CN103834796A - Method for online recognition of deviation of strip steel in furnace - Google Patents

Abstract

The present invention discloses a method for online recognition of deviation of a strip steel in a furnace. The method comprises: 1) collecting multiple strip steel images including a strip steel edge position; 2) adopting an attention selection method to treat the collected strip steel images to obtain the significant feature pictures of the strip steel images, wherein the significant feature pictures comprise the edge position information of the strip steel; 3) carrying out a gray stretching treatment on the significant feature pictures, and then carrying out a binaryzation treatment, wherein all the taken values in the value taking range are adopted to calculate the corresponding classes square error r2 so as to determine the first optimal gray threshold T1 and the second optimal gray threshold T2, the pixels in the pictures requiring the binaryzation treatment are divided into the target class, the background class and the interference class according to the first optimal gray threshold T1 and the second optimal gray threshold T2, the gray value of the target class pixels is set to 1, and the grey values of the background class pixels and the interference class pixels are set to 0; and 4) judging whether the deviation of the edge position of the strip steel in the binaryzation picture is more than the preset value. The technical scheme has characteristics of easy and accurate use.

Description

A kind of method of strip running deviation in ONLINE RECOGNITION stove

Technical field

The present invention relates to a kind of image-recognizing method, particularly a kind of band steel image-recognizing method.

Background technology

Cold rolling is extremely important step on output strip line, and the quality of this step directly has influence on the quality of finished product and the production efficiency of whole unit.Along with the renewal of band steel cold-tandem rolling processing unit, tend to and Bao Youkuan with the specification of steel, in cold rolled strip steel production process, in annealing furnace, on roller, the strip running deviation of high-speed cruising can cause being with steel wooden dipper song or wipe the limit portion equipment that touches, impact band steel become a useful person rate, damage unit equipment, affect the steady running of unit.The reason of strip running deviation mainly contains following several aspect: 1) transport roller factor: geometrical shape work in-process or the operation of transporting roller are rear tapered, and axis is not parallel when mounted to transport roller, and the rough degree that transports roller surface is inhomogeneous on axis; 2) band steel factor: strip section is inhomogeneous, and transverse gage inequality, is camber, or band steel specification is excessively thin, wide; 3) tension factor: the tension control fluctuation with steel is large and unstable, batch the tension force difference on front pinch roll both sides, or the belt tension of strap rolling-assisting device is improper; 4) system factor: the zero shift of Controlling System servo-valve or valve core of servo valve stop up, causes hydro-cylinder to realize correction according to instruction; 5) operation reason: operator's double pendulum regulates unbalanced, causes strip running deviation.In addition, the factor that sideslip is batched in impact also has whether the reeling machine speed of electrical control mate with unit speed, whether pinch roll roll gap belt tightness degree whether even, strap rolling-assisting device before reeling machine is suitable, whether sideslip etc. of travelling belt.

At present, the inherent tape steel of annealing furnace deviation phenomenon relies on eye-observation to complete.Personal monitoring's mode cannot provide accurate sideslip data, fault statistics and warning function, and meanwhile, the easy fatigue of human eye causes generation undetected, false retrieval.Therefore, each Iron And Steel Company all wishes the improvement by accelerating systems technology, realizes precisely, strip running deviation ONLINE RECOGNITION and early warning easily, avoid on-line rolling equipment to occur fault in production, prevent that Finished Steel from quality problems in batch occurring, improve production cost and production efficiency.

Summary of the invention

The object of this invention is to provide the method for strip running deviation in a kind of ONLINE RECOGNITION stove, it can be in cold-rolling process, realize ONLINE RECOGNITION and the early warning of strip running deviation, to remind field personnel's strip running deviation, thereby reduce the impact of strip running deviation for rolling line equipment, avoid output strip line that part or overall failure stoppage in transit occur, and then both ensured production security and stability, guaranteed again the quality of finished steel plate and the production capacity of output strip line.

In order to realize the object of foregoing invention, the invention provides the method for strip running deviation in a kind of ONLINE RECOGNITION stove, it comprises:

(1) gather the band steel image that several comprise strip edge edge position;

(2) adopt attention selection method according to following step, the band steel image collecting to be processed, obtain the significant characteristics figure with steel image, in significant characteristics figure, include the marginal position information with steel:

(2a), according to the band steel image that collects, using ICA(independent component analysis, IndependentComponent Analysis) method obtains basis funciton storehouse A, in this basis funciton storehouse, each basis funciton a _irepresent a band steel characteristics of image.Wherein, the particular content of ICA method can be consulted the open source literature of " based on the feature of information maximization ", it is published in the meeting of neural information processing systems in 2005 years, author is N.D. Bruce and J.K. Qiao Qiaosi (N.D.Bruce and J.K.Tsotsos, " Saliencybased on information maximization; " in Conf.Neural Information ProcessingSystems, 2005);

(2b) band steel image is divided into several image blocks, each image block is projected to basis funciton storehouse A, establishes x _{j, k}denotation coordination is the image block of (j, k), obtains image block x according to projection formula _{j, k}corresponding basis funciton a _icoefficient v _{i, j, k}:

x _j,k＝∑ _i?v _i,j,ka _i

Wherein, i represents coefficient v _{i, j, k}corresponding basis funciton a _isequence number;

(2c) calculate with single basis funciton a _ithe image block x weighing _jksimilarity degree p (w with peripheral region _{ij, k}=v _{i, jk}):

$p(w_{i,j,k}=v_{i,j,k})$

$=\frac{1}{\sqrt{2\mathrm{\π}}}\underset{\∀s,t\∈\mathrm{\ψ}}{\mathrm{\Σ}}\frac{1}{n}{e}^{-(v_{i,j,k}-v_{i,,s,t})2/2}$

In above formula, w _{i, j, k}represent the set of the coefficient of the basis funciton that image block in peripheral region Ψ centered by (j, k) is corresponding; E is the truth of a matter of natural logarithm; N is the number of cumulative; S, t is corresponding j, the variable of k, their span is in Ψ region, because i in formula remains unchanged, p (w _{i, j, k}=v _{i, j, k}) expressed physical significance is: only use single basis funciton a _iweigh characteristics of image that image block (j, k) locates and the similarity degree of peripheral region feature, when the feature of certain image block and peripheral region is when more similarity degree is larger, this place's significance is less; And in the time that the difference degree of certain image block and peripheral region is larger, this place's significance is larger;

(2d) calculate the image block x weighing with all basis funcitons _jksimilarity degree p with peripheral region _{j, k}:

p _j,k＝∏ _ip(w _i，j，k＝v _i，j，k)，

The weighing result of all basis funcitons is multiplied each other, the p that obtains _j,kphysical significance be: weigh characteristics of image that image block (j, j) locates and the similarity degree of peripheral region feature with all basis funcitons;

(2e) computed image piece x _jksignificance quantized value-log (p at place _{j, k}), obtaining significant characteristics figure, its principle is: because certain image block is more similar to the characteristics of image of peripheral region, significance is between the two just less;

(3) significant characteristics figure is done and carry out binary conversion treatment after gray scale stretch processing and be used for precisely determining strip edge edge position, because the result obtaining after binary conversion treatment has been eliminated noise and unnecessary information, consider in actual working environment factor, the reflective areas that the existence meeting of extraneous light forms in steel image, around between object, contrast is obviously for reflective areas in image and its, edge, reflective areas is more remarkable than strip edge edge, therefore, adopt two optimum gray thresholds to suppress the interference of reflective areas in image: establishing image total number-of-pixels is N, grey level range is [0, L-1], determine one first optimum gray threshold T ₁with one second optimum gray threshold T ₂, and T ₁<T ₂, according to the first optimum gray threshold T ₁with the second optimum gray threshold T ₂the pixel that will carry out in the image of binary conversion treatment divides three classes into, by gray-scale value in [T ₁+ 1, T ₂] pixel be divided into target class pixel, by gray-scale value in [0, T ₁] pixel be divided into background classes pixel, gray-scale value is in [T ₂+ 1, L-1] pixel be divided into and disturb class pixel, and the gray-scale value of target class pixel is put to 1, background classes pixel and disturb the gray-scale value of class pixel all to set to 0, obtains binary image, wherein, the first optimum gray threshold T ₁with one second optimum gray threshold T ₂determine according to following step:

If gray-scale value is the number of the pixel of i is Ni, calculate the Probability p of the pixel appearance that gray-scale value is i according to following formula _i:

Pi＝Ni/N（i＝0，1,2….L-1）

The average u of background classes pixel in computed image respectively ₁, target class pixel average u ₂with the average u that disturbs class pixel ₃:

$u_1=\frac{{\mathrm{\&Sigma;}}_{i=o}^{T_1}i*P_i}{{\mathrm{\&Sigma;}}_{i=o}^{T_1}{P}_i}$

${\mathrm{<figure-callout\; id="2"\; label="u"\; filenames="HDA00002455727900011.png"\; state="\{\{state\}\}">u</figure-callout>}}_2=\frac{{\mathrm{\&Sigma;}}_{i=T_1+1}^{T_2}i*P_i}{{\mathrm{\&Sigma;}}_{i=T_1+1}^{T_2}{P}_i}$

${\mathrm{<figure-callout\; id="3"\; label="u"\; filenames="HDA00002455727900011.png"\; state="\{\{state\}\}">u</figure-callout>}}_3=\frac{{\mathrm{\&Sigma;}}_{i={\mathrm{<figure-callout\; id="2"\; label="T"\; filenames="HDA00002455727900011.png"\; state="\{\{state\}\}">T</figure-callout>}}_2+1}^{L-1}i*P_i}{{\mathrm{\&Sigma;}}_{i={\mathrm{<figure-callout\; id="2"\; label="T"\; filenames="HDA00002455727900011.png"\; state="\{\{state\}\}">T</figure-callout>}}_2+1}^{L-1}{P}_i}$

Calculate the average u of pixel in the image of wanting binary conversion treatment ₀:

$u_0=u_1{\mathrm{\&Sigma;}}_{i=o}^{T_1}{P}_i+u_2{\mathrm{\&Sigma;}}_{i=T_1+1}^{T_2}{P}_i+u_3{\mathrm{\&Sigma;}}_{i={\mathrm{<figure-callout\; id="2"\; label="T"\; filenames="HDA00002455727900011.png"\; state="\{\{state\}\}">T</figure-callout>}}_2+1}^{L-1}{P}_i$

Calculate the inter-class variance r of background classes pixel, target class pixel and interference class pixel ²:

${\mathrm{<figure-callout\; id="2"\; label="r"\; filenames="HDA00002455727900011.png"\; state="\{\{state\}\}">r</figure-callout>}}^2={(u_1-u_0)}^2{\mathrm{\&Sigma;}}_{i=o}^{T_1}{P}_i+{(u_2-u_0)}^2{\mathrm{\&Sigma;}}_{i=T_1+1}^{T_2}{P}_i+{(u_3-u_0)}^2{\mathrm{\&Sigma;}}_{i={\mathrm{<figure-callout\; id="2"\; label="T"\; filenames="HDA00002455727900011.png"\; state="\{\{state\}\}">T</figure-callout>}}_2+1}^{L-1}{P}_i$

Travel through the first optimum gray threshold T ₁with the second optimum gray threshold T ₂all values in span are to calculate corresponding inter-class variance r ², inter-class variance r ²t corresponding to maximum value ₁and T ₂the first optimum gray threshold T that will determine for step (3) ₁with the second optimum gray threshold T ₂, the basic thought that carrys out the method for definite threshold by the maximum value of inter-class variance is: the inter-class variance of background and target has characterized the difference degree between background and target two classes.Wherein, inter-class variance is larger, and the difference between background and target is larger, if shown in visual effect, is exactly that background and target image contrast are more obvious.Because optimum gray threshold should make the variance maximum between background and target, thereby by obtaining an optimum gray-scale value threshold value, the pixel of input picture can be divided into background and target two classes, and the gray-scale value of target class pixel is set to 1, the gray-scale value of background classes pixel is set to 0, so just obtain needed bianry image, by searching for and calculate inter-class variance r ²maximum value and obtain optimum gray threshold T;

(4) if the deviation of strip edge edge position is greater than a preset value in the binary image of two of front and back, judgement band steel generation sideslip.

In ONLINE RECOGNITION stove of the present invention in the method for strip running deviation, comprise band steel production data and the image of strip edge edge positional information by collection, adopt attention selection method and binary conversion treatment method, avoid noise, the interference of reflective areas and unnecessary information, by to obtain image simplification be binary image in order to determine thereby whether the departure of strip edge edge position exceedes a preset value and judge and be with whether sideslip of steel, and then for by sideslip or the band steel of sideslip send early warning signal, field personnel can be adjusted sideslip band steel in time rapidly, avoid the generation of equipment failure and security incident.

Further, in ONLINE RECOGNITION stove of the present invention in the method for strip running deviation, T ₁span be [120,180].

Further, in ONLINE RECOGNITION stove of the present invention in the method for strip running deviation, T ₂span be [190,220].

Further, in ONLINE RECOGNITION stove of the present invention, in the method for strip running deviation, adopt the deviation of strip edge edge position in two binary images in difference shadow method judgement front and back.The ultimate principle of difference shadow method is: to gather first the edge picture receiving as benchmark, preserve an integer variable, and the initial value that this integer variable is set is 0, subtract each other with the second pictures and the first pictures, if because strip edge edge position in the short period of time does not change, two image subtraction result are all black picture; If strip edge edge position changes, two image subtraction result have comprised front and back change information, now there is the situation with the slight sideslip of steel, subtract each other result edge calculation change in location size by detection, be the distance that moves of marginal position and deposit above-mentioned integer variable in, wherein, strip edge edge position move right on the occasion of, be moved to the left as negative value; Continue again above-mentioned circulation taking the second pictures as benchmark; Along with integer variable is constantly cumulative, in the time detecting that the absolute value of integer variable is greater than a certain preset value, judge strip running deviation.It should be noted that, this preset value can need to be determined and arrange according to actual field.

The method of strip running deviation in ONLINE RECOGNITION stove of the present invention, by production data and the collection with steel image, and coordinates with attention selection method and binary conversion treatment method, its compared with prior art, the advantage possessing is as follows:

1) system ONLINE RECOGNITION strip running deviation, the undetected and false retrieval of avoiding artificial congnition to produce due to fatigue or parallax;

2) early warning signal that fast accurate identification strip running deviation also sends strip running deviation is in time to remind field personnel;

3) prevent that strip running deviation from causing band steel wooden dipper song or wiping and touch limit portion equipment and cause device systems fault;

4) ensure production security and stability;

5) guarantee the outgoing of finished steel plate and the efficiency production capacity of output strip line;

6) equipment cost is low, and operation is used simple, and on-the-spot easy for installation, recognition result reliability is strong.

Brief description of the drawings

Fig. 1 is system for realizing the method for strip running deviation in the ONLINE RECOGNITION stove of the present invention structured flowchart under a kind of embodiment.

Embodiment

Below in conjunction with specific embodiment and Figure of description, the method for strip running deviation in ONLINE RECOGNITION stove of the present invention is made further and being explained in detail, but this explanation does not form for improper restriction of the present invention.

As shown in Figure 1, in the present embodiment, comprise for the system of the method for strip running deviation in ONLINE RECOGNITION stove: 1) 8 industrial monitoring pick up cameras; 2) video frequency collection card; 3) monitoring host computer; 4) dedicated process machine.Wherein, industrial monitoring pick up camera 1 is all connected with video frequency collection card 3 by signal-transmitting cable 2, industrial monitoring pick up camera 1 is arranged in the vertical direction of the output strip line in annealing furnace in uniform way, makes the image of taking the photograph at least comprise edge and the beaming roller edge with steel one side; Video frequency collection card 3 has the various functions such as collection, signal mode number conversion, video preview and coding preservations of vision signal, and it passes through PCI(Peripheral Component Interconnect, Peripheral Component Interconnect standard) interface 4 is connected with monitoring host computer 5; Dedicated process machine 6 is connected with monitoring host computer 5 by Ethernet 7.The workflow of this system is: industrial monitoring pick up camera is produced band steel picture in real time, picture is transferred to video frequency collection card by signal-transmitting cable to be processed, and by pci interface, decoded video signal is passed to monitoring host computer by video frequency collection card, the software program of supervisory system moves on monitoring host computer, and user can manage whole system by function software.In ONLINE RECOGNITION stove, strip running deviation is realized by the monitoring software operating on monitoring host computer, intercepts several band steel images the picture that this monitoring software is taken from industrial monitoring pick up camera.The band steel image of the method for strip running deviation based on above-mentioned intercepting in ONLINE RECOGNITION stove in the present embodiment, judges whether sideslip of band steel by attention selection method and binary conversion treatment method, and included step is:

(1) monitoring software intercepts the band steel image that a width comprises strip edge edge position every the picture of taking from pick up camera for 2 seconds;

(2) adopt attention selection method according to following step, the band steel image collecting to be processed, obtain the significant characteristics figure with steel image, in these significant characteristics figure, include the marginal position information with steel:

(2a), according to the band steel image that collects, using ICA(independent component analysis, IndependentComponent Analysis) method obtains basis funciton storehouse A, in this basis funciton storehouse, each basis funciton a _irepresent a band steel characteristics of image;

(2b) band steel image is divided into several image blocks, each image block is projected to basis funciton storehouse A, establishes x _j,kdenotation coordination is the image block of (j, k), obtains image block x according to projection formula _j,kcorresponding basis funciton a _icoefficient v _{i, j, k}:

x _j,k＝∑ _iv _i，j，ka _i

(2c) calculate with single basis funciton a _ithe image block x weighing _{j, k}similarity degree p (w with peripheral region _{i, j, k}=v _{i, j, k}):

$p(w_{i,j,k}=v_{i,j,k})$

$=\frac{1}{\sqrt{2\mathrm{\&pi;}}}\underset{\&ForAll;s,t\&Element;\mathrm{\&psi;}}{\mathrm{\&Sigma;}}\frac{1}{n}{e}^{-(v_{i,j,k}-v_{i,,s,t})2/2}$

In formula, w _{i, j, k}represent the set of the coefficient of the basis funciton that image block in peripheral region Ψ centered by (j, k) is corresponding; E is the truth of a matter of natural logarithm; N is the number of cumulative; S, t is corresponding j, the variable of k, their span is in Ψ region, when the feature of certain image block and peripheral region is when more similarity degree is larger, this place's significance is less; And in the time that the difference degree of certain image block and peripheral region is larger, this place's significance is larger;

(2d) calculate the image block x that all basis funcitons are weighed _jksimilarity degree p with peripheral region _{j, k}: p _{j, k}=∏ _ip (w _{i, j, k}=v _{i, j, k}), the weighing result of all basis funcitons is multiplied by mutually and weighs the characteristics of image located of image block (j, k) and the similarity degree of peripheral region feature;

(2e) computed image piece x _jksignificance quantized value-log (p at place _{j, k}), obtain significant characteristics figure;

(3) after being done to gray scale stretch processing, significant characteristics figure carries out binary conversion treatment: establishing image total number-of-pixels is N, and grey level range is [0, L-1], according to on-site actual situations, in conjunction with many experiments data, determines the first optimum gray threshold T ₁span be [120,180] and the second optimum gray threshold T ₂span be [190,220], according to the first optimum gray threshold T ₁with the second optimum gray threshold T ₂the pixel that will carry out in the image of binary conversion treatment divides three classes into, by gray-scale value in [T ₁+ 1, T ₂] pixel be divided into target class pixel, by gray-scale value in [0, T ₁] pixel be divided into background classes pixel, gray-scale value is in [T ₂+ 1, L-1] pixel be divided into and disturb class pixel, and the gray-scale value of target class pixel is put to 1, background classes pixel and disturb the gray-scale value of class pixel all to set to 0, obtains binary image; Wherein, the first optimum gray threshold T ₁with one second optimum gray threshold T ₂determine according to following step:

If gray-scale value is the number of the pixel of i is Ni, calculate the Probability p of the pixel appearance that gray-scale value is i according to following formula _i:

Pi＝Ni/N（i＝0，1,2….L-1）

The average u of background classes pixel in computed image respectively ₁, target class pixel average u ₂with the average u that disturbs class pixel ₃:

$u_1=\frac{{\mathrm{\&Sigma;}}_{i=o}^{T_1}i*P_i}{{\mathrm{\&Sigma;}}_{i=o}^{T_1}{P}_i}$

${\mathrm{<figure-callout\; id="2"\; label="u"\; filenames="HDA00002455727900011.png"\; state="\{\{state\}\}">u</figure-callout>}}_2=\frac{{\mathrm{\&Sigma;}}_{i=T_1+1}^{T_2}i*P_i}{{\mathrm{\&Sigma;}}_{i=T_1+1}^{T_2}{P}_i}$

${\mathrm{<figure-callout\; id="3"\; label="u"\; filenames="HDA00002455727900011.png"\; state="\{\{state\}\}">u</figure-callout>}}_3=\frac{{\mathrm{\&Sigma;}}_{i={\mathrm{<figure-callout\; id="2"\; label="T"\; filenames="HDA00002455727900011.png"\; state="\{\{state\}\}">T</figure-callout>}}_2+1}^{L-1}i*P_i}{{\mathrm{\&Sigma;}}_{i={\mathrm{<figure-callout\; id="2"\; label="T"\; filenames="HDA00002455727900011.png"\; state="\{\{state\}\}">T</figure-callout>}}_2+1}^{L-1}{P}_i}$

Calculate the average u of pixel in the image of wanting binary conversion treatment ₀:

$u_0=u_1{\mathrm{\&Sigma;}}_{i=o}^{T_1}{P}_i+{\mathrm{<figure-callout\; id="2"\; label="u"\; filenames="HDA00002455727900011.png"\; state="\{\{state\}\}">u</figure-callout>}}_2{\mathrm{\&Sigma;}}_{i=T_1+1}^{T_2}{P}_i+{\mathrm{<figure-callout\; id="3"\; label="u"\; filenames="HDA00002455727900011.png"\; state="\{\{state\}\}">u</figure-callout>}}_3{\mathrm{\&Sigma;}}_{i={\mathrm{<figure-callout\; id="2"\; label="T"\; filenames="HDA00002455727900011.png"\; state="\{\{state\}\}">T</figure-callout>}}_2+1}^{L-1}{P}_i$

Calculate the inter-class variance r of background classes pixel, target class pixel and interference class pixel ²:

${\mathrm{<figure-callout\; id="2"\; label="r"\; filenames="HDA00002455727900011.png"\; state="\{\{state\}\}">r</figure-callout>}}^2={(u_1-u_0)}^2{\mathrm{\&Sigma;}}_{i=o}^{T_1}{P}_i+{(u_2-u_0)}^2{\mathrm{\&Sigma;}}_{i=T_1+1}^{T_2}{P}_i+{(u_3-u_0)}^2{\mathrm{\&Sigma;}}_{i={\mathrm{<figure-callout\; id="2"\; label="T"\; filenames="HDA00002455727900011.png"\; state="\{\{state\}\}">T</figure-callout>}}_2+1}^{L-1}{P}_i$

Travel through the first optimum gray threshold T ₁interior and the second optimum gray threshold T in span [120,180] ₂institute in the span [190,220] likely value to calculate corresponding inter-class variance r ², inter-class variance r ²t corresponding to maximum value ₁and T ₂for this step the first optimum gray threshold T to be determined ₁with the second optimum gray threshold T ₂;

(4) adopting before and after difference shadow method judgement the deviation of strip edge edge position: A in two binary images) first edge picture that collection is received be as benchmark, integer variable=0 is set, B) the second pictures and the first pictures are subtracted each other, if strip edge edge position does not change, two image subtraction result are all black picture; If strip edge edge position changes, two image subtraction result have front and back change information, and the situation with the slight sideslip of steel has now just occurred, and detect and subtract each other result, edge calculation change in location size, be the distance (being pixel count in the present embodiment) that marginal position moves, and count above-mentioned integer variable, wherein, in the time that strip edge edge position moves right, miles of relative movement count on the occasion of, in the time that strip edge edge position is moved to the left, miles of relative movement is counted negative value; C) proceed A taking the second pictures as benchmark again) and the B) circulation of step; D), if strip edge edge position continues to change, along with integer variable is constantly cumulative, in the time detecting that the absolute value of integer variable is greater than a certain preset value, judge strip running deviation.Wherein, this preset value can need to be determined and arrange according to actual field.

In addition, the quantity of pick up camera can be disposed to cover whole piece output strip line in stove according to situ production situation, is used for obtaining the realtime graphic of complete strip edge edge.

In addition, monitoring host computer can adopt the T5500 of Dell workstation, and dedicated process machine can adopt the product of the RS90/220 of Hitachi.

It should be noted that, the hardware configuration adopting for the system of the method for strip running deviation in ONLINE RECOGNITION stove in the present embodiment can be adjusted or design according to practical situation and field working conditions.In ONLINE RECOGNITION stove of the present invention, the core of the method for strip running deviation is by attention selection method and binary processing method, the front and back the change of divergence amount that gathers image is compared with the warning threshold value that sets in advance, thereby make rapidly and accurately with the whether identification judgement of sideslip of steel.Therefore, the realization of technical scheme of the present invention is not limited to structure and the setting of strip running deviation system in ONLINE RECOGNITION stove.

Be noted that above lifted be only the specific embodiment of the present invention, obviously the invention is not restricted to above embodiment, have many similar variations thereupon.If all distortion that those skilled in the art directly derives or associates from content disclosed by the invention, all should belong to protection scope of the present invention.

Claims (4)

1. a method for strip running deviation in ONLINE RECOGNITION stove, is characterized in that, comprising:

(1) gather the band steel image that several comprise strip edge edge position;

(2) adopt attention selection method according to following step, the band steel image collecting to be processed, obtain the significant characteristics figure with steel image, in described significant characteristics figure, include the marginal position information with steel:

(2a) according to the band steel image collecting, use ICA method to obtain basis funciton storehouse A, in described basis funciton storehouse, each basis funciton a _irepresent a band steel characteristics of image;

(2b) band steel image is divided into several image blocks, each image block is projected to basis funciton storehouse A, establishes x _{j, k}denotation coordination is the image block of (j, k), obtains image block x according to projection formula _{j, k}corresponding basis funciton a _icoefficient v _{i, j, k}:

x _j,k＝∑ _iv _i，j，ka _i

(2c) calculate with single basis funciton a _ithe image block x weighing _{j, k}similarity degree p (w with peripheral region _{ij, k}=v _{ij, k}):

$p(w_{i,j,k}=v_{i,j,k})$

$=\frac{1}{\sqrt{2\mathrm{\&pi;}}}\underset{\&ForAll;s,t\&Element;\mathrm{\&psi;}}{\mathrm{\&Sigma;}}\frac{1}{n}{e}^{-(v_{i,j,k}-v_{i,,s,t})2/2}$

In above formula, w _{i, j, k}represent with * j, k) centered by peripheral region Ψ in the set of coefficient of basis funciton corresponding to image block; E is the truth of a matter of natural logarithm; N is the number of cumulative; S, t is corresponding j, the variable of k, their span is in Ψ region;

(2d) calculate the image block x weighing with all basis funcitons _{j, k}similarity degree p with peripheral region _j,k:

p _j，k＝∏ _ip(w _i，j，k＝v _i，j,k)

(2e) computed image piece x _{j, k}significance quantized value-log (p at place _{j, k}), obtain significant characteristics figure;

(3) after being done to gray scale stretch processing, significant characteristics figure carries out binary conversion treatment: establishing image total number-of-pixels is N, and grey level range is [0, L-1], determines one first optimum gray threshold T ₁with one second optimum gray threshold T ₂, and T ₁<T ₂, according to the first optimum gray threshold T ₁with the second optimum gray threshold T ₂the pixel that will carry out in the image of binary conversion treatment divides three classes into, by gray-scale value in [T ₁+ 1, T ₂] pixel be divided into target class pixel, by gray-scale value in [0, T ₁] pixel be divided into background classes pixel, gray-scale value is in [T ₂+ 1, L-1] pixel be divided into and disturb class pixel, and the gray-scale value of target class pixel is put to 1, background classes pixel and disturb the gray-scale value of class pixel all to set to 0, obtains binary image; Wherein, described the first optimum gray threshold T ₁with one second optimum gray threshold T ₂determine according to following step:

If gray-scale value is the number of the pixel of i is Ni, calculate the Probability p of the pixel appearance that gray-scale value is i according to following formula _i:

Pi＝Ni/N（i＝0，1,2….L-1）

The average u of background classes pixel in computed image respectively ₁, target class pixel average u ₂with the average u that disturbs class pixel ₃:

$u_1=\frac{{\mathrm{\&Sigma;}}_{i=o}^{T_1}i*P_i}{{\mathrm{\&Sigma;}}_{i=o}^{T_1}{P}_i}$

$u_2=\frac{{\mathrm{\&Sigma;}}_{i=T_1+1}^{T_2}i*P_i}{{\mathrm{\&Sigma;}}_{i=T_1+1}^{T_2}{P}_i}$

$u_3=\frac{{\mathrm{\&Sigma;}}_{i=T_2+1}^{L-1}i*P_i}{{\mathrm{\&Sigma;}}_{i=T_2+1}^{L-1}{P}_i}$

Calculate the average u of pixel in the image of wanting binary conversion treatment ₀:

$u_0=u_1{\mathrm{\&Sigma;}}_{i=o}^{T_1}{P}_i+u_2{\mathrm{\&Sigma;}}_{i=T_1+1}^{T_2}{P}_i+u_3{\mathrm{\&Sigma;}}_{i=T_2+1}^{L-1}{P}_i$

Calculate the inter-class variance r2 of background classes pixel, target class pixel and interference class pixel:

$r^2={(u_1-u_0)}^2{\mathrm{\&Sigma;}}_{i=o}^{T_1}{P}_i+{(u_2-u_0)}^2{\mathrm{\&Sigma;}}_{i=T_1+1}^{T_2}{P}_i+{(u_3-u_0)}^2{\mathrm{\&Sigma;}}_{i=T_2+1}^{L-1}{P}_i$

Travel through the first optimum gray threshold T ₁with the second optimum gray threshold T ₂all values in span are to calculate corresponding inter-class variance r ², inter-class variance r ²t corresponding to maximum value ₁and T ₂the first optimum gray threshold T that will determine for step (3) ₁with the second optimum gray threshold T ₂;

(4) if the deviation of strip edge edge position is greater than a preset value in the binary image of two of front and back, judgement band steel generation sideslip.

2. the method for strip running deviation in ONLINE RECOGNITION stove as claimed in claim 1, is characterized in that T ₁span be [120,180].

3. the method for strip running deviation in ONLINE RECOGNITION stove as claimed in claim 1 or 2, is characterized in that T ₂span be [190,220].

4. the method for strip running deviation in ONLINE RECOGNITION stove as claimed in claim 1, is characterized in that, adopts the deviation of strip edge edge position in two binary images in difference shadow method judgement front and back.

Images