CN105405149A - Composite texture feature extraction method for flotation froth image - Google Patents

Abstract

The invention discloses a compound texture feature extraction method of a flotation foam image. Firstly, the area neighborhood set of all central pixels is obtained in the gray quantization matrix of the foam image; then, for all central pixels, a three-dimensional data table is constructed and a nested grayscale frequency table is obtained; again, an improved neighborhood is obtained Gray level correlation matrix; Finally, a new composite texture feature is obtained. This feature combines the size, texture and roughness of the foam, and has high stability and separability in reflecting the texture of the flotation foam. According to the extracted composite texture features, it is easy to distinguish the flotation foam images of different working conditions under different ore grades, and has a high recognition accuracy of working conditions. The invention is simple and effective, and has great significance for guiding the identification of foam working conditions on the mineral flotation site.

Description

A composite texture feature extraction method for flotation foam images

technical field

The invention belongs to the technical fields of image processing, pattern recognition and mineral flotation, in particular to a compound texture feature extraction method of flotation foam images.

Background technique

The apparent characteristics of foam are a comprehensive reflection of mineral flotation conditions, and are considered to be closely related to the flotation effect. How to accurately extract the foam appearance characteristics closely related to key production indicators in the flotation process is the key to realize the identification of flotation conditions. For a long time, experienced workers on site have adjusted the working conditions by observing the state of the foam surface. Due to artificial arbitrariness and subjectivity, the flotation state is unstable and it is difficult to adjust to the optimal state, resulting in low utilization of mineral resources, resulting in Waste of resources.

In recent years, the control and optimization of the flotation process based on machine vision has become a research hotspot at home and abroad, among which the recognition of operating conditions based on machine vision is one of the main contents of the research. The apparent characteristics of foam are closely related to the flotation conditions and are indicators of flotation performance. The foam appearance characteristics commonly used in the identification of flotation intelligent working conditions mainly include foam color, size, texture, etc. Among them, the texture feature is one of the most important characteristics of the foam appearance, which integrates the characteristics of the size and shape of the foam, and has It is widely used in the identification of flotation conditions due to its characteristics of not being affected by ambient light. Neighborhood gray level correlation matrix is a commonly used texture feature extraction method based on statistics. It has the characteristics of rotation invariance, no influence with illumination changes, and fast calculation speed. The secondary statistical features such as roughness and fineness based on the neighborhood gray correlation matrix can better reflect the texture characteristics of the foam surface. Since the traditional neighborhood gray level correlation matrix method only considers the number of the same or similar gray values of the central pixel and its neighboring pixels, but does not consider the number of differences between them and the size of the difference, therefore, lost A large number of spatial distribution attributes that can reflect the difference of image pixels are difficult to fully and accurately reflect the texture information of the foam surface. In addition, under different ore grade conditions, the froth image will be quite different, and the texture feature extraction method of the traditional neighborhood gray correlation matrix without considering the influence of ore grade is difficult to adapt to the accurate identification of froth flotation conditions demand.

Therefore, it is necessary to design a composite texture feature extraction method for flotation foam images in order to more comprehensively capture the foam surface texture information under different ore grade conditions.

Contents of the invention

The technical problem to be solved by the present invention is to provide a method for extracting composite texture features of flotation froth images, which can more comprehensively reflect the differences in texture features of froth images under different ore grades, and has good feature validity and reliability. Separation, in order to meet the needs of subsequent identification accuracy of working conditions.

A method for extracting composite texture features of flotation foam images, comprising the following steps:

Step 1: Read the RGB foam image according to the foam video obtained at the foam flotation site, grayscale the RGB image to obtain a grayscale image matrix; quantify the grayscale image to obtain a quantization matrix; obtain all centers in the quantization matrix A set of area neighborhoods of pixels;

Step 2: Calculate the absolute difference between the pixel point in the neighborhood of each face and the gray value of the central pixel point; count the number of each absolute difference; for all the central pixel points in the foam image, build a three-dimensional data table, in the three-dimensional data table The elements are the gray value of the central pixel, the number of absolute differences, and the absolute difference;

Step 3: Construct a nested grayscale frequency table based on a three-dimensional data table;

Step 4: Obtain an improved neighborhood gray-scale correlation matrix according to the constructed nested gray-scale frequency table;

Step 5: According to the obtained improved neighborhood gray correlation matrix, obtain a new secondary statistical feature—composite texture.

The specific steps of obtaining the surface neighborhood sets of all central pixels of the quantized foam image matrix in the step 1 are as follows:

Step 1 Read the RGB foam image according to the foam video obtained at the foam flotation site, convert the RGB image to grayscale, and obtain the grayscale image matrix A(x,y):

A _k×m (x,y)＝0.290×R(x,y)+0.587×G(x,y)+0.114×B(x,y)

In the formula, k×m is the resolution of the foam grayscale image, (x, y) represents the coordinates of any pixel in the foam grayscale image, x=0,1,...,k-1, y=0, 1,...,m-1; R(x,y), G(x,y), and B(x,y) are the R, G, and B matrices of the foam image, respectively.

Step 2 Quantize the grayscale image matrix A(x,y) to obtain the quantization matrix M(x,y):

In the formula, M _g is the quantization series of the foam image, is rounded down.

Step 3 In the quantization matrix M(x, y), take all the central pixel points (x _c , y _c ) as the center and D as the radius to obtain a set of surface neighborhoods:

V _D (x _c ,y _c )＝{(u,v)|(u,v)∈M _k×m ,0<ρ((x _c ,y _c ),(u,v))≤D}

where x _c ∈ D, D+1,..., k-1-D, y _c ∈ D, D+1,..., m-1-D; {(u, v)|( )} means satisfying the given The set of points (u, v) under the condition (·), ρ((x _c , y _c ), (u, v)) represents the pixel point (u, v) and the center pixel point (x _c ,y _c ) distance between:

ρ((x _c ,y _c ),(u,v))＝max(|x _c -u|,|y _c -v|)

In the formula, max(|x _c -u|, |y _c -v|) represents the maximum value among the pixel distances |x _c -u|, |y _c -v|.

The specific steps of constructing the three-dimensional data table of the central pixel point of the foam image in the step two are as follows:

Step 1 Calculate the absolute difference i between the gray value of the pixel (u, v) in the neighborhood of each face and the gray value of the central pixel (x _c , y _c ) in the face neighborhood set V _D (x _c , y _c ):

i=|f(u,v)-f(x _c ,y _c )|=|f(u,v)-g _c |

In the formula, f(u,v) and f(x _c ,y _c )=g _c represent the gray value of the pixel point (u,v) and the central pixel point (x _c ,y _c ) respectively, g _c =0, 1, . . . , M _g −1, i=0, 1, . . . , M _g −1.

Step 2: Count the number of absolute difference i

${\mathrm{<span\; class="notranslate">\; the\; s</span>}}_{{\mathrm{<span\; class="notranslate">\; g</span>}}_{\mathrm{<span\; class="notranslate">\; c</span>}}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; =</span><span\; class="notranslate">\; \#</span><span\; class="notranslate">\; \{</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; u</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; v</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; |</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; u</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; v</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&Element;</span>{\mathrm{<span\; class="notranslate">\; V</span>}}_{\mathrm{<span\; class="notranslate">\; D.</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; c</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; the\; y</span>}}_{\mathrm{<span\; class="notranslate">\; c</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; \}</span>$

In the formula, # represents the number of points (u, v) in the statistical set {(u, v)|(·)} that meet the given condition (·).

Step 3: Construct a three-dimensional data table of each central pixel point of the foam image;

Define the three-dimensional data table f _D (x _c ,y _c ,i) of a certain central pixel point (x _c ,y _c ) as follows:

${\mathrm{<span\; class="notranslate">\; f</span>}}_{\mathrm{<span\; class="notranslate">\; D.</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; c</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; the\; y</span>}}_{\mathrm{<span\; class="notranslate">\; c</span>}}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; t</span>}\mathrm{<span\; class="notranslate">\; a</span>}\mathrm{<span\; class="notranslate">\; b</span>}\mathrm{<span\; class="notranslate">\; l</span>}\mathrm{<span\; class="notranslate">\; e</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; g</span>}}_{\mathrm{<span\; class="notranslate">\; c</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; the\; s</span>}}_{{\mathrm{<span\; class="notranslate">\; g</span>}}_{\mathrm{<span\; class="notranslate">\; c</span>}}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; )</span>$

In the formula The elements in a table are g _c , i, a three-dimensional data table with a size of 1×1×M _g .

Construct a corresponding three-dimensional data table for all central pixel points in the foam image.

In the step 3, the method of constructing a nested grayscale frequency table based on a three-dimensional data table is as follows:

Define the nested grayscale frequency table F _D (x′ _c ,y′ _c ,i) as follows:

F _D (x _c ',y _c ',i)＝table(f _D (x _c ,y _c ,i))

where x _c '=x _c -D, x _c '∈0,1,...,k-1-2D, y _c '=y _c -D,y _c '∈0,1,...,m-1- 2D. The element of a certain cell position (x _c ', y _c ') in the nested grayscale frequency table F _D (x′ _c ,y′ _c , i) is

Embed each three-dimensional data table f _D (x _c , y _c , i) constructed in step 2 for all central pixel points (x _c , y _c ) into the corresponding cell positions in the nested grayscale frequency table to construct A nested grayscale frequency table of size (k-2D)×(m-2D)×M _g .

In the step four, an improved neighborhood gray correlation matrix is obtained:

Define the improved neighborhood correlation matrix Q _D (g, i) as follows:

where g=0,1,...,M _g -1, i=0,1,...,M _g -1, q _D (g,i) is the matrix Q _D (g,i) at position (g,i ) at the element.

In the nested grayscale frequency table F _D (x′ _c ,y′ _c , i) obtained in step 3, the absolute difference between the gray value of the pixel in the neighborhood of the statistical surface and the gray value of the central pixel g _c is i All the time The sum of q _D (g _c ,i):

In the formula, Σ represents the sum of all s _{g, I} in the set {s _{g, i} |( )} satisfying the given condition ( ), && represents the logical AND operation.

Taking q _D (g _c , i) as the element of the matrix Q _D (g, i) at the position ( _g = _{g c} , i), an improved neighborhood gray level correlation matrix Q _D ( g, i).

A new composite texture feature is calculated in the step five:

According to the improved neighborhood gray-level correlation matrix, a new secondary statistical feature, that is, the composite texture feature CT is defined as:

$\mathrm{<span\; class="notranslate">\; C</span>}\mathrm{<span\; class="notranslate">\; T</span>}<span\; class="notranslate">\; =</span>\frac{\underset{\mathrm{<span\; class="notranslate">\; g</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 0</span>}}{\overset{{\mathrm{<span\; class="notranslate">\; m</span>}}_{\mathrm{<span\; class="notranslate">\; g</span>}}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}}{\mathrm{<span\; class="notranslate">\; g</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; (</span>\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 0</span>}}{\overset{{\mathrm{<span\; class="notranslate">\; m</span>}}_{\mathrm{<span\; class="notranslate">\; g</span>}}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}}{\mathrm{<span\; class="notranslate">\; Q</span>}}_{\mathrm{<span\; class="notranslate">\; D.</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; g</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; )</span>}{\underset{\mathrm{<span\; class="notranslate">\; g</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 0</span>}}{\overset{{\mathrm{<span\; class="notranslate">\; m</span>}}_{\mathrm{<span\; class="notranslate">\; g</span>}}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}}\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 0</span>}}{\overset{{\mathrm{<span\; class="notranslate">\; m</span>}}_{\mathrm{<span\; class="notranslate">\; g</span>}}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}}<span\; class="notranslate">\; \&lsqb;</span>{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}{\mathrm{<span\; class="notranslate">\; Q</span>}}_{\mathrm{<span\; class="notranslate">\; D.</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; g</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&rsqb;</span>}<span\; class="notranslate">\; .</span>$

Beneficial effect

The present invention provides a composite texture feature extraction method of flotation foam images. Compared with the traditional neighborhood gray level correlation matrix, it not only counts the same number of neighborhood pixels and center pixels, but also counts both The different sizes and corresponding numbers can more comprehensively reflect the texture information of the foam surface. At the same time, on the basis of the improved neighborhood gray level correlation matrix, a new secondary statistical feature—composite texture feature is defined, which integrates the size, texture and roughness of the foam, and can describe the texture of the foam image more comprehensively. Structural properties of features. Experiments prove that the new composite texture feature extracted by the present invention has better feature validity and separability, and can reflect the changing relationship between concentrate grade and foam texture feature under different ore grade conditions. Due to the fast calculation speed, it can be effectively used for online identification of mineral flotation conditions.

Description of drawings

Figure 1 shows the foam images of five different working conditions, in which Figure a is excellent grade, Figure b is good grade, Figure c is medium grade, Figure d is deviation grade, and Figure e is extremely low grade;

Fig. 2 is a schematic diagram (D=1) of the plane neighborhood of the central pixel point (x _c , y _c );

Fig. 3 is a simple description diagram (D=1, Mg=8) of improving the process of solving the neighborhood gray correlation matrix; wherein, Fig. (a) is the gray quantization matrix A _k×m ( _x , y) of the image; Figure (b) is a three-dimensional data table f _D (x _c ,y _c ,i); Figure (c) is a nested grayscale frequency table F _D (x _c ',y _c ',i); Figure (d) is Improved neighborhood gray level correlation matrix Q _D (g,i);

Fig. 4 is the interval distribution map of composite texture features under each concentrate grade grade;

Fig. 5 is the distribution map of composite texture features for 5 different working conditions under the condition that the ore grade is 2.

detailed description

The present invention will be described in further detail below in conjunction with the accompanying drawings and specific implementation examples.

The specific embodiment of the present invention is described below in conjunction with accompanying drawing, there are 5 kinds of foams of different normal working conditions in a certain antimony roughing site, and they are divided according to the grade of concentrate ore, and these 5 kinds of grades are respectively: excellent class (grade value ≥40.5), good (grade value: [37.5-40.5)), medium (grade value: [33-37.5)), deviation (grade value: [27.5-33)), very poor (grade value <27.5). The foam images of five different working conditions are shown in Fig. 1. At the same time, the ore grade can be divided into 5 grades according to the experience of the on-site antimony roughing workers, which are: high (grade value ≥ 2.14), high (grade value: [1.75.-2.14)), medium (grade value: [ 1.33-1.75)), lower (grade value: [1.08-1.33)), low (grade value <1.08), as shown in the numbers in the lower left corner of the foam image in Figure 1, 1, 2, ..., 5 respectively represent The ore grade is high, high, ..., low. From the foam images, it can be found that under different ore grades, even for the same concentrate grade, the foam images of antimony roughing still show obvious differences.

The first step is to read the RGB foam image according to the antimony rough selection foam video collected at the gold-antimony flotation site, convert the RGB foam image into a grayscale matrix, and perform 16-level quantization on the grayscale matrix, that is, grayscale The matrices with grayscale values from 0 to 255 are quantized to matrices with grayscale values from 0 to 15. Finally, obtain the surface neighborhood set of all central pixels of the quantized foam image matrix, and the specific steps are as follows:

Step 1 Read the RGB foam image according to the foam video obtained at the foam flotation site, convert the RGB image to grayscale, and obtain the grayscale image matrix A(x,y):

A _k×m (x,y)=0.290×R(x,y)+0.587×G(x,y)+0.114×B(x,y) Formula 1

In the formula, k×m (k=600, m=800) is the resolution of the foam grayscale image, (x, y) represents the coordinates of any pixel in the foam grayscale image, x=0,1,... ,k-1, y=0,1,...,m-1; R(x,y), G(x,y), B(x,y) are the R, G, B matrix of the foam image respectively .

Step 2 Perform 16-level quantization on the grayscale image matrix A(x,y) to obtain the quantization matrix M(x,y):

Formula 2

In the formula, _Mg ( _Mg =16) is the quantization series of the foam image, is rounded down.

Step 3 In the quantization matrix M(x, y), the face neighborhood set centered on all the central pixel points (x _c , y _c ) and radius D=1 is:

V _D (x _c ,y _c )={(u,v)|(u,v)∈M _k×m ,0<ρ((x _c ,y _c ),(u,v))≤D} formula 3

where x _c ∈ D, D+1,..., k-1-D, y _c ∈ D, D+1,..., m-1-D; {(u, v)|( )} means satisfying the given The set of points (u, v) under the condition (·), ρ((x _c , y _c ), (u, v)) represents the pixel point (u, v) and the center pixel point (x _c ,y _c ) distance between:

ρ((x _c ,y _c ),(u,v))=max(|x _c -u|,|y _c -v|) Formula 4

In the formula, max(|x _c -u|, |y _c -v|) represents the maximum value among the pixel distances |x _c -u|, |y _c -v| (as shown in Figure 2).

The second step is to construct a three-dimensional data table of the central pixel of the rough selected image, and the specific steps are as follows:

Step 1 Calculate the absolute difference i between the gray value of the pixel (u, v) in the neighborhood of each face and the gray value of the central pixel (x _c , y _c ) in the face neighborhood set V _D (x _c , y _c ):

i=|f(u,v)-f(x _c ,y _c )|=|f(u,v)-g _c |Formula 5

In the formula, f(u,v) and f(x _c ,y _c )=g _c represent the gray value of the pixel point (u,v) and the central pixel point (x _c ,y _c ) respectively, g _c =0, 1, . . . , M _g −1, i=0, 1, . . . , M _g −1.

Step 2 count the number of absolute difference i

${\mathrm{the\; s}}_{g_c,i}=\#\left\{(u,v)\right|(u,v)\&Element;V_{\mathrm{D.}}(x_c,{\mathrm{the\; y}}_c),i\}$ Formula 6

In the formula, # represents the number of points (u, v) in the statistical set {(u, v)|(·)} that meet the given condition (·).

Step 3: Construct a three-dimensional data table of each central pixel point of the foam image;

Define the three-dimensional data table f _D (x _c ,y _c ,i) of a certain central pixel point (x _c ,y _c ) as follows:

$f_{\mathrm{D.}}(x_c,{\mathrm{the\; y}}_c,i)=table(g_c,{\mathrm{the\; s}}_{g_c,i},i)$ Formula 7

In the formula The elements in a table are g _c , i, a three-dimensional data table with a size of 1×1×15, i takes 0,1,…,15 in turn.

Construct a corresponding three-dimensional data table for all central pixel points in the foam image. The surface neighborhood set of the central pixel point is shown in Figure 3(a), and the corresponding three-dimensional data table is shown in Figure 3(b).

The third step is to construct the nested grayscale frequency table of antimony rough selection image;

Define the nested grayscale frequency table F _D (x′ _c ,y′ _c ,i) as follows:

F _D (x _c ',y _c ',i)＝table(f _D (x _c ,y _c ,i))Formula 8

where x _c '=x _c -D, x _c '∈0,1,...,k-1-2D, y _c '=y _c -D,y _c '∈0,1,...,m-1- 2D. The element of a certain cell position (x _c ', y _c ') in the nested grayscale frequency table F _D (x′ _c ,y′ _c , i) is

Embed each three-dimensional data table f _D (x _c , y _c , i) constructed for all central pixel points (x _c , y _c ) in the second step into the corresponding cell position in the nested grayscale frequency table, and construct Generate a nested grayscale frequency table with a size of 588×788×16. As shown in Figure 3(c).

The fourth step is to obtain an improved neighborhood gray correlation matrix:

Define the improved neighborhood correlation matrix Q _D (g, i) as follows:

Formula 9

where g=0,1,...,M _g -1, i=0,1,...,M _g -1, q _D (g,i) is the matrix Q _D (g,i) at position (g,i ) at the element.

In the nested grayscale frequency table F _D (x′ _c ,y′ _c , i) obtained in the third step, the absolute difference between the gray value of the pixel in the neighborhood of the statistical surface and the gray value of the center pixel g _c is i time all The sum of q _D (g _c ,i):

Formula 10

In the formula, Σ represents the sum of all s _{g, I} in the set {s _{g, i} |( )} satisfying the given condition ( ), && represents the logical AND operation.

Take q _D (g _c , i) as the element of matrix Q _D (g, i) at position (g=g _c , i), and obtain a 16×16 improved neighborhood gray level correlation matrix Q _D (g, i). As shown in Figure 3(d).

The fifth step is to calculate a new composite texture feature:

According to the improved neighborhood gray-level correlation matrix, a new secondary statistical feature, that is, the composite texture feature CT is defined as:

$CT=\frac{\underset{g=0}{\overset{m_g-1}{\&Sigma;}}{g}^2\left(\underset{i=0}{\overset{m_g-1}{\&Sigma;}}{Q}_{\mathrm{D.}}\right(g,i\left)\right)}{\underset{g=0}{\overset{m_g-1}{\&Sigma;}}\underset{i=0}{\overset{m_g-1}{\&Sigma;}}\&lsqb;{(i+1)}^2{Q}_{\mathrm{D.}}(g,i)\&rsqb;}$ Formula 11

The numerator of the composite texture feature reflects the sum average feature. When the concentrate grade is higher, that is, the antimony content is higher, the numerator is larger; the denominator reflects the thickness of the image texture. The higher the concentrate grade and the finer the image texture, the larger the numerator. The larger the composite texture feature value is. CT combines thickness features and average features, which can well reflect information such as the size, texture, and roughness of the foam surface.

Accompanying drawing 4 has provided the interval distribution of composite texture characteristic under each concentrate grade grade, thus can be seen, when being in different concentrate grade grades, composite texture characteristic value is distributed in different intervals, and when concentrate grade grade is When it is higher, the characteristic value of the composite texture is the largest. Therefore, the composite texture characteristic can reflect the change of the concentrate grade and can be used as an indicator of the concentrate grade.

Accompanying drawing 5 shows the distribution map of the extracted composite texture features for 5 different working conditions when the ore grade is 2 (higher). It can be seen from Figure 5 that under the same ore grade, the composite texture feature can clearly distinguish different working conditions. Therefore, the composite texture feature has good mode separability, which can be used for working condition identification and provide guidance for the control and optimization of antimony roughing process.

Neighborhood gray-level correlation matrix has the advantages of invariance to space rotation, no influence with light and fast calculation speed, which just fits the characteristics of complex flotation site environment and huge amount of flotation data, making it a commonly used method in froth flotation. texture feature extraction method. However, the traditional neighborhood gray-level correlation matrix only considers the number of neighboring pixels that are the same as or similar to the central pixel, but does not consider the number of differences between them and the size of the difference. Therefore, a large number of pixels that can reflect the image The spatial distribution attribute of pixel difference is difficult to fully reflect the texture information of the foam surface.

Therefore, the present invention aims at the problem that the traditional neighborhood gray correlation matrix ignores the difference between the neighborhood pixels and the central pixel, and proposes a composite texture feature extraction method for flotation foam images, which not only counts The number of domain pixels is the same as that of the center pixel, and the difference between the two and the corresponding number are counted. On the basis of the improved neighborhood gray correlation matrix, a new texture feature that combines the size, texture and roughness of the foam—composite texture feature is proposed to more fully express the foam surface under different ore grade conditions Texture information makes it easier to accurately identify froth flotation conditions.

Claims (6)

1. a composite texture feature extraction method of flotation foam image, is characterized in that, comprises the following steps: Step 1: Read the RGB foam image according to the foam video obtained at the mineral froth flotation site, grayscale the RGB foam image to obtain a grayscale image matrix; quantify the grayscale image to obtain a quantization matrix; obtain the quantization matrix A collection of surface neighborhoods of all central pixels; Step 2: Calculate the absolute difference between the pixel points in the neighborhood of each surface and the gray value of the central pixel point; count the number of each absolute difference; construct a three-dimensional data table for all central pixel points in the image, and the elements in the three-dimensional data table Respectively, the gray value of the center pixel, the number of absolute differences, and the absolute difference; Step 3: Construct a nested grayscale frequency table based on a three-dimensional data table; Step 4: Obtain an improved neighborhood gray-scale correlation matrix according to the constructed nested gray-scale frequency table; Step 5: Calculate a new composite texture feature based on the obtained improved neighborhood gray level correlation matrix. 2. the composite texture feature extraction method of flotation froth image according to claim 1, is characterized in that, described step one concrete steps are as follows: Step 1: Read the RGB foam image according to the foam video obtained at the foam flotation site, convert the RGB image to grayscale, and obtain the grayscale image matrix A(x,y): A _k×m (x,y)＝0.290×R(x,y)+0.587×G(x,y)+0.114×B(x,y) In the formula, k×m is the resolution of the foam grayscale image, (x, y) represents the coordinates of any pixel in the foam grayscale image, x=0,1,...,k-1, y=0, 1,...,m-1; R(x,y), G(x,y), and B(x,y) are the R, G, and B matrices of the foam image, respectively; Step 2: Quantize the grayscale image matrix A(x,y) to obtain the quantization matrix M(x,y): In the formula, M _g is the quantization series of the foam image, is rounded down; Step 3: In the quantization matrix M(x, y), take all the central pixel points (x _c , y _c ) as the center and D as the radius to obtain a set of surface neighborhoods: V _D (x _c ,y _c )＝{(u,v)|(u,v)∈M _k×m ,0<ρ((x _c ,y _c ),(u,v))≤D} Among them, x _c ∈ D, D+1,..., k-1-D; y _c ∈ D, D+1,..., m-1-D, {(u, v)|( )} means that the given ρ((x _c ,y _c ),(u,v)) means the pixel point (u,v) in the neighborhood and the central pixel point (x _c ,y _c ) distance between: ρ((x _c ,y _c ),(u,v))＝max(|x _c -u|,|y _c -v|) In the formula, max(|x _c -u|, |y _c -v|) represents the maximum value among the pixel distances |x _c -u|, |y _c -v|. 3. the composite texture feature extraction method of flotation froth image according to claim 2, is characterized in that, described step 2 specific steps are as follows: Step 1: Calculate the absolute difference i between the gray value of the pixel point (u, v) and the gray value of the central pixel point (x _c , y _c ) in each area of the face neighborhood set V _D (x _c , y _c ) : i=|f(u,v)-f(x _c ,y _c )|=|f(u,v)-g _c | In the formula, f(u,v) and f(x _c ,y _c )=g _c represent the gray value of the pixel point (u,v) and the central pixel point (x _c ,y _c ) respectively, g _c =0, 1,..., _Mg -1, i=0,1,..., _Mg -1; Step 2: Count the number of absolute difference i ${\mathrm{<span\; class="notranslate">\; the\; s</span>}}_{{\mathrm{<span\; class="notranslate">\; g</span>}}_{\mathrm{<span\; class="notranslate">\; c</span>}}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; =</span><span\; class="notranslate">\; \#</span><span\; class="notranslate">\; \{</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; u</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; v</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; |</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; u</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; v</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&Element;</span>{\mathrm{<span\; class="notranslate">\; V</span>}}_{\mathrm{<span\; class="notranslate">\; D.</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; c</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; the\; y</span>}}_{\mathrm{<span\; class="notranslate">\; c</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; \}</span>$ In the formula, # indicates the number of points (u, v) in the statistical set {(u, v)|( )} satisfying the given condition ( ); Step 3: Construct a three-dimensional data table of each central pixel point of the foam image; Define the three-dimensional data table f _D (x _c ,y _c ,i) of a certain central pixel point (x _c ,y _c ) as follows: ${\mathrm{<span\; class="notranslate">\; f</span>}}_{\mathrm{<span\; class="notranslate">\; D.</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; c</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; the\; y</span>}}_{\mathrm{<span\; class="notranslate">\; c</span>}}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; t</span>}\mathrm{<span\; class="notranslate">\; a</span>}\mathrm{<span\; class="notranslate">\; b</span>}\mathrm{<span\; class="notranslate">\; l</span>}\mathrm{<span\; class="notranslate">\; e</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; g</span>}}_{\mathrm{<span\; class="notranslate">\; c</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; the\; s</span>}}_{{\mathrm{<span\; class="notranslate">\; g</span>}}_{\mathrm{<span\; class="notranslate">\; c</span>}}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; )</span>$ In the formula The elements in a table are g _c , i, a three-dimensional data table whose size is 1×1×M _g ; Construct a corresponding three-dimensional data table for all central pixel points in the foam image. 4. the composite texture feature extraction method of flotation froth image according to claim 3, is characterized in that, described step 3 builds the nested gray scale frequency table based on three-dimensional data table and is specially: Define the nested grayscale frequency table F _D (x′ _c ,y′ _c ,i) as follows: F _D (x _c ',y _c ',i)＝table(f _D (x _c ,y _c ,i)) where x _c '=x _c -D, x _c '∈0,1,...,k-1-2D; y _c '=y _c -D,y _c '∈0,1,...,m-1- 2D, the element of a certain cell position (x _c ', y _c ') in the nested grayscale frequency table F _D (x′ _c ,y′ _c , i) is Embed each three-dimensional data table f _D (x _c , y _c , i) constructed in step 2 for all central pixel points (x _c , y _c ) into the corresponding cell positions in the nested grayscale frequency table to construct A nested grayscale frequency table of size (k-2D)×(m-2D)×M _g . 5. the composite texture feature extraction method of flotation froth image according to claim 4, is characterized in that, described step 4 obtains a kind of improved neighborhood gray scale correlation matrix and is specifically: Define the improved neighborhood correlation matrix Q _D (g, i) as follows: where g=0,1,...,M _g -1, i=0,1,...,M _g -1, q _D (g,i) is the matrix Q _D (g,i) at position (g,i ) at the element; In the nested grayscale frequency table F _D (x′ _c ,y′ _c , i) obtained in step 3, when the absolute difference between the gray value of the neighborhood pixel and the central pixel gray value g _c is i all The sum of q _D (g _c ,i): In the formula, Σ represents the sum of all s _{g, I} in the set {s _{g, i} |( )} satisfying the given condition ( ), && represents the logical AND operation; Taking q _D (g _c , i) as the element of the matrix Q _D (g, i) at the position ( _g = _{g c} , i), an improved neighborhood gray level correlation matrix Q _D ( g, i). 6. the composite texture feature extraction method of flotation froth image according to claim 5, is characterized in that, described step 5 calculates a kind of new composite texture feature to be specially: According to the improved neighborhood gray-level correlation matrix, a new secondary statistical feature, that is, the composite texture feature CT is defined as: $\mathrm{<span\; class="notranslate">\; C</span>}\mathrm{<span\; class="notranslate">\; T</span>}<span\; class="notranslate">\; =</span>\frac{\underset{\mathrm{<span\; class="notranslate">\; g</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 0</span>}}{\overset{{\mathrm{<span\; class="notranslate">\; m</span>}}_{\mathrm{<span\; class="notranslate">\; g</span>}}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}}{\mathrm{<span\; class="notranslate">\; g</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; (</span>\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 0</span>}}{\overset{{\mathrm{<span\; class="notranslate">\; m</span>}}_{\mathrm{<span\; class="notranslate">\; g</span>}}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}}{\mathrm{<span\; class="notranslate">\; Q</span>}}_{\mathrm{<span\; class="notranslate">\; D.</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; g</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; )</span>}{\underset{\mathrm{<span\; class="notranslate">\; g</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 0</span>}}{\overset{{\mathrm{<span\; class="notranslate">\; m</span>}}_{\mathrm{<span\; class="notranslate">\; g</span>}}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}}\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 0</span>}}{\overset{{\mathrm{<span\; class="notranslate">\; m</span>}}_{\mathrm{<span\; class="notranslate">\; g</span>}}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}}<span\; class="notranslate">\; \&lsqb;</span>{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}{\mathrm{<span\; class="notranslate">\; Q</span>}}_{\mathrm{<span\; class="notranslate">\; D.</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; g</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&rsqb;</span>}<span\; class="notranslate">\; .</span>$