CN108052950A - A kind of segmentation of electric melting magnesium furnace dynamic flame and feature extracting method based on MIA - Google Patents

Abstract

A MIA-based dynamic flame segmentation and feature extraction method of an electric fused magnesium furnace of the present invention comprises: collecting video images of the flame, converting the RGB image into a two-dimensional matrix; adopting the PCA method to reduce the dimension of the two-dimensional matrix; After dimensioning, the matrix is normalized to [0,255] to obtain a score histogram; compare the score histograms of different working conditions, find out the areas with obvious changes in the figure and mark them, and map the marked areas back to The flame segmentation image was obtained from the original RGB image; five characteristic data including the size of the flame brightness area, the number of flame color types, the average color of the flame area, the average color of the entire image, and the flame brightness value were calculated by using the marked area in the score histogram. This method can effectively segment the flame area, and its segmentation effect is good. By calculating five kinds of feature data for the segmented image, and applying the result to the working condition classification model, a higher classification accuracy is obtained.

Description

A dynamic flame segmentation and feature extraction method for fused magnesium furnace based on MIA

technical field

The invention belongs to the technical field of pattern recognition and artificial intelligence, and in particular proposes an MIA-based method for dynamic flame segmentation and feature extraction of an electric fused magnesium furnace.

Background technique

Fused magnesia has the characteristics of high purity, high melting point, strong insulation performance and compact structure. It is widely used in various industries and fields such as chemical industry, construction, home appliances, metallurgy, and military affairs. It is a good refractory raw material.

When electric arc furnace is used to smelt and produce fused magnesia, the brightness, color, and shape changes of the flame at the furnace mouth corresponding to smelting conditions, feeding conditions, exhaust conditions, and under-burning conditions are all different. Among them, under the smelting condition, the flame brightness is moderate, the flame color is relatively rich, and the flame shape changes generally; under the feeding condition, the flame brightness is relatively dark, the flame color is average, and the shape change is slow; under the exhaust condition, the flame brightness is relatively bright , the color of the flame is relatively simple, and the shape of the flame changes quickly. Therefore, the change of the flame at the furnace mouth is an important basis for judging the working conditions.

At present, the information contained in the flame at the furnace mouth needs to be manually inspected and obtained through the experience of "watching the fire" when going to the production line. However, there are the following problems in manual inspection: 1) The accuracy of the judgment is related to the experience and status of the operator, and it is easy to miss or misdetect; 2) The on-site production environment is harsh (strong light, high temperature, dust, etc.), and the labor intensity is high , high risk, not suitable for long-term on-site inspection by workers; 3) The boundary between the flame at the furnace mouth and the surrounding smoke and dust is blurred, and the process of "watching the fire" is easily disturbed. Therefore, enterprises need a method that can separate the flame area of the furnace mouth from the image, and express the characteristic information such as flame color, brightness, shape, etc. in specific numbers, and use it in the subsequent identification of working conditions.

With the application of industrial cameras, the accuracy of video image acquisition is gradually improving. Industrial cameras are used to collect images in industrial production, and then the region of interest is obtained through image segmentation methods, and the characteristic data of the region are calculated. This process has been widely used in industrial production such as natural gas fuel oil quality monitoring and rotary kiln combustion status. However, because the flame is dynamic and has no fixed and regular shape, it is difficult to determine the characteristics of the flame such as the frequency of change, color, and brightness for different industrial backgrounds. In a single industrial process, dynamic flame images are based on multiple It exists in one form and is closely combined with the background environment, so it is difficult to use the idea of threshold method to make the image segmentation algorithm suitable for all working conditions.

Contents of the invention

The embodiment of the present invention provides a dynamic flame segmentation and feature extraction method of an electric fused magnesium furnace based on MIA, which can effectively segment the dynamic flame area, calculate 5 kinds of feature data for the segmented image, and apply the result to the subsequent On the working condition classification model, a higher classification accuracy is obtained.

The invention provides a method for dynamic flame segmentation and feature extraction of an electric fused magnesium furnace based on MIA, comprising the following steps:

Step 1: Collect the video image of the flame, and convert the RGB image of a single frame into a two-dimensional matrix;

Step 2: Carry out dimensionality reduction processing on the two-dimensional matrix by principal component analysis, and map the selected principal elements back to the RGB image space for verification through image reconstruction technology, so as to ensure that the dimensionality-reduced image can replace the original image;

Step 3: Normalize the matrix with two columns of vectors after dimension reduction to [0,255], then use the two values of each row of data in the matrix as the position information in the XY coordinate system, and count the matrix with the same coordinates The number of pixels of the value, so as to obtain the score histogram;

Step 4: Repeat steps 1-3 for single-frame RGB images of different working conditions, compare the obtained score histograms of different working conditions, and find out the areas with obvious changes in the score histograms of different working conditions The area marked is processed, and the marked area in the score histogram is mapped back to the original RGB image to obtain the flame segmentation image;

Step 5: Fine-tune the marked area in the score histogram to obtain the determined marked area and accurate flame segmentation image, and segment all single-frame RGB images in the video image according to the determined marked area to obtain the Flame segmentation image;

Step 6: Through the feature extraction formula, the five characteristic data of the size of the flame brightness area, the number of flame color types, the average color of the flame area, the average color of the entire image, and the flame brightness value are calculated respectively with the help of the marked area in the score histogram.

The MIA-based dynamic flame segmentation and feature extraction method of the fused magnesium furnace of the present invention uses an industrial camera to obtain the process image of the fused magnesium production site, uses a fixed window to eliminate irrelevant areas far away from the flame in the image, and then uses multivariable Image Analysis (MIA) method, which segments the flame area of interest from the background and calculates the relevant characteristic data. This method can effectively segment the flame area of the furnace mouth, and its segmentation effect is much better than that based on the threshold method. By calculating 5 kinds of characteristic data for the segmented image, and applying the result to the subsequent working condition classification model, obtain higher classification accuracy.

Description of drawings

Fig. 1 is the flowchart of a kind of MIA-based electric fused magnesium furnace dynamic flame segmentation and feature extraction method of the present invention;

Fig. 2 is the flame image of the first working condition of the present invention;

Fig. 3 is the flame image of the second working condition of the present invention;

Fig. 4 is the flame image of the third working condition of the present invention;

Fig. 5 is the flame image of the fourth working condition of the present invention;

Fig. 6 is the scoring histogram of the first working condition of the present invention;

Fig. 7 is the scoring histogram of the second working condition of the present invention;

Fig. 8 is the scoring histogram of the third working condition of the present invention;

Fig. 9 is the scoring histogram of the fourth working condition of the present invention;

Fig. 10 is a schematic diagram of marking in the determined marked area of the score histogram;

Fig. 11 is a schematic diagram of marking in the attachment area of the determined marked area of the score histogram;

Fig. 12 is a schematic diagram of marking and mapping back to the original RGB image using the marked region determined in the score histogram;

Fig. 13 is a schematic diagram of labeling and mapping back to the original RGB image using the marked area attached to the determined marked area;

14 is a flame segmentation image obtained by segmenting a group of images based on MIA-based dynamic flame segmentation and feature extraction method of the fused magnesium furnace of the present invention;

Fig. 15 is a trend diagram of five kinds of feature data extracted by the MIA-based dynamic flame segmentation and feature extraction method of the fused magnesium furnace of the present invention.

Detailed ways

This paper proposes a method based on multivariate image analysis (Multivariate Image Analysis, MIA) for the dynamic flame image segmentation and feature extraction method of the fused magnesium furnace mouth. Use the industrial camera to obtain the process image of the fused magnesium production site, use a fixed window to eliminate the irrelevant areas far away from the flame in the image, and then use the multivariate image analysis (MIA) method to segment the flame area of interest from the background and calculate related feature data.

A kind of MIA-based electric fused magnesium furnace dynamic flame segmentation and feature extraction method of the present invention as shown in Figure 1 comprises the following steps:

Step 1: Collect the video image of the flame, and convert the RGB image of a single frame into a two-dimensional matrix; Step 1 specifically includes:

Step 1.1: Segment the video image collected by the industrial camera with a fixed window, eliminate the interference caused by the peripheral area of the flame to the image segmentation, reduce the operating cost of the computer, and increase the single-frame graphics processing speed;

Step 1.2: Transform a single flame RGB image into a matrix in a two-dimensional space through matrix transformation.

The image after the fixed window segmentation has 3 channels in the RGB space, that is, the dimension is m×n×3, and each channel is stretched row by row, such as the R channel, which becomes mn rows of data after stretching, then three After the channel is stretched, it becomes a matrix of mn×3, namely

Among them, I _(m,n,3) represents the original RGB image, and I _1(m,n,3) represents the matrix after expanding I _(m,n,3) into two dimensions.

Step 2: Use principal component analysis (PCA) to reduce the dimensionality of the two-dimensional matrix, and map the selected principal components back to the RGB image space for verification through image reconstruction technology, so as to ensure that after dimensionality reduction The image can replace the original image; the step 2 specifically includes:

Step 2.1: Reduce the dimension of the above mn×3 two-dimensional matrix by principal component analysis, and select a certain number of principal components to ensure that the matrix after dimensionality reduction can represent 99% of the information of the original data;

The core algorithm of principal component analysis is as follows:

Among them, PC is the number of main components, p _a represents the loading vector, and t _a represents the score vector, which is the matrix after dimensionality reduction.

In the process of finding p _a = (1,...,PC), t _a (a=1,...,PC), the problem of too large row dimension needs to be considered, a mn=512×512 The image space corresponds to 262,144 rows. For so many row-dimensional matrices formed after multivariate image expansion, a core algorithm is needed to simplify the calculation. The core idea of the above algorithm is to de-meanize the expanded image first for each attribute, then construct a kernel matrix I ₁ ^T I ₁ , and perform singular value decomposition on the obtained low-dimensional matrix C(3,3) ( Singular value decomposition, SVD), arrange the eigenvectors into a matrix according to the size of the corresponding eigenvalues from large to small, and each column of the obtained eigenvector matrix is the loading vector p _a =(1,...,PC). On this basis, the score vector t _a (a=1,...,PC) is calculated according to t _a =I ₁ p _a .

Step 2.2: Map the selected pivots back to the RGB image space through image reconstruction technology, and confirm that the dimensionality-reduced image can replace the original image through observation and comparison.

In specific implementation, the value of the pivot can be determined by the following formula, so that the data after dimension reduction is greater than the required original image information, and it is determined that the image after dimension reduction can replace the original image, where PC (PC≤3) is the principal element λ _k is the eigenvalue of the covariance matrix obtained through principal component analysis.

Among them, g is the threshold value, indicating the percentage of the selected pivot that can represent the original image information, and generally takes g≥0.95. For the score vector t _a (a=1,...,PC), t ₁ contains the most original graphics information, t ₂ contains the most information, and so on. The first PC dominant principal components are selected to reconstruct the original multivariate image, while the residual matrix E is ignored, thereby eliminating most of the unstructured noise in the original image.

After the above operations are completed, the original image can be reconstructed according to the score vector t _a and loading vector p _a to obtain a new RGB image matrix through the following formula, and the image distortion after compression can be observed by comparing the results.

Use the first two pivots, the first pivot, the second pivot and the third pivot for image reconstruction. After PCA processing on the original furnace mouth flame image, the first two eigenvalues are taken into the calculation, and the result shows that it is greater than 99.5%, that is, the result of reconstruction based on the first two principal elements contains more than 99.5% of the information of the original image. Then reconstruct according to each pivot, and their reconstructed images represent the size of the original image information just as their eigenvalues are from large to small. Among them, the first two principal component reconstruction results are very close to the original image and contain the most information. The first pivot, the second pivot and the third pivot correspond to the increasingly serious distortion of the original image. When the first two pivots are selected to represent the original information, then the value corresponding to the third pivot is the residual matrix E.

Step 3: Normalize the matrix with two columns of vectors after dimension reduction to [0,255], then use the two values of each row of data in the matrix as the position information in the XY coordinate system, and count the matrix with the same coordinates The number of pixels of the value, so as to obtain the score histogram;

During specific implementation, the matrix with two column vectors after dimensionality reduction is normalized to [0,255] by the following formula:

Among them, s _i represents the two column vectors after normalization, t _i represents the two column vectors before normalization, t _{i, min} represents the smallest element in the column vector t _i , t _{i, max} represents the column vector t _i largest element.

During specific implementation, the following formula is used to obtain the score histogram:

Among them, TT _{x, y} is the number of the same pixel points whose coordinates are (x, y) in the score histogram, b is a certain row of the matrix, s _{1, b} represents the element of the bth row of the vector s ₁ , s _{2, b} represents the element of row b of vector _s2 .

Step 4: Repeat steps 1-3 for single-frame RGB images of different working conditions, compare the obtained score histograms of different working conditions, and find out the areas with obvious changes in the score histograms of different working conditions The area marked is processed, and the marked area in the score histogram is mapped back to the original RGB image to obtain the flame segmentation image;

The following will take the furnace mouth flame in the production process of the electric fused magnesium furnace as an example for verification.

First of all, according to the operator's experience, we selected 4 typical flame shapes of the magnesium furnace mouth from a large amount of video data, representing 4 working conditions of feeding, normal smelting, under-firing and exhaust respectively. As shown in Figure 2 to Figure 5.

Then, repeat steps 1-3 to obtain the score histograms of the four working conditions, as shown in Figures 6 to 9. The above four pictures are the projections of the flame at the furnace mouth in the xy coordinate system after MIA processing, where the x-axis corresponds to s ₁ , and the y-axis corresponds to s ₂ , which reflect the color and brightness information of the flame at the furnace mouth in two The clustering situation in the three-dimensional space, the brighter the color area, the more the number of furnace mouths with this color.

Finally, by comparing the distribution of density maps in the four score histograms, we found that in the area of x-axis [0-50] and y-axis [0-200], the change of the score histogram changes with the color and brightness of the picture most obvious.

Step 5: Fine-tune the marked area in the score histogram to obtain the determined marked area and accurate flame segmentation image, and segment all single-frame RGB images in the video image according to the determined marked area to obtain the Flame segmentation image; said step 5 includes:

Step 5.1: Fine-tune the marked area in the score histogram, and obtain a new flame segmentation image. If the segmentation effect is significantly improved, keep the new marked area, and obtain a definite marked area and an accurate flame through multiple adjustments. Flame segmentation image;

Step 5.2: mark a verification area around the determined marked area by the verification method, map the verified area back to the original RGB image, if the segmented image corresponding to the verified area is located around the flame image, it indicates that the determined marked area conforms to Flame splitting requirements.

During the specific implementation, in order to verify that the area obtained in step 5.1 is the flame area of interest, we mark another area on the score histogram, which is close to the previous area of interest, and at the same time show the marked color area in the original image. As shown in Figure 10, mark a s1[0,50] interval and s2[0,200] interval on the score histogram, and Figure 12 shows the result of mapping back to the original image, that is, the segmentation result image, and Figure 11 shows the results in Figure 10 Near the marked area, re-mark a polygonal area, and map it back to the original image to get Figure 13, which is the smoke image around the flame. If the segmentation image corresponding to the verification area is located around the flame image, it indicates that the determined marked area meets the requirements of flame segmentation.

Step 6: Through the feature extraction formula, the five characteristic data of the size of the flame brightness area, the number of flame color types, the average color of the flame area, the average color of the entire image, and the flame brightness value are calculated respectively with the help of the marked area in the score histogram.

Among them, the size of the flame brightness region refers to the size of the region of interest obtained after segmentation. Each pixel value of the original RGB image is projected into the score histogram, so the size of the region of interest in the score histogram is the marked region The sum of all values calculates the size of the flame brightness area according to the following formula:

Among them, A represents the size of the flame brightness area, TT _{x, y} is the number of the same pixel points whose coordinates are (x, y) in the score histogram, M _i,j = 1 corresponds to the marked area in the score histogram.

Each coordinate with a value greater than 1 in the marked area in the score histogram in step 6 corresponds to a color of the original RGB image, and the number of flame colors is the number of coordinates with a value greater than 0 in the marked area in the score histogram and, calculate the number of flame colors according to the following formula:

Among them, C represents the number of flame colors, TT _{x, y} is the number of the same pixel points whose coordinates are (x, y) in the score histogram, M _i,j = 1 corresponds to the marked area in the score histogram.

In step 6, the average value of the color of the flame area is calculated according to the following formula:

Among them, s _f represents the average value of the color of the flame area, TT _{x, y} is the number of the same pixel points whose coordinates are (x, y) in the score histogram, and A represents the size of the flame brightness area;

Calculate the color average of the entire image according to the following formula:

Among them, s _m is the average color of the entire image, TT _{x, y} is the number of the same pixel points with coordinates (x, y) in the score histogram, and M×N is the size of the entire image.

The flame brightness value calculated in step 6 is specifically:

First, restore the score histogram to an RGB image according to the following formula:

Among them, t _1,max represents the largest element in the column vector t ₁ , t _1,min represents the smallest element in the column vector t ₁ , t _2,max represents the largest element in the column vector t ₂ , t _2,min represents the column The smallest element in vector t ₂ , p ₁ is the first column of loading vector p _a =(1,...,PC), p ₂ is the second column of loading vector p _a =(1,...,PC) List.

Second, the grayscale image is calculated according to the following formula:

L _xy ＝[R,G,B] _xy [0.299,0.587,0.114] ^T (13)

Finally, the flame brightness value is calculated according to the following formula:

Among them, TT _{x, y} is the number of the same pixel points whose coordinates are (x, y) in the score histogram.

Use the MIA-based dynamic flame segmentation and feature extraction method of the fused magnesia furnace of the present invention to process a section of about 58 seconds of fused magnesia furnace production video data, and save the 1500 images obtained by segmentation, and the region of interest obtained by segmentation That is, the flame image result of each frame is shown in Figure 14. Compared with the commonly used image classification methods, the MIA-based dynamic flame segmentation and feature extraction method of the fused magnesium furnace is used to achieve better segmentation results of the image area of interest. For the marked area in the score histogram of each frame image, use the formula (6) (7) (8) (9) (14) to calculate the characteristic data, and get the trend diagram of the five flame characteristic data as shown in Figure 15 Show. Corresponding to the video image and the image of the region of interest, observing the change of the curve, it is found that the change trend of these five characteristic data has a strong correlation with the change of the video image. Specifically reflected in the on-site environment, several working conditions of the fused magnesium furnace show periodic changes, and the size, color, and shape of the image of interest obtained by segmentation also show periodic changes, and this change is more intuitive. It is an approximate periodic change in the five characteristic data trend graphs. Therefore, these five types of feature data are of great significance for subsequent classification using machine learning to ensure high accuracy.

The above descriptions are only preferred embodiments of the present invention, and are not intended to limit the idea of the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention should be included in the present invention. within the scope of protection.

Claims (10)

1. a kind of segmentation of electric melting magnesium furnace dynamic flame and feature extracting method based on MIA, which is characterized in that including following step Suddenly：

Step 1：The video image of flame is gathered, and the RGB image of single frames is converted into two-dimensional matrix；

Step 2：Dimension-reduction treatment is carried out to the two-dimensional matrix using Principal Component Analysis Method, and passes through image reconstruction technique to choose Pivot map back RGB image space and verified, to ensure that the image after dimensionality reduction can replace original image；

Step 3：By the matrix normalization with two column vectors after dimensionality reduction between [0,255], then in matrix per line number According to two values as the location information under XY coordinate systems, number of pixels of the matrix with same coordinate value is counted, so as to obtain Obtain score block diagram；

Step 4：Step 1-3 is repeated for the RGB image of the single frames of different operating modes, by the score block diagram of the different operating modes of acquisition It is compared, finds out in the score block diagram of different operating modes and change apparent region, place is marked to changing apparent region Reason, and the area maps being labeled in the score block diagram are returned into original RGB image, to obtain flame segmentation figure picture；

Step 5：The region being labeled in score block diagram is finely adjusted, to obtain definite marked region and accurate flame Segmentation figure picture, and single frames RGB image all in video image is split according to definite marked region and is obtained per frame Flame segmentation figure picture；

Step 6：By feature extraction formula, it is big that flame luminance area is calculated respectively by marked area in score block diagram 5 kinds of small, flame color species number, flame region color average, entire image color average and flame brightness value features Data.

2. the electric melting magnesium furnace dynamic flame segmentation based on MIA and feature extracting method, feature exist as described in claim 1 In the step 1 includes：

Step 1.1：Window segmentation is fixed in the video image that industrial camera is collected, and rejects flame peripheral region to image Interference caused by segmentation；

Step 1.2：The matrix being transformed to single frames flame RGB image by matrixing under two-dimensional space.

3. the electric melting magnesium furnace dynamic flame segmentation based on MIA and feature extracting method, feature exist as described in claim 1 In the step 2 includes：

Step 2.1：Above-mentioned two-dimensional matrix is carried out by dimensionality reduction by Principal Component Analysis Method, chooses a certain amount of pivot quantitative commitments drop Matrix after dimension can characterize the information of initial data 99%；

Step 2.2：The pivot of selection is mapped back by RGB image space by image reconstruction technique, is compared by observing, determines drop Image after dimension can replace original image.

4. the electric melting magnesium furnace dynamic flame segmentation based on MIA and feature extracting method, feature exist as described in claim 1 In, in the step 3 using following formula by the matrix normalization with two column vectors after dimensionality reduction between [0,255]：

<mrow> <msub> <mi>s</mi> <mi>i</mi> </msub> <mo>=</mo> <mi>R</mi> <mi>o</mi> <mi>u</mi> <mi>n</mi> <mi>d</mi> <mrow> <mo>(</mo> <mfrac> <mrow> <msub> <mi>t</mi> <mi>i</mi> </msub> <mo>-</mo> <msub> <mi>t</mi> <mrow> <mi>i</mi> <mo>,</mo> <mi>min</mi> </mrow> </msub> </mrow> <mrow> <msub> <mi>t</mi> <mrow> <mi>i</mi> <mo>,</mo> <mi>m</mi> <mi>a</mi> <mi>x</mi> </mrow> </msub> <mo>-</mo> <msub> <mi>t</mi> <mrow> <mi>i</mi> <mo>,</mo> <mi>min</mi> </mrow> </msub> </mrow> </mfrac> <mo>&amp;times;</mo> <mn>255</mn> <mo>)</mo> </mrow> <mo>,</mo> <mrow> <mo>(</mo> <mi>i</mi> <mo>=</mo> <mn>1</mn> <mo>,</mo> <mn>2</mn> <mo>)</mo> </mrow> </mrow>

Wherein, s_iRepresent two column vectors after normalization, t_iRepresent two column vectors before normalization, t_i,minRepresent column vector t_iIn Minimum element, t_i,maxRepresent column vector t_iThe element of middle maximum.

5. the electric melting magnesium furnace dynamic flame segmentation based on MIA and feature extracting method, feature exist as described in claim 1 In using following formula acquisition score block diagram in the step 3：

<mfenced open = "" close = ""> <mtable> <mtr> <mtd> <mrow> <msub> <mi>TT</mi> <mrow> <mi>x</mi> <mo>,</mo> <mi>y</mi> </mrow> </msub> <mo>=</mo> <msub> <mi>&amp;Sigma;</mi> <mi>b</mi> </msub> <mn>1</mn> </mrow> </mtd> <mtd> <mrow> <mo>(</mo> <mo>&amp;ForAll;</mo> <mi>b</mi> <mo>,</mo> <mi>x</mi> <mo>=</mo> <msub> <mi>s</mi> <mrow> <mn>1</mn> <mo>,</mo> <mi>b</mi> </mrow> </msub> <mo>,</mo> <mi>y</mi> <mo>=</mo> <msub> <mi>s</mi> <mrow> <mn>2</mn> <mo>,</mo> <mi>b</mi> </mrow> </msub> <mo>,</mo> <mi>x</mi> <mo>,</mo> <mi>y</mi> <mo>=</mo> <mn>0</mn> <mo>,</mo> <mn>...</mn> <mo>,</mo> <mn>255</mn> <mo>)</mo> </mrow> </mtd> </mtr> </mtable> </mfenced>

Wherein, TT_x,yFor the number of similary pixel that coordinate in score block diagram is (x, y), b is certain row of matrix, s_1,bTable Show vectorial s₁The element of b rows, s_2,bRepresent vector s₂The element of b rows.

6. the electric melting magnesium furnace dynamic flame segmentation based on MIA and feature extracting method, feature exist as described in claim 1 In the step 5 includes：

Step 5.1：The region being labeled in score block diagram is finely adjusted, and obtains new flame segmentation figure picture, if segmentation Effect improves significantly, then retains new marked region, and definite marked region and accurate fire are obtained by repeatedly adjusting Flame segmentation figure picture；

Step 5.2：By proof method in described definite one piece of validation region of marked region surrounding markings, validation region is mapped Original RGB image is returned, if the corresponding segmentation figure image position of validation region shows the definite mark around flame image Region meets the requirement of flame segmentation.

7. the electric melting magnesium furnace dynamic flame segmentation based on MIA and feature extracting method, feature exist as described in claim 1 In the step 6 Flame luminance area size refers to the size of the gained area-of-interest after over-segmentation, original RGB image Each pixel value be projected in score block diagram, so area-of-interest size in score block diagram be mark The sum of all numerical value in region calculate flame luminance area size according to the following formula：

<mfenced open = "" close = ""> <mtable> <mtr> <mtd> <mrow> <mi>A</mi> <mo>=</mo> <munder> <mo>&amp;Sigma;</mo> <mrow> <mi>x</mi> <mo>,</mo> <mi>y</mi> </mrow> </munder> <msub> <mi>TT</mi> <mrow> <mi>x</mi> <mo>,</mo> <mi>y</mi> </mrow> </msub> </mrow> </mtd> <mtd> <mrow> <mo>(</mo> <mo>&amp;ForAll;</mo> <mo>(</mo> <mrow> <mi>x</mi> <mo>,</mo> <mi>y</mi> </mrow> <mo>)</mo> <mo>,</mo> <msub> <mi>M</mi> <mrow> <mi>x</mi> <mo>,</mo> <mi>y</mi> </mrow> </msub> <mo>=</mo> <mn>1</mn> <mo>,</mo> <mi>x</mi> <mo>,</mo> <mi>y</mi> <mo>=</mo> <mn>0</mn> <mo>,</mo> <mn>...</mn> <mo>,</mo> <mn>255</mn> <mo>)</mo> </mrow> </mtd> </mtr> </mtable> </mfenced>

Wherein, A represents flame luminance area size, TT_x,yFor of similary pixel that coordinate in score block diagram is (x, y) Number, M_i,jMarked region in=1 pair of reserved portion block diagram.

8. the electric melting magnesium furnace dynamic flame segmentation based on MIA and feature extracting method, feature exist as described in claim 1 In each coordinate of the numerical value in marked region more than 0 in the step 6 in score block diagram corresponds to original RGB figures A kind of color of picture, flame color species number be score block diagram in marked region in numerical value more than 1 coordinate number and, Flame color species number is sought according to the following formula：

<mfenced open = "" close = ""> <mtable> <mtr> <mtd> <mrow> <mi>C</mi> <mo>=</mo> <munder> <mo>&amp;Sigma;</mo> <mrow> <mi>x</mi> <mo>,</mo> <mi>y</mi> </mrow> </munder> <mrow> <mo>(</mo> <msub> <mi>TT</mi> <mrow> <mi>x</mi> <mo>,</mo> <mi>y</mi> </mrow> </msub> <mo>&gt;</mo> <mn>0</mn> <mo>)</mo> </mrow> </mrow> </mtd> <mtd> <mrow> <mo>(</mo> <mo>&amp;ForAll;</mo> <mo>(</mo> <mrow> <mi>x</mi> <mo>,</mo> <mi>y</mi> </mrow> <mo>)</mo> <mo>,</mo> <msub> <mi>M</mi> <mrow> <mi>x</mi> <mo>,</mo> <mi>y</mi> </mrow> </msub> <mo>=</mo> <mn>1</mn> <mo>,</mo> <mi>x</mi> <mo>,</mo> <mi>y</mi> <mo>=</mo> <mn>0</mn> <mo>,</mo> <mn>...</mn> <mo>,</mo> <mn>255</mn> <mo>)</mo> </mrow> </mtd> </mtr> </mtable> </mfenced>

Wherein, C represents flame color species number, TT_x,yFor the number of similary pixel that coordinate in score block diagram is (x, y), M_i,jMarked region in=1 pair of reserved portion block diagram.

9. the electric melting magnesium furnace dynamic flame segmentation based on MIA and feature extracting method, feature exist as described in claim 1 In flame region color average is calculated in the step 6 according to the following formula：

<mfenced open = "" close = ""> <mtable> <mtr> <mtd> <mrow> <msub> <mi>s</mi> <mi>f</mi> </msub> <mo>=</mo> <mfrac> <mrow> <msub> <mi>&amp;Sigma;</mi> <mrow> <mi>x</mi> <mo>,</mo> <mi>y</mi> </mrow> </msub> <msub> <mi>TT</mi> <mrow> <mi>x</mi> <mo>,</mo> <mi>y</mi> </mrow> </msub> <mi>x</mi> </mrow> <mi>A</mi> </mfrac> </mrow> </mtd> <mtd> <mrow> <mo>(</mo> <mo>&amp;ForAll;</mo> <mo>(</mo> <mrow> <mi>x</mi> <mo>,</mo> <mi>y</mi> </mrow> <mo>)</mo> <mo>,</mo> <msub> <mi>M</mi> <mrow> <mi>x</mi> <mo>,</mo> <mi>y</mi> </mrow> </msub> <mo>=</mo> <mn>1</mn> <mo>,</mo> <mi>x</mi> <mo>,</mo> <mi>y</mi> <mo>=</mo> <mn>0</mn> <mo>,</mo> <mn>...</mn> <mo>,</mo> <mn>255</mn> <mo>)</mo> </mrow> </mtd> </mtr> </mtable> </mfenced>

Wherein, s_fRepresent flame region color average, TT_x,yFor similary pixel that coordinate in score block diagram is (x, y) Number, A represent flame luminance area size；

Entire image color average is calculated according to the following formula：

<mfenced open = "" close = ""> <mtable> <mtr> <mtd> <mrow> <msub> <mi>s</mi> <mi>m</mi> </msub> <mo>=</mo> <mfrac> <mrow> <msub> <mo>&amp;Sigma;</mo> <mrow> <mi>x</mi> <mo>,</mo> <mi>y</mi> </mrow> </msub> <msub> <mi>TT</mi> <mrow> <mi>x</mi> <mo>,</mo> <mi>y</mi> </mrow> </msub> <mi>x</mi> </mrow> <mrow> <mi>m</mi> <mo>&amp;times;</mo> <mi>n</mi> </mrow> </mfrac> </mrow> </mtd> <mtd> <mrow> <mo>(</mo> <mo>&amp;ForAll;</mo> <mo>(</mo> <mrow> <mi>x</mi> <mo>,</mo> <mi>y</mi> </mrow> <mo>)</mo> <mo>,</mo> <msub> <mi>M</mi> <mrow> <mi>x</mi> <mo>,</mo> <mi>y</mi> </mrow> </msub> <mo>=</mo> <mn>1</mn> <mo>,</mo> <mi>x</mi> <mo>,</mo> <mi>y</mi> <mo>=</mo> <mn>0</mn> <mo>,</mo> <mn>...</mn> <mo>,</mo> <mn>255</mn> <mo>)</mo> </mrow> </mtd> </mtr> </mtable> </mfenced>

Wherein, s_mFor entire image color average, TT_x,yFor of similary pixel that coordinate in score block diagram is (x, y) Number, m × n are the size of entire image.

10. the electric melting magnesium furnace dynamic flame segmentation based on MIA and feature extracting method, feature exist as described in claim 1 In flame brightness value is calculated in the step 6 is specially：

First, score block diagram is reduced by RGB image according to following equation：

<mrow> <msub> <mrow> <mo>&amp;lsqb;</mo> <mi>R</mi> <mo>,</mo> <mi>G</mi> <mo>,</mo> <mi>B</mi> <mo>&amp;rsqb;</mo> </mrow> <mrow> <mi>x</mi> <mi>y</mi> </mrow> </msub> <mo>=</mo> <msub> <mi>t</mi> <mn>1</mn> </msub> <mrow> <mo>(</mo> <mi>x</mi> <mo>)</mo> </mrow> <msubsup> <mi>p</mi> <mn>1</mn> <mi>T</mi> </msubsup> <mo>+</mo> <msub> <mi>t</mi> <mn>2</mn> </msub> <mrow> <mo>(</mo> <mi>y</mi> <mo>)</mo> </mrow> <msubsup> <mi>p</mi> <mn>2</mn> <mi>T</mi> </msubsup> <mo>,</mo> <mrow> <mo>(</mo> <mi>x</mi> <mo>,</mo> <mi>y</mi> <mo>=</mo> <mn>0</mn> <mo>,</mo> <mn>...255</mn> <mo>)</mo> </mrow> </mrow>

<mrow> <msub> <mi>t</mi> <mn>1</mn> </msub> <mrow> <mo>(</mo> <mi>x</mi> <mo>)</mo> </mrow> <mo>=</mo> <mfrac> <mrow> <mi>x</mi> <mo>&amp;times;</mo> <mrow> <mo>(</mo> <msub> <mi>t</mi> <mrow> <mn>1</mn> <mo>,</mo> <mi>m</mi> <mi>a</mi> <mi>x</mi> </mrow> </msub> <mo>-</mo> <msub> <mi>t</mi> <mrow> <mn>1</mn> <mo>,</mo> <mi>m</mi> <mi>i</mi> <mi>n</mi> </mrow> </msub> <mo>)</mo> </mrow> </mrow> <mn>255</mn> </mfrac> <mo>+</mo> <msub> <mi>t</mi> <mrow> <mn>1</mn> <mo>,</mo> <mi>m</mi> <mi>i</mi> <mi>n</mi> </mrow> </msub> </mrow>

<mrow> <msub> <mi>t</mi> <mn>2</mn> </msub> <mrow> <mo>(</mo> <mi>y</mi> <mo>)</mo> </mrow> <mo>=</mo> <mfrac> <mrow> <mi>y</mi> <mo>&amp;times;</mo> <mrow> <mo>(</mo> <msub> <mi>t</mi> <mrow> <mn>2</mn> <mo>,</mo> <mi>m</mi> <mi>a</mi> <mi>x</mi> </mrow> </msub> <mo>-</mo> <msub> <mi>t</mi> <mrow> <mn>2</mn> <mo>,</mo> <mi>m</mi> <mi>i</mi> <mi>n</mi> </mrow> </msub> <mo>)</mo> </mrow> </mrow> <mn>255</mn> </mfrac> <mo>+</mo> <msub> <mi>t</mi> <mrow> <mn>2</mn> <mo>,</mo> <mi>m</mi> <mi>i</mi> <mi>n</mi> </mrow> </msub> </mrow>

Wherein, t_1,maxRepresent column vector t₁The element of middle maximum, t_1,minRepresent column vector t₁The element of middle minimum, t_2,maxRepresent row Vectorial t₂The element of middle maximum, t_2,minRepresent column vector t₂The element of middle minimum, p₁To load vector p_aThe of=(1 ..., PC) One row, p₂To load vector p_aThe secondary series of=(1 ..., PC)；

Secondly, gray level image is calculated according to the following formula：

L_xy=[R, G, B]_xy[0.299,0.587,0.114]^T

Finally, flame brightness value is calculated according to the following formula：

<mfenced open = "" close = ""> <mtable> <mtr> <mtd> <mrow> <mi>F</mi> <mo>=</mo> <munder> <mo>&amp;Sigma;</mo> <mrow> <mi>x</mi> <mo>,</mo> <mi>y</mi> </mrow> </munder> <msub> <mi>TT</mi> <mrow> <mi>x</mi> <mi>y</mi> </mrow> </msub> <msub> <mi>L</mi> <mrow> <mi>x</mi> <mi>y</mi> </mrow> </msub> <mo>,</mo> </mrow> </mtd> <mtd> <mrow> <mo>(</mo> <mo>&amp;ForAll;</mo> <mo>(</mo> <mrow> <mi>x</mi> <mo>,</mo> <mi>y</mi> </mrow> <mo>)</mo> <mo>,</mo> <msub> <mi>M</mi> <mrow> <mi>x</mi> <mo>,</mo> <mi>y</mi> </mrow> </msub> <mo>=</mo> <mn>1</mn> <mo>,</mo> <mi>x</mi> <mo>,</mo> <mi>y</mi> <mo>=</mo> <mn>0</mn> <mo>,</mo> <mn>...</mn> <mo>,</mo> <mn>255</mn> <mo>)</mo> </mrow> </mtd> </mtr> </mtable> </mfenced>

Wherein, TT_x,yFor the number of similary pixel that coordinate in score block diagram is (x, y).