray distribution of image variance yields respectively

In the formula (1) Be the mean pixel number of every each gray level of width of cloth image, n is the total number of pixels of image.In the formula (2)

Be the variance of each image gray levels number of pixels, Number of pixels for each each gray level of image;

Step 3: a plurality of intensity profile variance yields of obtaining of Step 2 relatively, and to choose the image setting that the intensity profile variance yields mediates from these a few width of cloth images be optimized image, and it is advanced 3 * 3 mean filters, as pending image F ₁

The second, the gray scale level line extracts.Since in the blast furnace infrared image distribution trend of grey scale pixel value be increase gradually to central point by the edge and do not have a background image.Therefore, profile extraction, the edge extracting of outer isocontour extraction of gradation of image of blast furnace red and general pattern are different.Cut the feature extracting method of threshold range to F in this employing based on statistical method and dynamic variation ₁Image extracts the gray scale level line.Compare with traditional threshold segmentation methods such as fixing segmentation threshold and bimodals, based on statistics and dynamically variation cut the method for threshold value, more can adapt to cutting apart to blast furnace charge level infrared image under the multi-state.Concrete algorithm flow is as follows:

Step 1: determine image F ₁Left hand edge mid point A and coboundary mid point B;

Step 2: along mid point A and B image is carried out horizontal and vertical scanning, record horizontal pixel number N _AWith vertical pixel count N _B, and each pixel corresponding gray;

Step 3: to N _A+ N _BIndividual grey scale pixel value is added up.Determine segmentation threshold M _k(k=1,2,3,4,5), and M _kBe to be incremented to 255 gradually since 0, up to satisfying following condition, and the record corresponding M _kAnd threshold range.

The IF gray-scale value is less than M ₁Number of pixels N ₁∈ ((N _A+ N _B)/5-50, (N _A+ N _B)/5+50)

First threshold range of THEN is (M ₁-5, M ₁);

The IF gray-scale value is less than M ₂Number of pixels N ₂∈ (2* (N _A+ N _B)/5-50,2* (N _A+ N _B)/5+50)

Second threshold range of THEN is (M ₂-5, M ₂);

The IF gray-scale value is less than M ₃Number of pixels N ₃∈ (2* (N _A+ N _B)/5-50,2* (N _A+ N _B)/5+50)

The 3rd threshold range of THEN is (M ₃-5, M ₃);

The IF gray-scale value is less than M ₄Number of pixels N ₄∈ (2* (N _A+ N _B)/5-50,2* (N _A+ N _B)/5+50)

The 4th threshold range of THEN is (M ₄-5, M ₄);

ELSE M ₅=(255-M ₄)/2, promptly the 5th threshold range is (M ₅-5, M ₅).

Determine 5 segmentation threshold scopes successively, be respectively (M ₁-5, M ₁), (M ₂-5, M ₂), (M ₃-5, M ₃), (M ₄-5, M ₄), (M ₅-5, M ₅), because M _k(k=1,2,3,4,5) change along with the picture difference, and therefore corresponding threshold range also is dynamic change.

Step 4: to F ₁Carry out the image segmentation of many threshold ranges.To F ₁In each pixel scan, obtain gray-scale value x (i, j), carry out following binary conversion treatment then:

IF?x(i，j)∈(M _k-5，M _k)，k＝1，2，3，4，5

THEN?x(i，j)＝255；

ELSE?x(i，j)＝0。

Wherein, (i j) is the gray-scale value of pixel in the image to x.Cut apart by above-mentioned, image becomes several width of cloth and has the isocontour bianry image F of coarse gray scale ₂

The 3rd, image overlay through the processing of second step, can be extracted the gray scale contour map picture that obtain the different information of several reflection charge level temperature fields from the optimized image of choosing, these images are adopted pixel exclusive disjunction superposition algorithm, can obtain the characteristic image of the multiple information of a width of cloth concentrated expression charge level temperature field.

Consider that the pixel exclusive disjunction can bring new stationary noise, the present invention adopts the filtering method of square error and minimum estimate can obtain good denoising effect.Therefore, next carry out second layer Filtering Processing based on square error and minimum method.Determine each grey scale pixel value x (i, j) local mean value μ earlier _IjWith local variance sigma _Ij ²,

${\mathrm{\μ}}_{\mathrm{ij}}=\frac{1}{\mathrm{card}\left(\mathrm{\φ}\right)}\×\underset{\mathrm{\θ}\∈\mathrm{\φ}}{\mathrm{\Σ}}{x}_{\mathrm{ij}}^{\mathrm{\θ}}---\left(3\right)$

${\mathrm{\σ}}_{\mathrm{ij}}^2=\frac{1}{\mathrm{card}\left(\mathrm{\φ}\right)}\×\underset{\mathrm{\θ}\∈\mathrm{\φ}}{\mathrm{\Σ}}{(x_{\mathrm{ij}}^{\mathrm{\θ}}-{\mathrm{\μ}}_{\mathrm{ij}})}^2---\left(4\right)$

${\mathrm{\δ}}^2=\frac{1}{N\×M}\×\underset{i=1}{\overset{N}{\mathrm{\Σ}}}\underset{j=1}{\overset{M}{\mathrm{\Σ}}}{\mathrm{\σ}}_{\mathrm{ij}}^2---\left(5\right)$

φ={ 0 °, 45 °, 90 °, 135 °, 180 °, 225 °, 275 °, 315 ° } wherein, card (φ) is a set of computations member number.With stationary noise variance in formula (5) presentation video.Determine at last after the filtering pixel x in the image (i, gray-scale value y j) (i, j),

$y(i,j)={\mathrm{\μ}}_{\mathrm{ij}}+\frac{{\mathrm{\σ}}_{\mathrm{ij}}^2-{\mathrm{\δ}}^2}{{\mathrm{\σ}}_{\mathrm{ij}}^2}\×(x_{\mathrm{ij}}-{\mathrm{\μ}}_{\mathrm{ij}})---\left(6\right)$

By on image after the Threshold Segmentation, carrying out the filtering method of square error and minimum estimate, more can thoroughly remove isolated noise point than carrying out a filtering before the Flame Image Process, obtain filtered image F ₃

The 4th, the correction of gray scale level line.Extract the gray scale level line that obtains in previous step, since more coarse, also can there be some problems such as some isolated noise points, burr, local fracture.Therefore to the image F after cutting apart ₃Carry out the isocontour comprehensive repairing of gray scale.Concrete correction algorithm is as follows:

Step 1: breakaway poing is searched for.Along each bar gray scale level line pointwise track-while-scan, determine two reference Point C and D, to the D line, determine the isocontour trend of gray scale by C, with 24 neighborhoods is standard, with the D point is the center, carries out 7 angle scannings in the direction of gray scale level line trend, judges the gray-scale value that indicate gray pixels point of Fig. 4 in gray scale level line trend, when gray-scale value is not 0 entirely, then think breakaway poing, then forward Step 3 to, otherwise forward Step 2 to.

Step 2: burr is searched for.In the scintigram 4 on E and the F line, and the point beyond 24 neighborhoods, be 0 point, if gray scale then think jagged puts 0 to the gray-scale value of respective point if there is gray-scale value.If gray-scale value all is 255 then thinks there is not burr, forwards Step 4 to.

Step 3: breakaway poing is repaired, gray-scale value in 10 points that indicate grey among Fig. 4 be 255 be changed to 0, forward Step 2 to.

Step 4: move along gray scale level line trend, determine next group reference point, and forward Step 1 to.

By to the isocontour correction of gray scale, obtaining a width of cloth has the isocontour image F of coherent gray scale ₄, be also referred to as isotherm.

At last, isotherm temperature calibration.The method of demarcating is according to the pixel of zones of different and the corresponding relation of gray-scale value, does demarcation for respectively every gray scale level line, and concrete demarcating steps is as follows:

Step 1: search near the shortest point of distance of the thermopair of distance on 5 isotherms, note the coordinate (i of 5 points _m, j _m), (m=1,2,3,4,5) and the coordinate (i of difference corresponding thermocouples in image _k, j _k), (k=1,2,3,4,5);

Step 2: obtain 5 coordinate points respectively at image F ₁In grey scale pixel value x (i _m, j _m), (m=1,2,3,4,5) and 5 thermopairs are at image F ₁The grey scale pixel value x ' (i of middle corresponding point _k, j _k), (k=1,2,3,4,5), 5 corresponding electric thermo-couple temperature value t _ζ(ζ=1,2,3,4,5);

Step 3: the temperature value T that calculates 5 points according to formula (7) _δ, (δ=1,2,3,4,5)

$T_{\mathrm{\δ}}=\frac{t_{\mathrm{\ζ}}\·x(i_m,j_m)}{x^{\′}(i_k,j_k)}---\left(7\right)$

In the formula, δ=ζ=m=k=1,2,3,4,5, by the linear temperature calibrating method, calculate 5 temperature values, successively as 5 isothermal demarcation temperature values;

Step 4: according to the detection data such as distance of tedge temperature, cross temperature and charge level, utilize experimental formula that isothermal demarcation temperature value is carried out gamma correction.

Along with successively decreasing highly gradually, rule of thumb calculate and revise temperature value by formula (8) and (9) to the temperature of furnace roof for the temperature of supposing charge level,

$\mathrm{\ΔT}=\frac{({\stackrel{\‾}{T}}_{\mathrm{top}}-{\stackrel{\‾}{T}}_{\mathrm{\δ}})\·d_2}{d_1}---\left(8\right)$

d ₂＝L-d ₁ (9)

Wherein, Δ T is for revising temperature value, T _TopBe 4 thermopair mean values of furnace roof, T _δBe 5 isotherm temperature-averaging values based on the linear temperature calibration, d ₁Be the distance of furnace roof to cross temperature, d ₂Be the distance of cross temperature to charge level, L is a blast-furnace line;

Step 5: blast furnace charge level temperature field isotherm, gray scale-temperature map resolve into linear scaled temperature value T _δWith error compensation part Δ T, promptly

T′ _δ＝T _δ+ΔT (10)

Wherein, δ=1,2,3,4,5.

After the selection of process optimized image, the extraction of gray scale level line, the correction of gray scale level line, a series of processing of isotherm temperature calibration, realize the isothermal extraction of blast furnace charge level temperature field, the isothermal map picture that obtains having the temperature mark at last can characterize the principal character information of blast furnace charge level temperature field, and has characteristics such as clear, directly perceived.

Claims (1)

1. intelligent extraction method of blast burden temperature field isotherm, it is characterized in that: utilize optimized image selection strategy, the gray scale level line extracting method of gradation of image distribution variance quality comprehensive index, the image processing method of gray scale level line patch algorithm that infrared image is handled, use isotherm temperature calibration method again, obtain characterizing the isotherm of temperature field principal character information, concrete steps are as follows:

(1) gathers different blast furnace images constantly, i.e. F _{T-4 τ}, F _{T-3 τ}, F _{T-2 τ}, F _T-τ, F _t, F wherein _{T-i τ}Be t-i τ images acquired constantly, τ=500ms adds up the number of pixels of each each gray level of width of cloth image, obtains

Wherein

The number of pixels of the gray level that is respectively every width of cloth image from 0～255, i=0～4, according to gradation of image distribution variance quality comprehensive index,

${\mathrm{\μ}}_{f_{t-\mathrm{n\τ}}}=\frac{1}{256}n---\left(1\right)$

${\mathrm{\σ}}_{f_{t-\mathrm{n\τ}}}^2=\frac{1}{256}\underset{m=0}{\overset{255}{\mathrm{\Σ}}}{(g_{f_{t-\mathrm{n\τ}}m}-{\mathrm{\μ}}_{f_{t-\mathrm{n\τ}}m})}^2---\left(2\right)$

Wherein

In every width of cloth gray distribution of image variance yields formula (1)

Be the mean pixel number of every each gray level of width of cloth image, n is the total number of pixels of image, in the formula (2) For. the variance of each image gray levels number of pixels,

Be the number of pixels of each each gray level of image, choosing the image setting that the intensity profile variance yields mediates from these a few width of cloth images is optimized image, carries out mean filter then, tentatively removes the overriding noise signal;

(2) by the prime number on the horizontal vertical center line of picture is scanned, determine horizontal pixel number N _AWith vertical pixel count N _B, according to the method for statistics, pixel count satisfies following condition under certain threshold value:

The IF gray-scale value is less than M ₁Number of pixels N ₁∈ ((N _A+ N _B)/5-50, (N _A+ N _B)/5+50)

First threshold range of THEN is (M ₁-5, M ₁);

The IF gray-scale value is less than M ₂Number of pixels N ₂∈ (2* (N _A+ N _B)/5-50,2* (N _A+ N _B)/5+50)

Second threshold range of THEN is (M ₂-5, M ₂);

The IF gray-scale value is less than M ₃Number of pixels N ₃∈ (2* (N _A+ N _B)/5-50,2* (N _A+ N _B)/5+50)

The 3rd threshold range of THEN is (M ₃-5, M ₃);

The IF gray-scale value is less than M ₄Number of pixels N ₄∈ (2* (N _A+ N _B)/5-50,2* (N _A+ N _B)/5+50)

The 4th threshold range of THEN is (M ₄-5, M ₄);

ELSE M ₅=(255-M ₄)/2, promptly the 5th threshold range is (M ₅-5, M ₅); Determine 5 segmentation threshold scopes successively, be respectively (M ₁-5, M ₁), (M ₂-5, M ₂), (M ₃-5, M ₃), (M ₄-5, M ₄), (M ₅-5, M ₅), M wherein _k, k=1,2,3,4,5, the dynamic change along with the picture difference, by dynamic variable threshold value image being cut apart acquisition has several isocontour images;

(3) a few width of cloth images in the step (2) are superposeed, obtain a width of cloth and have many isocontour images of gray scale, adopt square error and minimum value filtering method that the image after superposeing is carried out filtering once more then,

Determine each grey scale pixel value x (i, j) local mean value μ earlier _IjWith local variance

${\mathrm{\μ}}_{\mathrm{ij}}=\frac{1}{\mathrm{card}\left(\mathrm{\φ}\right)}\×\underset{\mathrm{\θ}\∈\mathrm{\φ}}{\mathrm{\Σ}}{x}_{\mathrm{ij}}^{\mathrm{\θ}}---\left(3\right)$

${\mathrm{\σ}}_{\mathrm{ij}}^2=\frac{1}{\mathrm{card}\left(\mathrm{\φ}\right)}\×\underset{\mathrm{\θ}\∈\mathrm{\φ}}{\mathrm{\Σ}}{(x_{\mathrm{ij}}^{\mathrm{\θ}}-{\mathrm{\μ}}_{\mathrm{ij}})}^2---\left(4\right)$

${\mathrm{\δ}}^2=\frac{1}{N\×M}\×\underset{i=1}{\overset{N}{\mathrm{\Σ}}}\underset{j=1}{\overset{M}{\mathrm{\Σ}}}{\mathrm{\σ}}_{\mathrm{ij}}^2---\left(5\right)$

Wherein φ=0 °, 45 °, 90 °, 135 °, 180 °, 225 °, 275 °, 315 ° }, card (φ) is a set of computations member number, with stationary noise variance in formula (5) presentation video, determine again after the filtering that (i, (i j), obtains image F after the filtering to gray-scale value y j) to pixel x in the image ₃

$y(i,j)={\mathrm{\μ}}_{\mathrm{ij}}+\frac{{\mathrm{\σ}}_{\mathrm{ij}}^2-{\mathrm{\δ}}^2}{{\mathrm{\σ}}_{\mathrm{ij}}^2}\×(x_{\mathrm{ij}}-{\mathrm{\μ}}_{\mathrm{ij}})---\left(6\right)$

(4) look like to exist burr and breakaway poing at the gray scale contour map that obtains in the step (3), respectively each bar gray scale level line is carried out burr and breakaway poing search, reject the isocontour burr of gray scale earlier and fill up breakaway poing again;

(5) according to temperature and the gray scale corresponding relation between the thermopair recently in the cross temperature of the point on each bar gray scale level line, demarcate temperature value respectively, then temperature value is revised, obtained to have the blast furnace charge level temperature field isothermal map picture of temperature mark at last.

Images