A method and system for evaluating the surface state of a cold bed crystallizer

Abstract

The present application belongs to the technical field of image processing, and particularly relates to a cold bed crystallizer surface state evaluation method and system, which comprises the following steps: collecting a crystallizer inner surface original image; calculating a background estimation value of a target pixel point; subtracting the gray value of each pixel point from the background estimation value to obtain a detail image; for each pixel point in the detail image: determining the information entropy of the pixel point, clustering all the information entropies, and taking the maximum information entropy in the cluster with the most pixel points as the reference entropy; determining the dynamic regularization parameter of the pixel point according to the difference between the information entropy of the pixel point and the reference entropy; filtering and threshold segmenting the detail image by using the dynamic regularization parameter, counting the percentage of the defect pixel points in the segmented image, and determining whether the crystallizer surface state is qualified. The present application realizes the self-adaptive adjustment of the regularization parameter and improves the evaluation accuracy of the inner surface state of the crystallizer.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image processing. More particularly, the present application relates to a cold bed crystallizer surface state evaluation method and system. BACKGROUND

[0002] In the smelting process of metal materials, the cold bed crystallizer is the core mold component that determines the quality of the ingot. The inner cavity surface of the cold bed crystallizer directly contacts the high-temperature molten metal during the smelting process and causes it to solidify and form. The smoothness of the inner surface of the crystallizer has a decisive influence on the surface quality of the ingot. If there are scratches, pits, cracks or attachments left by previous smelting tasks on the surface of the crystallizer, these defects are extremely easy to reproduce when the ingot solidifies, resulting in surface defects of the ingot, which seriously affects the product performance and yield.

[0003] At present, the automatic detection method for the surface state of the crystallizer, such as the guided filter (Guided Filter) algorithm, uses a guide image to calculate the filter weight, which can smooth the image while maintaining the strong edges in the guide image, thereby distinguishing the surface texture and defects.

[0004] However, the guided filter algorithm relies on the gray scale information in its neighborhood range and a fixed regularization parameter when calculating the output value of each pixel point. The regularization parameter controls the sensitivity of the filter to the edge and is used to determine the retention and smoothing of the edge. Considering that there are regular machining textures on the inner surface of the crystallizer, if the regularization parameter value is set too small, the filter will be too sensitive and will retain all the machining textures as edges while retaining the defects, resulting in false positives. If the regularization parameter value is set too large, the filter will successfully smooth out the textures, but it will also smooth out the weak real defects at the same time, resulting in false negatives, which will affect the product quality. SUMMARY

[0005] To solve the technical problems of false positives and false negatives caused by the inability of the regularization parameter in the traditional guided filter to adapt, the present application provides solutions in the following aspects.

[0006] In a first aspect, the present application provides a cold bed crystallizer surface state evaluation method, comprising: collecting a crystallizer inner surface original image; taking any pixel point in the original image as a target pixel point, extracting the median of the gray values of all neighborhood pixel points in the neighborhood window of the target pixel point; determining a weight value according to the deviation of the gray value of each neighborhood pixel point from the median; obtaining a background estimation value of the target pixel point by weighted average of the gray values of all neighborhood pixel points according to the weight value; subtracting the gray value of each pixel point in the original image from the background estimation value to obtain the gray value of each pixel point in the detail image; for each pixel point in the detail image: statistics the probability of each gray level in the neighborhood window of the pixel point; determining the information entropy according to the probability of each gray level; clustering all information entropies, and taking the cluster containing the most pixel points as a texture cluster; taking the maximum value of the information entropy in the texture cluster as a reference entropy; determining the dynamic regularization parameter of the pixel point according to the difference between the information entropy of the pixel point and the reference entropy; filtering the detail image by using the dynamic regularization parameter of each pixel point, thresholding the filtered image, and statistics the percentage of defect pixel points in the thresholded binary image, and determining that the state is unqualified in response to the percentage being greater than a set threshold.

[0007] The present application first obtains a detail image by background estimation, calculates the information entropy of each pixel point in the detail image, automatically identifies the texture cluster containing the most pixel points by clustering all information entropies, and defines the maximum information entropy in the texture cluster as a reference entropy; generates a dynamic regularization parameter for each pixel point according to the difference between the information entropy of each pixel point and the reference entropy, ensures that the texture region obtains a larger parameter and the defect region obtains a smaller parameter, and this method effectively smooths the texture region while retaining defect details, thereby improving the accuracy of the evaluation of the inner surface state of the crystallizer.

[0008] Preferably, the weight value is equal to the reciprocal of the absolute difference between the gray value of each neighborhood pixel point and the median; and when the absolute difference is 0, the weight value is 1.

[0009] The present application uses the reciprocal of the absolute difference between the gray value of each pixel point and the normal gray level of its neighborhood as the weight, so that the pixel points with larger deviation from the normal gray level obtain lower weight, and the pixel points close to the normal gray level obtain higher weight, which can effectively reduce the interference of defect and other abnormal pixel points on the background estimation, make the background extraction purer, and ensure that the subsequent detail image only contains high-frequency textures and defects.

[0010] Preferably, the background estimation value of the target pixel point satisfies the expression: ; in the expression, is the background estimation value of the target pixel point; is the gray value of the i-th neighborhood pixel point in the neighborhood window of the target pixel point; and is the number of neighborhood pixel points in the neighborhood window of the target pixel point. a weight value of a target pixel point in a neighborhood window of the target pixel point; a weight value of a target pixel point in a neighborhood window of the target pixel point; an index value and a number of pixel points in a neighborhood window of the target pixel point.

[0011] The application realizes weighted average of the neighborhood pixel points by the weight value, ensures that the background estimation value can accurately reflect low-frequency light change, and effectively eliminates the interference of the defect region with high gray difference.

[0012] Preferably, the probability of each gray level in the neighborhood window of the pixel point comprises: according to the gray value of all pixel points in the neighborhood window of each pixel point, the number of pixel points of each gray level is counted, and the probability of each gray level is calculated, and the probability is equal to the ratio of the number of pixel points of each gray level to the number of all pixel points in the neighborhood window.

[0013] Preferably, the information entropy satisfies the expression: ; in the expression, is the information entropy of the target pixel point; is the information entropy of the target pixel point; is the probability of the target pixel point in the neighborhood window of the target pixel point; is the probability of the target pixel point in the neighborhood window of the target pixel point; is the probability of the target pixel point in the neighborhood window of the target pixel point; is the index value and the total number of gray levels; is a logarithmic function.

[0014] The application can evaluate the complexity of the gray value distribution in the neighborhood of the pixel point in the detail image by calculating the information entropy. The gray distribution of the regular texture region is relatively simple, and the entropy value is low. The irregularity of the defect region makes the gray distribution more chaotic, and the entropy value is high. This index provides reliable data basis for distinguishing texture and defects.

[0015] Preferably, the algorithm for clustering all information entropies is a DBSCAN clustering algorithm.

[0016] Preferably, the dynamic regularization parameter of the pixel point satisfies the expression: ; in the expression, is the dynamic regularization parameter of the target pixel point; is the dynamic regularization parameter of the target pixel point; is the reference entropy; is the information entropy of the target pixel point; is the information entropy of the target pixel point; is a hyperbolic tangent function.

[0017] The present application maps the difference between the information entropy of the pixel points and the reference entropy to the range of 0-1 by using the hyperbolic tangent function, so as to obtain a dynamic regularization parameter, ensure that the defect area obtains a smaller parameter to retain the edge details, and the texture area obtains a larger parameter to realize the smoothing processing, and provide an accurate control signal for subsequent adaptive filtering.

[0018] Preferably, the algorithm for threshold segmentation is the Otsu method, and the white pixel points in the segmented binary image are the defect pixel points.

[0019] Preferably, the percentage of the defect pixel points is equal to the percentage of the number of the white pixel points in the segmented binary image to the number of all pixel points.

[0020] In the second aspect, the present application provides a cold bed crystallizer surface state evaluation system, comprising a processor and a memory, the memory stores computer program instructions, when the computer program instructions are executed by the processor, the above-mentioned cold bed crystallizer surface state evaluation method is realized.

[0021] By adopting the above technical scheme, the above-mentioned cold bed crystallizer surface state evaluation method is generated into a computer program and stored in the memory, so as to be loaded and executed by the processor, so as to manufacture a terminal device according to the memory and the processor, and facilitate use.

[0022] The present application has the following advantages:

[0023] (1) The present application can more accurately extract a detail image containing only high-frequency texture and defects by subtracting the gray value of each pixel point in the original image from the background estimation value;

[0024] (2) The present application automatically marks the maximum entropy value in the maximum clustering cluster as the reference entropy based on the prior feature that the normal texture pixel points account for the most by clustering the information entropy of all pixel points in the detail image, without manually setting a threshold, and provides an adaptive texture complexity evaluation reference for subsequent parameter calculation;

[0025] (3) The present application guides the subsequent filtering processing according to the dynamic regularization parameter of each pixel point, so that the filtering behavior can dynamically adjust the strength of edge smoothing, improves the accuracy of the crystallizer inner surface defect detection, and overcomes the limitations of the traditional fixed parameter guided filtering. BRIEF DESCRIPTION OF DRAWINGS

[0026] Figure 1 is a flow chart schematically showing a cold bed crystallizer surface state evaluation method in the present application. DETAILED DESCRIPTION

[0027] With reference to the drawings of the embodiments of the present application, the technical solutions in the embodiments of the present application will be clearly and completely described. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments of the present application, all the other embodiments obtained by a person of ordinary skill in the art without creative effort should fall within the scope of the present application.

[0028] The specific implementation of the present application will be described in detail below with reference to the drawings.

[0029] The embodiments of the present application disclose a cold bed crystallizer surface state evaluation method, referring to Figure 1 , comprising steps S1-S5:

[0030] S1, collect the original image of the inner surface of the crystallizer.

[0031] It should be noted that the smoothness of the inner surface of the cold bed crystallizer has an important influence on the surface quality of the ingot. Once there is a defect in the inner surface, it will affect the forming quality. Therefore, the inner surface image of the crystallizer needs to be collected before the process operation to evaluate whether the surface state meets the standard, thereby reducing the rate of defective products.

[0032] Specifically, a rotary-lifting mechanism is installed in the area where the cold bed is placed in the crystallizer. A face array camera and a coaxial ring light source are arranged above the mechanism. Before the process operation, the mechanism is adjusted to a position parallel to the center plane of the crystallizer, and the inner surface image of the crystallizer is collected. The collected image is subjected to grayscale processing to obtain the grayscale value of each pixel point.

[0033] At this point, the original image of the inner surface of the crystallizer and the grayscale value of each pixel point are obtained.

[0034] S2, any pixel point in the original image is taken as a target pixel point, and the background estimation value of the target pixel point is calculated. The grayscale value of each pixel point is subtracted from the background estimation value to obtain the grayscale value of each pixel point in the detail image.

[0035] It should be noted that the crystallizer is usually a curved surface made of metal materials such as copper or aluminum. The metal material and the curved surface have high reflectivity under the light source, resulting in high-light areas in the collected image. In fact, these high-light areas represent low-frequency uneven lighting, i.e. slow and large-scale changes in lighting, which can mask the real detail features and affect the subsequent filtering effect. Therefore, the low-frequency component of the image, i.e. the background lighting, needs to be extracted to restore the enhanced image containing only high-frequency details.

[0036] Specifically, the neighborhood window size is , wherein, The local analysis range of each pixel point is defined, which is used to capture the gray scale change in the local range, and the value cannot be too small to avoid capturing the gray scale change, and the analysis process is to obtain other pixel points in the neighborhood window centered on each pixel point, is a single value; the present application will is set to 13, and the implementer can adjust it as appropriate; it should be noted that for the pixel points in the edge area that cannot obtain a complete neighborhood window, the gray scale value of the pixel points themselves is used for completion processing.

[0037] Considering that the normal inner wall texture of the crystallizer is composed of dense grooves, which appears as rapid alternation of gray scale values in the local range on the image, the average gray scale value in the local range will smooth out all high-frequency texture details; but if there is a defect area, its gray scale value may also be a low-frequency signal, to avoid this part of the signal from polluting the low-frequency light extraction, a weighted average method based on neighborhood statistical characteristics is introduced, which can smooth the high-frequency texture and exclude defect interference, and ensure accurate extraction of the pure background.

[0038] Any pixel point in the original image is taken as a target pixel point, and the median of the gray scale values of all neighborhood pixel points in the neighborhood window of the target pixel point is extracted; the weight value of each neighborhood pixel point is calculated, wherein the weight value is equal to the reciprocal of the absolute difference between the gray scale value of each neighborhood pixel point and the median; it should be noted that if the absolute difference is 0, it indicates that the pixel point is a normal pixel point in the neighborhood, and the weight value of the pixel point is taken as 1.

[0039] The gray scale values of all neighborhood pixel points are weighted and averaged according to the weight values to obtain the background estimation value of the target pixel point; the background estimation value satisfies the expression:

[0040]

[0041] In the formula, is the background estimation value of the target pixel point; is the gray scale value of the first neighborhood pixel point in the neighborhood window of the target pixel point; is the weight value of the first neighborhood pixel point in the neighborhood window of the target pixel point; is the index value and number of the pixel points in the neighborhood window of the target pixel point.

[0042] Among them, reflects the weight value of the first neighborhood pixel point in the neighborhood window of the target pixel point, and the weight value is inversely proportional to the absolute difference between the gray scale value of the first neighborhood pixel point and the median, and the smaller the weight value, the smaller the absolute difference between the gray scale value of the first The more the gray value of the neighboring pixel deviates from the normal level of the neighborhood, the higher the possibility that the neighboring pixel is a defect. By assigning a smaller weight to the neighboring pixel, the contribution of the defect to the calculation of the background estimation value is reduced. Conversely, a neighboring pixel with a gray value close to the median is assigned a larger weight value, thereby dominating the calculation of the background estimation value.

[0043] The background estimation value of each pixel in the original image is obtained.

[0044] The gray value of each pixel in the original image is subtracted from the background estimation value to obtain the gray value of each pixel in the detail image.

[0045] S3, for each pixel in the detail image: statistics of the probability of each gray level in the neighborhood window of the pixel; and determining the information entropy according to the probability of each gray level.

[0046] It should be noted that the high-frequency detail image obtained contains normal textures and abnormal defects, and the processing texture distribution on the inner surface of the actual crystallizer is relatively regular, i.e. the gray value changes stably in a local range, and the abnormal defect breaks the stability of the gray value in the original region. Therefore, the information entropy is evaluated by analyzing the distribution change of the gray value in the neighborhood window of each pixel in the detail image, wherein the local information entropy of the normal texture region is stable at a lower level, and the local information entropy of the defect region is higher, thereby providing data basis for subsequent filtering.

[0047] Specifically, for each pixel in the detail image: according to the gray value of all pixels in the neighborhood window of each pixel, the number of pixels of each gray level is counted, and the probability of each gray level is calculated, which is equal to the ratio of the number of pixels of each gray level to the number of all pixels in the neighborhood window.

[0048] According to the probability of each gray level in the neighborhood window of each pixel, the information entropy of each pixel is determined; the information entropy satisfies the expression:

[0049]

[0050] In the formula, is the information entropy of the i-th pixel; is the probability of the j-th gray level in the neighborhood window of the i-th pixel; is the index value and total number of the gray level; is the logarithmic function. wherein, reflects the information entropy of the i-th pixel in the detail image;

[0051] wherein, reflects the information entropy of the i-th pixel in the detail image; ​​​The complexity of grayscale value changes within the neighborhood window of a pixel. The larger the value, the more chaotic the grayscale distribution within the neighborhood window of that pixel, and the higher the probability of it being a defect; conversely, the more consistent the grayscale distribution within the neighborhood window of that pixel, and the higher the probability of it being a regular texture area.

[0052] At this point, the information entropy of each pixel in the detailed image has been obtained.

[0053] S4. Cluster all information entropy and select the cluster containing the most pixels as the texture cluster; use the maximum information entropy in the texture cluster as the baseline entropy; determine the dynamic regularization parameter of the pixel based on the difference between the information entropy of the pixel and the baseline entropy.

[0054] It should be noted that, considering that defects on the inner surface of the actual crystallizer are caused by wear, scratches, or deposits, the proportion of defective pixels is usually very small, and the vast majority of the inner wall area is normal texture. Therefore, clustering algorithms can be used to cluster the information entropy of each pixel in the acquired detail image. The cluster containing the most pixels has a higher probability of being a texture cluster. The regularization parameter of the guided filter is used to control the sensitivity of the filter to edges. Too large or too small a parameter value will affect the filtering effect. If it is set too small, too much texture will be retained; if it is set too large, real defects will be smoothed out, leading to inaccurate subsequent state evaluation. Therefore, the regularization parameter needs to be adaptively adjusted.

[0055] Specifically, the information entropy of all pixels is used as clustering samples, and DBSCAN clustering is performed to obtain the clustering results. It should be noted that the DBSCAN clustering algorithm is a well-known technology and will not be described in detail here. The cluster with the most pixels in the cluster is taken as the texture cluster. The maximum value of the information entropy of the pixels in the texture cluster is extracted as the baseline entropy.

[0056] The dynamic regularization parameter is determined based on the difference between the information entropy of each pixel and the baseline entropy; the dynamic regularization parameter satisfies the expression:

[0057]

[0058] In the formula, For the first Dynamic regularization parameters for each pixel; As the baseline entropy; For the first Information entropy of each pixel; It is the hyperbolic tangent function.

[0059] in, Reflecting the The deviation of the information entropy of the nth pixel from the baseline entropy; if this value is less than 0 and the smaller the value, the more likely the nth pixel is to be deviating from the baseline entropy. The complexity of the gray scale change of the region where the first pixel point is located is far more than the complexity of the maximum gray scale change of the normal texture region, which means that the first pixel point is more likely to belong to the abnormal defect region, and a smaller regularization parameter is needed to ensure that the edge information is retained in the subsequent processing. The complexity of the gray scale change of the region where the first pixel point is located is far more than the complexity of the maximum gray scale change of the normal texture region, which means that the first pixel point is more likely to belong to the abnormal defect region, and a smaller regularization parameter is needed to ensure that the edge information is retained in the subsequent processing. The complexity of the gray scale change of the region where the first pixel point is located is far more than the complexity of the maximum gray scale change of the normal texture region, which means that the first pixel point is more likely to belong to the abnormal defect region, and a smaller regularization parameter is needed to ensure that the edge information is retained in the subsequent processing. The function is to map the value of the first pixel point in the detail image to the range of 0-1. The function is to map the value of the first pixel point in the detail image to the range of 0-1.

[0060] At this point, the dynamic regularization parameter of each pixel point in the detail image is obtained.

[0061] S5, using the dynamic regularization parameter of each pixel point to filter the detail image, threshold segmentation is performed on the filtered image, and the percentage of defect pixel points in the segmented binary image is counted. In response to the percentage being greater than a set threshold, the state is determined to be unqualified.

[0062] It should be noted that the guided filter can smooth the image while preserving the strong edges in the guide image, thereby distinguishing between texture and defects. In order to ensure the processing effect of the filter, the filter is combined with the adaptively adjusted regularization parameter to ensure that the filter can both smooth the texture and retain the defects, and the effect of the filter has been significantly enhanced in the defect region, and the normal texture has been effectively smoothed, the gray scale contrast has been significantly improved, and the surface defect region can be segmented by threshold segmentation on the filtered image, thereby significantly improving the accuracy of surface state evaluation.

[0063] Specifically, the dynamic regularization parameter of each pixel point is replaced with the regularization parameter in the traditional guided filter algorithm to filter the detail image and obtain a filtered image. The filtered image is threshold segmented by Otsu method to obtain a segmented binary image, wherein the defect region is the region where the white pixel points are located, and the normal region is the region where the black pixel points are located. Therefore, the white pixel points are recorded as defect pixel points. The number of white pixel points is counted, and the percentage of the number of white pixel points in the total number of pixel points is calculated. If the percentage is greater than a preset threshold, the inner surface state of the crystallizer is determined to be unqualified, and an alarm is issued. Otherwise, the inner surface state of the crystallizer is determined to be qualified, and subsequent process operations can be performed.

[0064] The guided filter algorithm and the Otsu method are well-known technologies and will not be described here. The specific value of the preset threshold can be set according to the actual application scenario and requirements, and the value range of the preset threshold is [1%, 10%]. This range is set considering that actual surfaces are allowed to have minor flaws, but not significant damage. The present application sets the preset threshold to 5%.

[0065] The embodiment of the present application also discloses a cold bed crystallizer surface state evaluation system, comprising a processor and a memory, the memory storing computer program instructions, and the computer program instructions are executed by the processor to realize the cold bed crystallizer surface state evaluation method according to the present application.

[0066] The above system also comprises other components such as a communication bus and a communication interface which are well known to those skilled in the art, and the setting and functions thereof are known in the art, so they will not be described here.

Claims

1. A method for evaluating the surface condition of a cold bed crystallizer, characterized in that, include: Acquire the original image of the inner surface of the crystallizer; take any pixel in the original image as the target pixel, and extract the median gray value of all neighboring pixels in the neighborhood window of the target pixel; The weight value is determined based on the deviation of the gray value of each neighboring pixel from the median; the gray values ​​of all neighboring pixels are then weighted and averaged according to the weight value to obtain the background estimate of the target pixel. Subtract the estimated background value from the gray value of each pixel in the original image to obtain the gray value of each pixel in the detail image; For each pixel in the detailed image: Calculate the probability of each gray level within the neighborhood window of the pixel; determine the information entropy based on the probability of each gray level; cluster all information entropies and take the cluster containing the most pixels as the texture cluster; take the maximum information entropy in the texture cluster as the baseline entropy; determine the dynamic regularization parameter of the pixel based on the difference between the information entropy of the pixel and the baseline entropy. The detailed image is filtered using the dynamic regularization parameters of each pixel. The filtered image is then segmented by thresholding, and the percentage of defective pixels in the segmented binary image is counted. If the percentage is greater than a set threshold, the image is determined to be unqualified.

2. The method for evaluating the surface condition of a cold bed crystallizer according to claim 1, characterized in that, The weight value is equal to the reciprocal of the absolute difference between the gray value and the median of each neighboring pixel; and when the absolute difference is 0, the weight value is set to 1.

3. The method for evaluating the surface condition of a cold bed crystallizer according to claim 1, characterized in that, The estimated background value of the target pixel satisfies the expression: ； In the formula, The background estimate for the target pixel; Within the neighborhood window of the target pixel, the first The grayscale value of each neighboring pixel; Within the neighborhood window of the target pixel, the first The weight values ​​of each neighboring pixel; This refers to the index value and number of pixels within the neighborhood window of the target pixel.

4. The method for evaluating the surface condition of a cold bed crystallizer according to claim 1, characterized in that, The probability of each gray level within the neighborhood window of the statistical pixel includes: Based on the gray values ​​of all pixels within the neighborhood window of each pixel, the number of pixels at each gray level is counted, and the probability of each gray level is calculated. The probability is equal to the ratio of the number of pixels at each gray level to the number of pixels in the neighborhood window.

5. The method for evaluating the surface condition of a cold bed crystallizer according to claim 1, characterized in that, The information entropy satisfies the expression: ； In the formula, For the first Information entropy of each pixel; For the first Within the neighborhood window of the nth pixel The probability of each gray level; This refers to the index value and total number of gray levels; This is a logarithmic function.

6. The method for evaluating the surface condition of a cold bed crystallizer according to claim 1, characterized in that, The algorithm used to cluster all information entropy is the DBSCAN clustering algorithm.

7. The method for evaluating the surface condition of a cold bed crystallizer according to claim 1, characterized in that, The dynamic regularization parameters of the pixel satisfy the expression: ； In the formula, For the first Dynamic regularization parameters for each pixel; As the baseline entropy; For the first Information entropy of each pixel; It is the hyperbolic tangent function.

8. The method for evaluating the surface condition of a cold bed crystallizer according to claim 1, characterized in that, The threshold segmentation algorithm is the Otsu method, and the white pixels in the segmented binary image are defective pixels.

9. The method for evaluating the surface condition of a cold bed crystallizer according to claim 1, characterized in that, The percentage of defective pixels is equal to the percentage of white pixels in the segmented binary image out of all pixels.

10. A surface condition assessment system for a cold bed crystallizer, characterized in that, include: A processor and a memory, the memory storing computer program instructions that, when executed by the processor, implement a method for evaluating the surface condition of a cold bed crystallizer according to any one of claims 1-9.

Images