A metal smelting equipment detection method based on image

By decomposing and screening the grayscale images of metal smelting equipment and determining the optimal threshold in combination with outlier analysis, the problem of inaccurate determination of singular value thresholds in the prior art is solved, and detection accuracy and safety are improved.

CN118691602BActive Publication Date: 2024-12-03BAOJI TOWIN RARE METALS CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411155289.4
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-08-22
Publication Date
2024-12-03
Estimated Expiration
2044-08-22

AI Technical Summary

Technical Problem

The prior art cannot accurately determine the singular value threshold in metal smelting equipment detection, resulting in poor noise reduction effect and affecting detection accuracy and safety.

Method used

By decomposing the grayscale image at the edge of the melting furnace singular value, reconstructing multiple images, and filtering through priority and noise interference, combining outlier analysis to determine the best singular value threshold, reduce noise interference and improve detection accuracy .

Benefits of technology

It improves the accuracy of metal smelting equipment detection, reduces the impact of noise on the damaged area, and enhances the accuracy and safety of detection.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118691602B_ABST
    Figure CN118691602B_ABST
Patent Text Reader

Abstract

The present invention relates to the field of equipment detection, and in particular to an image-based metal smelting equipment detection method, the method comprising: graying the edge image in the smelting furnace to obtain a gray image, performing singular value decomposition on the gray image to obtain multiple singular values, and superimposing the singular values ​​in order from small to large to obtain multiple reconstructed images; dividing the reconstructed image into blocks to obtain multiple block areas of the reconstructed image; binarizing the reconstructed image to obtain a binary image, and calculating the preferred degree of each binary image; calculating the noise interference degree of the block area in each remaining reconstructed image, and the proportion of the number of block areas whose noise interference degree is less than the interference degree threshold; calculating the damage degree of the pixel points in the remaining reconstructed image, and performing outlier analysis on the pixel points whose damage degree in the reconstructed image is greater than the damage threshold. The present invention improves the denoising effect of the image and improves the detection accuracy of the smelting furnace.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of equipment detection, and in particular to an image-based metal smelting equipment detection method. Background Art

[0002] Metal smelting equipment is required in the smelting process of metals. Metal smelting equipment is usually used to extract metals from their ores or waste and refine them into a form that can be used for manufacturing. Metal smelting equipment usually includes electric arc furnaces, induction heating furnaces, high-frequency melting furnaces, crucible furnaces and smelting furnaces. During the manufacturing and use of metal smelting equipment, it is necessary to detect defects inside the smelting equipment to ensure the safety of production equipment and operators. When performing defect detection inside the smelting furnace, due to the influence of noise and surface roughness, the image needs to be denoised, and the denoising process usually uses the singular value decomposition method, which decomposes the image into multiple singular values, and then selects an appropriate number of singular values ​​for reconstruction to obtain a denoised image.

[0003] A Chinese patent document with announcement number CN111582205B discloses a denoising method based on a multi-resolution singular value decomposition model, comprising: constructing a noisy signal y(n) into a matrix A0, decomposing the matrix A0 using an L-partition method to obtain L singular values, where L≥3, and the components corresponding to the L singular values ​​include an approximate signal A and a detail signal D; setting the singular values ​​corresponding to the detail signal D to zero, and reconstructing the singular values ​​corresponding to the approximate signal A into a Hankel matrix for decomposition at the next level, and so on, to perform noise separation.

[0004] However, in the above scheme, the singular value decomposition method cannot accurately determine the singular value threshold to obtain the best noise reduction effect, resulting in poor noise reduction effect, making it impossible to accurately detect the damaged area inside the smelting furnace, resulting in low detection efficiency and the risk of accidents. Summary of the invention

[0005] In order to solve the problem that the singular value threshold cannot be accurately determined during the detection of metal smelting equipment, the present invention provides an image-based metal smelting equipment detection method.

[0006] In a first aspect, the present invention provides an image-based metal smelting equipment detection method, which adopts the following technical solution:

[0007] Performing singular value decomposition on the grayscale image of the inner edge of the smelting furnace to obtain multiple singular values, and superimposing the singular values ​​in order from small to large to obtain multiple reconstructed images;

[0008] Divide each reconstructed image into blocks to obtain multiple block regions of the reconstructed image;

[0009] Binarize the reconstructed image to obtain a binary image, calculate the preference of each binary image, and use the reconstructed image corresponding to the binary image whose priority is greater than the priority threshold as the first group of reconstructed images. The preference is positively correlated with the number of four-connected domains in the binary image.

[0010] Calculate the noise interference degree of the block area in each image in the first group of reconstructed images, as well as the number ratio of the block areas whose noise interference degree is less than the interference degree threshold, and use the reconstructed images whose number ratio is less than the ratio threshold as the second group of reconstructed images. The noise interference degree is positively correlated with the product of the grayscale mean and grayscale variance of the pixel points in the block area.

[0011] The damage degree of each pixel in the second group of reconstructed images is calculated. The damage degree represents the probability that the pixel is a damaged area. Outlier analysis is performed on the pixels in the reconstructed images whose damage degree is greater than the damage threshold. The reconstructed image with the least outliers is segmented to obtain the damaged area.

[0012] The effect is that a reconstructed image is obtained by performing singular value decomposition and reconstruction on the image, and then the reconstructed image of the detail feature angle is screened out by priority, and the reconstructed image with less noise interference is screened out according to the noise interference degree. Finally, outlier analysis is performed on the screened reconstructed image, and the singular value of the reconstructed image with the least outliers is the optimal singular value threshold. Finally, the reconstructed image with the least outliers is segmented to obtain the damaged area, which improves the detection accuracy of the smelting furnace and reduces the influence of noise on the damaged area.

[0013] Preferably, the calculation formula for the preference of the binary image is:

[0014]

[0015] in, express The preference of the binary image corresponding to the reconstructed image by stacking the singular values ​​of the layers, represents the number of four-connected domains in a binary image, Indicates the reconstructed image corresponding to the binary image. The grayscale variance of the block area is Indicates the number of block areas in the image. () is the normalization function.

[0016] The effect is that by calculating the optimization degree of the binary image, the reconstructed image with more detailed features can be screened out, which facilitates the subsequent analysis of the reconstructed image with more detailed features and improves the detection accuracy of the smelting furnace.

[0017] Preferably, the calculation formula for the noise interference degree of the block area is:

[0018]

[0019] in, Indicates the block area The noise interference degree, Indicates the block area The grayscale mean of the inner pixels, Indicates the block area The grayscale variance of the inner pixel, Indicates the block area Adjacent block areas The grayscale mean of the inner pixels, Indicates the block area The adjacent block area The grayscale variance of the inner pixel, is the block region vector, and its direction is the sum of the unit vectors of the pixel gradient direction in the block region. For the The first A vector of pixels, The direction is the block area The The gradient direction of each pixel is Indicates the number of pixels in the block area.

[0020] Preferably, the calculation formula for the noise interference degree of the block area is:

[0021]

[0022] in, Indicates the block area The noise interference degree, Indicates the block area The grayscale mean of the inner pixels, Indicates the block area The grayscale variance of the inner pixel, Indicates the block area Adjacent block areas The grayscale mean of the inner pixels, Indicates the block area The adjacent block area The grayscale variance of the inner pixel, is the block region vector, and its direction is the sum of the unit vectors of the pixel gradient direction in the block region. For the The first A vector of pixels, The direction is the block area The The gradient direction of each pixel is Indicates the number of pixels in the block area. () is the normalization function, Representation vector With vector The sine of the angle.

[0023] The effect is that the noise interference degree of the block area can be calculated through the above formula, and the noise interference degree reflects the influence of noise on the damaged area, so as to facilitate the screening of the reconstructed image which is greatly affected by the noise.

[0024] Preferably, the calculation formula for the damage degree of a pixel is:

[0025]

[0026] in, Represents pixel The degree of damage, Represents the mean value of noise interference of all pixels in the image. Represents pixel In the 8 neighborhoods of The noise interference of each pixel is Represents pixel and the 8th neighborhood The absolute value of the noise interference difference of each pixel, Represents pixel The vector is the gradient direction of the pixel point. The size of is the noise interference degree of the block area where the pixel point is located. Represents pixel Among the neighboring pixels pixel vectors, Represents pixel The number of neighboring pixels.

[0027] The effect is that by calculating the damage degree of a pixel point, the probability that the pixel point is a damaged area can be determined, which facilitates the subsequent segmentation and identification of the damaged area and improves the accuracy of identifying the damaged area.

[0028] Preferably, the method for analyzing the reconstructed image with the least outliers to obtain the damaged area is: performing outlier analysis on the pixels by using a lof algorithm.

[0029] Preferably, the method for analyzing the reconstructed image with the least outliers to obtain the damaged area is: clustering the pixels to obtain the damaged area.

[0030] The effect is that the outlier analysis of the pixels is performed through the LOF algorithm to obtain a reconstructed image with the least outliers. The image at this time is the reconstructed image with the best noise reduction effect. By analyzing this reconstructed image, the detection accuracy of the smelting furnace can be improved.

[0031] Preferably, the Otsu threshold method is used to perform threshold segmentation on the reconstructed image to obtain the damaged area.

[0032] The present invention has the following technical effects: by performing singular value decomposition on the image of the smelting furnace, and then superimposing and reconstructing the singular values ​​to obtain multiple reconstructed images, then screening the corresponding reconstructed images by priority and noise interference, and finally performing outlier analysis on the screened reconstructed images, the singular value of the reconstructed image with the least outliers is the best singular value threshold, and the reconstructed image at this time is the image with the best noise reduction effect. The reconstructed image with the least outliers is segmented by the lof algorithm to obtain the damaged area, which reduces the influence of noise on the damaged area and improves the detection accuracy of the smelting furnace. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] By reading the detailed description below with reference to the accompanying drawings, the above and other purposes, features and advantages of the exemplary embodiments of the present invention will become readily understood. In the accompanying drawings, several embodiments of the present invention are shown in an exemplary and non-restrictive manner, and the same or corresponding reference numerals represent the same or corresponding parts.

[0034] Figure 1 It is a method flow chart of an image-based metal smelting equipment detection method of the present invention. DETAILED DESCRIPTION

[0035] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0036] It should be understood that when the terms "first", "second", etc. are used in the claims, descriptions, and drawings of the present invention, they are only used to distinguish different objects, rather than to describe a specific order. The terms "include" and "comprise" used in the description and claims of the present invention indicate the presence of the described features, wholes, steps, operations, elements, and / or components, but do not exclude the presence or addition of one or more other features, wholes, steps, operations, elements, components, and / or their collections.

[0037] The embodiment of the present invention discloses a metal smelting equipment detection method based on an image, which is used to solve the problem that the singular value threshold cannot be accurately determined during the detection of the metal smelting equipment. Figure 1 , including steps S1 to S6, which are specifically as follows:

[0038] S1. Perform singular value decomposition on the grayscale image of the inner edge of the smelting furnace to obtain multiple singular values, and superimpose the singular values ​​in order from small to large to obtain multiple reconstructed images.

[0039] Use a high-definition industrial camera to collect the edge image of the smelting furnace, grayscale the image to obtain a grayscale image, so as to reduce the dimension of the data, better capture the main information of the image, and simplify the singular value decomposition process; perform singular value decomposition on the grayscale image, and superimpose the decomposed singular values ​​one by one from small to large to obtain a reconstructed image. When the number of singular values ​​is superimposed from small to large, the singular values ​​corresponding to the low-frequency information will be retained, and the high-frequency information will be suppressed. The performance effect is that as the number of superimposed singular values ​​increases, the details of the edge image of the smelting furnace gradually increase, and the influence of roughness and noise on defects also increases with the increase of details.

[0040] The influence of the rough surface changes on the damage of the inner edge of the melting furnace in the inner edge image of the melting furnace is analyzed. In the process of superposition and reorganization from small to large singular values, the more singular values ​​there are, the more details of the rough pixels are distributed and the more chaotic the distribution is. At the same time, as the number of superposition of singular values ​​decreases, the details of the rough texture on the inner edge of the melting furnace will gradually merge.

[0041] For grayscale images, the matrix is ​​constructed using grayscale values. , the matrix Decomposed into the product of three matrices, namely ; Where: is non-zero dimensional matrix, is an orthogonal matrix, is an orthogonal matrix, is an orthogonal matrix The transpose of is a diagonal matrix, The elements on the diagonal are singular values. When reconstructing an image, the grayscale matrix of the reconstructed image is , then we have . is the diagonal matrix during reconstruction, the number of elements on the diagonal is less than The number of elements on the diagonal of a diagonal matrix. The decomposition process is prior art and will not be described in detail here.

[0042] For example, for the matrix Diagonal matrix after singular value decomposition , then the singular values ​​of the grayscale image after decomposition are 1, 2, 3, 4, and 5 respectively. The singular values ​​are reconstructed in the manner of (1, 2), (1, 2, 3), and (1, 2, 3, 4), and 3 reconstructed images are obtained respectively; when the singular values ​​are reconstructed in the manner of (1, 2): , the gray matrix of the reconstructed image is ; Similarly, when the singular values ​​are reconstructed according to (1, 2, 3): ; When the singular values ​​are reconstructed according to (1, 2, 3, 4): .

[0043] S2. Divide the reconstructed image into blocks to obtain a plurality of block regions of the reconstructed image.

[0044] For example, the reconstructed image is The size of the blocks is divided into multiple blocks to obtain multiple block areas.

[0045] S3. Binarize the reconstructed image to obtain a binary image, calculate the preference of each binary image, and use the reconstructed image corresponding to the binary image whose priority is greater than the priority threshold as the first group of reconstructed images. The preference is positively correlated with the number of four-connected domains in the binary image.

[0046] According to the number of four-connected domains in the binary image, the preference of each binary image is calculated. The preference calculation formula is:

[0047] in, express The preference of the binary image corresponding to the reconstructed image by stacking the singular values ​​of the layers, represents the number of four-connected domains in a binary image, Indicates the reconstructed image corresponding to the binary image. The grayscale variance of the block area is Indicates the number of block areas in the image. () is the normalization function.

[0048] The larger the value, the more details there are in the reconstructed image, which can distinguish images with too small number of singular values ​​and insufficient details. The larger the value, the more detailed features there are in the corresponding reconstructed image. The more singular values ​​there are, the more details of the coarse pixels are and the more chaotic the distribution is. express The first binary image corresponding to the superposition of the singular values ​​of the layers The grayscale variance of each block. The larger the value, the more detailed features there are in the block area. Reflects the number of detail features in the entire reconstructed image. The larger the value, the more detail features there are in the block.

[0049] Exemplarily, the priority threshold is , when the priority of the binary image is greater than When the priority of the binary image is less than , the corresponding reconstructed image is discarded.

[0050] S4. Calculate the noise interference degree of the block area in each image in the first group of reconstructed images, and the proportion of the block areas whose noise interference degree is less than the interference degree threshold, and use the reconstructed images whose proportion is less than the proportion threshold as the second group of reconstructed images. The noise interference degree is positively correlated with the product of the grayscale mean and grayscale variance of the pixel points in the block area.

[0051] In one embodiment, the calculation formula of the noise interference degree of the block area is:

[0052]

[0053] in, Indicates the block area The noise interference degree, Indicates the block area The grayscale mean of the inner pixels, Indicates the block area The grayscale variance of the inner pixel, Indicates the block area Adjacent block areas The grayscale mean of the inner pixels, Indicates the block area The adjacent block area The grayscale variance of the inner pixel, is the block region vector, and its direction is the sum of the unit vectors of the pixel gradient direction in the block region. For the The first A vector of pixels, The direction is the block area The The gradient direction of each pixel is Indicates the number of pixels in the block area. () is the normalization function, Representation vector With vector The sine of the angle.

[0054] For example, the block area Take as an example, obtain the gradient direction of the pixel points in the block area, sum the unit vectors of the gradient directions of all the pixels in the block area, and use the direction of the obtained vector as the vector direction of the block area, recorded as The gradient direction and magnitude of the k-th pixel in the block area are taken as the direction and magnitude of the k-th pixel vector, recorded as .

[0055] in, It reflects the influence of noise on the damaged area. The larger the value is, the greater the probability that the block area is affected by noise and the lower the damage performance is. and The larger the value, the greater the possibility of noise influence in the block area, and the less obvious the damaged area. Because the more noise there is, the more chaotic the grayscale performance in the block area. and The larger the value is, the greater the probability that noise will be present in the block area.

[0056] pass It reflects the possibility of the existence of a damaged area in the block area. The larger the value, the greater the probability of the existence of a damaged area in the block area. In the reconstructed image, because the noise is random, the smaller the noise interference, the higher the approximation of the pixels in a small area. Compare tile areas Block area with neighbors The larger the value, the greater the probability of noise influence in the block area, and the less obvious the damaged area.

[0057] Indicates the block area Inner pixel The sin value of the angle between the vector direction of the pixel and the vector direction of the block area. The larger the value, the greater the difference in direction between the pixel point and the block area, the greater the possibility of the block being affected by noise, and the lower the damage performance. It reflects the disorder of the direction of pixels in the block area and the difference of pixels in the block area. The larger the value, the greater the possibility of noise influence in the block area and the lower the damage performance.

[0058] Exemplarily, by calculating the noise interference degree of each block area in the reconstructed image corresponding to each number of singular values, the number of blocks in the reconstructed image whose noise interference degree is greater than 0.8 is calculated, and images in which the ratio of the number of block areas with a noise interference degree greater than 0.8 to the total number of block areas is greater than 60% are screened out, and the screened out images are removed to remove images with excessive noise interference, that is, the singular value threshold with poor noise reduction effect is removed, and the remaining reconstructed images are used for analysis in subsequent steps.

[0059] When there is a loss feature defect in the reconstructed image, the grayscale value of the pixel corresponding to the loss feature is lower, and the grayscale value of the pixel corresponding to the rough surface of the edge of the smelting furnace is higher. When the damaged area is not affected by noise, the grayscale feature uniformity of the pixels in the damaged area is higher, and the grayscale directions of the pixels in the damaged area are approximately the same. When the damaged area is affected by noise, the grayscale feature uniformity of the pixels corresponding to the damaged area is lower. At the same time, noise is random. The smaller the noise interference, the higher the similarity of the pixels in a small area. Therefore, the possibility of the pixels being affected by noise is judged by the similarity of the pixels in the block area.

[0060] In other embodiments, the calculation formula of the noise interference degree of the block area is:

[0061]

[0062] in, Indicates the block area The noise interference degree, Indicates the block area The grayscale mean of the inner pixels, Indicates the block area The grayscale variance of the inner pixel, Indicates the block area Adjacent block areas The grayscale mean of the inner pixels, Indicates the block area The adjacent block area The grayscale variance of the inner pixel, is the block region vector, and its direction is the sum of the unit vectors of the pixel gradient direction in the block region. For the The first A vector of pixels, The direction is the block area The The gradient direction of each pixel is Indicates the number of pixels in the block area. () is the normalization function, Representation vector With vector The sine of the angle.

[0063] S5. Calculate the damage degree of each pixel in the second group of reconstructed images, where the damage degree indicates the probability that the pixel is a damaged area.

[0064] The calculation formula of pixel damage degree is:

[0065]

[0066] in, Represents pixel The degree of damage, Represents the mean value of noise interference of all pixels in the image. Represents pixel In the 8 neighborhoods of The noise interference of each pixel is Represents pixel and the 8th neighborhood The absolute value of the noise interference difference of each pixel, Represents pixel The vector is the gradient direction of the pixel point. The size of is the noise interference degree of the block area where the pixel point is located. Represents pixel Among the neighboring pixels pixel vectors, Represents pixel The number of neighboring pixels. The direction of is the gradient direction of the pixel point. The size of is the noise interference degree of the block area where the pixel point is located. For example, the pixel point in the image For example, the noise interference degree of the block area where it is located is taken as the noise interference degree of the pixel point, and the gradient direction of the pixel point and the corresponding noise interference degree are taken as the direction and magnitude of the pixel point vector, so as to obtain the pixel point Pixel vector , the neighboring pixel of the pixel The pixel vector is recorded as .

[0067] The larger the value of is, the greater the probability that the pixel is a damaged area. Reflects the change range between pixel p and neighboring pixels. The larger the value, the greater the impact of the damaged area. Because the grayscale of the surrounding pixels close to the damaged area changes faster, Reflects the difference between the pixel and its neighbors. The larger the value, the greater the impact of the damage. express The norm of .

[0068] Because the damaged area is small and the normal surface occupies most of the reconstructed image, the damage has a small impact range. Reflects the difference between the pixel and the overall reconstructed image. The larger the value, the greater the impact of the pixel on the damage. It reflects the difference between the pixel and its neighboring pixels. The larger the value, the greater the impact of the damage.

[0069] S6. Perform outlier analysis on the pixel points in the reconstructed image whose damage degree is greater than the damage threshold, and analyze the reconstructed image with the least outliers to obtain the damage area.

[0070] In one embodiment, a method for analyzing the reconstructed image with the least outliers to obtain the damaged area is: performing outlier analysis on the pixels using a lof algorithm.

[0071] In the reconstructed image, pixels with loss features appear in blocks. The distance relationship between pixels in the damaged area is used to determine whether the pixels with possible damage are noise or defects. For pixels with a damage degree greater than 0.8 in the reconstructed image, the density of the coordinates of each pixel is calculated using the lof algorithm, and the reconstructed image with the least outliers is selected. The Otsu method is used for threshold segmentation to obtain the precise location of the damaged area, thereby enabling more accurate detection of the damaged area in the smelting furnace.

[0072] In other embodiments, a method for analyzing the reconstructed image with the least outliers to obtain the damaged area is: clustering the pixels to obtain the damaged area.

[0073] The reconstructed image is obtained by performing singular value decomposition and reconstruction on the image, and then the reconstructed image of the detail feature angle is screened out by priority, and the reconstructed image with less noise interference is screened out according to the noise interference degree. Finally, outlier analysis is performed on the screened reconstructed image. The singular value of the reconstructed image with the least outliers is the optimal singular value threshold, which has the best noise reduction effect. Finally, the reconstructed image with the least outliers is analyzed to obtain the damaged area, which improves the detection accuracy of the smelting furnace and reduces the influence of noise on the damaged area.

[0074] The above are all preferred embodiments of the present invention, and are not intended to limit the protection scope of the present invention. Therefore, any equivalent changes made based on the structure, shape, and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A metal smelting equipment detection method based on an image, characterized in that: Includes steps: Performing singular value decomposition on the grayscale image of the inner edge of the smelting furnace to obtain multiple singular values, and superimposing the singular values ​​in order from small to large to obtain multiple reconstructed images; Divide each reconstructed image into blocks to obtain multiple block regions of the reconstructed image; Binarize the reconstructed image to obtain a binary image, calculate the preference of each binary image, and use the reconstructed image corresponding to the binary image whose preference is greater than the preference threshold as the first group of reconstructed images. The preference is positively correlated with the number of four-connected domains in the binary image. The calculation formula for the preference of a binary image is: in, express The preference of the binary image corresponding to the reconstructed image by stacking the singular values ​​of the layers, represents the number of four-connected domains in a binary image, Indicates the reconstructed image corresponding to the binary image. The grayscale variance of the block area is Indicates the number of block areas in the image. () is the normalization function; Calculate the noise interference degree of each image block area in the first group of reconstructed images, as well as the number ratio of the block areas whose noise interference degree is less than the interference degree threshold, and use the reconstructed images whose number ratio is less than the ratio threshold as the second group of reconstructed images. The size of the noise interference degree is positively correlated with the product of the grayscale mean and grayscale variance of the pixel points in the block area; Calculate the damage degree of each pixel in the second group of reconstructed images. The damage degree indicates the probability that the pixel is a damaged area. Perform outlier analysis on the pixels in the reconstructed images whose damage degree is greater than the damage threshold. Analyze the reconstructed images with the least outliers to obtain the damaged area. The calculation formula of pixel damage degree is: in, Represents pixel The degree of damage, Represents the mean value of noise interference of all pixels in the image. Represents pixel In the 8 neighborhoods of The noise interference of each pixel is Represents pixel and the 8th neighborhood The absolute value of the noise interference difference of each pixel, Represents pixel The vector is the gradient direction of the pixel point. The size of is the noise interference degree of the block area where the pixel point is located. Represents pixel Among the neighboring pixels pixel vectors, Represents pixel The number of neighboring pixels.

2. The image-based metal smelting equipment detection method according to claim 1, characterized in that: The calculation formula of the noise interference degree of the block area is: in, Indicates the block area The noise interference degree, Indicates the block area The grayscale mean of the inner pixels, Indicates the block area Grayscale variance of inner pixels, Indicates the block area Adjacent block areas The grayscale mean of the inner pixels, Indicates the block area Adjacent block areas The grayscale variance of the inner pixel, is the block region vector, and its direction is the sum of the unit vectors of the pixel gradient direction in the block region. For the The first A vector of pixels, The direction is the block area The The gradient direction of each pixel is Indicates the number of pixels in the block area. () is the normalization function, Representation vector With vector The sine of the angle.

3. The image-based metal smelting equipment detection method according to claim 1, characterized in that: The calculation formula of the noise interference degree of the block area is: in, Indicates the block area The noise interference degree, Indicates the block area The grayscale mean of the inner pixels, Indicates the block area The grayscale variance of the inner pixel, Indicates the block area Adjacent block areas The grayscale mean of the inner pixels, Indicates the block area Adjacent block areas The grayscale variance of the inner pixel, is the block region vector, and its direction is the sum of the unit vectors of the pixel gradient direction in the block region. For the The first A vector of pixels, The direction is the block area The The gradient direction of each pixel is Indicates the number of pixels in the block area. () is the normalization function, Representation vector With vector The sine of the angle.

4. The image-based metal smelting equipment detection method according to any one of claims 1 to 3, characterized in that: The method for analyzing the reconstructed image with the least outliers to obtain the damaged area is: performing outlier analysis on the pixels through the lof algorithm to obtain the damaged area.

5. The image-based metal smelting equipment detection method according to any one of claims 1 to 3, characterized in that: The method for analyzing the reconstructed image with the least outliers to obtain the damaged area is: clustering the pixels to obtain the damaged area.

6. The image-based metal smelting equipment detection method according to any one of claims 1 to 3, characterized in that: The reconstructed image was segmented by Otsu threshold method to obtain the damaged area.