An image enhancement system and its safety monitoring system for the image of a steelmaking project area

Abstract

The present invention discloses an image enhancement system for images in a steelmaking engineering area and its safety monitoring system, belonging to the technical field of image processing. According to the differences of pixel points at the same positions between different engineering area images, the present invention obtains the soot possibility of the pixel points; optimizes the soot possibility according to the texture change characteristics of the pixel points to obtain the optimized soot possibility; obtains the optimal filtering window size of the pixel points according to the optimized soot possibility of the pixel points and the filtering change value, and further obtains the enhanced image; meanwhile, uses a neural network to obtain the soot concentration of the enhanced image; and conducts safety monitoring of the steelmaking project according to the soot concentration. By adjusting the filtering window, the present invention avoids the problems of image blurring and loss of image edge information, and improves the accuracy of the steelmaking project safety monitoring system.

Description

Technical Field

[0001] The present invention belongs to the technical field of image processing, and more specifically, relates to an image enhancement system and a safety monitoring system for images in a steelmaking project area. Background Art

[0002] With the development of the times, the demand for steel in the market is constantly increasing, and more and more enterprises have entered the steel production industry. A large amount of dust, such as iron powder, particulate matter, and harmful chemical substances, is often generated in the steelmaking project. Long-term exposure to a high-concentration dust environment can cause damage to employees' respiratory systems, eyes, and skin, leading to health problems such as respiratory diseases and allergic reactions. By monitoring the dust concentration, the dust concentration level in the working environment can be grasped, and corresponding measures can be taken to reduce the impact of dust on employees' health. Therefore, it is of great significance to conduct safety monitoring on the dust concentration in the steelmaking project.

[0003] When conducting safety monitoring on a steelmaking project through a neural network, the quality of the steelmaking project image greatly affects the monitoring efficiency and accuracy. When using traditional window filtering to enhance the steelmaking project image, the size of the filtering window often needs to be set manually. A larger filtering window has a stronger smoothing effect, but an overly large window may blur the details in the image, making the image become blurred; a smaller filtering window helps to retain the edge information of the image, but a smaller window may not be able to effectively remove the noise in the image. The image enhancement effect of the traditional window filtering algorithm is not ideal, which in turn affects the efficiency and accuracy of the steelmaking project safety monitoring system. Summary of the Invention

[0004] 1. Problems to be Solved

[0005] In view of at least some of the above problems existing in the prior art, the present invention proposes an image enhancement system and a safety monitoring system for images in a steelmaking project area, aiming to solve the problem that the image enhancement effect of the traditional window filtering algorithm is not ideal, thereby affecting the efficiency and accuracy of the steelmaking project safety monitoring system.

[0006] 2. Technical Solutions

[0007] To solve the above problems, the technical solutions adopted by the present invention are as follows:

[0008] An image enhancement system for images in a steelmaking project area of the present invention includes an image acquisition module for acquiring engineering area images of the steelmaking project area; wherein, the engineering area images include real-time images to be processed and their temporally adjacent historical images;

[0009] An initial soot analysis module, which is used to establish an initial filtering window centered on a target pixel point in the image to be processed according to a preset size, and establish a historical window in the corresponding area of the initial filtering window in the historical image;

[0010] According to the difference in pixel values at the same position between the initial filtering window and the historical window, obtain the soot possibility of the target pixel point;

[0011] Change the target pixel point to obtain the soot possibility of each pixel point in the image to be processed;

[0012] An optimized soot analysis module, which is used to obtain the texture category feature value of each pixel point in the window according to the pixel value distribution in the initial filtering window and the historical window;

[0013] According to the change characteristics of the texture category feature values of pixel points between engineering area images, obtain the soot blurriness of each pixel point in the initial filtering window;

[0014] According to the correlation between the difference in soot blurriness between each neighborhood pixel point and the target pixel point in the initial filtering window and the change characteristics of the texture category feature values, obtain the distribution feature weight of the neighborhood pixel point;

[0015] Combine the distribution feature weight of the neighborhood pixel point and the corresponding soot possibility to optimize the soot possibility of the target pixel point, obtain the initial optimized soot possibility, and obtain the optimized soot possibility according to the initial optimized soot possibility and the soot blurriness of the target pixel point;

[0016] An image enhancement module, which is used to obtain the filtering change value before and after filtering of the target pixel point in the initial filtering window; obtain the optimal filtering window size according to the optimized soot possibility and the filtering change value of the target pixel point;

[0017] Change the target pixel point to obtain the optimal filtering window size of each pixel point in the image to be processed;

[0018] Enhance the image to be processed according to the optimal filtering window size to obtain an enhanced image.

[0019] Further, the method for obtaining the soot possibility of the target pixel point includes:

[0020] Obtain the first pixel value difference between the pixel points of the initial filtering window and the pixel points at the corresponding positions of each historical window, and use the sum of the second pixel value differences of all pixel points in the initial filtering window as the initial soot possibility of the target pixel point under the corresponding historical window;

[0021] Accumulate the initial soot possibilities under all historical windows to obtain the soot possibility of the target pixel point.

[0022] Further, the method for obtaining the texture category feature value includes:

[0023] Obtain the second pixel value difference between the neighborhood pixel points and the target pixel point in the window; classify the second pixel value difference according to numerical values to obtain different categories, and each category corresponds to a preset texture category feature value.

[0024] Further, the method for obtaining the soot blurriness of the pixel point includes:

[0025] Construct a texture category feature value sequence of the corresponding pixel point from the texture category feature value of the pixel point in the initial filtering window and the texture category feature values of the corresponding pixel points in all the historical windows;

[0026] Obtain the variance of the texture category feature value of the pixel point according to the texture category feature value sequence;

[0027] Multiply the range in the texture category feature value sequence of the pixel point, the time difference corresponding to the range, and the variance of the texture category feature value, and use the product as the soot blurriness of the pixel point;

[0028] When there are multiple ranges composed of multiple groups of maximum and minimum values in the texture category feature value sequence, take the range corresponding to the group of maximum and minimum values closest to the acquisition moment corresponding to the image to be processed for soot blurriness calculation.

[0029] Further, the method for obtaining the distribution feature weight of the pixel point includes:

[0030] Obtain the correlation coefficient of the texture category feature value sequence between each neighborhood pixel point and the target pixel point;

[0031] Negatively map and normalize the absolute value of the difference between the soot blurriness of the central pixel point and the neighborhood pixel points in the initial filtering window, and then multiply it by the corresponding correlation coefficient to obtain the distribution feature weight of the pixel point.

[0032] Further, the method for obtaining the optimized soot possibility includes:

[0033] Multiply the soot possibility of the pixel point in the initial filtering window by its corresponding distribution feature weight to obtain the weighted soot possibility of this pixel point;

[0034] Sum the weighted soot possibilities of all pixel points in the initial filtering window and multiply it by the soot possibility of the target pixel point to obtain the initial optimized soot possibility of the target pixel point;

[0035] Multiply the initial optimized soot possibility of the target pixel point by the soot blurriness of the target pixel point to obtain the optimized soot possibility of the target pixel point.

[0036] Further, the method for obtaining the filtering variation value includes:

[0037] Filter the initial filtering window, divide the difference in pixel values of the central pixel points before and after filtering by the maximum value of the pixel values of the two pixel points, and use the quotient as the filtering variation value.

[0038] Further, the method for obtaining the optimal filtering window size includes:

[0039] When the filtering variation value of the initial filtering window is less than or equal to the preset variation threshold, the filtering window size is not changed, and the preset size is the optimal filtering window size;

[0040] When the filtering variation value of the initial filtering window is greater than the preset variation threshold, normalize the optimized soot possibility of the target pixel point to obtain the normalized optimized soot possibility;

[0041] If the normalized optimized soot possibility is less than the preset optimization threshold, add the natural number 1 to the normalized optimized soot possibility, and then multiply by the preset size to obtain the size close to the optimal filtering window size;

[0042] If the normalized optimized soot possibility is greater than or equal to the preset optimization threshold, subtract the normalized optimized soot possibility from the natural number 1, and then multiply by the preset size to obtain the size close to the optimal filtering window size;

[0043] Take the odd number closest to the size close to the optimal filtering window size to obtain the optimal filtering window size.

[0044] The present invention also provides a safety monitoring system for a steelmaking project, which includes a safety monitoring module and the above image enhancement system;

[0045] Among them, the safety monitoring module uses a pre-trained neural network to obtain the soot concentration of the enhanced image; and performs safety monitoring of the steelmaking project according to the soot concentration.

[0046] 3. Beneficial effects

[0047] Compared with the prior art, the beneficial effects of the present invention are:

[0048] (1) An image enhancement system for an image of a steelmaking project area of the present invention obtains the soot possibility of a target pixel point through the difference in pixel values of pixel points at the same position in a preset window of the target pixel point in the project area image, making full use of the characteristic that the spatial position of soot changes with time and causes the image to change, so that the obtained soot possibility is more reasonable.

[0049] (2) An image enhancement system for the steelmaking engineering area image of the present invention obtains the soot blurriness of each pixel point. The soot blurriness can reflect the change degree of the texture class feature value of the pixel point over time. Subsequently, optimizing the soot possibility through the soot blurriness can improve the accuracy of the soot possibility. At the same time, the distribution feature weight characterizes the similarity degree of the change between the neighboring pixel points and the target pixel point over time. The more similar the change between the neighboring pixel points and the target pixel point is, the greater the similarity degree is, the smaller the possibility that the target pixel point is a noise pixel point is, and the greater the distribution feature weight is. Subsequently, optimizing the soot possibility with the distribution feature weight can reduce the error of the soot possibility; among them, the distribution feature weight of the neighboring pixel points and the corresponding soot possibility can reflect the overall soot possibility within the neighborhood. Optimizing the soot possibility of the target pixel point with the overall soot possibility within the neighborhood can reduce the influence of noise points and reduce the possibility that noise points are misjudged as soot pixel points, so as to make the subsequent adjustment of the window size more accurate.

[0050] (3) An image enhancement system for the steelmaking engineering area image of the present invention obtains the corresponding optimal filtering window size by combining the filtering change value before and after filtering of the target pixel point in the initial filtering window and optimizing the soot possibility, protecting the area insensitive to image enhancement in the image and obtaining a more accurate optimal filtering window size; at the same time, enhancing the image according to the optimal filtering window size of the pixel point can better retain the edge information of the image while removing the noise points in the image, improving the image enhancement effect and being beneficial to the monitoring of the enhanced image by the neural network, thereby improving the monitoring accuracy of the steelmaking engineering safety monitoring system. Brief Description of the Drawings

[0051] Figure 1 It is a frame schematic diagram of a steelmaking engineering area safety monitoring system of the present invention.

[0052] In the figure: 101, image acquisition module; 102, initial soot analysis module; 103, optimized soot analysis module; 104, image enhancement module; 105, safety monitoring module. Detailed Embodiments

[0053] The present invention will be further described below in conjunction with specific embodiments.

[0054] Embodiment 1

[0055] Refer to Figure 1 As shown, an image enhancement system for the steelmaking engineering area image of this embodiment includes an image acquisition module 101, an initial soot analysis module 102, an optimized soot analysis module 103, and an image enhancement module 104.

[0056] Among them, the image acquisition module 101 is used to acquire the engineering area image of the steelmaking project area; the engineering area image includes the real-time image to be processed and its temporally adjacent historical images.

[0057] Specifically in this embodiment, through an industrial camera and a fixed light source, the soot image of the steelmaking project area during the steelmaking process is collected; the collected image is an RGB image, and the RGB image is grayscale processed by the weighted grayscale method to obtain the grayscale image of the soot image in the steelmaking project. The real-time image to be processed and its temporally adjacent historical images are selected as the engineering area images of the steelmaking project area.

[0058] Compared with the color image, the grayscale image has a smaller memory, faster operation speed, and can highlight the target area; the weighted grayscale method can adjust the weight according to the scene requirements, can better adapt to different image processing scenarios, and performing weighted grayscale processing on the collected image can improve the monitoring response speed and accuracy of the monitoring system.

[0059] In addition, in this embodiment, 10 engineering area images that are temporally adjacent and continuous to the image to be processed are selected as historical images before the acquisition time of the image to be processed.

[0060] The initial soot analysis module 102 is used to establish an initial filtering window centered on the target pixel point in the image to be processed according to a preset size, and establish a historical window in the corresponding area of the initial filtering window in the historical image; according to the pixel value difference at the same position between the initial filtering window and the historical window, obtain the soot possibility of the target pixel point; change the target pixel point to obtain the soot possibility of each pixel point in the image to be processed.

[0061] Specifically in this implementation, an initial filtering window is established centered on the target pixel point in the image to be enhanced according to a preset size, and a historical window is established in the corresponding area of the initial filtering window in the historical image. Among them, the window size is set to K*K, K = 9, and the adjustment range of the filtering window size is [3, 17], and each odd number within the interval can be used as the filtering window size. Of course, the implementer can set the initial filtering window size according to actual needs.

[0062] In addition, considering that the bilateral filtering algorithm can filter out noise and retain the image edge information. Therefore, in this embodiment, the bilateral filtering algorithm can be used for improvement to achieve image enhancement of the image to be processed. It should be noted that the bilateral filtering algorithm is a well-known technical means for those skilled in the art and will not be elaborated here. Of course, the implementer can select other suitable window filtering algorithms according to actual needs.

[0063] The described optimized soot analysis module 103 is used to obtain the texture category feature value of each pixel point within the window based on the pixel value distribution in the initial filtering window and the historical window; obtain the soot blurriness of each pixel point within the initial filtering window according to the change feature of the texture category feature value between pixel points in the engineering area images; obtain the distribution feature weight of the neighboring pixel points according to the correlation between the difference in soot blurriness and the change feature of the texture category feature value between each neighboring pixel point and the target pixel point within the initial filtering window; optimize the soot possibility of the target pixel point by combining the distribution feature weight of the neighboring pixel points and the corresponding soot possibility to obtain the initial optimized soot possibility, and obtain the optimized soot possibility based on the initial optimized soot possibility and the soot blurriness of the target pixel point.

[0064] The described image enhancement module 104 is used to obtain the filtering change value before and after filtering of the target pixel point in the initial filtering window; obtain the optimal filtering window size according to the optimized soot possibility and the filtering change value of the target pixel point; change the target pixel point to obtain the optimal filtering window size of each pixel point in the image to be processed; enhance the image to be processed according to the optimal filtering window size to obtain the enhanced image.

[0065] Embodiment 2

[0066] In the steelmaking project, soot is generated by the reaction of impurities in fuels and raw materials at high temperatures, and during the steelmaking process, the furnace top, furnace mouth, and other positions are often opened to discharge the soot. Since soot usually moves with the air in the air, in the engineering images, the distribution of soot should gradually weaken from the emission port to the surroundings, that is, the concentration is the highest at the emission port, and the soot concentration gradually attenuates around the emission port. Therefore, due to the diffusion of soot, there are differences between the image to be processed and the historical image. The greater the difference, the greater the probability that the corresponding position is caused by soot diffusion. Therefore, the soot possibility of the pixel points is obtained according to the pixel point difference between the image to be processed and the historical image. Taking the change features of the pixel points in the initial filtering window and the historical window as the local change features of the pixel points, and reflecting the soot possibility of the pixel points according to the local change features of the pixel points in different images, fully utilizes the characteristic that the soot position changes over time, the obtained soot possibility is more in line with the application scenario, has a more accurate impact on adjusting the filtering window size subsequently, improves the image enhancement effect, and improves the accuracy of the steelmaking project safety monitoring system.

[0067] In this embodiment, the method for obtaining the soot possibility of the target pixel point includes:

[0068] Obtain the first pixel value difference between the pixel points of the initial filtering window and the pixel points at the corresponding positions of each historical window, and take the sum of the second pixel value differences of all pixel points within the initial filtering window as the initial soot possibility of the pixel point under the corresponding historical window; accumulate the initial soot possibilities under all historical windows to obtain the soot possibility of the pixel point. The calculation formula for the soot possibility of the pixel point is as follows:

[0069]

[0070] where N is the number of historical images of the image to be processed, q is the q-th pixel point of the image to be processed, M is the neighborhood range of point q, that is, the number of pixel points within the initial filtering window, M = K * K, i represents the i-th historical image, G q , j represents the gray value of the j-th pixel point within the neighborhood range of point q in the image to be processed, represents the corresponding pixel point q of point q in the i-th relevant historical image ′ the gray value of the j-th pixel point within the neighborhood of point, F q represents the soot possibility of point q. Similarly, by changing the target pixel point, the soot possibility of each pixel point in the image to be processed can be obtained.

[0071] As the image changes over time, the greater the sum of the differences between the pixel points in the image to be processed and the corresponding pixel points in the historical images, the greater the soot possibility of the target pixel point, and the more likely it is to be a soot pixel point; conversely, the target pixel point is more likely to be a background pixel point.

[0072] Of course, the implementer can select other statistical information such as variance and range to obtain the soot possibility.

[0073] Embodiment 3

[0074] Since the environment in the steelmaking project is relatively complex, there is a high possibility of noise points in the collected soot images, and the noise points are randomly non-repetitive, and the changes of the noise points in the images at different times are relatively large. Therefore, the soot possibility of the pixel points obtained by the initial soot analysis module 102 will be affected by the noise points, and the noise points will be misjudged as soot pixel points. Therefore, directly judging the soot pixel points based on the soot possibility of the pixel points will affect the subsequent processing, resulting in inaccurate adjustment of the window size, poor image enhancement effect, and reduced accuracy of the steelmaking project safety monitoring system.

[0075] Therefore, in this embodiment, the soot possibility of the pixel points is optimized by means of the local texture information of the pixel points and the distribution characteristics of the soot pixel points, so as to improve the accuracy of the pixel points as the soot possibility, reduce the possibility of misjudgment of the noise points, and further improve the effect of the subsequent bilateral filtering and enhance the monitoring accuracy of the steelmaking project safety monitoring system.

[0076] The specific optimization process includes:

[0077] (1) Obtain the texture category feature value of each pixel point in the window based on the pixel value distribution in the initial filtering window and the historical window. According to the diffusion principle, the smoke concentration in different areas of the window is different, and the pixels have certain texture information. There are differences in the pixel value distribution between the pixels in the window. Based on this texture information, the texture category feature value of the pixel point can be obtained; the texture category feature value can reflect the texture information of the area around the pixel point, so the pixel point can be classified according to the texture category feature value of the pixel point.

[0078] The specific method for obtaining the texture category feature value includes:

[0079] According to the pixel value distribution in the initial filtering window and the historical window, the texture category feature value of each pixel point in the window is obtained, and the second pixel value difference between the neighborhood pixel point and the target pixel point in the window is obtained; the second pixel value difference is classified according to the numerical value to obtain different categories, and each category corresponds to a preset texture category feature value.

[0080] Specifically, the grayscale value difference between all the neighborhood pixels and the target pixel in the window is obtained, the empirical grayscale value difference threshold is set to 5, the grayscale value difference interval step is set to 10, and the difference interval is open on the left and closed on the right; when the grayscale value difference is in the grayscale value difference interval (5, 15], the texture category feature value of the corresponding neighborhood pixel is 1; when the grayscale value difference is in the grayscale value difference interval (15, 25], the texture category feature value of the corresponding neighborhood pixel is 2; and so on; when the grayscale value difference is less than the grayscale value difference threshold, the texture category feature value of the corresponding neighborhood pixel is 0. In other embodiments of the present invention, the implementer can set other appropriate grayscale value difference thresholds and grayscale value difference intervals according to actual needs.

[0081] (2) Obtain the smoke blur of each pixel in the initial filtering window based on the change characteristics of the texture category eigenvalues ​​of the pixels between the images of the project area. The texture category eigenvalue of the pixel represents the texture information of the pixel. The change of the texture category eigenvalue between different images can reflect the texture change characteristics of the pixel. The more drastic the change of the texture category eigenvalue, the more drastic the change of the pixel texture, which reflects the greater the smoke concentration, and the greater the smoke blur of the corresponding pixel.

[0082] Specifically, the texture class eigenvalue of the pixel points in the initial filtering window and the texture class eigenvalues of the corresponding pixel points in all historical windows form a texture class eigenvalue sequence of the corresponding pixel points; the variance of the texture class eigenvalues of the pixel points is obtained according to the texture class eigenvalue sequence; the range in the texture class eigenvalue sequence of the pixel points, the time difference corresponding to the range, and the variance of the texture class eigenvalues are multiplied, and the product is used as the soot blurriness of the pixel points; when there are multiple ranges composed of maximum and minimum values in the texture class eigenvalue sequence, the range corresponding to the group of maximum and minimum values closest to the acquisition time of the image to be processed is taken for calculating the soot blurriness. The calculation formula for the soot blurriness of pixel points is as follows:

[0083] R = f(Z) * |Z max - Z min | * |T max - T min |

[0084] Wherein, R is the soot blurriness of the pixel points, f(Z) represents the variance of the texture class eigenvalues corresponding to the pixel points, Z max represents the maximum value in the texture class eigenvalue sequence, Z min represents the minimum value in the texture class eigenvalue sequence, T max represents the acquisition time of the image corresponding to Z max , and T min represents the acquisition time of the image corresponding to Z min .

[0085] In the calculation formula for the soot blurriness of pixel points, the larger f(Z) is, the greater the change in the corresponding pixel points between the initial filtering window and the historical windows is reflected, that is, the greater the soot concentration is, so the greater the soot blurriness of the corresponding pixel points is; the larger |Z max - Z min | * |T max - T min | is, the greater the change in the texture class eigenvalues with time is, and the greater the soot blurriness of the pixel points is. The soot blurriness obtained according to the texture class eigenvalues of the pixel points and their corresponding image acquisition times can reflect the degree of differential change of the image over time, characterize the change degree of the soot concentration, and then further process the soot possibility of the target pixel points through the soot blurriness, so that the obtained optimized soot possibility is more accurate, thereby improving the image enhancement effect and the detection accuracy of the steelmaking project safety monitoring system.

[0086] (3) According to the correlation between the difference in soot blur degree between each neighborhood pixel point and the target pixel point within the initial filtering window and the change characteristic value of the texture category feature, obtain the distribution feature weight of the neighborhood pixel points. The correlation between the difference in soot blur degree between the neighborhood pixel points and the target pixel point and the change characteristic value of the texture category feature characterizes the similarity of the changes between the neighborhood pixel points and the target pixel point. The smaller the difference and the greater the correlation, the more referenceable the soot possibility of the corresponding neighborhood pixel point, and the greater the distribution feature weight.

[0087] Specifically, obtain the Pearson correlation coefficient of the texture category feature value sequence between each neighborhood pixel point and the target pixel point; map the absolute value of the negative correlation of the soot blur degree difference between the central pixel point and the neighborhood pixel points within the initial filtering window and normalize it, and then multiply it by the corresponding Pearson correlation coefficient to obtain the distribution feature weight of the neighborhood pixel points. Implementers can select other methods such as the Spearman correlation coefficient to obtain the correlation coefficient. The calculation formula of the distribution feature weight includes:

[0088] G i = exp(-|R i -R q |)*|H i |

[0089] where, G i is the distribution feature weight of the i-th neighborhood pixel point of the target pixel point q, R i is the soot blur degree of the i-th neighborhood pixel point, R q is that of the target pixel point q of the target pixel point q, H i is the Pearson correlation coefficient of the texture category feature value sequence between the i-th neighborhood pixel point and the target pixel point q.

[0090] In the calculation formula of the distribution feature weight, the larger the value of the Pearson correlation coefficient, the more similar the change characteristics of the corresponding neighborhood pixel point and the target pixel point, and the greater the distribution feature weight; the smaller |R i -R q |, the more similar the pixel value distributions of the corresponding neighborhood pixel point and the target pixel point, and the larger exp(-|R i -R q |) after the negative correlation mapping, and the greater the distribution feature weight. The distribution feature weight obtained through the correlation of the texture information changes between the neighborhood pixel points and the target pixel point and the pixel value distribution can reflect the distribution characteristics of the neighborhood pixel points. The greater the distribution feature weight, the more similar the changes of the pixel points within the neighborhood in different engineering area images over time. In subsequent processing, the distribution feature weight is used to further process the soot possibility, so that the obtained optimized soot possibility is more accurate, and the window size adjusted according to the optimized soot possibility is more appropriate, thereby improving the image enhancement effect and the detection accuracy of the steelmaking project safety monitoring system.

[0091] (4) Optimize the soot possibility of the target pixel by combining the distribution feature weight of the neighboring pixel points and the corresponding soot possibility to obtain the initial optimized soot possibility, and obtain the optimized soot possibility according to the initial optimized soot possibility and the soot blurriness of the target pixel point. The distribution feature weight of the neighboring pixel points and the corresponding soot possibility can reflect the overall soot possibility in the neighborhood. Optimizing the soot possibility of the target pixel point with the overall soot possibility in the neighborhood can reduce the influence of noise points and reduce the possibility of misjudging noise points as soot pixel points; further optimization is carried out by combining the soot blurriness representing the texture change characteristics of the target pixel point itself, and the obtained optimized soot possibility is more accurate.

[0092] Specifically, multiply the soot possibility of the pixel points in the initial filter window by their corresponding distribution feature weights to obtain the weighted soot possibility of this pixel point; sum the weighted soot possibilities of all pixel points in the initial filter window and multiply by the soot possibility of the target pixel point to obtain the initial optimized soot possibility of the target pixel point; multiply the initial optimized soot possibility of the target pixel point by the soot blurriness of the target pixel point to obtain the optimized soot possibility of the target pixel point. The calculation formula for the optimized soot possibility of the target pixel point is as follows:

[0093]

[0094] where, W′ q represents the optimized soot possibility of the target pixel point q, F q the soot possibility of the target pixel point q, R q the soot blurriness of the target pixel point q, G i represents the distribution feature weight of the i-th neighboring pixel point among the neighboring pixel points of the target pixel point q, F i represents the soot possibility of the i-th neighboring pixel point, G i *F i represents the weighted soot possibility of the i-th neighboring pixel point, and M represents the number of neighboring pixel points.

[0095] In the calculation formula for the optimized soot possibility of the target pixel, the greater the soot possibility of the target pixel q, the more likely the target pixel is a soot pixel, and the greater the corresponding optimized soot possibility; the greater the soot blur of the target pixel, the more drastic the change between the engineering area images, the higher the soot concentration, and the greater the corresponding optimized soot possibility; the more similar the changes in the neighboring pixels of the target pixel, the greater the soot possibility of the neighboring pixels, the greater the weighted soot possibility of the neighboring pixels, and the greater the sum of the weighted soot possibilities, reflecting that the overall changes in the pixels in the neighborhood are similar and have a greater soot possibility, representing the overall soot possibility of the neighborhood. The greater the overall soot possibility of the neighborhood, the greater the optimized soot possibility of the corresponding target pixel. Combining the soot possibility, soot blur, and weighted soot possibility of the neighboring pixels of the target pixel, the obtained optimized soot possibility reduces the influence of noise points, reduces the possibility of misjudging noise points, and the window size adjusted according to the optimized soot possibility is more appropriate, thereby improving the effect of subsequent bilateral filtering and enhancing the monitoring accuracy of the steelmaking project safety monitoring system.

[0096] Example 4

[0097] In the steelmaking project, some regions in the acquired images are insensitive to the image enhancement algorithm, that is, the pixel value changes of the pixels in the corresponding regions before and after image enhancement are not obvious. At this time, adjusting the window size will instead affect the characteristics of these regions. Therefore, in this embodiment, the filtering change value before and after filtering of the target pixel in the initial filtering window is obtained, and the filtering change value is used as one of the necessary factors for obtaining the optimal filtering window size.

[0098] Specifically, filter the initial filtering window, divide the difference in pixel values of the central pixel before and after filtering by the maximum value of the pixel values of the two pixels, and take the quotient as the filtering change value. The calculation formula for the filtering change value is as follows:

[0099]

[0100] Among them, Y q represents the filtering change value of the target pixel q, G q is the gray value of point q before filtering, G q is the gray value of point q after filtering, and max( ) represents selecting the maximum value among them.

[0101] In the calculation formula for the filtering change value, the greater the change in the gray value of the target pixel before and after filtering and the greater the proportion of the maximum gray value before and after filtering, the greater the filtering change value, reflecting that the target pixel is more sensitive to image enhancement and requires a more accurate filtering result by adjusting the window size; on the contrary, the window size is not adjusted to protect the characteristics of the target pixel with low sensitivity to image enhancement.

[0102] Furthermore, since the optimized soot possibility can represent the possibility that the target pixel is a soot pixel, but cannot reflect the sensitivity of the target pixel to image enhancement, while the filtering change value can represent the sensitivity of the target pixel, the characteristics of both are combined to obtain the optimal filtering window size based on the optimized soot possibility and the filtering change value of the target pixel.

[0103] Preferably, the preset change threshold is 0.5. When the filtering change value of the initial filtering window is less than or equal to the preset change threshold, the filtering window size remains unchanged, and the preset size is the optimal filtering window size; when the filtering change value of the initial filtering window is greater than the preset change threshold, the optimized soot possibility of the target pixel is normalized to obtain the normalized optimized soot possibility. If the normalized optimized soot possibility is less than the preset optimization threshold, then add the natural number 1 to the normalized optimized soot possibility and multiply by the preset size to obtain the size close to the optimal filtering window size; if the normalized optimized soot possibility is greater than or equal to the preset optimization threshold, then subtract the normalized optimized soot possibility from the natural number 1 and multiply by the preset size to obtain the size close to the optimal filtering window size; take the closest odd number of the size close to the optimal filtering window size to obtain the optimal filtering window size. The normalization process is already well-known to those skilled in the art and will not be elaborated here. In other embodiments of the present invention, the implementer can set the change threshold according to requirements. The calculation formula for the optimal filtering window size includes:

[0104]

[0105] where K q ′ is the optimal filtering window size of the target pixel q, K is the initial preset filtering window size, F′ q represents the normalized optimized soot possibility of the target pixel q, F′ q is obtained by normalizing W′ q , Y q represents the filtering change value of the target pixel q, K*(1 + F′ q ) represents the size close to the optimal filtering window size, and B() represents taking the closest odd integer of the part inside the parentheses.

[0106] In the calculation formula of the optimal filtering window size, when the filtering change value of the target pixel point is less than or equal to the preset change threshold, in order to protect the characteristic that the sensitivity of the target pixel point to image enhancement is not strong, the filtering window size is not adjusted, and the preset filtering window size is the optimal filtering window size of the target pixel point; when the filtering change value of the target pixel point is greater than the preset change threshold, there are two cases. One is that the possibility of normalized optimization of soot is relatively large, that is, the possibility that the target pixel point is a soot pixel point is relatively large, and the filtering window size is reduced, which is beneficial to retaining the detail information and edge information of the target pixel point, and then improving the judgment accuracy of the subsequent soot concentration; the other case is that the possibility of normalized optimization of soot is relatively small, that is, the possibility that the target pixel point is a soot pixel point is relatively small, and the filtering window size is increased, which is beneficial to removing image noise points and improving the image smoothing effect. Processing the image according to the optimal filtering window size can improve the image enhancement quality, thereby improving the accuracy of the steelmaking project safety monitoring system.

[0107] Further, by changing the target pixel point, the optimal filtering window size of each pixel point in the image to be processed is obtained. Processing the image according to the optimal filtering window size of each pixel point in the image can improve the image enhancement quality, thereby improving the accuracy of the steelmaking project safety monitoring system.

[0108] In this embodiment, the image to be processed is enhanced using the bilateral filtering algorithm according to the optimal filtering window size of each pixel point therein to obtain an enhanced image. The bilateral filtering algorithm is already well-known to those skilled in the art and will not be elaborated here. In other embodiments of the present invention, the implementer can select other window filtering algorithms such as non-local means filtering to process the image to obtain an enhanced image.

[0109] Embodiment 5

[0110] This embodiment provides a steelmaking project safety monitoring system. On the basis of the above image processing system, a safety monitoring module 105 is further provided. The safety monitoring module 105 is used to obtain the soot concentration of the enhanced image using a pre-trained neural network; and perform steelmaking project safety monitoring according to the soot concentration.

[0111] Specifically, the enhanced image is input into a pre-trained neural network. The loss function during the neural network training process is the mean square error loss function, and the output is the current soot concentration. For those skilled in the art, the training process of the network is a well-known technology and will not be elaborated here. The soot concentration judgment threshold 0.6 is set according to the empirical value. When the soot concentration is greater than or equal to the set judgment threshold, the staff is reminded to perform relevant operations such as soot removal to avoid harm to the safety of the staff. In other embodiments of the present invention, the implementer can set a suitable soot concentration judgment threshold according to needs.

[0112] A safety monitoring system for steelmaking projects according to the present invention utilizes the characteristic that the spatial position of soot changes over time, resulting in changes in the image. By the pixel value differences of the pixel points at the same positions in the preset window of the target pixel points in the image of the project area, the soot possibility of the target pixel points is obtained, and then the soot possibilities of all pixel points are obtained.

[0113] Meanwhile, in order to avoid the influence of noise in the image, and through the local texture information features of the target pixel points, the change features of the texture class feature values of each pixel point are obtained, and then the soot blurriness of each pixel point is obtained; by combining the differences in soot blurriness between each neighboring pixel point and the target pixel point, the correlation of the change features of the texture class feature values, and the soot blurriness of the target pixel point, the soot possibility of the target pixel point is optimized, reducing the misjudgment possibility of misidentifying noise points as soot pixel points. The obtained optimized soot possibility is more accurate and persuasive, and the optimal filtering window size obtained by combining the filtering change value of the target pixel point is more accurate. The image is enhanced according to the optimal filtering window size of the pixel points, removing the noise points in the image while better retaining the edge information of the image, improving the image enhancement effect, being beneficial to the monitoring of the enhanced image by the neural network, and improving the monitoring accuracy of the safety monitoring system for steelmaking projects.

[0114] The above has schematically described the present invention and its implementation manners. This description is not restrictive, and what is shown in the drawings is only one of the implementation manners of the present invention, and the actual structure is not limited thereto. Therefore, if those of ordinary skill in the art are inspired by it and design similar structural manners and embodiments to this technical solution without creative work without departing from the gist of the present invention, they shall fall within the protection scope of the present invention.

Claims

1. An image enhancement system for the images in the steelmaking engineering area, characterized in that: It includes an image acquisition module (101) for acquiring an engineering area image of the steelmaking engineering area; wherein, the engineering area image includes a real-time image to be processed and its temporally adjacent historical images; An initial soot analysis module (102) for establishing an initial filtering window centered on a target pixel point in the image to be processed according to a preset size, and establishing a historical window in the corresponding area of the initial filtering window in the historical image; Obtain the soot possibility of the target pixel point according to the pixel value difference at the same position between the initial filtering window and the historical window; Change the target pixel point to obtain the soot possibility of each pixel point in the image to be processed; An optimized soot analysis module (103) for obtaining the texture category feature value of each pixel point in the window according to the pixel value distribution in the initial filtering window and the historical window; Obtain the soot blurriness of each pixel point in the initial filtering window according to the change feature of the texture category feature value between pixel points in the engineering area images; Obtain the distribution feature weight of the neighborhood pixel points according to the correlation between the difference in soot blurriness between each neighborhood pixel point and the target pixel point in the initial filtering window and the change feature of the texture category feature value; Optimize the soot possibility of the target pixel point by combining the distribution feature weight of the neighborhood pixel points and the corresponding soot possibility to obtain the initial optimized soot possibility, and obtain the optimized soot possibility according to the initial optimized soot possibility and the soot blurriness of the target pixel point; An image enhancement module (104) for obtaining the filtering change value before and after filtering of the target pixel point in the initial filtering window; obtaining the optimal filtering window size according to the optimized soot possibility and the filtering change value of the target pixel point; Change the target pixel point to obtain the optimal filtering window size of each pixel point in the image to be processed; Enhance the image to be processed according to the optimal filtering window size to obtain an enhanced image.

2. The image enhancement system for a steelmaking project area image according to claim 1, wherein: The method for obtaining the soot possibility of the target pixel point includes: Obtain the first pixel value difference between the pixel points of the initial filtering window and the pixel points at the corresponding positions of each historical window, and take the sum of the second pixel value differences of all pixel points in the initial filtering window as the initial soot possibility of the target pixel point under the corresponding historical window; Accumulate the initial soot possibilities under all historical windows to obtain the soot possibility of the target pixel point.

3. An image enhancement system for an image of a steelmaking engineering area according to claim 1, characterized in that: The method for obtaining the texture category feature value includes: Obtain the second pixel value difference between the neighborhood pixel points and the target pixel point in the window; classify the second pixel value difference according to the numerical value to obtain different categories, and each category corresponds to a preset texture category feature value.

4. An image enhancement system for an image of a steelmaking engineering area according to claim 1, characterized in that: The method for obtaining the soot blurriness of the pixel point includes: Construct a texture category feature value sequence of the corresponding pixel points from the texture category feature values of the pixel points in the initial filtering window and the texture category feature values of the corresponding pixel points in all historical windows; Obtain the variance of the texture category feature value of the pixel point according to the texture category feature value sequence; Multiply the range in the texture category feature value sequence of the pixel point, the time difference corresponding to the range, and the variance of the texture category feature value, and take the product as the soot blurriness of the pixel point; When there is a range formed by multiple sets of maximum and minimum values in the texture category eigenvalue sequence, the range corresponding to the set of maximum and minimum values closest to the acquisition moment corresponding to the image to be processed is taken for soot blurriness calculation.

5. An image enhancement system for a steelmaking project area image according to claim 1, characterized in that: The method for obtaining the distribution feature weight of the pixel points includes: Obtaining the correlation coefficient of the texture category eigenvalue sequence between each neighborhood pixel point and the target pixel point; Negatively correlating and normalizing the absolute value of the difference in soot blurriness between the central pixel point and the neighborhood pixel points within the initial filtering window, and then multiplying by the corresponding correlation coefficient to obtain the distribution feature weight of the pixel points.

6. The image enhancement system for the steelmaking project area image according to claim 1, characterized in that: The method for obtaining the optimized soot possibility includes: Multiplying the soot possibility of the pixel points within the initial filtering window by their corresponding distribution feature weights to obtain the weighted soot possibility of this pixel point; Summing up the weighted soot possibilities of all pixel points within the initial filtering window and multiplying by the soot possibility of the target pixel point to obtain the initial optimized soot possibility of the target pixel point; Multiplying the initial optimized soot possibility of the target pixel point by the soot blurriness of the target pixel point to obtain the optimized soot possibility of the target pixel point.

7. An image enhancement system for a steelmaking project area image according to claim 1, characterized in that: The method for obtaining the filtering change value includes: Filtering the initial filtering window, dividing the difference in pixel values of the central pixel point before and after filtering by the maximum value of the pixel values of the two pixel points, and taking the quotient as the filtering change value.

8. An image enhancement system for an image of a steelmaking engineering area according to claim 1, characterized in that: The method for obtaining the optimal filtering window size includes: When the filtering change value of the initial filtering window is less than or equal to the preset change threshold, the filtering window size is not changed, and the preset size is the optimal filtering window size; When the filtering change value of the initial filtering window is greater than the preset change threshold, normalizing the optimized soot possibility of the target pixel point to obtain the normalized optimized soot possibility; If the normalized optimized soot possibility is less than the preset optimization threshold, adding the normalized optimized soot possibility to the natural number 1 and then multiplying by the preset size to obtain the size close to the optimal filtering window size; If the normalized optimized soot possibility is greater than or equal to the preset optimization threshold, subtracting the normalized optimized soot possibility from the natural number 1 and then multiplying by the preset size to obtain the size close to the optimal filtering window size; Taking the odd number closest to the size close to the optimal filtering window size to obtain the optimal filtering window size.

9. A safety monitoring system for steelmaking projects, characterized in that: Including a safety monitoring module (105) and the image enhancement system according to any one of claims 1-8; Wherein, the safety monitoring module (105) obtains the soot concentration of the enhanced image by using a pre-trained neural network; and performs safety monitoring of the steelmaking project according to the soot concentration.

Images