A steel billet flower head automatic identification system and method

By segmenting, preprocessing and annotating the billet morphological image data, combined with the convolutional neural network recognition model, the problem of insufficient identification of billet fine flower head defects and edge defects in traditional technology is solved, achieving a more efficient and accurate recognition effect.

CN118570244BActive Publication Date: 2024-11-15SHANDONG PROVINCE QINGZHOU ELECTRIC TRANSFORMER PLANT
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411061336.9
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-08-05
Publication Date
2024-11-15
Estimated Expiration
2044-08-05

AI Technical Summary

Technical Problem

Traditional billet automatic identification technology is difficult to accurately identify the fine flower head defects of the billet, and the subtle defects of the edge profile are insufficient, resulting in low efficiency and poor accuracy.

Method used

By obtaining billet morphological image data, segmenting, preprocessing and annotating, generating final morphological data, and using the convolutional neural network identification model to extract, locate and classify high-level features, output classification and annotation results, and generate billet flower head data report.

Benefits of technology

The precise identification of fine flower head defects of steel billets and edge defects is realized, and the efficiency and accuracy of the automatic identification system are improved.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118570244B_ABST
    Figure CN118570244B_ABST
Patent Text Reader

Abstract

The present invention relates to the technical field of data recognition, and discloses a system and method for automatic recognition of steel billet flower heads. The method for automatic recognition of steel billet flower heads obtains steel billet shape data by scanning the steel billet shape, which is convenient for subsequent recognition of image data features, and then segments the steel billet shape data to obtain initial shape data, pre-processes the initial shape data one by one to obtain secondary shape data, and annotates the secondary shape data one by one, and integrates the annotated data into final shape data, and then inputs the final shape data into a recognition model. The recognition model can accurately identify subtle steel billet flower head defects and subtle steel billet edge defects, outputs classification and annotation results through the recognition model, and finally generates a steel billet flower head data report based on the classification results, thereby making the automatic recognition of steel billet flower heads more accurate and efficient.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of data recognition, and more specifically, to a system and method for automatically identifying a steel billet flower head. Background Art

[0002] Data recognition technology plays an increasingly important role in modern society, and its application field is very wide. Therefore, in the process of billet casting, data recognition technology is used to automatically identify the billet head. The existing data recognition technology has defects in identifying billet head. The traditional billet automatic recognition technology is difficult to identify various types of subtle head defects, and the subtle defects of the billet edge contour are often ignored and cannot be identified, which makes the overall efficiency of the traditional automatic recognition technology low and the accuracy poor.

[0003] Therefore, the present invention obtains the steel billet morphology image data, performs flower head recognition model training after data processing, and obtains the training results to determine the direction of the next experimental operation.

[0004] The steel billet morphology image data is segmented to obtain initial morphology data, the initial morphology data are preprocessed one by one to obtain secondary morphology data, the secondary morphology data are labeled one by one, and the labeled data are integrated into final morphology data, the final morphology data is input into the recognition model and the result is output for related research.

[0005] In summary, how to identify subtle flower head defects in steel billets more accurately and efficiently is a problem that needs to be solved and optimized urgently in the steel billet automatic identification system. Summary of the invention

[0006] The present invention provides a system and method for automatically identifying steel billet flower heads, which solve the technical problem in the related art that it is difficult to accurately and efficiently identify subtle flower head defects in steel billets.

[0007] In a first aspect, a steel billet flower head automatic identification system is provided, the system comprising:

[0008] Data acquisition module: used to scan the shape of the steel billet to obtain the shape data of the steel billet;

[0009] Data processing module: it is used to segment the steel billet morphology data to obtain initial morphology data, pre-process the initial morphology data one by one to obtain secondary morphology data, annotate the secondary morphology data one by one, and integrate the annotated data into final morphology data;

[0010] Flower head recognition module: It is used to input the final morphological data into the recognition model, and the recognition model includes:

[0011] Extraction layer: Based on the input final morphological data, high-level feature data is gradually extracted through multiple convolutional layers;

[0012] Positioning layer: performing comparison and positioning based on the extracted high-level feature data;

[0013] Classification layer: input the high-level feature data after positioning, identify it, classify it according to the category of the sub-data in the final morphological data, and output the classification and annotation results;

[0014] Report generation module: It is used to generate a steel billet head data report based on the classification results.

[0015] In a second aspect, a method for automatically identifying a steel billet flower head comprises:

[0016] Scan the billet morphology to obtain billet morphology data, including:

[0017] Adjust the current, voltage and scanning parameters of industrial X-ray scanning equipment to obtain high-resolution scanners;

[0018] The billet morphology is scanned in multiple directions by industrial X-rays based on a high-resolution scanner to obtain CT projection image data of the billet morphology.

[0019] Further, the billet morphology data is segmented to obtain initial morphology data, the initial morphology data are pre-processed one by one to obtain secondary morphology data, the secondary morphology data are annotated one by one, and the annotated data are integrated into final morphology data, including:

[0020] The billet morphology CT projection image data is segmented by setting multiple thresholds, and the segmented billet morphology CT projection image data is equally divided to obtain initial morphology data;

[0021] Selecting RGB color models for the initial morphological data one by one, and using the G+R component in the RGB color model and combining the threshold to binarize the color image, so as to obtain a binary image with clear contour and internal texture;

[0022] A 31x31 template is used and combined with a threshold to filter the image to obtain denoised image data;

[0023] The sum of the grayscale values ​​of pixels in a 2x2 template is used in combination with the edge detection method of the threshold interval to obtain a refined and continuous image edge;

[0024] Strengthening the edge contour of the steel billet obtained based on edge detection, combining the image edge, and integrating it into sub-morphological data;

[0025] The secondary morphological features are annotated one by one to obtain the annotated data, and the annotated data are integrated into the final morphological data.

[0026] Further, the edge contour of the steel billet obtained based on edge detection is enhanced, combined with the image edge, and integrated into sub-morphological data, including:

[0027] The skeleton method is combined with the standard deviation of the gray value in the circle area, and the boundary points are iterated by linear interpolation to extract the edge contour vector of the steel billet.

[0028] Linearly fitting edge contour segments of the steel billet edge contour vectors, and sharpening the linearly fitting edge contour segments to increase the gray gradient of the edge;

[0029] Based on the grayscale gradient, the horizontal and vertical edges of the steel billet are divided by sub-pixel using the Canny algorithm to obtain the steel billet contour defect data;

[0030] The edge defect data is combined with the image edge data to form sub-morphological data.

[0031] Furthermore, the sub-morphological features are annotated one by one to obtain annotated data, and the annotated data are integrated into the final morphological data, including:

[0032] Extracting the steel billet profile defect data and feature vectors of the image edge to obtain edge feature data;

[0033] Based on the edge feature vector The analysis is performed, where J is the objective function representing the sum of square errors within the cluster; k represents the number of clusters; represents the i-th cluster, The i-th cluster center, x represents a sample point in the data set; Represents the sample point x to the cluster center The square of the Euclidean distance of to obtain the analysis results;

[0034] Automatically labeling different classes of the edge feature vectors based on the analysis results specifically includes:

[0035] Set to detect edge defects with protrusions or depressions larger than 0.13mm and mark them; obtain edge feature defect information through edge feature vector texture and mark them; mark them according to whether the edge feature vector type is common or uncommon;

[0036] The labeled data are integrated into final morphological data, and the three sub-data in the final morphological data respectively represent the three types of labeled edge feature vectors.

[0037] Further, the final morphological data is input into the recognition model, including:

[0038] Input: high-level feature data subsample set ;Label ;

[0039] Initialization: Identify the model rate η, set the number of convolution operation neurons, and initialize the connection weights and biases between the model networks to a small random threshold. Each training sample is represented by a sample and a corresponding label sequence in the form of x, t; x is the network input value vector and t is the target output value.

[0040] Furthermore, the final morphological data is input into the recognition model, which also includes:

[0041] Extraction layer: respectively intercept n sub-images of fixed size from the input final morphological data, and perform multi-layer convolution operations on the n sub-images according to the size and number of convolution kernels to extract high-level feature data;

[0042] Positioning layer: sub-sampling the high-level feature data, removing interference information by reducing the resolution, and converting it into low-resolution features using a weighted sum operation to locate the high-level feature data;

[0043] Classification layer: input the high-level feature data after positioning, identify it, classify it according to the category of the sub-data in the final morphological data, and output the classification and annotation results.

[0044] The sampling set The data is input to the extraction layer. After the extraction layer has undergone data iteration, it perceives the sample set through each neuron, and then passes it to the positioning layer to integrate the local information to obtain the overall information. The overall information is then passed to the classification layer for clustering recognition. Finally, the sample set is classified according to the label. Classification output.

[0045] Further, the classification result is used to generate a billet head data report, including:

[0046] Performing statistics on the output classification results to obtain statistical data;

[0047] Reconstructing the CT projection image data based on the high-level feature data into three-dimensional X-ray image data to obtain a three-dimensional image of the billet flower head;

[0048] Use visualization tools to generate charts based on statistical data and three-dimensional image data of billet heads to display the types and distribution of head defects;

[0049] Prepare a data report based on the acquired flower head absence information and save the report in PDF format.

[0050] In a third aspect, a steel billet flower head automatic identification system comprises:

[0051] one or more data identification processors;

[0052] A data identification processor is used to identify data in one or more programs. When the one or more programs are executed and identified by the one or more processors, the one or more processors implement the steps of the automatic identification method for the steel billet flower head.

[0053] In a fourth aspect, a computer-readable storage medium stores a program, which, when executed by a processor, implements the steps of a method for automatically identifying a steel billet flower head.

[0054] The beneficial effects of the present invention are:

[0055] The above scheme of the present invention obtains billet morphology data by scanning the billet morphology, which is convenient for subsequent identification of image data features, and then segments the billet morphology data to obtain initial morphology data, pre-processes the initial morphology data one by one to obtain secondary morphology data, and annotates the secondary morphology data one by one, and integrates the annotated data into final morphology data, and then inputs the final morphology data into the recognition model. The recognition model can accurately identify subtle defects in the billet and subtle defects on the edge of the billet, and outputs the classification and annotation results through the recognition model. Finally, the classification results are used to generate a billet head data report, thereby making the automatic identification of billet heads more accurate and efficient. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] The accompanying drawings are used for further understanding of the present invention and are used to explain the present invention together with the embodiments of the present invention, but do not constitute a limitation of the present invention.

[0057] Figure 1 Schematic diagram of a steel billet flower head automatic identification system provided by an embodiment of the present invention;

[0058] Figure 2 It is a principle block diagram of each module in the steel billet flower head automatic identification system provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0059] The subject matter described herein will now be discussed with reference to example implementations. It should be understood that the discussion of these implementations is only to enable those skilled in the art to better understand and implement the subject matter described herein, and the functions and arrangements of the elements discussed may be changed without departing from the scope of protection of the contents of this specification. Various examples may omit, replace, or add various processes or components as needed. In addition, the features described in some examples may also be combined in other examples.

[0060] like Figure 1 As shown, in this embodiment, a steel billet flower head automatic identification system 20 is provided, and the system includes the following modules:

[0061] Data acquisition module 201: used to scan the shape of the steel billet to obtain the shape data of the steel billet;

[0062] Data processing module 202: used for segmenting the steel billet morphology data to obtain initial morphology data, pre-processing the initial morphology data one by one to obtain secondary morphology data, annotating the secondary morphology data one by one, and integrating the annotated data into final morphology data;

[0063] Flower head recognition module 203: It is used to input the final morphological data into a recognition model, and the recognition model includes:

[0064] Extraction layer: Based on the input final morphological data, high-level feature data is gradually extracted through multiple convolutional layers;

[0065] Positioning layer: performing comparison and positioning based on the extracted high-level feature data;

[0066] Classification layer: input the high-level feature data after positioning, identify it, classify it according to the category of the sub-data in the final morphological data, and output the classification and annotation results;

[0067] The report generation module 204 is used to generate a steel billet head data report based on the classification result.

[0068] In a preferred embodiment of the present invention, the data acquisition module 201 scans and collects image data of the billet head through an X-ray scanner to ensure that comprehensive head information is obtained; the collected data will be transmitted to the cloud storage system in real time for further processing, and real-time data collection and transmission ensure that the system can respond and process quickly; the data processing module 202 pre-processes the collected raw data, including image denoising, enhancement, grayscale and other operations to improve image quality and remove redundant information, and then extracts key features of the billet head, such as edge contour, texture pattern, etc., through edge detection, texture analysis and other algorithms. Through pre-processing, the quality of data input is guaranteed, thereby improving the accuracy of subsequent recognition, and then accurately extracts the characteristic information of the billet head to provide accurate data support for the recognition module;

[0069] Then, the flower head recognition module 203 uses the extraction layer in the recognition model to extract high-level features from the preprocessed image; the positioning layer locates the specific position of the steel billet flower head, performs target detection and positioning, and then the classification layer classifies the extracted features. Therefore, the recognition model can automatically extract high-level features of the image, reduce human intervention, improve recognition accuracy, and accurately locate the position of the steel billet flower head through advanced target detection algorithms, improve recognition effects, and improve the ability to recognize complex and subtle defects; finally, the report generation module 204 integrates the recognition results with the original data to generate a comprehensive report, including information such as the location, type, and severity of the flower head, and visualizes the recognition results in the form of charts, images, etc., so that users can intuitively understand the recognition results, and then outputs the generated report file (such as PDF, Excel, etc.), and supports online viewing and downloading, provides a comprehensive analysis report, so that users can fully understand the situation of the steel billet flower head, and through visual display, complex recognition results are easy to understand and analyze. The report is output in PDF format, which is convenient for users to archive and share.

[0070] like Figure 2 As shown, the principle content of a steel billet flower head automatic identification system module includes the following method steps:

[0071] Step 11, scanning the shape of the steel billet to obtain steel billet shape data;

[0072] Step 12, segmenting the steel billet morphology data to obtain initial morphology data, preprocessing the initial morphology data one by one to obtain secondary morphology data, annotating the secondary morphology data one by one, and integrating the annotated data into final morphology data;

[0073] Step 13, inputting the final morphological data into a recognition model, the recognition model comprising:

[0074] Extraction layer: Based on the input final morphological data, high-level feature data is gradually extracted through multiple convolutional layers;

[0075] Positioning layer: performing comparison and positioning based on the extracted high-level feature data;

[0076] Classification layer: input the high-level feature data after positioning, identify it, classify it according to the category of the sub-data in the final morphological data, and output the classification and annotation results;

[0077] Step 14, generating a steel billet head data report based on the classification result.

[0078] In the embodiment of the present invention, the surface of the steel billet is scanned in all directions by an X-ray laser scanner in step 11, and the morphological data of the steel billet, including the morphological contour of the steel billet, the surface texture of the steel billet, the flower head defect and other detailed information, are obtained, and these data are digitally stored, so as to fully capture the details of the steel billet surface, ensure the accuracy of subsequent processing, and provide a high-quality original data basis for subsequent data processing and recognition; then, in step 12, the scanned steel billet morphological data is segmented according to a predetermined rule into small pieces of data suitable for processing, and each small piece of data is subjected to preprocessing operations such as denoising, smoothing, and enhancement to improve data quality, and the preprocessed data is annotated one by one, and the specific position and shape of the steel billet flower head are annotated, and the annotated data is integrated to form a complete final morphological data, and the data is made to reach a unified standard by using image segmentation and preprocessing, which is convenient for subsequent processing, and the data is accurately annotated one by one to ensure the accuracy and completeness of the data, thereby improving the training effect of the recognition model, and then all the annotated data are integrated to provide a complete data input for the recognition model;

[0079] Then, the extraction layer in step 13 extracts the high-level feature data, which can automatically extract the key features in the image, reduce human intervention, and improve recognition accuracy. Through multi-layer convolution, more abstract and high-level features are gradually extracted to improve the model's recognition ability of complex flower heads. Then, the positioning layer uses the regional proposal network (RPN) target detection algorithm to compare the extracted high-level feature data, so that the position of the billet flower head can be accurately located, providing accurate input data for classification, and improving the speed and accuracy of detection. Finally, the located high-level feature data is input into the classifier through the classification layer for identification and classification. The classifier can use the support vector machine algorithm for classification, thereby accurately classifying the billet flower heads, identifying different types of defects and flower heads, and being able to identify complex and subtle defects, thereby improving the overall recognition accuracy.

[0080] Finally, through step 14, the classification results output by the classification layer are integrated with the original morphological data to generate a report containing detailed identification information. By generating a visual report on the billet head data, including information such as defect location, type, severity, etc., the report is generated in PDF format and output, which supports online viewing and downloading to provide a detailed identification report, so that users can fully understand the situation of the billet head, and display the identification results in a visual way to make the report more intuitive and easy to understand. The report is output in PDF format for easy archiving and sharing by users.

[0081] In a preferred embodiment of the present invention, the above step 11 may include:

[0082] Step 111, adjusting the current, voltage and scanning parameters of the industrial X-ray scanning equipment to obtain a high-resolution scanner;

[0083] Step 112, scanning the steel billet morphology in multiple directions by industrial X-rays based on a high-resolution scanner to obtain CT projection image data of the steel billet morphology.

[0084] In an embodiment of the invention, in step 111, the X-ray source voltage parameter is adjusted between 420-450kV to maintain the X-ray penetration ability under high voltage; the current parameter is adjusted between 80-100mA to maintain a stable X-ray intensity; the exposure time is adjusted between 0.13s; the image signal-to-noise ratio is kept stable; the scanning speed is adjusted between 50-80mm / s to ensure that the scanning speed is moderate, the scanning path is set to be linear, and the scanning range is selected according to the size of the steel billet; the scanning resolution is adjusted between 0.51mm to ensure the clarity of the image details. Therefore, by accurately adjusting the current, voltage and scanning parameters, it is ensured that the acquired scanned image has the best clarity and resolution, and it can be adjusted according to different types of steel billet materials and thicknesses to ensure consistent scanning effects. The optimized scanning parameter settings can improve scanning efficiency and reduce scanning time while ensuring high resolution.

[0085] Then, in step 112, by changing the angle of the X-ray source, the steel billet is scanned at multiple angles to obtain projection image data in different directions. When the X-ray penetrates the steel billet and forms a projection image on the detector, different density and thickness regions will form different grayscale values. The projection image data captured by the detector is then transmitted to the computer for image processing and reconstruction to generate a CT image of the internal structure of the steel billet. Therefore, the internal and external morphology of the steel billet can be fully captured through multi-directional scanning to provide complete morphological data. According to the high-resolution CT projection image, the internal defects of the steel billet, such as pores, cracks, etc., can be clearly detected to improve quality control, so that the multi-angle projection data can be reconstructed into a high-precision three-dimensional morphological model through an algorithm, providing an accurate data basis for subsequent analysis and processing.

[0086] In a preferred embodiment of the present invention, the above step 12 may include:

[0087] Step 121, segmenting the billet morphology CT projection image data by setting multiple thresholds, and equally dividing the segmented billet morphology CT projection image data to obtain initial morphology data;

[0088] Step 122, selecting RGB color models for the initial morphological data one by one, and using the G+R component in the RGB color model and combining with the threshold to binarize the color image to obtain a binary image with clear contours and internal textures;

[0089] Step 123, using a 30x30 template and combining a threshold to filter the image to obtain denoised image data;

[0090] Step 124, using the sum of the grayscale values ​​of pixels in the 20x20 template combined with the edge detection method of the threshold interval to obtain a refined and continuous image edge;

[0091] Step 125, performing enhancement processing on the edge contour of the steel billet obtained based on edge detection, combining the image edge, and integrating into sub-morphological data;

[0092] Step 126 , the secondary morphological features are annotated one by one to obtain annotated data, and the annotated data are integrated into the final morphological data.

[0093] In an embodiment of the invention, in step 121, multiple thresholds are set to 80, 90, 100, 110, and 120, and profile line segmentation is performed on the CT projection image data to divide the image into different areas, and intervals of different grayscale values ​​are obtained, so that the selected threshold is determined to be 100, and the image effect obtained is the best, so as to facilitate subsequent image processing, so as to better distinguish the internal structure of the steel billet, and the segmented image data is equally divided and differentiated, that is, the image is divided into several equal-sized parts, so as to facilitate subsequent processing and analysis. Through multi-threshold segmentation, different parts of the steel billet can be distinguished more finely, and the accuracy of image processing can be improved. In addition, the equal division and differentiation can make the data more regular, which provides convenience for subsequent processing steps.

[0094] Then in step 122, the RGB color model in the initial morphological data is selected one by one, and the red (R), green (G) and blue (B) components in the image are extracted respectively. The G+R component sum method is used, that is, the green and red components in the image are added, and then combined with the set threshold value of 100, a binarization process is performed to generate a binary image. The G+R component sum can enhance the contrast of the image, making the contour and internal texture clearer. The binarization process simplifies the image into black and white colors, which is convenient for subsequent analysis and processing.

[0095] Then, in step 123, the image is filtered using a 31x31 template. The template is slid on the image to process each pixel to remove noise. In the filtering process, the threshold 530 is combined to more effectively remove the noise in the image while retaining important image details. The noise in the image is effectively removed, and the image quality is improved. While removing noise, important details in the image are retained as much as possible to improve the clarity of the image. The specific steps are as follows:

[0096] Set the side length to 31x31 for a square. The odd value is set to facilitate the determination of the center position of the template. Specifically, the odd length ensures that the center of the template falls exactly on a pixel point, which facilitates symmetry operations and calculations. According to the formula: , where w represents the width of the image, h is an intermediate variable, and D is the offset distance between the center position of the template and the origin; since the template has a certain width, in order to ensure that the template can cover all pixels at any position in the image and avoid the phenomenon of hanging, this offset distance D ensures that no matter where the template moves to the image, the template can completely cover the pixels in the image area without exceeding the image boundary or leaving uncovered areas; then apply for a new memory space to store the new image data initialized to 0, and initially place the template in the lower left corner of the image. By counting the number of pixels with a gray value of 0 in the template coverage area and comparing it with the set value, if it is less than the set value, the gray value of the point in the new image is set to 255, otherwise it remains at 0, thus completing the first template filtering cycle. Each time the template moves to a position, the same algorithm as the first cycle is executed until the cycle ends.

[0097] Then, in step 124, the sum of the grayscale values ​​of the pixels in the 2x2 template is used in combination with the edge detection method of the threshold interval to obtain a refined and continuous image edge, the sum of the grayscale values ​​of the pixels in the template, and the grayscale values ​​of the pixels in the template are summed using the 2x2 template to determine the edge in the image. The specific steps are: the edge extraction algorithm uses the 2x2 template to traverse the image, and the sum of the grayscale values ​​of the 4 pixels covered by the template is calculated and compared with the set range (T1, T2), T1=255xn, T2=4x255; if it is within the range, the corresponding point of the new image is set to 255 (white), otherwise it is 0 (black), different n values ​​(0, 1, 2) determine the thickness of the edge, and by combining the set threshold interval 100530, the calculation result is judged to determine that the area belongs to the edge of the image. Using a smaller template can more accurately detect the edge of the image, making the edge more refined and continuous. The method combined with the threshold interval can effectively reduce false detection and ensure the accuracy of edge detection.

[0098] According to step 125, the edge contour of the steel billet obtained by edge detection is enhanced to enhance the clarity and contrast of the edge. The processed edge data is integrated to generate sub-morphological data as the basis for subsequent analysis. The edge is made clearer through the enhanced processing to facilitate subsequent morphological analysis. The edge data is integrated into sub-morphological data to provide a unified data basis for further processing and analysis.

[0099] Specifically in another preferred embodiment, step 125 further includes:

[0100] Step 1251, using the skeleton method combined with the standard deviation of the gray value of the circle area, and using linear interpolation iteration to obtain the boundary points to extract the edge contour vector of the steel billet;

[0101] Step 1252, performing linear fitting of edge contour segments on the edge contour vector of the steel billet, and performing sharpening processing on the linear fitting edge contour segments to increase the gray gradient of the edge;

[0102] Step 1253, dividing the horizontal and vertical edges of the steel billet by sub-pixel using the Canny algorithm based on the grayscale gradient to obtain steel billet contour defect data;

[0103] Step 1254, integrating the edge defect data with the image edge data into sub-morphological data.

[0104] In this embodiment, the foreground object in the binary image is thinned into a one-pixel wide skeleton using the skeleton method through step 1251, while maintaining the connectivity of the object, and gradually removing edge pixels until only the skeleton remains. The points can be deleted by marking: for each pixel (x, y), mark whether it can be deleted; then delete all pixels marked as deletable, repeat until there are no deletable pixels, and the resulting skeleton image is represented as S (x, y), where (x, y) is the coordinate of the skeleton point. Then, combined with the standard deviation of the grayscale value in the circular area around each skeleton point, the standard deviation of the grayscale value in the circular area around each skeleton point is calculated on the basis of the skeleton point to enhance specific features or identify defects. The circular area is first defined, and for each skeleton point (x, y), a circular area with a radius of r is defined:

[0105] , where (i, j) is the pixel coordinate within the circular area; then:

[0106] Determine the grayscale standard deviation; where σ r (x, y) is the standard deviation of the gray value in the circular area at position (x, y); |Ω r (x, y)| is the number of pixels in the circular area; (i, j) is the pixel coordinates in the circular area; I grey (i, j) is the grayscale value of the grayscale image at position (i, j); μ r (x, y) is the mean grayscale value in the circular area at position (x, y).

[0107] Finally, the boundary points are obtained through linear interpolation iteration. The Canny edge detection algorithm is used to obtain the initial boundary points of the image, and then the initial boundary point set is extracted from the edge detection result. Then the interpolation function is set. The linear interpolation formula is: , where x0 and x1 are two known points, I(x0) and I(x1) are the corresponding pixel values, according to: ; In the formula, λ is the step size parameter, which controls the interpolation distance. is the gradient direction of the boundary point, are the boundary point coordinates, is the new boundary point coordinate. In each iteration, linear interpolation is performed to obtain new boundary points according to the position and gradient of the current boundary points to extract the edge contour vector of the steel billet to simplify and refine the shape in the image, extract the main structure and contour, and the standard deviation of the grayscale value of the circular area selects a circular area in the steel billet image, calculates the standard deviation of the pixel grayscale value in the area, and judges the change of the edge. At the same time, the linear interpolation iteration will iteratively calculate based on the result of the standard deviation using the linear interpolation method to accurately locate the boundary points, thereby improving the accuracy of edge detection, ensuring the accurate extraction of the steel billet edge contour, and reducing noise interference.

[0108] Step 1252 performs linear fitting on the extracted edge contour points to generate continuous edge contour line segments. The number of line segments is set as n. In the plane rectangular pixel coordinate system of COR, the positive direction of R is vertically downward, and the positive direction of C is horizontally rightward. The minimum and maximum values ​​of the coordinates in the R and C directions in the straight line segment, as well as the maximum deviation differences △R and △C, R min = {RB, RE}; R max = {RB, RE}; C min = {CB, CE}; C max = {CB, CE}; ΔR = R max -R min ; ΔC = C max -C min , where RB; RE; CB; CE are the starting and ending coordinate values ​​of the i-th straight line segment in the R and C directions; linear fitting is used to improve the smoothness and continuity of the edge segment, making the edge clearer and facilitating subsequent processing and analysis.

[0109] Step 1253 uses the grayscale gradient enhanced in the previous step as a basis, and uses the Canny algorithm to perform edge division at the sub-pixel level to ensure the fineness and accuracy of the detection, accurately divide the horizontal and vertical edges of the billet, and extract the contour defect data of the billet for easy identification and processing, through feature enhancement:

[0110] , I enhanced (x, y) is the value of the enhanced image at position (x, y), I gray (x, y) is the grayscale value of the grayscale image at position (x, y), σ r (x, y) is the standard deviation of the gray value in the circle at position (x, y), and then use: ;I defect (x, y) is the binary value (0 or 1) of the defect image at position (x, y), σ r(x, y) is the standard deviation of the gray value in the circular area at position (x, y). r is the threshold for defect recognition, thereby identifying defect features.

[0111] Then, step 1254 uses the steel billet contour defect data extracted in the previous step to integrate the edge defect data with the overall image edge data to generate sub-morphological data, providing more comprehensive and detailed steel billet morphological information, facilitating quality inspection and defect analysis, and improving the accuracy and efficiency of inspection.

[0112] Finally, in step 126, the features in the secondary morphological data are annotated one by one, such as edges, defects, and textures. The annotated data are integrated to generate complete final morphological data. Then, through annotation, various features in the billet can be clearly identified to facilitate quality control and analysis. Finally, the annotated data is integrated to form complete final morphological data, providing comprehensive data support for subsequent analysis and decision-making.

[0113] Specifically in another preferred embodiment, step 126 also includes:

[0114] Step 1261, extracting the steel billet contour defect data and the feature vector of the image edge to obtain edge feature data;

[0115] Step 1262, based on the edge feature vector The analysis is performed, where J is the objective function representing the sum of square errors within the cluster; k represents the number of clusters; represents the i-th cluster, The i-th cluster center, x represents a sample point in the data set; Represents the sample point x to the cluster center The square of the Euclidean distance of to obtain the analysis results;

[0116] Step 1263, based on the analysis result, automatically labeling the different classes of the edge feature vectors, specifically including: setting the edge defects with protrusions or depressions greater than 0.13 mm to be labeled; obtaining edge feature defect information through edge feature vector texture and labeling; labeling according to whether the edge feature vector type is common or uncommon;

[0117] Step 1264, integrating the labeled data into final morphological data, wherein the three sub-data in the final morphological data respectively represent the three types of labeled edge feature vectors.

[0118] In this embodiment, by using in step 1261:

[0119] ; In the formula, θ is the edge direction, and G is the edge gradient strength; the edge image is processed to extract the feature vector of each edge point. These feature vectors may include information such as position, direction, gradient strength, etc., thereby providing a detailed description of the surface profile defects of the billet, which is helpful for subsequent defect analysis and classification. The extracted edge feature data is the basis for subsequent analysis and annotation, which improves the accuracy of data processing. Then, step 1262 is used to analyze the extracted edge feature vectors, including statistical feature vector distribution, calculation of similarity between feature vectors, etc., and clustering algorithms (such as K-means algorithm) are used to classify edge feature vectors to identify different types of defect areas; then, step 1263 is set to detect edge defects with protrusions or depressions of more than 0.13 mm for annotation, specifically: , where Δh is the height deviation, h current is the height of the current edge point, h average It is the average height of the surrounding points. By calculating the height deviation of each edge point, it is determined whether it meets the annotation standard; through the texture information in the edge feature vector, the edge feature defects are identified and annotated:

[0120] ;

[0121] In the formula, Texture is the texture feature value, which represents the texture feature calculated at the position (x, y), (x, y) is the position coordinate of the current pixel, i and j are the offsets used to define the neighborhood range, ranging from -k to k, k is the range of the neighborhood, which determines the size of the neighborhood considered when calculating the texture feature, I(x+i, y+j) is the pixel value at the position (x+i, y+j), and I(x, y) is the pixel value at the position (x, y). According to the edge feature vector type, common and uncommon defects are distinguished and marked. Common defects can be pre-defined. Common defect types include pores: small holes formed by bubbles generated by molten steel during solidification. These pores usually reduce the mechanical properties and service life of steel; inclusions: non-metallic substances included in the molten steel, such as oxides, sulfides, etc. These inclusions will cause the toughness and plasticity of steel to decrease; cracks: cracks caused by uneven cooling or stress concentration, common ones are transverse cracks, longitudinal cracks and corner cracks. Cracks can seriously affect the strength and durability of steel; Porosity: Loose areas caused by incomplete solidification of molten steel. This defect can reduce the density and strength of steel.

[0122] Uncommon defects are identified by anomaly detection algorithms. Uncommon defect types include: Shrinkage cavities are cavities formed by volume shrinkage of molten steel during solidification. These cavities are usually located in the central area of ​​the billet; Segregation: Abnormal concentration of components in a local area due to uneven distribution of components. This defect can lead to uneven performance of the steel; Cold shut: Two streams of molten steel fail to completely merge together due to insufficient flow of molten steel, forming a clear seam or dividing line; Interlayer: A layered defect inside or on the surface of the billet, usually caused by the rolling process. This defect can cause the steel to be prone to delamination and fracture during use; Dissolved gases (such as hydrogen and nitrogen) in the molten steel precipitate during solidification to form bubbles or gas inclusions; Surface pitting: Small pits or pitting on the surface of the billet, usually caused by surface oxidation or corrosion; Shrinkage network: A special type of shrinkage that presents a network-like distribution. Usually occurs in the thick part of the casting, resulting in more pores inside the material; solidification cracks: cracks caused by internal stress during the solidification of molten steel, usually occur before solidification is completed; finally, the different types of annotated data are integrated through step 1264 to form the final shape data. The final shape data includes three types of annotated edge feature vectors: raised or depressed defects, texture feature defects, common and uncommon defects, and finally the final shape data is formatted to ensure data consistency and readability, which is convenient for subsequent data analysis and report generation, so that data integration and formatting improve the readability and consistency of the data, and provide a good foundation for subsequent data analysis and report generation. The final shape data can comprehensively and systematically reflect the surface defects of the billet, which is convenient for quality control and production adjustment.

[0123] In a preferred embodiment of the present invention, the above step 13 may include:

[0124] Step 131, setting a high-level feature data sub-sampling set ;Label ;

[0125] Step 132, initializing the model: identifying the model rate η, setting the number of convolution operation neurons, and initializing the connection weights and biases between the model networks to a small random threshold, wherein each training sample is represented by a sample and a corresponding label sequence pair in the form of x, t; wherein x is the network input value vector, and t is the target output value;

[0126] Step 133, extraction layer: respectively intercepting n sub-images of fixed size from the input final morphological data, and performing multi-layer convolution operations on the n sub-images according to the size and number of convolution kernels to extract high-level feature data;

[0127] Step 134, positioning layer: sub-sampling the high-level feature data, removing interference information by reducing the resolution, and converting it into low-resolution features using a weighted sum operation to locate the high-level feature data;

[0128] Step 135, classification layer: input the located high-level feature data, identify it, classify it according to the category of the sub-data in the final morphological data, and output the classification labeling result.

[0129] The sampling set The data is input to the extraction layer. After the extraction layer has undergone data iteration, it perceives the sample set through each neuron, and then passes it to the positioning layer to integrate the local information to obtain the overall information. The overall information is then passed to the classification layer for clustering recognition. Finally, the sample set is classified according to the label. Classification output.

[0130] In the embodiment of the present invention, the sampling set is Input to the extraction layer in step 133, so that the extraction layer intercepts N images of standard size A×A based on the sub-data in the final morphological data, by: ; In the formula, l represents the number of layers, k is the convolution kernel, M represents the jth feature map, and b represents the bias; multi-level convolution operations are performed on N images to obtain multiple feature maps, that is, the previous layer is convolved with multiple trainable convolution kernels, the size of the convolution kernel is K×K, and a bias is added. After an excitation function, the feature maps of the same number as the convolution kernel are obtained. The size of the feature map is (A-K+l)×(A-K+l) to obtain high-level feature data. When the extraction layer passes through: ; In the formula, f represents the output activation function, w represents the weight matrix, and b represents the bias. After data iteration, each neuron perceives the sample set and then passes it to the positioning layer;

[0131] By positioning the layer in step 135: ; In the formula, down() represents the downsampling function, which performs weighted summation on the n×n region of the input image of this layer, and β and b correspond to the feature map of each output respectively; the high-level feature data image is sub-sampled, the high-resolution representation of the extraction layer is converted to low resolution, and the interference information is screened, that is, the weighted summation operation is performed on the m×n neighborhood in each A×A feature image, and a bias is added, and then an excitation function is passed to obtain the same number of feature maps as the input, and the feature map size is (A-K+l) / m, in order to locate the high-level feature data, integrate the local information to obtain the overall information, and then pass the overall information to the classification layer for clustering recognition, and finally make the sampling set according to the annotation label Classification output.

[0132] In a preferred embodiment of the present invention, the above step 14 may include:

[0133] Step 141, performing statistics on the output classification results to obtain statistical data;

[0134] Step 142, reconstructing the CT projection image data into three-dimensional X-ray image data based on the high-level feature data to obtain a three-dimensional image of the billet tip;

[0135] Step 143, using a visualization tool to generate a chart based on the statistical data and the three-dimensional image data of the steel billet flower head to display the flower head defect type and distribution;

[0136] Step 144, compile a data report based on the acquired flower head absence information, and save the report in PDF format.

[0137] In the embodiment of the present invention, the results output by the classification model are collected through step 141, including information such as defect type, location, size, etc., and the classification results are sorted and formatted to facilitate subsequent statistical analysis. The number, proportion, spatial distribution, etc. of various defects are then counted, and the defect information is classified and summarized using data aggregation and summary technology, specifically:

[0138] ; In the formula, 1 is an indicator function, which takes a value of 1 when the condition is true; N is the total number of samples; the statistical data provides an overall overview of the billet flower head defects, which helps to understand the distribution and type of defects, provides basic data for subsequent three-dimensional reconstruction and visualization, and improves the systematicness and comprehensiveness of data processing; then, in step 142, multi-angle projection image data of the billet flower head is obtained from the CT scanning device, each projection image represents the billet cross section at a different angle, and then a reconstruction algorithm (such as a filtered back projection algorithm) is used to reconstruct the multi-angle CT projection data into a three-dimensional image: ;

[0139] Where f(x, y, z) is the reconstructed 3D image, p(θ, t) is the projection data, h is the filter function, and θ is the projection angle. The 3D reconstructed image provides detailed internal structural information of the billet head, which is helpful for comprehensive defect analysis. In addition, by combining statistical data, defects can be more accurately located and quantified, thereby improving the accuracy and reliability of detection.

[0140] Then, step 143 is used to integrate the statistical data and the 3D reconstructed image data, prepare the visualization input data, and use visualization tools (such as Matplotlib, Plotly) to generate charts to display the defect type, quantity and distribution. The visualization charts intuitively display the defects of the billet head, which is convenient for quick understanding and analysis, so that the combination of 3D images and data charts provides richer information, which is helpful for production decision-making and quality control. The generated charts include bar charts, pie charts, heat maps and 3D renderings; finally, step 144 is used to write a detailed data report based on the statistical data and 3D images. The report content includes detection overview, data statistics, 3D image analysis, conclusions and suggestions, etc., and then the report generation tool is used to format the report content and save it as a PDF file. The generated PDF report provides a systematic document record for the billet head defect detection results, so that the report can be easily shared and archived, which is conducive to the continuous improvement of production management and quality control.

[0141] An embodiment of the present invention further provides a data identification device based on a system and method for automatically identifying steel billet flower heads, including: one or more data identification processors. When the computer program is run by the processor, the automatic steel billet flower head identification method as described above is executed to identify data in one or more programs. All implementation methods in the above method embodiments are applicable to this embodiment and can achieve the same technical effect.

[0142] The embodiment of the present invention also provides a computer-readable storage medium storing instructions, which, when executed on a computer, enable the computer to execute the method described above. All implementations in the above method embodiment are applicable to this embodiment and can achieve the same technical effect.

[0143] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0144] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0145] In the embodiments provided by the present invention, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0146] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0147] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0148] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as USB flash drives, mobile hard disks, ROM, RAM, magnetic disks, or optical disks.

[0149] In addition, it should be pointed out that in the apparatus and method of the present invention, it is obvious that each component or each step can be decomposed and / or recombined. These decompositions and / or recombinations should be regarded as equivalent schemes of the present invention. Moreover, the steps of performing the above series of processing can naturally be performed in chronological order according to the order of description, but it is not necessary to perform them in chronological order, and some steps can be performed in parallel or independently of each other. For those of ordinary skill in the art, it can be understood that all or any steps or components of the method and apparatus of the present invention can be implemented in hardware, firmware, software or a combination thereof in any computing device (including processors, storage media, etc.) or a network of computing devices, which can be achieved by those of ordinary skill in the art using basic programming skills after reading the description of the present invention.

[0150] Therefore, the purpose of the present invention can also be achieved by running a program or a group of programs on any computing device. The computing device can be a well-known general device. Therefore, the purpose of the present invention can also be achieved by simply providing a program product containing a program code that implements the method or device. That is to say, such a program product also constitutes the present invention, and the storage medium storing such a program product also constitutes the present invention. Obviously, the storage medium can be any well-known storage medium or any storage medium developed in the future. It should also be pointed out that in the device and method of the present invention, it is obvious that each component or each step can be decomposed and / or recombined. These decompositions and / or recombinations should be regarded as equivalent schemes of the present invention. In addition, the steps of performing the above-mentioned series of processing can naturally be performed in chronological order according to the order of description, but it is not necessary to perform them in chronological order. Some steps can be performed in parallel or independently of each other.

[0151] An example of the present specific implementation mode is described above, but the present embodiment is not limited to the above-mentioned specific implementation mode, which is merely illustrative and not restrictive. A person skilled in the art may make many forms inspired by the present embodiment, all of which are protected by the present embodiment.

Claims

1. A steel billet flower head automatic identification system (20), It is characterized in that The system includes the following modules: Data acquisition module (201): used for scanning the shape of the steel billet to obtain the shape data of the steel billet; A data processing module (202): used for segmenting the steel billet morphology data to obtain initial morphology data, pre-processing the initial morphology data one by one to obtain secondary morphology data, annotating the secondary morphology data one by one, and integrating the annotated data into final morphology data; Flower head recognition module (203): used to input the final morphological data into a recognition model, the recognition model comprising: Extraction layer: Based on the input final morphological data, high-level feature data is gradually extracted through multiple convolutional layers; Positioning layer: performing comparison and positioning based on the extracted high-level feature data; Classification layer: input the high-level feature data after positioning, identify it, classify it according to the category of the sub-data in the final morphological data, and output the classification and annotation results; A report generation module (204): used to generate a steel billet head data report based on the classification result; The steel billet morphology data is segmented to obtain initial morphology data, the initial morphology data is pre-processed one by one to obtain secondary morphology data, the secondary morphology data is annotated one by one, and the annotated data is integrated into final morphology data, including: The billet morphology CT projection image data is segmented by setting multiple thresholds, and the segmented billet morphology CT projection image data is equally divided to obtain initial morphology data; the equal division is to divide the image data into a number of equal-sized parts; Selecting RGB color models for the initial morphological data one by one, and using the G+R component in the RGB color model and combining the threshold to binarize the color image, so as to obtain a binary image with clear contour and internal texture; A 31x31 template is used and combined with a threshold to filter the image to obtain denoised image data; The sum of the grayscale values ​​of pixels in a 2x2 template is used in combination with the edge detection method of the threshold interval to obtain a refined and continuous image edge; Strengthening the edge contour of the steel billet obtained based on edge detection, combining the image edge, and integrating it into sub-morphological data; The sub-morphological features are annotated one by one to obtain annotated data, and the annotated data are integrated into the final morphological data; The skeleton method is combined with the standard deviation of the gray value in the circle area, and the boundary points are iterated by linear interpolation to extract the edge contour vector of the steel billet. Linearly fitting edge contour segments of the steel billet edge contour vectors, and sharpening the linearly fitting edge contour segments to increase the gray gradient of the edge; Based on the grayscale gradient, the horizontal and vertical edges of the steel billet are divided by sub-pixel using the Canny algorithm to obtain the steel billet contour defect data; Integrate the edge defect data into sub-morphological data in combination with the image edge data; Extracting the steel billet profile defect data and feature vectors of the image edge to obtain edge feature data; Based on the edge feature vector Analyze, among which, J represents the sum of squared errors within the cluster for the objective function; k represents the number of clusters; represents the i-th cluster, The i-th cluster center, x represents a sample point in the data set; Represents the sample point x to the cluster center The square of the Euclidean distance of to obtain the analysis results; Automatically labeling different classes of the edge feature vectors based on the analysis results specifically includes: Set to detect edge defects with protrusions or depressions larger than 0.13mm and mark them; obtain edge feature defect information through edge feature vector texture and mark them; mark them according to whether the edge feature vector type is common or uncommon; Integrate the labeled data into final morphological data, wherein the three sub-data in the final morphological data respectively represent the three types of labeled edge feature vectors; The final morphological data is input into the recognition model, including: Input: high-level feature data subsample set ;Label ; Initialization: Identify the model rate η, set the number of convolution operation neurons, and initialize the connection weights and biases between the model networks to a small random threshold. Each training sample is represented by a sample and a corresponding label sequence in the form of x, t; x is the network input value vector, and t is the target output value; Extraction layer: respectively intercept n sub-images of fixed size from the input final morphological data, and perform multi-layer convolution operations on the n sub-images according to the size and number of convolution kernels to extract high-level feature data; Positioning layer: sub-sampling the high-level feature data, removing interference information by reducing the resolution, and converting it into low-resolution features using a weighted sum operation to locate the high-level feature data; Classification layer: input the high-level feature data after positioning, identify it, classify it according to the category of the sub-data in the final morphological data, and output the classification and annotation results; The sampling set The data is input to the extraction layer. After the extraction layer has undergone data iteration, it perceives the sample set through each neuron, and then passes it to the positioning layer to integrate the local information to obtain the overall information. The overall information is then passed to the classification layer for clustering recognition. Finally, the sample set is classified according to the label. Classification output; The classification results are used to generate a billet head data report, including: Performing statistics on the output classification results to obtain statistical data; Reconstructing the CT projection image data based on the high-level feature data into three-dimensional X-ray image data to obtain a three-dimensional image of the billet flower head; Use visualization tools to generate charts based on statistical data and three-dimensional image data of billet heads to display the types and distribution of head defects; Prepare a data report based on the acquired flower head absence information and save the report in PDF format.

2. A method for automatically identifying a steel billet flower head, according to the contents of each module in the automatic steel billet flower head identification system of claim 1, It is characterized in that Scan the billet morphology to obtain billet morphology data, including: Adjust the current, voltage and scanning parameters of industrial X-ray scanning equipment to obtain high-resolution scanners; The billet morphology is scanned in multiple directions by industrial X-rays based on a high-resolution scanner to obtain CT projection image data of the billet morphology.

3. A steel billet flower head automatic identification system, It is characterized in that Also includes: one or more data identification processors; A data identification processor is used to identify data in one or more programs. When the one or more programs are executed and identified by the one or more processors, the one or more processors implement the steps of the automatic identification method for steel billet flower heads as described in claim 2.

4. A computer-readable storage medium, It is characterized in that The computer-readable storage medium stores a program, which, when executed by a processor, implements the steps of the method for automatically identifying the steel billet flower head as described in claim 2.