A method for adaptively acquiring a sintering machine tail section analysis image

Abstract

This invention relates to an adaptive method for acquiring analytical images of the tail section of a sintering machine, comprising: image acquisition; image preprocessing; image clustering; image validity verification; and image cropping for analysis. The advantages of this invention are: it is non-manual, enabling automatic, real-time, long-term, and continuous effective image acquisition of the dynamic tail section of a sintering machine and automatically segmenting the sample to be analyzed; multi-spatial clustering effectively suppresses the low-brightness image blurring caused by tail section image jitter and flame flicker, and the image quality impact caused by the halo effect of the red-fire layer; image validity verification eliminates invalid images, extracting only clear and valid images per trolley cycle; the final analytical image extraction method suppresses the interference of loose slag red-fire and fractured red-fire layers on the analytical image extraction; and it can be integrated into the technical output of a sintering machine tail section image analysis system.

Description

Technical Field

[0001] This invention relates to the field of sintering machine technology, and in particular to an adaptive method for acquiring cross-sectional analysis images of the tail section of a sintering machine. Background Technology

[0002] In sintering production, the sintering machine tail section is the endpoint of the sintering process, and the tail section image provides the final and most comprehensive reflection of the sintering process conditions. Through the tail section image, characteristics of various process parameters such as over-burning, under-burning, moisture content, and FeO can be obtained. Therefore, accurate acquisition of the tail section image is a prerequisite and guarantee for accurate analysis of the results. The tail section image exhibits a certain periodicity, but this periodicity is related to various factors, such as changes in the trolley speed and the effects of over-burning and under-burning, which can all cause variations in this periodicity.

[0003] Currently, automatic image acquisition methods include manual triggering, trigger-based triggering, and timed triggering. For example, patent publication number CN101216881A, "An Automatic Image Acquisition Method and Device," uses a manual triggering method. This method can handle abnormal situations, eliminate various influencing factors, and acquire accurate images. However, its drawback is that it relies on manual operation, making automatic and continuous image acquisition impossible, and it cannot acquire dynamic images. Another example is the authorization announcement number CN2634428Y, "An Automatic Displacement Reading Device Based on an Image," which uses a detector and... The device consists of two parts and provides a method for dynamically, in real-time, continuously, and for long-term detection of the building facilities being inspected. However, this device is a time-triggered automatic reading device and cannot collect images of periodically occurring objects, nor can it automatically adjust to periodic changes in object characteristics. The "Highway Toll Collection System with Automatic Image Capture Function" with publication number CN1266231A uses a trigger-triggered acquisition method. This system can automatically, in real-time, continuously, and for long-term capture of vehicles entering and leaving the area. However, the disadvantage is that this system cannot be used in harsh production environments. Summary of the Invention

[0004] The purpose of this invention is to provide a method for adaptively acquiring cross-sectional analysis images of the tail section of a sintering machine, solving the problem of automatically, in real-time, for a long period of time, and continuously acquiring and adaptively segmenting and analyzing images of the tail section in the harsh environment of sintering production. Moreover, it can operate independently without relying on other basic automation methods and without contact.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] A method for adaptively acquiring cross-sectional analysis images of the tail section of a sintering machine, specifically including the following steps:

[0007] S1. Image Acquisition: Real-time images of the red section of the sintering machine tail are continuously acquired through an image acquisition device;

[0008] S2. Image preprocessing: Image preprocessing involves converting the acquired real-time image to grayscale and removing the background.

[0009] S3. Image clustering: The image processed in step S2 is divided into three clustering spaces using an adaptive clustering method: low grayscale clustering space, medium grayscale clustering space, and high grayscale clustering space.

[0010] S4. Image validity verification: including image validity feature extraction and image validity feature analysis;

[0011] S5. Image analysis and cropping: This involves analyzing the image boundary thresholds and cropping the image between the upper and lower boundary thresholds as the image to be analyzed.

[0012] In step S2, background removal includes absolute background removal and relative background removal. Absolute background removal removes images below a fixed threshold from the converted grayscale image, while relative background removal removes background images below a red-to-red boundary threshold.

[0013] In step S4, image validity feature extraction includes weak red flame index extraction, strong red flame index extraction, stomatal index extraction, steady state index extraction, location index extraction, and air redness index extraction, specifically including the following:

[0014] The weak red flame index extraction is to calculate the average gray value of the low gray value cluster space divided in step S3; the strong red flame index extraction is to calculate the average gray value of the medium gray value cluster space divided in step S3; and the porosity index extraction is to calculate the average gray value of the high gray value cluster space divided in step S3.

[0015] The location index extraction involves calculating the number of rows corresponding to the baseline of the red flame band in the image after processing in step S2, i.e., traversing each row of the image. The calculation formula is as follows:

[0016] Row highlight cluster count * Row highlight cluster width ①

[0017] The row highlight cluster count represents the count of points in the row that belong to the high grayscale cluster space, and the row highlight cluster width represents the difference between the maximum and minimum horizontal position of the points in the row that belong to the high grayscale cluster space. The row count is the maximum value of Formula ①.

[0018] Steady-state index extraction is the average of the absolute values ​​of the differences between the vertical position of a point on the image and the localization index after processing in step S2. Air red index extraction is the ratio of the number of points in the high grayscale cluster space divided in step S3 to the total number of points in the three cluster spaces.

[0019] In step 4, the image validity feature analysis involves numerically analyzing and binarizing the weak red flame index, strong red flame index, pore index, positioning index, steady state index, and red flame index, and then performing an AND operation to obtain the image validity analysis result. When the validity analysis result is equal to 1, the image is determined to be valid and can be further segmented; when the validity analysis result is equal to 0, the image is determined to be invalid, and the process returns to step S1.

[0020] In step S5, the image boundary threshold is analyzed: the analysis proceeds line by line from the baseline of the red zone upwards and to both sides, and the calculation formula is as follows:

[0021] Row low-brightness cluster count / (row low-brightness cluster count + row medium-brightness cluster count) < 0.9 ②

[0022] In Formula ②, the red-flame band baseline represents the row corresponding to the positioning index, the row low-brightness cluster count represents the count of points in the low-grayscale cluster space, and the row medium-brightness cluster count represents the count of points in the medium-grayscale cluster space; the first row value of the boundary position of Formula ② is taken as the boundary threshold of the analyzed image.

[0023] Compared with the prior art, the beneficial effects of the present invention are:

[0024] This method is non-manual, enabling automatic, real-time, long-term, and continuous effective image acquisition of the dynamic sintering machine tail section and automatic segmentation of the sample to be analyzed. Multi-spatial clustering effectively suppresses the low-brightness image blurring caused by tail section image jitter and flame flicker, as well as the image quality impact caused by the halo effect of the red-fire layer. Image validity verification eliminates invalid images, extracting only clear and valid images per trolley cycle. The final image extraction method effectively suppresses interference from loose slag red-fire and fractured red-fire layers on the extracted images. It can be integrated into the sintering machine tail section image analysis system for technical output and can also be applied to other application scenarios requiring periodic image acquisition to achieve corresponding system technical output. It accurately acquires the images to be analyzed, thereby reducing the amount of data analyzed, improving system operating efficiency, reducing energy consumption, and enhancing the effectiveness of the tail section image detection system. Attached Figure Description

[0025] Figure 1 This is a flowchart of the method for detecting the validity of the cross-sectional image of the tail section of a sintering machine.

[0026] Figure 2 A schematic diagram of the image cropping results. Detailed Implementation

[0027] The present invention will now be described in detail with reference to the accompanying drawings, but it should be noted that the implementation of the present invention is not limited to the following embodiments.

[0028] The following embodiments are implemented based on the technical solution of the present invention, providing detailed implementation methods and specific operation processes. However, the scope of protection of the present invention is not limited to the following embodiments. Unless otherwise specified, the methods used in the following embodiments are conventional methods.

[0029]

Example 1

[0030] See Figure 1 A method for adaptively acquiring cross-sectional analysis images of the tail section of a sintering machine, specifically including the following steps:

[0031] Step S1, Image Acquisition: The image acquisition device continuously acquires real-time images of the red cross section of the sintering machine tail. The acquisition cycle can be fixed or vary according to the production cycle.

[0032] Step S2, Image Preprocessing: This involves converting the acquired raw image to grayscale and removing the background. Background removal includes absolute background removal and relative background removal. Absolute background removal removes images below a fixed threshold in the converted grayscale image. The fixed threshold is 60. Relative background removal removes background images below the red and red boundary threshold.

[0033] The formula for calculating the red-fire boundary threshold is as follows:

[0034] Red Fire Boundary Threshold = V avg-step1 -V sdv-step1 ×K hbv ①

[0035] In formula ①, V avg-step1 V represents the average grayscale value of the remaining image after absolute background removal. sdv-step1 K is the standard deviation of the gray levels of the remaining image after absolute background removal. hbv The coefficient for calculating the red-fire boundary threshold is 0.5.

[0036] Step S3, Image Clustering: The image processed in step S2 is divided into three clustering spaces using an adaptive clustering method: low grayscale clustering space, medium grayscale clustering space, and high grayscale clustering space.

[0037] Adaptive clustering methods typically employ fuzzy C-means clustering.

[0038] Step S4, Image Validity Verification, includes two processes: Image Validity Feature Extraction and Image Validity Feature Analysis.

[0039] Image validity feature extraction includes weak redness index extraction, strong redness index extraction, stomatal index extraction, steady-state index extraction, location index extraction, and air redness index extraction, specifically including the following:

[0040] 1) The weak red flame index extraction is the average gray value of the low gray-level cluster space divided in step S3, and the calculation formula is as follows:

[0041]

[0042] In Formula ②, f(x,y) is the gray value of point (x,y) on the image, and Ns is the number of points in the low gray-level clustering space;

[0043] 2) The strong redness index is extracted by taking the average gray level of the mid-gray level cluster space divided in step S3. The calculation formula is as follows:

[0044]

[0045] In formula ③, Nm is the number of points in the medium gray-scale clustering space;

[0046] 3) The stomatal index is extracted by taking the average gray level of the high-gray-level cluster space divided in step S3. The calculation formula is as follows:

[0047]

[0048] In formula ④, Nl is the number of points in the high grayscale clustering space;

[0049] 4) The positioning index extraction is the number of rows corresponding to the red fire band baseline on the image after step S2. That is, it traverses each row of the image and takes the row count with the maximum value of (row highlight cluster count * row highlight cluster width). Here, the row highlight cluster count represents the count of points in the high grayscale cluster space of the row, and the row highlight cluster width represents the difference between the maximum and minimum values ​​of the horizontal position of the points in the high grayscale cluster space of the row.

[0050] 5) The steady-state index extraction is calculated by averaging the absolute values ​​of the differences between the vertical position of a point in the image and the localization index after processing in step S2. The calculation formula is as follows:

[0051]

[0052] In formula ⑤, N A This represents the number of points in the preprocessed image.

[0053] 6) The air red index is extracted as the ratio of the number of points in the high-grayscale cluster space divided in step S3 to the total number of points in the three cluster spaces. The calculation formula is as follows:

[0054]

[0055] Image validity feature analysis is used to determine the weak red flame index, strong red flame index, stomatal index, steady state index, positioning index and red flame index, and to determine the determination result by binarization. Specifically, it includes the following:

[0056] 1) Weak Red Flame Index Analysis

[0057]

[0058] In formula ⑥, D 弱红火 Generally, 100 is used, U 弱红火 Generally, 160 is used;

[0059] 2) Strong Red Flame Index Analysis

[0060]

[0061] In formula ⑦, D 强红火 Generally, 160 is chosen, U 强红火 Generally, 200 is used;

[0062] 3) Stomatal index analysis

[0063]

[0064] In formula ⑧, D 气孔 Generally, 220 is chosen, U 气孔 Generally, 255 is used;

[0065] 4) Positioning Index Analysis

[0066]

[0067] In formula ⑨, U 定位 Generally, the vertical position of the upper edge of the red-hot zone in the image is taken when the trolley flips over and the sinter falls, with a value of 500; D 定位 Generally, after the sintered ore falls, when the cross-section of the red-fire layer of the next vehicle can be clearly seen, the average value of the upper and lower edges of the red-fire zone is taken as 50. These two values ​​are affected by the camera placement and camera resolution and need to be set according to the actual situation.

[0068] 5) Steady-state index analysis

[0069]

[0070] In formula 10, D 稳态 Generally, 10 is taken, U 稳态 100 can be selected. This value is affected by the camera resolution and needs to be set according to the actual situation.

[0071] 6) Analysis of the redness index

[0072]

[0073] formula In the middle, D 气红 Generally, 0.1 is taken, U 气红 0.5 is acceptable;

[0074] The formula for calculating the effectiveness analysis results is as follows:

[0075] When the validity analysis result is equal to 1, the image is considered valid and can be further segmented; when the validity analysis result is equal to 0, the image is considered invalid and the process returns to step S1.

[0076] Step S5: Image analysis and cropping: This involves determining the image boundary thresholds and then cropping the image between the upper and lower boundary thresholds as the image to be analyzed.

[0077] See Figure 2 The method for determining the boundary threshold of the analysis image is to analyze line by line from the red fire band baseline upwards and downwards, and take the first line value of the boundary position where (row low brightness cluster count / (row low brightness cluster count + row medium brightness cluster count)) < 0.9 as the boundary threshold of the analysis image. Here, the red fire band baseline represents the line corresponding to the positioning index, the row low brightness cluster count represents the count of points in the low grayscale cluster space of the line, and the row medium brightness cluster count represents the count of points in the medium grayscale cluster space of the line.

[0078] The final image to be analyzed is shown below:

[0079] Analyze the image = image f(x,y) ((x,y)∈Preprocessed image, upper boundary threshold≤y≤low boundary threshold).

[0080] This invention is non-manual, capable of automatically, in real-time, long-term, and continuous effective image acquisition of the tail section of a dynamic sintering machine and automatically segmenting the sample to be analyzed. Through multi-spatial clustering, it effectively suppresses the low-brightness image blurring caused by tail section image jitter and flame flicker, as well as the image quality impact caused by the halo effect of the red-fire layer. Image validity verification eliminates invalid images, extracting only clear and valid images per trolley cycle. The final image extraction method effectively suppresses the interference of loose slag red-fire and fractured red-fire layers on the extracted images. It can be integrated into the sintering machine tail section image analysis system for technology output, and can also be applied to other application scenarios requiring periodic image acquisition to achieve corresponding system technology output. It accurately acquires the images to be analyzed, thereby reducing the amount of data analyzed, improving system operating efficiency, reducing energy consumption, and enhancing the effectiveness of the tail section image detection system.

Claims

1. A method for adaptively acquiring cross-sectional analysis images of the tail section of a sintering machine, characterized in that, Specifically, the following steps are included: S1. Image Acquisition: Real-time images of the red section of the sintering machine tail are continuously acquired through an image acquisition device; S2. Image preprocessing: Image preprocessing involves converting the acquired real-time image to grayscale and removing the background. S3. Image clustering: The image processed in step S2 is divided into three clustering spaces using an adaptive clustering method: low grayscale clustering space, medium grayscale clustering space, and high grayscale clustering space. S4. Image validity verification: including image validity feature extraction and image validity feature analysis; S5. Image analysis and cropping: This involves analyzing the image boundary thresholds and cropping the image between the upper and lower boundary thresholds as the image to be analyzed. The image validity feature extraction includes weak red flame index extraction, strong red flame index extraction, stomatal index extraction, steady state index extraction, positioning index extraction, and air redness index extraction, specifically including the following: The weak red flame index extraction is to calculate the average gray value of the low gray value cluster space divided in step S3; the strong red flame index extraction is to calculate the average gray value of the medium gray value cluster space divided in step S3; and the porosity index extraction is to calculate the average gray value of the high gray value cluster space divided in step S3. Steady-state index extraction is the average of the absolute values ​​of the differences between the vertical position of a point in the image and the localization index after processing in step S2. Air red index extraction is the ratio of the number of points in the high grayscale cluster space divided in step S3 to the total number of points in the three cluster spaces. The location index extraction involves calculating the number of rows corresponding to the baseline of the red flame band in the image after processing in step S2, i.e., traversing each row of the image. The calculation formula is as follows: Row highlight cluster count * Row highlight cluster width ① In Formula ①, the row highlight cluster count represents the count of points in the row that belong to the high grayscale cluster space, and the row highlight cluster width represents the difference between the maximum and minimum horizontal position values ​​of the points in the row that belong to the high grayscale cluster space. The row count is the maximum value of Formula ①. The image boundary threshold for analysis is calculated by analyzing line by line from the baseline of the red zone upwards and downwards, using the following formula: (Row low-brightness cluster count / (Row low-brightness cluster count + Row medium-brightness cluster count)) < 0.9 ② In Formula ②, the red-flame band baseline represents the row corresponding to the positioning index, the row low-brightness cluster count represents the count of points in the low-grayscale cluster space, and the row medium-brightness cluster count represents the count of points in the medium-grayscale cluster space; the first row value of the boundary position of Formula ② is taken as the boundary threshold of the analyzed image.

2. The method for adaptively acquiring cross-sectional analysis images of the sintering machine tail section according to claim 1, characterized in that, In step S2, the background removal includes absolute background removal and relative background removal. Absolute background removal removes images below a fixed threshold in the grayscale image, while relative background removal removes background images below a red-fire boundary threshold.

3. The method for adaptively acquiring cross-sectional analysis images of the sintering machine tail section according to claim 1, characterized in that, In step S4, the image validity feature analysis involves numerically analyzing and binarizing the weak red flame index, strong red flame index, pore index, positioning index, steady state index, and red flame index, and then performing an AND operation to obtain the image validity analysis result. When the validity analysis result is equal to 1, the image is determined to be valid and can be further segmented. If the validity analysis result is 0, the image is determined to be invalid, and the process returns to step S1.

Images