Blast furnace molten iron stream diameter on-line measuring method and system based on improved hough transform

Abstract

Embodiments of the present disclosure provide a blast furnace molten iron stream diameter online measurement method and system based on improved Hough transform. Applied to smelting intelligent manufacturing technology field, the method comprises: acquiring a real-time molten iron stream video stream of a blast furnace tapping hole site, pre-processing the molten iron stream video stream to obtain a group of first images; based on the non-rigid of the molten iron stream, setting that the molten iron stream boundary in each frame of the first image satisfies the parabolic characteristic, determining a dynamic region of interest in the first image according to the parabolic characteristic; using adaptive bilateral filtering to strengthen the edge of the dynamic region of interest, and performing edge detection on the dynamic region of interest after edge strengthening; using an improved Hough algorithm to extract two boundary parabolas of the continuous boundary of the molten iron stream; and calculating the vertical diameter between the continuous boundaries of the molten iron stream according to the boundary parabolas. The present disclosure has high intelligence degree, can automatically identify abnormal tapping state, and provides decision support for process adjustment.

Description

TECHNICAL FIELD

[0001] The present disclosure relates to the technical field of smelting intelligent manufacturing, and particularly relates to a blast furnace molten iron stream diameter online measurement method and system based on an improved Hough transform. BACKGROUND

[0002] Blast furnace tapping is an important link of blast furnace production, and continuous, orderly and stable slag-iron discharge can ensure the production efficiency of the blast furnace. The diameter of the molten iron stream is determined by the diameter of the drill bit selected for the tapping hole, and its change trend in the tapping process is affected by the slag-iron ratio, slag composition, ram quality, opening depth, furnace pressure and slag-iron fluidity. From the change of the slag-iron stream diameter in the tapping process, various parameters such as the slag-iron distribution and flow state in the furnace, the quality of the front sealing operation, and the quality of the ram can be calculated.

[0003] In the current blast furnace environment, the diameter of the molten iron stream is mainly estimated by manual visual observation, which has the following problems:

[0004] (1) The traditional visual measurement method has low precision and poor stability due to the harsh environment of high temperature, strong light and much dust in the blast furnace tapping site;

[0005] (2) There are strong subjectivity, low precision, and inability to continuously monitor, and the high-temperature and strong-radiation environment poses a threat to the health of the operators, posing a serious safety hazard.

[0006] In a patent with application number CN202110437293.X, a blast furnace tapping hole state monitoring method is disclosed, which obtains the size information of the tapping hole through the molten iron area and time parameters, but this method can only indirectly reflect the tapping hole state and cannot directly measure the stream diameter, and is greatly affected by strong light and dust. In another patent with application number CN201910388544.2, a blast furnace slag-iron state online detection method is disclosed, which processes the image of the iron channel in the tapping area through an imaging device, but can only identify three basic tapping states and cannot provide accurate stream diameter data.

[0007] The molten iron stream in the blast furnace has the characteristics of high temperature, strong light radiation, fast dynamic change, and complex surface texture, and the conventional edge detection and measurement algorithm cannot be directly applied. Moreover, the molten iron stream appears as a non-rigid parabolic boundary in the image, and its shape is in real-time change due to the influence of factors such as the shape of the tapping hole and the fluctuation of the furnace pressure, so it is impossible to realize real-time online measurement of the stream diameter. Therefore, it is of great industrial application value to develop a machine vision method that can accurately measure the diameter of the molten iron stream in the harsh environment of the blast furnace.

[0008] Based on the above problems, the present disclosure provides a blast furnace molten iron stream diameter online measurement method based on an improved Hough transform algorithm. SUMMARY

[0009] The present disclosure provides a solution to the problem that the diameter of a molten iron stream in a blast furnace cannot be calculated in real time and accurately due to non-rigidity in the prior art.

[0010] According to a first aspect of the present disclosure, there is provided a method for online measurement of the diameter of a molten iron stream in a blast furnace based on an improved Hough transform algorithm. The method comprises:

[0011] acquiring a real-time molten iron stream video stream at a tap hole of the blast furnace, pre-processing the molten iron stream video stream to obtain a set of first images;

[0012] based on the non-rigidity of the molten iron stream, setting the boundary of the molten iron stream in each frame of the first images to satisfy a parabolic characteristic, and determining a dynamic region of interest in the first images according to the parabolic characteristic;

[0013] performing edge enhancement on the dynamic region of interest using adaptive bilateral filtering, and performing edge detection on the dynamic region of interest after edge enhancement using an improved Canny operator to obtain a continuous boundary;

[0014] extracting two boundary parabolas of the continuous boundary of the molten iron stream using an improved Hough algorithm;

[0015] calculating a vertical diameter between the continuous boundary of the molten iron stream according to the boundary parabolas, and converting the vertical diameter into an actual diameter.

[0016] As described above, the aspect and any possible implementation manner further provide an implementation manner, and the pre-processing comprises: performing optical filtering on the collected molten iron stream video stream, and then performing image enhancement processing; the image enhancement processing comprises illumination compensation based on the Retinex theory and dust processing based on the symmetric zero-area algorithm.

[0017] As described above, the aspect and any possible implementation manner further provide an implementation manner, and the determining of the dynamic region of interest in the first images according to the parabolic characteristic comprises:

[0018] determining the molten iron stream region in each frame of the first images using an adaptive threshold method;

[0019] dynamically adjusting the molten iron stream region according to the parabolic characteristic to obtain the dynamic region of interest.

[0020] As described above, the aspect and any possible implementation manner further provide an implementation manner, and the performing of edge detection on the dynamic region of interest after edge enhancement using the improved Canny operator comprises:

[0021] calculating the gradient amplitude and direction of the dynamic region of interest;

[0022] determining the adaptive gradient high threshold and the adaptive gradient low threshold according to the mean value and the standard deviation of the gradient amplitude values;

[0023] obtaining candidate edge points by using the high threshold, the low threshold and non-maximum suppression, and connecting the candidate edge points and the points corresponding to the adaptive gradient high threshold and the adaptive gradient low threshold to form a continuous boundary.

[0024] According to the aspect and any possible implementation manner described above, further provided is an implementation manner that the obtaining candidate edge points by using the high threshold, the low threshold and non-maximum suppression comprises:

[0025] checking the gradient direction of each pixel point of the dynamic region of interest by using non-maximum suppression, and retaining the local maximum value point in the gradient direction to obtain a group of discrete candidate edge points.

[0026] According to the aspect and any possible implementation manner described above, further provided is an implementation manner that the improved Hough algorithm specifically comprises the following steps:

[0027] discretizing the parameter space of the parabola parameters;

[0028] for each candidate edge point, selecting the parameter of the parameter space in which the gradient direction and the tangent direction of the parabola are consistent to vote;

[0029] performing hierarchical clustering on the peaks formed after the parameter voting of the parameter space to obtain the boundary parabolas of the two sides of the molten iron stream;

[0030] performing fitting optimization on the boundary parabolas of the two sides of the molten iron stream.

[0031] According to the aspect and any possible implementation manner described above, further provided is an implementation manner that the performing hierarchical clustering on the peaks formed after the parameter voting of the parameter space to obtain the boundary parabolas of the two sides of the molten iron stream comprises:

[0032] performing 3D Gaussian smoothing on the parameters of the parameter space;

[0033] finding the local maximum value points satisfying V p V thresh , wherein V p represents the voting number of the peak point, and V thresh represents a preset threshold value for screening peaks;

[0034] taking each local maximum value point as an initial cluster, aggregating other clusters closest to the initial cluster until the spacing between the initial clusters is greater than a preset minimum separation threshold value;

[0035] The initial cluster after the aggregation is selected as the two boundary parabolas of the molten iron stream.

[0036] According to the aspect and any possible implementation manner described above, an implementation manner is further provided, and the vertical diameter between the continuous boundaries of the molten iron stream is calculated according to the boundary parabolas, including:

[0037] N horizontal coordinates are selected at equal intervals in the dynamic interested region, and the vertical coordinates on the two boundary parabolas corresponding to each horizontal coordinate are calculated respectively, wherein N is a non-zero natural number;

[0038] The absolute value of the difference between the two vertical coordinates is calculated to obtain the vertical diameter;

[0039] The vertical diameter is converted into an actual physical diameter according to a calibration parameter; the calibration parameter includes a pitch angle of a video acquisition device for obtaining a real-time molten iron stream video stream at a blast furnace tap hole site and a pixel size of the molten iron stream video stream;

[0040] The actual physical diameter is dynamically weighted according to a position to obtain an actual diameter.

[0041] According to the aspect and any possible implementation manner described above, an implementation manner is further provided, and the aspect further includes:

[0042] Time sequence feature extraction is performed on the time sequence data of the actual diameter; the time sequence feature includes a statistical feature, a dynamic feature and a frequency domain feature;

[0043] The time sequence feature is input into a constructed random forest classification model, and an abnormal state and a confidence degree are output.

[0044] According to a second aspect of the present disclosure, a blast furnace molten iron stream diameter online measurement system based on an improved Hough transform algorithm is provided, including:

[0045] A video stream acquisition module is configured to obtain a real-time molten iron stream video stream at a blast furnace tap hole site, and to perform preprocessing on the molten iron stream video stream to obtain a group of first images;

[0046] An interested confirmation module is connected with the video stream acquisition module and is configured to set a non-rigid of the molten iron stream based on the molten iron stream, so that the molten iron stream boundary in each frame of the first image satisfies a parabolic characteristic, and to determine a dynamic interested region in the first image according to the parabolic characteristic;

[0047] A continuous boundary confirmation module is connected with the interested confirmation module and is configured to perform edge enhancement on the dynamic interested region by using an adaptive bilateral filter, to perform edge detection on the dynamic interested region after the edge enhancement by using an improved Canny operator, and to obtain a continuous boundary;

[0048] A parabola confirmation module, connected with the continuous boundary confirmation module, is configured to extract two boundary parabolas of the continuous boundary of the molten iron stream by using an improved Hough algorithm.

[0049] A stream diameter calculation module, connected with the parabola confirmation module, is configured to calculate a vertical diameter between the continuous boundaries of the molten iron stream according to the boundary parabolas, and convert the vertical diameter into an actual diameter.

[0050] The present disclosure has the following beneficial effects:

[0051] The present disclosure removes dust from the collected molten iron stream video stream, first performs peak searching analysis by using a symmetric zero area algorithm, and judges whether each image of the video stream has a dust covered area; for the area with double peak characteristics, an improved OTSU algorithm is used for segmentation;

[0052] In the present disclosure, the boundary of the molten iron stream is set as a parabola model based on the non-rigid characteristics, and the dynamic region of interest of the molten iron stream is determined based on the characteristics that the molten iron stream is always in dynamic motion, focusing on the effective area and reducing the subsequent calculation amount;

[0053] The present disclosure considers that the overall trajectory of the molten iron stream has relative stability, but will be disturbed by the change of the internal gas pressure of the high horse, and will produce a certain jumping, so the improved Canny operator is used for edge detection, and the adaptive gradient threshold is dynamically calculated to enhance the overall adaptability;

[0054] The present disclosure extracts the boundary parabola by using an improved Hough algorithm, wherein the parameter space in the Hough algorithm is discretized according to the physical characteristics of the molten iron stream, reducing invalid voting; and the gradient direction consistency constraint is introduced, combined with the weighted voting strategy, which improves the fitting accuracy of the parabola while reducing the extraction time of the parabola;

[0055] The present disclosure uses an improved Hough transform algorithm, the stream diameter measurement error is less than 1mm, which meets the various control requirements derived by the industry subsequently; has strong real-time performance, the processing time of a single frame of image is less than 50ms, which can realize real-time online measurement of the diameter of the molten iron stream; has strong environmental adaptability, effectively overcomes the high temperature strong light and dust interference through optical filtering and image enhancement algorithm; the present disclosure is a completely non-contact measurement, which avoids the safety risk of manual approach to high temperature molten iron; the present disclosure has high intelligence, can automatically identify abnormal tapping state, and provides decision support for process adjustment.

[0056] It should be understood that the content described in the summary section is not intended to limit the key or important features of the embodiments of the present disclosure, nor to limit the scope of the present disclosure. Other features of the present disclosure will become apparent through the following description. BRIEF DESCRIPTION OF DRAWINGS

[0057] The above and other features, advantages, and aspects of embodiments of the present disclosure will become more apparent by describing in detail exemplary embodiments thereof with reference to the attached drawings. The following drawings are provided to assist in understanding the present disclosure and are not intended to limit the present disclosure. The same or similar components have the same reference numbers throughout the drawings, in which:

[0058] Figure 1 A flow chart of a blast furnace molten iron stream diameter online measurement method based on improved Hough transform is shown.

[0059] Figure 2 A schematic diagram of a blast furnace molten iron stream field photo is shown. DETAILED DESCRIPTION

[0060] To make the objectives, technical solutions, and advantages of the embodiments of the present disclosure clearer, the technical solutions in the embodiments of the present disclosure will be described clearly and completely below with reference to the drawings in the embodiments of the present disclosure. Obviously, the described embodiments are some but not all of the embodiments of the present disclosure. Based on the embodiments in the present disclosure, all other embodiments obtained by a person of ordinary skill in the art without creative work fall within the protection scope of the present disclosure.

[0061] In addition, the term "and / or" herein merely describes an association relationship of associated objects, which means that there can be three relationships, for example, A and / or B can mean that there are three cases of A alone, A and B together, and B alone. In addition, the character " / " herein generally represents an "or" relationship between the front and rear associated objects.

[0062] In the present disclosure, a blast furnace molten iron stream diameter online measurement method based on improved Hough transform is provided, as shown in Figure 1 , which comprises:

[0063] S1, acquiring a real-time molten iron stream video stream of a blast furnace tap hole site, pre-processing each frame image in the molten iron stream video stream to obtain a group of first images;

[0064] Referring to Figure 2 , the tap hole is surrounded by a high-temperature, high-brightness, and high-smoke environment, and the iron stream state cannot be directly observed by the naked eye on site, so a high-definition industrial camera needs to be used for video acquisition.

[0065] It should be noted that the high-definition industrial camera needs to be arranged on both sides of the molten iron stream, and the included angle between its optical axis and the flow direction of the stream should be as close to 90° as possible, and generally at least greater than 45°.

[0066] To observe the molten iron stream, a high-definition industrial camera needs to be installed at a position close to the taphole. The high-definition industrial camera must be cooled and regularly dusted, so a protective cover made of high-temperature-resistant material is used to surround the high-definition industrial camera, and a unique cooling structure is configured to ensure that the high-definition industrial camera can work continuously and stably. The temperature of the molten iron stream ranges from 1450°C to 1550°C. The molten iron stream appears dazzling white and has high-temperature infrared radiation. Conventional optical systems cannot directly collect and process it. The present disclosure configures a filter to suppress strong light and infrared rays.

[0067] The point where the device is installed needs to ensure that the molten iron stream can be observed in real time. After the high-definition industrial camera is installed, it starts collecting on-site images, i.e., the molten iron stream video stream.

[0068] The preprocessing of the molten iron stream video stream specifically includes the following contents:

[0069] The collected molten iron stream video stream is optically filtered to remove most visible light and near-infrared interference. Then, image enhancement processing is performed, which includes illumination compensation and dust processing.

[0070] (1) Illumination compensation

[0071] An illumination estimation and compensation algorithm based on the Retinex theory is used to eliminate the effects of uneven illumination. The formula is as follows:

[0072] R(x, y) = log I(x, y) - log [F(x, y) * I(x, y)],

[0073] where R(x, y) represents the final extracted reflection component, i.e., the target information after removing uneven illumination; I(x, y) represents the pixel value of the original image at coordinates (x, y), such as brightness or color channel value; F(x, y) represents the weight value of the Gaussian filter kernel at (x, y), which is used to simulate local illumination (low-frequency component); * represents the convolution operation of image I and Gaussian kernel F to obtain the estimation of the illumination component; x and y represent the spatial coordinates of the pixels in the image, i.e., the row and column positions of the pixels in the image; x usually represents the horizontal direction, i.e., the column index, and y usually represents the vertical direction, i.e., the row index.

[0074] (2) Dust removal

[0075] A symmetric zero-area algorithm is used for peak searching analysis to determine whether each image of the video stream has a dust-covered area. The traditional OTSU algorithm maximizes the inter-class variance σ 2 to determine the best segmentation threshold T, with the formula being:

[0076] ,

[0077] where current threshold the variance between foreground and background; represents the proportion of pixels in the foreground, i.e. the molten iron stream area or the area without dust, represents the proportion of pixels in the background, i.e. the dust or non-molten iron stream area; is the candidate gray threshold value traversed by the algorithm in the calculation process; is the proportion of pixels in the foreground, i.e. the proportion of the number of pixels with a gray value less than or equal to to the total number of pixels; is the proportion of pixels in the background, i.e. the proportion of the number of pixels with a gray value greater than to the total number of pixels; is the average gray value of the foreground, is the average gray value of the background.

[0078] For the area with bimodal characteristics, the low gray peak corresponds to the dust area, and the high gray peak corresponds to the non-dust area. The bimodal distribution of the dust image may cause the optimal threshold of the traditional OTSU algorithm to deviate to the low gray peak, i.e. the dust area, thereby misclassifying the non-dust area as the dust area.

[0079] Therefore, the present disclosure uses an improved OTSU algorithm for segmentation, and the specific method is:

[0080] First, the histogram is analyzed by the symmetric zero-area algorithm to identify the dust-covered area, assist in determining the bimodal position, improve the robustness against dust interference, and ensure that the segmentation result is more consistent with the actual molten iron stream area. By analyzing the histogram to determine the bimodal position, the threshold value search is limited between the valley bottom between the bimodal and the high gray peak, avoiding the interference of the low gray peak; a weight factor is added in the threshold calculation, giving the high gray area a higher priority, preferentially retaining the high gray area, and ensuring that the molten iron stream area is not misclassified by dust noise.

[0081] Through the improved OTSU algorithm, the threshold search range is limited, the high gray area weight is optimized, and the dust area is pre-identified, effectively solving the misclassification problem of the traditional OTSU algorithm caused by the bimodal distribution of the dust.

[0082] In addition, each preprocessed first image can be angle corrected and scale calibrated. The angle correction is used to adjust the direction of the image through geometric transformation, and the scale calibration is used to determine the proportional relationship between the pixels in the image and the actual physical size.

[0083] S2, based on the non-rigid of the molten iron stream, set the molten iron stream boundary in each frame of the first image as a parabolic characteristic, and determine the dynamic region of interest in the first image according to the parabolic characteristic;

[0084] The shape of the molten iron flow changes over time, and the traditional rigid boundary model cannot accurately describe the boundary thereof. Therefore, the boundary of the molten iron flow is set as a parabolic characteristic, so as to effectively capture the flow pattern under the gravity domination. The parameters in the parabolic equation can be changed over time to reflect the dynamic adjustment of the flow boundary. The parabolic parameters only need three, thereby reducing the calculation amount and improving the real-time performance.

[0085] The parabolic model, the cubic curve model and the straight line model are compared and verified through experiments. The parabolic model can balance between accuracy and efficiency.

[0086] The pressure in the taphole area changes instantaneously due to the influence of the blast furnace gas pressure, blast volume, molten iron composition, iron channel design and other factors, the flow rate of the molten iron flow is changed, so that the molten iron flow is in a jumping state, i.e. a non-steady state. The molten iron flow is still in a high-temperature state after falling into the main channel. It is difficult to accurately obtain the dynamic interested region of the molten iron flow due to the great disturbance.

[0087] The present disclosure realizes the extraction of the dynamic interested region of the molten iron flow through the following method.

[0088] (1) Determine the molten iron flow region

[0089] Based on the high-temperature characteristics of the molten iron flow, the thermal radiation intensity is high, and the brightness in the image is high. Therefore, the molten iron flow region can be divided out. The adaptive threshold method is used to determine the molten iron flow region in each frame of the first image. The threshold value is dynamically calculated according to the brightness characteristics of the local region of the image, and the image is divided into foreground and background. The foreground is the molten iron flow region, and specifically:

[0090] ;

[0091] wherein T is the segmentation threshold value, I is the input gray image matrix, which represents the high-temperature region image of the molten iron flow obtained by the camera; max(I) is the maximum gray value in the image I, which reflects the intensity of the high-brightness region in the image; OTSU(I) is the best segmentation threshold value T calculated by the OTSU algorithm, which is used to distinguish the background and the molten iron region; mean(I) is the average gray value of the image I, which reflects the overall brightness level; α, β and γ are dynamic weight coefficients, which are used to balance the contribution of different threshold components, satisfy the normalization constraint, and α+β+γ=1. Specifically, α is the weight of the maximum gray value, β is the weight of the OTSU threshold value, and γ is the weight of the average gray value.

[0092] (2) Dynamically adjust the molten iron flow region

[0093] The molten iron stream has two boundaries, so the corresponding boundary parabolas have two, and the two parabolas correspond to different points as starting point coordinates, respectively.

[0094] The tap hole diameter is a certain value between 40mm and 65mm according to the different blast furnace volume. The coordinate system is constructed in the first image, and the origin of the coordinate system is located at the upper left corner of the first image. The starting point is not the origin, and the coordinate system of each frame of the first image is the same.

[0095] The starting point coordinates P of the dynamic region of interest are determined respectively s :

[0096] P s =(i, 0.5(y t(i) +y b(i) )),

[0097] wherein i represents the horizontal coordinate, y t(i) is the vertical coordinate value of the upper boundary parabola at i, and y b(i) is the vertical coordinate value of the lower boundary parabola at i, and P s is the starting point, because the molten iron stream is always in dynamic motion, the dynamic region of interest is dynamically expanded along the parabolic direction, that is, the monitoring area is adjusted in real time according to the adaptive threshold method with the starting point coordinates, and the dynamic region of interest is obtained.

[0098] The position of the molten iron stream region in each frame of the first image is detected in real time by the adaptive threshold method, and the horizontal direction range of the molten iron stream region is obtained after segmentation, and i is the starting horizontal coordinate of the region in the image coordinate system. Because each frame is recalculated, the dynamic adjustment is performed with the change of the position of the molten iron stream, y t(i) and y b(i) are the vertical coordinate values of the upper and lower boundary parabolas at the horizontal coordinate i fitted in the current frame, and each fitting is independent, corresponding to a new starting point of fitting, so the starting point coordinates are not fixed values. In the present disclosure, the starting point coordinates are defined as the dynamic midpoint of the left boundary of the stream region, that is, the vertical coordinate midpoint of the upper and lower boundaries at x=i.

[0099] In summary, the above operations are performed on each frame of the first image, because the molten iron stream is in constant motion, so the dynamic region of interest can be obtained, the dynamic region of interest is determined in real time, the effective region is focused, the subsequent calculation amount is reduced, and the processing speed is improved.

[0100] S3, edge enhancement is performed on the dynamic region of interest by using adaptive bilateral filtering, and edge detection is performed on the dynamic region of interest after edge enhancement by using an improved Canny operator.

[0101] (1) Adopting adaptive bilateral filtering, the edge information is preserved while smoothing the noise, that is, edge enhancement is performed:

[0102] ,

[0103] wherein, represents the adaptive output pixel value; u and v both represent pixels, represents the neighborhood window centered on pixel u , v is any pixel in the window; represents the adaptive normalization factor, represents the spatial Gaussian kernel, the weight decays with distance; represents the control of the edge preservation degree, represents the spatial distance of pixels u and v , represents the spatial standard deviation, which is used to control the spatial smoothing degree; represents the intensity (range) Gaussian kernel, the weight decays with the difference of pixel value; represents the range standard deviation, which controls the intensity weight decay rate; represents the gray difference of pixels u and v , represents the gray of pixel v, represents the gray of pixel u.

[0104] Through the above steps, the edge is preserved while the high-temperature dust noise is suppressed. Due to the jumping nature of the molten iron stream, that is, the phenomenon of unstable fluctuation appearing in the flow process, multi-scale edge enhancement needs to be performed on the parabola constituted by the stream in order to facilitate subsequent edge calculation.

[0105] (2) The improved Canny operator is used for edge detection, which specifically includes the following steps:

[0106] 1. Based on the coordinate system constructed when determining the starting point coordinates of the parabola, the gradient amplitude and direction of the dynamic region of interest are calculated:

[0107] , ,

[0108] wherein, G x and G y represent the gradient components in horizontal and vertical directions respectively, G represents the gradient amplitude, represents the gradient direction.

[0109] 2. The adaptive gradient high threshold and the adaptive gradient low threshold are determined according to the mean value and the standard deviation of the gradient amplitude:

[0110] , ,

[0111] wherein, represents the mean value of gradient amplitude, reflecting the overall edge strength; represents the standard deviation of gradient amplitude, used to measure the dispersion degree of gradient distribution; k represents an adjustment coefficient, which can be adjusted according to the noise level; high represents an adaptive gradient high threshold value, which is a threshold value for certain edge points, dynamically adapts to strong light and dust interference, and improves the accuracy; low represents an adaptive gradient low threshold value, which is a threshold value for suspected edge points.

[0112] The molten iron stream is a high-temperature fluid, which is disturbed by the change of the internal gas pressure of the blast furnace and will produce a certain fluctuation, but the overall trajectory has relative stability. Through the dynamic calculation of T high and T high , it is easier to adapt to the severe fluctuations of the gradient in the high-temperature scene, and the setting of the adjustment coefficient enhances the adaptability.

[0113] 3. The candidate edge points are obtained by using the high threshold value, the low threshold value and non-maximum suppression, and the candidate edge points and the points corresponding to the high threshold value and the low threshold value are connected to form a continuous boundary.

[0114] The gradient direction of each pixel point in the dynamic region of interest is checked by using non-maximum suppression, and the local maximum value point in the gradient direction is retained. The local maximum value point with a gradient between the high threshold value and the low threshold value is selected as a group of discrete candidate edge points. Through non-maximum suppression, the thick line effect of the edge can be eliminated, such as changing a 3-pixel wide bright band into a 1-pixel wide point chain, so as to obtain an accurate single-pixel edge.

[0115] The obtained candidate edge points and the double threshold values are connected, and the discrete points are connected into a complete curve, that is, a continuous boundary is obtained.

[0116] S4, two boundary parabolas of the molten iron stream continuous boundary are extracted by using the improved Hough algorithm.

[0117] The improved Hough algorithm specifically includes the following steps:

[0118] (1) Discretization of the parameter space of the parabola parameters

[0119] Discretize the parameter space (a, b, c) of the general equation y=ax 2 +bx+c of the parabola.

[0120] The speed of the molten iron stream from the high-pressure furnace cylinder through the molten iron channel to the outside open space is limited, the trajectory curvature is small, and the initial speed direction is close to the horizontal direction when it is sprayed. Considering the above physical characteristics, the parameter range is constrained:

[0121] a∈[-0.01,0.01], b∈[-1,1], c∈[0,H].

[0122] Wherein, H is the height of the first image, the discretization step is set according to the accuracy requirement, and the present disclosure is not specifically limited.

[0123] (2) For each candidate edge point, select the parameter space parameter whose gradient direction and parabola tangent direction are consistent to vote;

[0124] The present disclosure adopts a weighted voting strategy based on local consistency, which has smaller calculation amount and is not easily disturbed by noise compared with the global voting of edge points in the traditional Hough transform.

[0125] Based on the coordinate system constructed in the foregoing, the local gradient direction of each edge point e i =(x i , y i ) is calculated ; In the parameter space, only the parabola parameters that meet the direction consistency are voted:

[0126] ,

[0127] Wherein, represents the tangent angle of the parabola j at x i , and represents the angle tolerance.

[0128] The voting weight of each edge point is set as follows:

[0129] w i =G i ·C i ,

[0130] Wherein, G i is the gradient amplitude of the edge point, and C i is the local curvature consistency factor in the neighborhood of the edge point. The voting weight adjusts the contribution degree of each edge point in the parameter space voting, so that the edge point that is more reliable and more consistent with the local geometric feature has a greater influence on the final result, thereby improving the detection accuracy and noise resistance.

[0131] (3) The peaks formed after the parameter voting of the parameter space are hierarchically clustered to obtain the boundary parabolas of the two sides of the molten iron stream.

[0132] 1. Perform 2D Gaussian smoothing on the parameters in the parameter space to suppress false peaks caused by noise, making the true peaks more prominent.

[0133] 2. Find the condition that satisfies V p >V thresh The local maximum point, where V p V represents the number of votes at the peak point, reflecting the probability that this point might be a true boundary. thresh This indicates the preset threshold for filtering peak values. Only peak values ​​with a number of votes greater than this threshold will be considered as potential boundaries. Furthermore, the distance between local maxima and the detected peak points must be greater than the preset minimum separation threshold to avoid repeated detection of the same parabola.

[0134] 3. Take each local maximum point as an initial cluster, and aggregate the other clusters that are closest to it, until the distance between each initial cluster is greater than the preset minimum separation threshold.

[0135] 4. Select the cluster with the most votes after aggregation as the boundary parabolas on both sides of the molten iron flow.

[0136] (4) Fit and optimize the boundary parabolas on both sides of the molten iron stream.

[0137] In this disclosure, the RANSAC algorithm is used for optimization:

[0138] Randomly sample the smallest subset of samples from local maxima, such as fitting a parabola with 3 points; use a nonlinear optimization algorithm to estimate the parabola parameters, calculate all errors to the parabola, mark points with errors less than a preset value as interior points, and count the number of interior points; iterate multiple times, select the model with the most interior points, and refit the boundary parabola using all interior points.

[0139] S5. Calculate the vertical diameter between the continuous boundaries of the molten iron stream based on the boundary parabola, and convert the vertical diameter into the actual diameter.

[0140] N x-coordinates are selected at equal intervals within the dynamically interested region. i Where N is a non-zero natural number; calculate the ordinates on the boundary parabolas corresponding to each abscissa, and calculate the absolute value of the difference between the two ordinates to obtain the vertical diameter D. i :

[0141] D i =y b (xi) -y t (xi) ,

[0142] Among them, y t(xi) Indicates the parabola at the upper boundary at x i The ordinate value at point y b(xi) Indicates the lower boundary parabola at xi the ordinate value of the vertical diameter at the i-th position, x i represents the corresponding abscissa value at the i-th position.

[0143] convert the vertical diameter into the actual physical diameter D according to the calibration parameters reali , the calibration parameters including the pitch angle of the video acquisition device for obtaining the real-time molten iron stream video stream at the blast furnace tap hole site and the pixel size of the molten iron stream video stream:

[0144] D reali =D i ·s·cosα,

[0145] wherein s is the pixel size in mm / pixel and a is the pitch angle of the industrial camera.

[0146] Considering the fluid characteristics of the molten iron stream, the diameter measurement values at different positions are dynamically weighted to solve the measurement errors caused by factors such as viewing angle inclination, edge noise, and fluid fluctuation. The calculation formula of dynamic weighting is:

[0147]

[0148] wherein, is the actual diameter finally obtained, represents the vertical diameter measurement value (pixel unit) at the i-th position, s represents the pixel size (mm / pixel) determined by camera calibration; a represents the pitch angle of the camera, represents the dynamic weight of the i-th position. The dynamic weight is determined by the position reliability and the stream characteristics. In a specific embodiment, the weight proportion of the position reliability is 60%, which is set according to the edge detection confidence and the viewing angle geometric correction, and the weight proportion of the stream characteristics is 40%, which is set according to the fluid stability and spatial continuity.

[0149] In addition, the present disclosure also includes abnormal state detection. First, time sequence features are extracted from the time sequence data of the actual diameter, including statistical features, dynamic features, and frequency domain features; wherein the statistical features include mean value (representing the average level of stream diameter), variance (measuring the degree of diameter fluctuation, and a sudden increase in variance may indicate that the stream is unstable or broken), skewness (process distribution of diameter value, whether large or small), and kurtosis (local extreme value); the dynamic features include short-term change rate (diameter change rate between specific frames), and acceleration (reflecting the acceleration / deceleration trend of diameter change); the frequency domain features include the energy proportion of main frequency components (extracting periodic fluctuation characteristics through Fourier transform).

[0150] A classification model based on random forest is constructed, the extracted time sequence features are input into the constructed classification model, and abnormal states such as "running large stream", "stuck focus", "spattering", and "running gas" and confidence are identified.

[0151] Based on the above technical scheme, the improved Hough transform algorithm is adopted, the flow diameter measurement error is less than 1mm, and the industrial subsequent derived various control requirements are met;

[0152] The present disclosure has strong real-time performance, and the processing time of a single frame of image is less than 50ms, so that the real-time online measurement of the molten iron flow diameter can be realized;

[0153] The present disclosure has strong environmental adaptability, and the high temperature strong light and dust interference can be effectively overcome through optical filtering and image enhancement algorithm;

[0154] The present disclosure has good safety, and is a completely non-contact measurement, which avoids the safety risk of manual approach to high temperature molten iron;

[0155] The present disclosure has high intelligent degree, and can automatically identify abnormal tapping state, thereby providing decision support for process adjustment.

[0156] The present disclosure also provides a blast furnace molten iron flow diameter online measurement system based on improved Hough transform, comprising the following modules:

[0157] A video stream acquisition module is used to acquire a real-time molten iron flow video stream of a blast furnace tapping hole site, pre-process the molten iron flow video stream, and obtain a group of first images;

[0158] An interest confirmation module is connected with the video stream acquisition module, and is used to set the non-rigid of the molten iron flow based on the molten iron flow, so that the molten iron flow boundary in each frame of the first image satisfies the parabolic characteristic, and the dynamic region of interest in the first image is determined according to the parabolic characteristic;

[0159] A continuous boundary confirmation module is connected with the interest confirmation module, and is used to utilize adaptive bilateral filtering to strengthen the edge of the dynamic region of interest, utilize an improved Canny operator to perform edge detection on the dynamic region of interest after edge strengthening, and obtain a continuous boundary;

[0160] A parabola confirmation module is connected with the continuous boundary confirmation module, and is used to utilize an improved Hough algorithm to extract two boundary parabolas of the continuous boundary of the molten iron flow;

[0161] A flow diameter calculation module is connected with the parabola confirmation module, and is used to calculate the vertical diameter between the continuous boundaries of the molten iron flow according to the boundary parabolas, and convert the vertical diameter into an actual diameter.

[0162] Other contents can refer to the blast furnace molten iron flow diameter online measurement method based on the improved Hough transform, which will not be repeated here.

[0163] The present disclosure also provides an apparatus comprising: one or more processors; and a storage storing one or more programs configured to, when executed by the one or more processors, cause the one or more processors to implement the steps of the above method.

[0164] It should be noted that, for the above-mentioned method embodiments, in order to simply describe, they are all expressed as a series of action combinations, but those skilled in the art should know that the present disclosure is not limited to the order of the actions described, because according to the present disclosure, certain steps can be performed in other order or simultaneously. Secondly, those skilled in the art should know that the embodiments described in the specification are all optional embodiments, and the actions and modules involved are not necessarily required by the present disclosure.

[0165] Program code for carrying out methods of the present disclosure can be written in any combination of one or more programming languages. The program code can be provided to a processor or controller of a general purpose computer, special purpose computer, or other programmable data processing apparatus to produce a machine, such that the program code, when executed by the processor or controller, produces the functions / operations specified in the flowcharts and / or the block diagrams. The program code can be entirely on the machine, partially on the machine, partially on the machine and partially on a remote machine or server, or entirely on a remote machine or server.

[0166] In the context of the present disclosure, a machine-readable medium can be a tangible medium that contains or stores a program for use by or in connection with an instruction execution system, apparatus, or device. The machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination of the foregoing. More specific examples of the machine-readable storage medium will include one or more of: a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0167] To provide for interaction with a user, the systems and techniques described here can be implemented on a computer having a display device for displaying information to the user and a keyboard and a pointing device, e.g., a mouse or a trackball, by which the user can provide input to the computer. Other kinds of devices can be used to provide for interaction with a user as well; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form, including acoustic, speech, or tactile input.

[0168] It should be understood that the various forms of flow shown above can be re-ordered, added to, or deleted from without departing from the scope of the present disclosure. For example, the steps recited in the disclosure can be performed in parallel, in series, or in different orders, as long as the desired results of the technology disclosed in the present disclosure are achieved, which is not limited herein.

[0169] The specific embodiments described above are not intended to be limiting, and any and all modifications, variations, and equivalents that are within the spirit and scope of the present disclosure are intended to be included herein. Other embodiments can be apparent to those of ordinary skill in the art from this disclosure and from the practice of the embodiments described herein.

Claims

1. A blast furnace molten iron stream diameter online measurement method based on improved Hough transform, comprising: acquiring a real-time molten iron stream video stream of a blast furnace tap hole site, pre-processing the molten iron stream video stream to obtain a set of first images; based on the non-rigid setting of the molten iron stream, setting the molten iron stream boundary in each frame of the first image to satisfy the parabolic characteristic, and determining the dynamic region of interest in the first image according to the parabolic characteristic; using adaptive bilateral filtering to edge enhance the dynamic region of interest, and using an improved Canny operator to edge detect the edge enhanced dynamic region of interest to obtain a continuous boundary; the improved Canny operator for edge detection of the edge enhanced dynamic region of interest, comprising: calculating the gradient amplitude and direction of the dynamic region of interest; determining the adaptive gradient high threshold and the adaptive gradient low threshold according to the mean and standard deviation of the gradient amplitude; using the high threshold, the low threshold and the non-maximum suppression to obtain candidate edge points, and connecting the candidate edge points and the points corresponding to the adaptive gradient high threshold and the adaptive gradient low threshold into a continuous boundary; wherein the use of high threshold, low threshold and non-maximum suppression to obtain candidate edge points comprises: using non-maximum suppression to check the gradient direction of each pixel point in the dynamic region of interest, and retaining the local maximum value point in the gradient direction, and selecting the local maximum value point with the gradient between the high threshold and the low threshold as a set of discrete candidate edge points; using an improved Hough algorithm to extract two boundary parabolas of the continuous boundary of the molten iron stream; wherein the improved Hough algorithm specifically comprises the following steps: discretizing the parameter space of the parabolic parameters; for each candidate edge point, selecting the parameters of the parameter space whose gradient direction and parabolic tangent direction are consistent to vote; after voting the parameters of the parameter space, performing hierarchical clustering on the formed peaks to obtain the two boundary parabolas of the molten iron stream; fitting and optimizing the two boundary parabolas of the molten iron stream; calculating the vertical diameter between the continuous boundaries of the molten iron stream according to the boundary parabolas, and converting the vertical diameter into an actual diameter.

2. The method of claim 1, wherein, The pre-processing includes optical filtering of the collected molten iron stream video stream, and then image enhancement processing; the image enhancement processing includes illumination compensation based on Retinex theory and dust processing based on symmetric zero area algorithm.

3. The method of claim 1, wherein, The determination of the dynamic region of interest in the first image according to the parabolic characteristic comprises: determining the molten iron stream region in each frame of the first image using an adaptive threshold method; dynamically adjusting the molten iron stream region according to the parabolic characteristic to obtain a dynamic region of interest.

4. The method of claim 1, wherein, The hierarchical clustering of the peaks formed after voting the parameters of the parameter space to obtain the two boundary parabolas of the molten iron stream comprises: performing 2D Gaussian smoothing on the parameters of the parameter space; Finding local maximum points satisfying V p V thresh p V thresh represents the threshold of the screening peak preset;​ taking each local maximum value point as an initial cluster, aggregating the nearest other clusters until the spacing between the initial clusters is greater than a preset minimum separation threshold; selecting the most voted initial cluster after aggregation as the two boundary parabolas of the molten iron stream.

5. The method of claim 1, wherein, The vertical diameter between the continuous boundaries of the molten iron stream is calculated according to the boundary parabolas, and the vertical diameter is converted into an actual diameter, comprising: N equally spaced x-coordinates are selected in the dynamic region of interest, and the y-coordinates on the boundary parabolas of the two boundaries corresponding to each x-coordinate are calculated respectively, wherein N is a non-zero natural number; The absolute value of the difference between the two y-coordinates is calculated to obtain the vertical diameter; The vertical diameter is converted into an actual physical diameter according to a calibration parameter; the calibration parameter includes the pitch angle of a video acquisition device for obtaining a real-time molten iron stream video stream at a blast furnace tap hole site and the pixel size of the molten iron stream video stream; The actual physical diameter is dynamically weighted according to the position to obtain an actual diameter.

6. The method of claim 1, wherein, Further comprising: Time series feature extraction is performed on the time series data constructed by the actual diameter; The time series features include statistical features, dynamic features, and frequency domain features; The time series features are input into a constructed random forest classification model to output an abnormal state and a confidence level.

7. A blast furnace molten iron stream diameter online measurement system based on an improved Hough transform algorithm, for implementing the method of any one of claims 1 to 6, comprising the following modules: A video stream acquisition module for obtaining a real-time molten iron stream video stream at a blast furnace tap hole site, and pre-processing the molten iron stream video stream to obtain a set of first images; An interest confirmation module connected with the video stream acquisition module, for setting that the boundaries of the molten iron stream in each frame of the first images satisfy the parabolic characteristic based on the non-rigid of the molten iron stream, and determining a dynamic region of interest in the first images according to the parabolic characteristic; A continuous boundary confirmation module connected with the interest confirmation module, for performing edge enhancement on the dynamic region of interest using adaptive bilateral filtering, and performing edge detection on the edge-enhanced dynamic region of interest using an improved Canny operator to obtain a continuous boundary; A parabola confirmation module connected with the continuous boundary confirmation module, for extracting two boundary parabolas of the continuous boundary of the molten iron stream using an improved Hough algorithm; A stream diameter calculation module connected with the parabola confirmation module, for calculating the vertical diameter between the continuous boundaries of the molten iron stream according to the boundary parabolas, and converting the vertical diameter into an actual diameter.

Images