CN112734722A - Flame endpoint carbon content prediction method based on improved complete local binary pattern - Google Patents

Flame endpoint carbon content prediction method based on improved complete local binary pattern Download PDF

Info

Publication number
CN112734722A
CN112734722A CN202110023350.XA CN202110023350A CN112734722A CN 112734722 A CN112734722 A CN 112734722A CN 202110023350 A CN202110023350 A CN 202110023350A CN 112734722 A CN112734722 A CN 112734722A
Authority
CN
China
Prior art keywords
flame
local
image
clbp
amplitude
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN202110023350.XA
Other languages
Chinese (zh)
Other versions
CN112734722B (en
Inventor
孙文强
刘辉
李超
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Kunming University of Science and Technology
Original Assignee
Kunming University of Science and Technology
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Kunming University of Science and Technology filed Critical Kunming University of Science and Technology
Priority to CN202110023350.XA priority Critical patent/CN112734722B/en
Publication of CN112734722A publication Critical patent/CN112734722A/en
Application granted granted Critical
Publication of CN112734722B publication Critical patent/CN112734722B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/0002—Inspection of images, e.g. flaw detection
    • G06T7/0004—Industrial image inspection
    • C—CHEMISTRY; METALLURGY
    • C21—METALLURGY OF IRON
    • C21C—PROCESSING OF PIG-IRON, e.g. REFINING, MANUFACTURE OF WROUGHT-IRON OR STEEL; TREATMENT IN MOLTEN STATE OF FERROUS ALLOYS
    • C21C5/00—Manufacture of carbon-steel, e.g. plain mild steel, medium carbon steel or cast steel or stainless steel
    • C21C5/28—Manufacture of steel in the converter
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T5/00—Image enhancement or restoration
    • G06T5/40—Image enhancement or restoration using histogram techniques
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T5/00—Image enhancement or restoration
    • G06T5/50—Image enhancement or restoration using two or more images, e.g. averaging or subtraction
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/136—Segmentation; Edge detection involving thresholding
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/194—Segmentation; Edge detection involving foreground-background segmentation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/40—Analysis of texture
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T9/00—Image coding
    • C—CHEMISTRY; METALLURGY
    • C21—METALLURGY OF IRON
    • C21C—PROCESSING OF PIG-IRON, e.g. REFINING, MANUFACTURE OF WROUGHT-IRON OR STEEL; TREATMENT IN MOLTEN STATE OF FERROUS ALLOYS
    • C21C2300/00—Process aspects
    • C21C2300/06—Modeling of the process, e.g. for control purposes; CII
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/10—Image acquisition modality
    • G06T2207/10016—Video; Image sequence
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/10—Image acquisition modality
    • G06T2207/10024—Color image
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20081—Training; Learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20212—Image combination
    • G06T2207/20221—Image fusion; Image merging
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
    • Y02P10/00—Technologies related to metal processing
    • Y02P10/25—Process efficiency

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Chemical & Material Sciences (AREA)
  • Multimedia (AREA)
  • Manufacturing & Machinery (AREA)
  • Quality & Reliability (AREA)
  • Materials Engineering (AREA)
  • Metallurgy (AREA)
  • Organic Chemistry (AREA)
  • Facsimile Image Signal Circuits (AREA)
  • Image Analysis (AREA)

Abstract

The invention relates to a flame endpoint carbon content prediction method based on an improved complete local binary pattern, and belongs to the technical field of converter flame endpoint carbon content. Firstly, performing color space conversion on a flame image and fully considering texture information of each single channel; then, extracting image phase information by Local Phase Quantization (LPQ) as a supplementary combination of amplitude information in the CLBP to form a fusion feature CLBP _ MP so as to enhance the robustness of the CLBP algorithm structure, and weighting the image phase information by an improved color information weighting strategy so as to enhance the local contrast information of the flame image; finally, the carbon content is predicted by using a K nearest neighbor regression model, and the experimental result shows that the accuracy of the carbon content prediction within the error range of 0.02% is 83.9%. The method can effectively solve the problem that the carbon content cannot be accurately and precisely predicted due to the fact that the flame images with high similarity and similar carbon content are difficult to distinguish, and improves the local structure contrast information of the flame images.

Description

Flame endpoint carbon content prediction method based on improved complete local binary pattern
Technical Field
The invention relates to a flame endpoint carbon content prediction method based on an improved complete local binary pattern, and belongs to the technical field of converter flame endpoint carbon content.
Background
Steel plays an important role in national economic production activities, marks the national economic development degree, and converter steelmaking is widely applied due to the advantages of high productivity and low cost. The control of the end point carbon and temperature of the converter in the steel production process determines the quality of the tapped steel. Because various metal materials, non-metal materials, gases and other raw materials need to be added in the converter steelmaking production process, converter smelting is a complex physical and chemical change process, and particularly in a high-temperature environment, the physical and chemical reaction process is influenced by a plurality of factors and becomes more difficult to control, so that the standard requirement of tapping is difficult to achieve.
Aiming at the detection problem of the end point carbon content, the main technical methods are an artificial experience method, a sublance detection method, a furnace gas analysis method and the like, but the methods are greatly influenced by human subjective and objective factors, so the tapping requirements can not be met. The method based on image processing has the advantages of wide applicability, high precision and the like, and becomes a research hotspot in the field of converter steelmaking.
The flame image has the characteristics of multiple directions, multiple scales, random texture and the like, and the local characteristics of the flame contain more detailed information, which reflects the local specificity of the image, so that how to extract the key local characteristics of the flame image is the focus of research.
Disclosure of Invention
The invention aims to provide a flame endpoint carbon content prediction method based on an improved complete local binary pattern, and solves the problem that the carbon content cannot be accurately predicted due to the fact that flame images with similar carbon content are difficult to distinguish due to high similarity of the flame images.
The invention adopts the technical scheme that a flame endpoint carbon content prediction method based on an improved complete local binary pattern comprises the following specific steps:
step 1, acquiring furnace mouth flame video data shot in the actual production process of a steel mill;
step 2, when experimental data are manufactured, extracting the flame fire hole area of each training sample in the sample set from all times and unifying the size to obtain a flame image data set;
step 3, extracting color texture characteristics of the flame image by using an improved complete local binary pattern method;
and 4, inputting the color texture characteristics extracted in the step 3 into a K neighbor regression model to predict the carbon content.
Specifically, the color texture features of the flame image in the step 3 include a symbol feature, an amplitude feature, a central pixel gray value feature and a phase feature, and the extraction process specifically includes:
step 3.1, performing spatial conversion on the flame image and further dividing the flame image into three single-channel images, wherein each single-channel image is subjected to local difference operation to obtain a local area segmentation block of each flame single-channel image; then, further expanding the two window sizes of each flame single-channel image local area segmentation block;
3.2, respectively obtaining an amplitude component, a symbol component, a central pixel gray value and a phase component by two window scales under each flame single-channel image local area partition block;
3.3, multiplying two amplitude components obtained by two different window scales under each flame single-channel image local area partition block by a weight after binary coding, simultaneously obtaining a specific phase value by the phase component through binary coding, and finally measuring the correlation between the two vectors and reducing the dimension to combine the amplitude phase fusion characteristics CLBP _ MP by the final amplitude characteristic vector and the phase characteristic vector of each single-channel flame image through KCCA;
and 3.4, for each single-channel flame image, combining three operational characters of a sign characteristic CLBP _ S, an amplitude phase fusion characteristic CLBP _ MP and a central pixel gray value CLBP _ C by adopting a serial and parallel combined structure: firstly, serially connecting an amplitude and phase fusion characteristic CLBP _ MP and a central pixel gray value CLBP _ C in sequence to form a CLBP _ MP/C; then, the CLBP _ MP/C and the sign feature CLBP _ S are connected in parallel to form a 2-dimensional CLBP _ S _ MP/C histogram of each single-channel flame image;
and 3.5, splicing and combining the histograms of the three single channels of the flame image together, performing nonlinear dimensionality reduction by using local linear embedding, and inputting the reduced dimensions into a K nearest neighbor regression model to predict the carbon content.
Specifically, in step 3.3, the extraction step of the single-channel flame image amplitude-phase fusion feature CLBP _ MP is as follows:
step 3.3.1, two amplitude components obtained by two different window scales under a flame single-channel image local area segmentation block are multiplied by a weight value after being respectively subjected to binary coding,
Figure BDA0002889370480000021
is a value obtained by binary coding the amplitude component of the local region partition block and multiplying the binary coded amplitude component by a weight, WcolorTo enhance the weighting of the local contrast color information,
Figure BDA0002889370480000022
the amplitude component of the complete local binary pattern is subjected to binary coding; m ispIs a neighborhood sampling point of an image local region amplitude block, c is a central threshold of the image local region amplitude block, 2pIs binary coding, and p is the number of sampling points; meanwhile, the phase component is also subjected to binary coding to obtain a specific phase value;
step 3.3.2, combining with the weighting strategy of the color information, calculating the weight of the pixel point in the image, wherein the calculation formula is as follows:
Figure BDA0002889370480000031
in the formula (3), wherein Δ cpq,3x3Is the color distance, Δ c, at the 3 × 3 scalepq,5x5Is the color distance on the scale of 5 × 5, wherein
Figure BDA0002889370480000032
(HP,SP,VP) And (H)q,Sq,Vq) The method comprises the steps of dividing pixels p and pixels q of an image segmentation block at the corresponding position of each channel in HSV color space, wherein p is a neighborhood sampling point, q is a central threshold value, var is3×3(p, q) and var5×5(p, q) local variance at two scales, 3 × 3 and 5 × 5, Δ cpq,3x3×Δcpq,5x5The color distance values of 3 x 3 and 5 x 5 scales at the corresponding positions of the segmentation blocks of the respective local areas of the three single channels in the flame image HSV color space,
Figure BDA0002889370480000033
the local variance values of the flame image under a single channel under two scales of 3 x 3 and 5 x 5 are obtained;
step 3.3.3, finally, the final amplitude characteristic vector and the phase characteristic vector of the single channel are combined into amplitude phase fusion characteristic CLBP _ MP by measuring the correlation between the two vectors and reducing the dimension through kernel canonical correlation analysis KCCA, and the calculation formula is as follows:
Figure BDA0002889370480000034
in the formula (I), the compound is shown in the specification,
Figure BDA0002889370480000035
the total amplitude characteristic vector of all local segmentation blocks under the single channel is obtained;
Figure BDA0002889370480000036
for all local partitions of the block total phase eigenvector in the single channel, wherein
Figure BDA0002889370480000037
2j-1Is binary coding; KCCA is a nuclear canonical correlation analysis.
The invention has the beneficial effects that: aiming at the problems that the flame images corresponding to different carbon contents at the converter end point have high similarity and the flame images with similar carbon contents are difficult to distinguish, the flame image characteristics are extracted and analyzed by a carbon content prediction method based on an improved complete local binary pattern so as to realize the accurate prediction of the carbon content at the converter end point.
Drawings
FIG. 1 is a flow chart of a method for predicting the carbon content of an endpoint of a flame based on an improved complete local binary pattern according to the present invention;
FIG. 2 is an algorithmic graph based on an improved complete local binary pattern flame endpoint carbon content prediction method;
FIG. 3 is a schematic diagram of the complete local binary pattern CLBP feature extraction calculation principle.
Detailed Description
The invention is described in detail below with reference to the figures and specific embodiments.
Example 1: 1-3, a method for predicting carbon content based on an improved complete local binary pattern flame endpoint. The Complete Local Binary Pattern (CLBP) describes the gray value, the Sign information and the Magnitude information of the central pixel of an image by using a local difference Sign-Magnitude transform (LDSMT) and three texture descriptors CLBP _ C (CLBP _ Center), CLBP _ S (CLBP _ Sign) and CLBP _ M (CLBP _ Magnitude), respectively. The method comprises the following specific steps:
step 1, acquiring furnace mouth flame video data shot in the actual production process of a steel mill;
step 2, when experimental data are manufactured, extracting the flame fire hole area of each training sample in the sample set from all times and unifying the size to obtain a flame image data set;
step 3, extracting color texture characteristics of the flame image by using an improved complete local binary pattern method;
and 4, inputting the color texture characteristics extracted in the step 3 into a K neighbor regression model to predict the carbon content.
Further, in the step 3, the flame image single-channel local area feature extraction adopts a local phase quantization method to extract the local phase information of the flame image, and as a supplement to the amplitude information in the CLBP, the amplitude feature vector CLBP _ M and the phase information feature vector extracted by the LPQ are subjected to Kernel Canonical Correlation Analysis (KCCA) to measure the correlation between the amplitude feature vector CLBP _ M and the phase information feature vector extracted by the LPQ and are subjected to dimensionality reduction to be combined into the fusion feature CLBP _ MP.
In step 3, in order to solve the problem that the image texture features extracted by the CLBP algorithm lack local contrast information, a novel weighting strategy combining color information is provided starting from the color texture features of the flame image.
In the step 3, the fluctuation of the flame image of the converter, especially the flame of molten steel combustion, is large, and more interference factors exist. Therefore, for the characteristics of the flame image, such as multiple scales, multiple directions and the like, two scales with the window sizes of 3 × 3 and 5 × 5 are adopted, the radius R of the two scales is 1, the sampling points p are respectively 8 and 16, and the flame image is subjected to feature extraction on the two scales.
The extraction process of extracting the color texture features of the flame image by using the improved complete local binary pattern method in the step 3 specifically comprises the following steps:
and 3.1, performing spatial conversion on the flame image and further dividing the flame image into three single-channel images, wherein each single-channel image is subjected to local difference operation to obtain a local region segmentation block of each flame single-channel image. Aiming at the characteristics of multi-scale and multi-direction and the like of the flame image, further expanding the size of two window scales of the local area segmentation block of each flame single-channel image, and extracting corresponding CLBP characteristic values under different scale parameters by adopting two scales of 3 multiplied by 3 and 5 multiplied by 5;
3.2, respectively obtaining an amplitude component, a symbol component, a central pixel gray value and a phase component by two window scales under each flame single-channel image local area partition block;
step 3.3, firstly, respectively calculating the gray value, the symbol value and the amplitude value of the central pixel under different scales, and multiplying the gray value, the symbol value and the amplitude value of each central pixel by the weight value after binary coding; meanwhile, extracting corresponding phase components of a single-channel local area of the flame image under different scale parameters and obtaining a specific phase value through binary coding; finally, the amplitude characteristic vector CLBP _ M and the phase information characteristic vector extracted by LPQ are combined into a fusion characteristic CLBP _ MP by measuring the correlation between the amplitude characteristic vector CLBP _ M and the phase information characteristic vector by Kernel Canonical Correlation Analysis (KCCA) and reducing the dimension;
and 3.4, combining the sign eigenvector, the amplitude eigenvector and the central pixel value under different scale parameters in each single channel of the flame image by adopting a serial and parallel combined structure to use three operators CLBP _ S, CLBP _ M and CLBP _ C. Firstly, serially connecting amplitude phase information CLBP _ MP and central pixel gray value CLBP _ C in sequence to form CLBP _ MP/C; then, the 2-dimensional CLBP _ S _ MP/C histogram is formed by connecting the CLBP _ S with the CLBP _ S in parallel;
and 3.5, splicing and combining the histograms of the three single channels of the flame image together, performing nonlinear dimensionality reduction by using local linear embedding, and inputting the reduced dimensions into a K nearest neighbor regression model to predict the carbon content.
Step 3.1, the CLBP feature extraction step corresponding to the flame image single-channel local area is as follows:
step 3.1.1, defining the central pixel point as (x, y), the pixel value as g (x, y), the radius R of the scale parameter as 1, and the sampling points p as 8 and 16 respectively, and calculating neighborhood sampling points (x) around the central pixel point (x, y) under the scale parameterj,yj) To obtain the pixel value f (x) of the neighborhood sampling pointj,yj)；
Wherein the neighborhood samples (x)j,yj) The position calculation formula of (a) is as follows;
Figure BDA0002889370480000061
in the formula (1), R is the sampling radius of the neighborhood sampling points, P is the total number of the neighborhood sampling points, (x)j,yj) Is composed ofThe position of any neighborhood sampling point j (j is more than or equal to 0 and less than or equal to P-1) around the central pixel point, f (x)j,yj) Is the pixel value of sample point j;
step 3.1.2, obtaining the pixel values of P neighborhood sampling points from the step 3.1.1, and converting the pixel value g (x) of the central pixel point0,y0) As a threshold, P neighborhood sampling points of the center pixel are discriminated to obtain a P-bit 0/1 binary value, and the specific discrimination method is as follows:
Figure BDA0002889370480000062
in the formula (2), (x)j,yj) Is the position of any neighborhood sampling point j (j is more than or equal to 0 and less than or equal to P-1) around the central pixel point, f (x)j,yj) Is the pixel value of sample point j;
step 3.3, the CLBP feature extraction step corresponding to the flame image single-channel local area is as follows:
step 3.3.1, splicing the P-bit 0/1 binary values obtained after discrimination in the step 3.1.2 according to a clockwise sequence to obtain the amplitude and the symbol size of the central pixel point (x, y) of the local difference block of the image, and obtaining the symbol value and the amplitude after binary coding; carrying out binary coding on the gray value of each central pixel, the symbol value and the amplitude value, and multiplying the binary coded gray value, the symbol value and the amplitude value by the weight value respectively;
3.3.2, in order to solve the problem that the image texture features extracted by the CLBP algorithm lack local contrast information, fully describe the color features and the texture features of the flame image, starting from the color texture features of the flame image, and combining with a weighting strategy of color information, calculating the weight of pixel points in the image, wherein the calculation formula is as follows:
the calculation formula is as follows:
Figure BDA0002889370480000063
in the formula (3), wherein Δ cpq,3x3Is the color distance, Δ c, at the 3 × 3 scalepq,5x5Is the color distance on the scale of 5 × 5, wherein
Figure BDA0002889370480000071
(HP,SP,VP) And (H)q,Sq,Vq) The method comprises the steps of obtaining a pixel p and a pixel q of an image segmentation block at a corresponding position of each channel in the HSV color space, wherein p is a neighborhood sampling point, and q is a central threshold value. var3×3(p, q) and var5×5(p, q) are the local variance at two scales, 3 × 3 and 5 × 5, respectively. Δ cpq,3x3×Δcpq,5x5The color distance values of 3 x 3 and 5 x 5 scales at the corresponding positions of the segmentation blocks of the respective local areas of the three single channels in the flame image HSV color space,
Figure BDA0002889370480000072
the local variance values of the flame image under a single channel under two scales of 3 x 3 and 5 x 5 are obtained.
Step 3.3.3, respectively calculating the gray value, the symbol value and the amplitude value of the central pixel under different scales in the local area block of the flame single-channel image, and multiplying the gray value, the symbol value and the amplitude value of each central pixel by a weight value after binary coding, such as C (i) andp,gc)×Wcolordividing the value of the multiplication of the gray value of the central pixel of the block and the weight value for the local area, wherein C (i)p,gc) For the central pixel grey value in a local area block, WcolorIs the weight;
Figure BDA0002889370480000073
binary-coding the symbol components of the block for local region segmentation and multiplying the result by a weight, wherein
Figure BDA0002889370480000074
Is a binary coded symbol value, gpNeighborhood sampling points, g, for symbol blocks in local regions of an imagecFor the central threshold of the symbol block of the local area of the image, 2pIs binary coding, and p is the number of sampling points;
Figure BDA0002889370480000075
binary-coding the amplitude component of the local region partition block and multiplying the binary-coded amplitude component by a weight value
Figure BDA0002889370480000076
Is the binary coded amplitude, mpIs a neighborhood sampling point of an image local region amplitude block, c is a central threshold of the image local region amplitude block, 2pFor binary coding, p is the number of sample points.
Meanwhile, as shown in the dotted line frame shown in fig. 2, local phase information of the flame image is extracted by using a local phase quantization method, and a specific phase value is obtained by binary coding a phase component of the flame image.
And finally, measuring the correlation between the final amplitude characteristic vector and the final phase characteristic vector of the flame single-channel image by KCCA and combining the correlation and the dimension reduction into a fusion characteristic CLBP _ MP, wherein the formula (4) is as follows:
Figure BDA0002889370480000081
in the formula (4), WcolorThe weight value of the color information for enhancing the local contrast is provided by the invention;
Figure BDA0002889370480000082
the amplitude component is multiplied by the weight value after binary coding;
Figure BDA0002889370480000083
the total amplitude characteristic vector of all local segmentation blocks under the single channel is obtained;
Figure BDA0002889370480000084
the total phase eigenvector of all local segmentation blocks under the single channel; KCCA is a nuclear canonical correlation analysis. The KCCA is used for respectively carrying out linear projection on two groups of data, and reflecting the correlation between two groups of indexes through the correlation between comprehensive variables. First, at
Figure BDA0002889370480000085
And
Figure BDA0002889370480000086
finding a pair of projection vectors U ═ alpha in two groups of variablesTX and V ═ βTY, such that the linear combinations of the two groups have the largest correlation coefficient therebetween; then, a plurality of linear groups of variables with the maximum number of phase relations are selected from the two groups of random variables, and a projection matrix U is outputxAnd VyThe linear combination thus selected is referred to as a representative correlation variable.
Finally, the final CLBP _ S, CLBP _ M and CLBP _ C under the single channel of the flame adopt a serial and parallel combined structure to combine and use three operators of CLBP _ S, CLBP _ MP and CLBP _ C. Firstly, serially connecting amplitude phase information CLBP _ MP and central pixel gray value CLBP _ C in sequence to form CLBP _ MP/C; then, a 2-dimensional histogram of CLBP _ S _ MP/C is formed in parallel with CLBP _ S.
The prediction model in the step 4 adopts a K nearest neighbor regression model, and the prediction process specifically comprises the following steps:
as shown in fig. 2, the total eigenvector of three single channels (H channel, S channel, V channel) under the flame image is subjected to nonlinear dimensionality reduction using Local Linear Embedding (LLE) and input into a K-nearest neighbor regression model for carbon content prediction. The K-nearest neighbor algorithm is that N sample points contained in an R-dimensional space of flame image features are directly input in a training set, a target point P is given arbitrarily, K (K is more than or equal to 1 and less than or equal to N) nearest neighbor points of P in the R-dimensional space are searched, and most label types in the K nearest neighbor points are used as label values of P. When the carbon content is predicted, firstly, dividing a data set into ten parts, training 1 part of the ten parts in turn, and testing 1 part of the ten parts, wherein the average value of 10 times of results is used as an estimated value of algorithm precision; then, directly inputting N sample points of the flame image characteristics in a K-nearest neighbor training model, and weighting and summing the label values of the taken K (K is 5) sample points to be used as the carbon content category corresponding to the input image characteristics; and finally, solving the mean value of the predicted values of 10 groups of corresponding test sets by using a ten-fold cross-validation method. In the K-nearest neighbor regression model (neighbors. kneighborsregressor) function, except that the K value is manually adjusted, the other parameters are default parameter values.
The experimental result shows that the carbon content prediction accuracy of the invention is 83.9% within the error range of 0.02%. The method can effectively solve the problem that the carbon content cannot be accurately and precisely predicted due to the fact that the flame images with high similarity and similar carbon content are difficult to distinguish, and improves the local structure contrast information of the flame images.
While the present invention has been described in detail with reference to the embodiments shown in the drawings, the present invention is not limited to the embodiments, and various changes can be made without departing from the spirit of the present invention within the knowledge of those skilled in the art.

Claims (3)

1. A flame endpoint carbon content prediction method based on an improved complete local binary pattern is characterized by comprising the following steps: the method comprises the following specific steps:
step 1, acquiring furnace mouth flame video data shot in the actual production process of a steel mill;
step 2, when experimental data are manufactured, extracting the flame fire hole area of each training sample in the sample set from all times and unifying the size to obtain a flame image data set;
step 3, extracting color texture characteristics of the flame image by using an improved complete local binary pattern method;
and 4, inputting the color texture characteristics extracted in the step 3 into a K neighbor regression model to predict the carbon content.
2. The method for improving the complete local binary pattern flame endpoint carbon content prediction as claimed in claim 1, wherein the method comprises the following steps: the color texture features of the flame image in the step 3 comprise symbol features, amplitude features, central pixel gray value features and phase features, and the extraction process specifically comprises the following steps:
step 3.1, performing spatial conversion on the flame image and further dividing the flame image into three single-channel images, wherein each single-channel image is subjected to local difference operation to obtain a local area segmentation block of each flame single-channel image; then, further expanding the two window sizes of each flame single-channel image local area segmentation block;
3.2, respectively obtaining an amplitude component, a symbol component, a central pixel gray value and a phase component by two window scales under each flame single-channel image local area partition block;
3.3, multiplying two amplitude components obtained by two different window scales under each flame single-channel image local area partition block by a weight after binary coding, simultaneously obtaining a specific phase value by the phase component through binary coding, and finally measuring the correlation between the two vectors and reducing the dimension to combine the amplitude phase fusion characteristics CLBP _ MP by the final amplitude characteristic vector and the phase characteristic vector of each single-channel flame image through KCCA;
and 3.4, for each single-channel flame image, combining three operational characters of a sign characteristic CLBP _ S, an amplitude phase fusion characteristic CLBP _ MP and a central pixel gray value CLBP _ C by adopting a serial and parallel combined structure: firstly, serially connecting an amplitude and phase fusion characteristic CLBP _ MP and a central pixel gray value CLBP _ C in sequence to form a CLBP _ MP/C; then, the CLBP _ MP/C and the sign feature CLBP _ S are connected in parallel to form a 2-dimensional CLBP _ S _ MP/C histogram of each single-channel flame image;
and 3.5, splicing and combining the histograms of the three single channels of the flame image together, performing nonlinear dimensionality reduction by using local linear embedding, and inputting the reduced dimensions into a K nearest neighbor regression model to predict the carbon content.
3. The method for predicting the carbon content of the converter steelmaking endpoint based on the improved complete local binary pattern flame feature extraction as claimed in claim 2, wherein in the step 3.3, the extraction step of the single-channel flame image amplitude and phase fusion feature CLBP _ MP is as follows:
step 3.3.1, two amplitude components obtained by two different window scales under a flame single-channel image local area segmentation block are multiplied by a weight value after being respectively subjected to binary coding,
Figure FDA0002889370470000021
is a value obtained by binary coding the amplitude component of the local region partition block and multiplying the binary coded amplitude component by a weight, WcolorTo enhance the weighting of the local contrast color information,
Figure FDA0002889370470000022
the amplitude component of the complete local binary pattern is subjected to binary coding; m ispIs a neighborhood sampling point of an image local region amplitude block, c is a central threshold of the image local region amplitude block, 2pIs binary coding, and p is the number of sampling points; meanwhile, the phase component is also subjected to binary coding to obtain a specific phase value;
step 3.3.2, combining with the weighting strategy of the color information, calculating the weight of the pixel point in the image, wherein the calculation formula is as follows:
Figure FDA0002889370470000023
in the formula (3), wherein Δ cpq,3x3Is the color distance, Δ c, at the 3 × 3 scalepq,5x5Is the color distance on the scale of 5 × 5, wherein
Figure FDA0002889370470000024
(HP,SP,VP) And (H)q,Sq,Vq) The method comprises the steps of dividing pixels p and pixels q of an image segmentation block at the corresponding position of each channel in HSV color space, wherein p is a neighborhood sampling point, q is a central threshold value, var is3×3(p, q) and var5×5(p, q) local variance at two scales, 3 × 3 and 5 × 5, Δ cpq,3x3×Δcpq,5x5The color distance values of 3 x 3 and 5 x 5 scales at the corresponding positions of the segmentation blocks of the respective local areas of the three single channels in the flame image HSV color space,
Figure FDA0002889370470000025
is a fireThe local variance values of the flame image under the single channel under two scales of 3 x 3 and 5 x 5 of the local segmentation block;
step 3.3.3, finally, the final amplitude characteristic vector and the phase characteristic vector of the single channel are combined into amplitude phase fusion characteristic CLBP _ MP by measuring the correlation between the two vectors and reducing the dimension through kernel canonical correlation analysis KCCA, and the calculation formula is as follows:
Figure FDA0002889370470000031
in the formula (I), the compound is shown in the specification,
Figure FDA0002889370470000032
the total amplitude characteristic vector of all local segmentation blocks under the single channel is obtained;
Figure FDA0002889370470000033
for all local partitions of the block total phase eigenvector in the single channel, wherein
Figure FDA0002889370470000034
2j-1Is binary coding; KCCA is a nuclear canonical correlation analysis.
CN202110023350.XA 2021-01-08 2021-01-08 Flame endpoint carbon content prediction method based on improved complete local binary pattern Active CN112734722B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110023350.XA CN112734722B (en) 2021-01-08 2021-01-08 Flame endpoint carbon content prediction method based on improved complete local binary pattern

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110023350.XA CN112734722B (en) 2021-01-08 2021-01-08 Flame endpoint carbon content prediction method based on improved complete local binary pattern

Publications (2)

Publication Number Publication Date
CN112734722A true CN112734722A (en) 2021-04-30
CN112734722B CN112734722B (en) 2022-09-13

Family

ID=75591286

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110023350.XA Active CN112734722B (en) 2021-01-08 2021-01-08 Flame endpoint carbon content prediction method based on improved complete local binary pattern

Country Status (1)

Country Link
CN (1) CN112734722B (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117612078A (en) * 2023-10-08 2024-02-27 成都格理特电子技术有限公司 Image-based hearth flame detection method

Citations (16)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6205728B1 (en) * 1997-04-30 2001-03-27 Frank Sutelan Laminated composite building component
CN101698896A (en) * 2009-09-28 2010-04-28 南京理工大学 System and method for steel-making online end-point control through furnace mouth radiation information fusion
CN102392095A (en) * 2011-10-21 2012-03-28 湖南镭目科技有限公司 Termination point prediction method and system for converter steelmaking
CN102876838A (en) * 2012-10-30 2013-01-16 湖南镭目科技有限公司 System for detecting carbon content and temperature in converter
CN103714351A (en) * 2013-12-18 2014-04-09 五邑大学 Depth self learning-based facial beauty predicting method
CN106148637A (en) * 2015-04-10 2016-11-23 南京理工大学 The pneumatic steelmaking carbon content dynamic detection system of high stable
CN106153550A (en) * 2015-04-10 2016-11-23 南京理工大学 Converter steel-smelting molten steel carbon content based on SVM online Real-time and Dynamic Detection method
CN106153551A (en) * 2015-04-10 2016-11-23 南京理工大学 Converter steel-smelting molten steel carbon content based on SVM online Real-time and Dynamic Detection system
CN107025652A (en) * 2017-05-05 2017-08-08 太原理工大学 A kind of flame detecting method based on kinetic characteristic and color space time information
CN108319964A (en) * 2018-02-07 2018-07-24 嘉兴学院 A kind of fire image recognition methods based on composite character and manifold learning
CN109214420A (en) * 2018-07-27 2019-01-15 北京工商大学 The high texture image classification method and system of view-based access control model conspicuousness detection
CN109711345A (en) * 2018-12-27 2019-05-03 南京林业大学 A kind of flame image recognition methods, device and its storage medium
CN109975507A (en) * 2019-04-28 2019-07-05 华北理工大学 A kind of real-time determining method and system for making steel later period carbon content of molten steel and temperature value
CN110766023A (en) * 2018-07-27 2020-02-07 深圳市白麓嵩天科技有限责任公司 Complete local contrast binary pattern texture image processing method
CN111461003A (en) * 2020-03-31 2020-07-28 湖南大学 Coal-fired working condition identification method based on video image sequence feature extraction
US20200250822A1 (en) * 2019-02-01 2020-08-06 Essen Instruments, Inc. D/B/A Essen Bioscience, Inc. Label-Free Cell Segmentation Using Phase Contrast and Brightfield Imaging

Patent Citations (16)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6205728B1 (en) * 1997-04-30 2001-03-27 Frank Sutelan Laminated composite building component
CN101698896A (en) * 2009-09-28 2010-04-28 南京理工大学 System and method for steel-making online end-point control through furnace mouth radiation information fusion
CN102392095A (en) * 2011-10-21 2012-03-28 湖南镭目科技有限公司 Termination point prediction method and system for converter steelmaking
CN102876838A (en) * 2012-10-30 2013-01-16 湖南镭目科技有限公司 System for detecting carbon content and temperature in converter
CN103714351A (en) * 2013-12-18 2014-04-09 五邑大学 Depth self learning-based facial beauty predicting method
CN106148637A (en) * 2015-04-10 2016-11-23 南京理工大学 The pneumatic steelmaking carbon content dynamic detection system of high stable
CN106153550A (en) * 2015-04-10 2016-11-23 南京理工大学 Converter steel-smelting molten steel carbon content based on SVM online Real-time and Dynamic Detection method
CN106153551A (en) * 2015-04-10 2016-11-23 南京理工大学 Converter steel-smelting molten steel carbon content based on SVM online Real-time and Dynamic Detection system
CN107025652A (en) * 2017-05-05 2017-08-08 太原理工大学 A kind of flame detecting method based on kinetic characteristic and color space time information
CN108319964A (en) * 2018-02-07 2018-07-24 嘉兴学院 A kind of fire image recognition methods based on composite character and manifold learning
CN109214420A (en) * 2018-07-27 2019-01-15 北京工商大学 The high texture image classification method and system of view-based access control model conspicuousness detection
CN110766023A (en) * 2018-07-27 2020-02-07 深圳市白麓嵩天科技有限责任公司 Complete local contrast binary pattern texture image processing method
CN109711345A (en) * 2018-12-27 2019-05-03 南京林业大学 A kind of flame image recognition methods, device and its storage medium
US20200250822A1 (en) * 2019-02-01 2020-08-06 Essen Instruments, Inc. D/B/A Essen Bioscience, Inc. Label-Free Cell Segmentation Using Phase Contrast and Brightfield Imaging
CN109975507A (en) * 2019-04-28 2019-07-05 华北理工大学 A kind of real-time determining method and system for making steel later period carbon content of molten steel and temperature value
CN111461003A (en) * 2020-03-31 2020-07-28 湖南大学 Coal-fired working condition identification method based on video image sequence feature extraction

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
M. SHIMODA等: "Prediction method of unburnt carbon for coal fired utility boiler using image processing technique of combustion flame", 《IEEE》 *
李清荣等: "基于火焰彩色纹理特征的转炉炼钢碳含量预测", 《计算机集成制造系统》 *

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117612078A (en) * 2023-10-08 2024-02-27 成都格理特电子技术有限公司 Image-based hearth flame detection method

Also Published As

Publication number Publication date
CN112734722B (en) 2022-09-13

Similar Documents

Publication Publication Date Title
CN111860573A (en) Model training method, image class detection method and device and electronic equipment
CN108564085B (en) Method for automatically reading of pointer type instrument
CN106446004B (en) Digital pathological full-section image retrieval method
CN112819802B (en) Method for supervising and predicting blast furnace condition abnormality based on tuyere information deep learning
CN112734722B (en) Flame endpoint carbon content prediction method based on improved complete local binary pattern
CN112329558A (en) FeO content prediction method based on sintering machine tail section video
CN102713974B (en) Learning device, recognition device, study recognition system and study recognition device
CN117474918B (en) Abnormality detection method and device, electronic device, and storage medium
CN109584267B (en) Scale adaptive correlation filtering tracking method combined with background information
CN114926826A (en) Scene text detection system
Yang et al. Multisource Information Fusion for Autoformer: Soft Sensor Modeling of FeO Content in Iron Ore Sintering Process
CN107742114B (en) Hyperspectral image feature detection method and device
CN117611901A (en) Small sample image classification method based on global and local contrast learning
CN113420173A (en) Minority dress image retrieval method based on quadruple deep learning
CN114065798A (en) Visual identification method and device based on machine identification
CN110503631B (en) Remote sensing image change detection method
CN112581502A (en) Target tracking method based on twin network
CN109448030B (en) Method for extracting change area
Huang et al. Graphite classification of gray cast iron in metallographic via a deep learning approach
CN114581475A (en) Laser stripe segmentation method based on multi-scale saliency features
CN106033550B (en) Method for tracking target and device
Gui et al. A study on tongue image color description based on label distribution learning
JP2000020721A (en) Device for retrieving image and method therefor
CN113076438B (en) Classification method based on conversion from majority class to minority class under unbalanced data set
CN115600486A (en) CJS-SLLE dimension reduction-based converter steelmaking end point carbon temperature prediction method

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant