Rolled piece side face effective flange positioning method based on infrared imaging

Abstract

The invention discloses a rolled piece side face effective flange positioning method based on infrared imaging, and relates to the technical field of intelligent detection. The method for positioning the effective flange on the side surface of the rolled piece based on infrared imaging comprises the following specific operation steps: S1, mounting an infrared temperature measuring device within a certain range of the side surface of the high-temperature rolled piece, and acquiring temperature field data of the side surface of an object in real time by using an infrared spectrum detection device; according to the method for positioning the effective flange on the side face of the rolled piece based on infrared imaging, the position of the effective flange of the rolled piece is positioned by analyzing the temperature-position change rate of a continuous temperature field in the horizontal direction and the vertical direction, and meanwhile edge information in an image is extracted through a marginal detection algorithm according to image color information; and the result is compared with a temperature change rate analysis result, so that the detection precision and reliability are improved. In the process of vision measurement and positioning of the side face of the rolled piece, effective flange positioning of the rolled pieces of different sizes can be achieved by the camera under the fixed visual angle, and interference caused by the other side edge is avoided.

Description

Technical Field

[0001] The invention relates to the field of intelligent detection technology, and in particular to a method for positioning an effective flange on a side surface of a rolled piece based on infrared imaging. Background Art

[0002] The QST water-piercing equipment is a post-rolling water-piercing process equipment, but since it is used for the first time on heavy H-beams, there are many problems in the configuration and design of the equipment. In the prior art, heavy rolled pieces cannot be completely centered when leaving the QST equipment, which will cause lateral deviation, causing the center position of the rolled piece to deviate from the center line of the roller. As a result, the actual temperature measurement position of the temperature point measured by the point temperature measurement method before the hot saw is higher or lower than the actual position to be measured, and there is a predicted deviation between the feedback temperature of the temperature measurement point and the actual temperature of the stable measurement point required by the process. In addition, the water nozzles of the QST equipment are prone to clogging, especially the side nozzles on the transmission side. The clogging will seriously affect the temperature difference between the inside and outside of the H-beam flange, resulting in uneven cooling of the H-beam cross section. Uneven cooling can cause varying degrees of deformation of the finished flange.

[0003] In order to solve the above problems, existing technologies attempt to visually measure the side of the high-temperature rolled piece to determine the cooling state of the side of the rolled piece as a whole, so as to detect whether there are defects such as uneven cooling on its side. However, the problem of processing rolled pieces of different sizes under a fixed camera viewing angle is the main technical difficulty. The size and position of rolled pieces of different sizes in the image may be significantly different, and two flange images may appear on the side of the rolled piece, thereby affecting the positioning accuracy of the flange. In addition, the shape and posture of the rolled piece may also affect the visibility and positioning of the flange, which will interfere with the temperature zone detection.

[0004] At present, only traditional image processing technology is found to be applied to visual inspection, while infrared spectrum data is used for object temperature detection. No infrared spectrum data has been found to be applied to the field of image processing at home and abroad, nor has the infrared spectrum temperature field data algorithm been found to be applied to the field of visual edge positioning detection of high-temperature steel. Summary of the invention

[0005] 1. Technical issues to be resolved

[0006] In view of the deficiencies of the prior art, the present invention provides a method for locating the effective flange of a rolled piece on the side surface based on infrared imaging. The position of the effective flange of the rolled piece is located by analyzing the temperature-position change rate of the continuous temperature field in the horizontal and vertical directions. At the same time, the edge information in the image is extracted using an edge detection algorithm using the image color information, and compared with the temperature change rate analysis result, thereby improving the accuracy and reliability of the detection. In the process of visual measurement and positioning on the side surface of the rolled piece, the camera can locate the effective flange of rolled pieces of different sizes at a fixed viewing angle, avoiding interference caused by another measured edge. The problems existing in the prior art are solved. The purpose is to improve the efficiency of visual inspection of the surface of the rolled piece.

[0007] (II) Technical solution

[0008] To achieve the above objectives, the present invention is implemented through the following technical solutions: a method for locating the effective flange of the side of a rolled piece based on infrared imaging, the specific operations are as follows:

[0009] S1. Install an infrared temperature measuring device within a certain range on the side of the high-temperature rolled piece, and use an infrared spectrum detection device to obtain temperature field data on the side of the object in real time;

[0010] S2. Using the temperature field information, identify the high temperature area of ​​the rolled piece, and calculate the moving speed of the object. According to the relationship between the moving speed of the object and the temperature change, the temperature judgment threshold is adaptively adjusted to eliminate abnormal data and generate a temperature field map of the moving object and its surroundings.

[0011] S3, comprehensive analysis of temperature field data and visible light images. The acquired temperature field data and visible light images are first preprocessed. For visible light images, in order to simplify the color feature extraction of the image, color space conversion is performed. The spectral color data is statistically analyzed, and the threshold is set according to the statistical spectral range and color distance to perform image color segmentation.

[0012] S4. According to the analysis of the temperature-position change rate in the vertical direction of the continuous temperature field, the position where the temperature point data change rate suddenly changes is the edge point, that is, the position where the temperature point data changes the most. The marginal extraction operator is used to calculate the first-order partial derivatives in the horizontal and vertical directions of the temperature field data. First, a three-dimensional coordinate system is established to generate a three-dimensional surface function image z=f(x,y), where x represents the horizontal coordinate, y represents the vertical coordinate, and z represents the temperature value of the point. Partial derivatives are calculated in the horizontal and vertical directions, and extreme values ​​are calculated based on the partial derivative results. The extreme value points in the interval are the discrete points where the temperature point data change rate suddenly changes. In the image, we use gradient to represent the rate and direction of change of the temperature field. After obtaining the temperature field gradient, the gradient of the entire image is scanned, and each temperature point is checked. According to the two principles of whether the gradient of the point is the largest among the points with the same gradient direction and whether the gradient direction is perpendicular to the edge, the edge point is found; the coordinates of the discrete mutation points are gathered into a line through the trajectory filtering algorithm, which is the flange. The effective flange or invalid flange is determined according to the ordinate value of the discrete points in the flange. The two flanges with the lowest ordinate are effective flanges.

[0013] S5. Use the edge detection algorithm to extract edge information from the image at the same time, and compare it with the temperature change rate analysis results to further accurately locate the image. Combined with the color information in the object image, use the edge detection algorithm to calculate the gradient size and direction of the color parameters, find the gradient by calculating the derivatives of the color parameters in the horizontal and vertical directions, determine the degree and direction of change of the color parameters based on the gradient, and use the edge extraction operator to calculate the first-order derivatives in the horizontal and vertical directions to obtain the edge gradient and direction. The gradient direction is perpendicular to the edge. Secondly, based on the gradient size and direction, scan the gradient of the entire image, remove those non-edge numerical points, and retain possible edge points. Finally, compare the gradient value of the color point with the set threshold to determine the position and direction of the edge;

[0014] S6. During the positioning process, edge detection algorithms are used to extract edge information from the image, and compared with the temperature change rate analysis results to further accurately locate the object. Multi-scale analysis technology is used to analyze images of different scales to improve the accuracy and stability of positioning. The temperature field and visible light image are fused, and the two data sources are aligned through image registration technology to ensure the consistency of spatial position. The weighted average method is used to assign weights according to the confidence of the temperature field data and the visible light image. Combined with the gradient information and color information of the object's continuous temperature field, the effective flange position of the rolled piece is identified, the limited flange and effective side data are determined, and the invalid flange and interference area image data are eliminated;

[0015] S7. Use the continuous temperature field change rate to detect whether the rolled piece has entered the monitoring area, and automatically adjust the parameters of the infrared spectrum detection device according to the real-time detected rolled piece position and temperature field changes. Combine the temperature field data and visible light image data analysis results to determine the effective flange position of the rolled piece and generate a panoramic image of the rolled piece;

[0016] S8. After removing the invalid flange, the recognition accuracy of uneven cooling is improved. The temperature data clustering algorithm is used to identify the area of ​​uneven cooling. The temperature change rate is determined by calculating the temperature difference between adjacent temperature points. The clustering algorithm divides the temperature data into multiple categories to identify the area of ​​uneven cooling. Combined with the color and shape information in the visible light image, the uneven cooling phenomenon is compared and verified;

[0017] S9. During the temperature field data analysis, we build the machine neural network ES structure, design the learning algorithm, and continuously optimize the temperature change rate analysis and image fusion algorithm, which helps to improve the system's adaptability and detection accuracy. According to historical data and actual detection conditions, we adjust the algorithm parameters to ensure that the system can operate stably under different working conditions.

[0018] Preferably, an infrared temperature measuring device is installed on the side of the rolled piece to obtain temperature field data within a certain range. When installing the device, attention should be paid to the adjustment of angle and height to avoid blind areas. The device is fixed on a special bracket, which can be adjusted according to the height and width of the rolled piece to adapt to rolled pieces of different specifications;

[0019] The infrared spectrum detection device is used to obtain the temperature field data of the side of the rolled piece in real time. The data collected by the device is transmitted to the central control system through a high-speed data transmission interface. The sampling frequency is usually set to more than 10 times per second, and the specific frequency is adjusted according to the moving speed of the rolled piece and the cooling process requirements. The data transmission process uses shielded cables and anti-interference filters to ensure the stability and accuracy of data transmission, and to ensure the integrity and anti-interference ability of the signal.

[0020] Preferably, the acquired temperature field data is first preprocessed, including denoising, smoothing and normalization. The denoising process uses a median filter to eliminate random noise. The smoothing process uses a Gaussian filter to ensure the continuity and smoothness of the temperature field data. The normalization process converts the temperature data to a uniform scale for subsequent analysis;

[0021] A visible light camera is installed next to the infrared temperature measuring device to capture visible light images of the side of the rolled piece in real time. The visible light image undergoes image preprocessing steps, including denoising, graying, and edge detection. The visible light image is converted to color space, and the RGB space is converted to HSV space to simplify the color feature extraction of the image, so as to express the brightness and tone of the color and facilitate the contrast between colors. The contour information of the rolled piece is extracted using color filtering and edge detection algorithms. Color filtering filters out the red area on the surface of the rolled piece by setting the range of H values ​​(such as 0-10 and 160-180). The principle of converting RGB space to HSV space is:

[0022] Assume the input RGB values ​​are R, G, B, and the value range is 0, 255. First, they need to be normalized:

[0023]

[0024] Calculate the maximum and minimum values: C max =max(r,g,b),C min =min(r,g,b)

[0025] Calculate Hue: Let Δ = C max -C min ,

[0026]

[0027] Calculate Saturation:

[0028]

[0029] Calculate the value: V = C max .

[0030] Preferably, the temperature change rate analysis:

[0031] By analyzing the temperature-position change rate in the vertical direction of the continuous temperature field, the position where the temperature point data change rate suddenly changes is the edge point, and the flange position of the rolled piece is preliminarily located. First, a three-dimensional coordinate system is established to generate a three-dimensional surface function image z=f(x,y), where x represents the horizontal coordinate, y represents the vertical coordinate, and z represents the temperature value of the point. Partial derivatives are calculated in the horizontal and vertical directions, and the extreme values ​​are calculated based on the partial derivative results. The extreme value point in the interval is the position where the temperature point data change rate suddenly changes. The marginal extraction operator is used to calculate the first-order partial derivatives in the horizontal and vertical directions of the temperature field data to obtain the temperature field gradient and direction. In the image, we use the gradient to represent the degree and direction of the temperature field change. After obtaining the temperature field gradient, the gradient of the entire image is scanned, and each temperature point is checked. According to the two principles of whether the gradient of the point is the largest among the points with the same gradient direction and whether the gradient direction is perpendicular to the edge, the edge point is found. The coordinates of the discrete mutation points are gathered into a line through the trajectory filtering algorithm, which is the flange. The flange is judged as a valid flange or an invalid flange based on the ordinate value of the discrete point in the flange. The two flanges with the lowest ordinate values ​​are valid flanges.

[0032] 1) For three-dimensional space:

[0033] F(x,y,z)=0

[0034] 2) First-order partial derivative:

[0035]

[0036] 3) Differential operator of temperature field in horizontal direction:

[0037]

[0038] 4) Differential operator in the vertical direction of the temperature field:

[0039]

[0040] 5) Calculate the gradient and direction:

[0041]

[0042]

[0043] By calculating the rate of change of the unit temperature field, the area with significant temperature change is identified and the position of the flange is preliminarily determined.

[0044] Visible light image analysis:

[0045] For visible light color information, edge detection algorithms are used to extract edge information in the image, and compared with the results of temperature change rate analysis to further accurately locate. Combined with the color information in the object image, the edge detection algorithm is used to calculate the gradient size and direction of the color parameters. The gradient is found by calculating the derivatives of the color parameters in the horizontal and vertical directions. The degree and direction of change of the color parameters are determined according to the gradient. The edge extraction operator is used to calculate the first-order derivatives in the horizontal and vertical directions to obtain the edge gradient and direction. The gradient direction is perpendicular to the edge. Secondly, based on the gradient size and direction, the gradient of the entire image is scanned to remove those non-edge numerical points and retain possible edge points. Finally, the double threshold method is used to determine the real and potential edges. The position and direction of the edge are determined by comparing the gradient value of the color point with the set threshold (high threshold and low threshold).

[0046] The color change rate calculation formula is: where G is the color gradient, X is the vertical coordinate, and Y is the vertical coordinate.

[0047] 1) Differential operator in the horizontal direction of the parameter:

[0048]

[0049] 2) Differential operator in the vertical direction of the parameter:

[0050]

[0051] 3) Calculate the gradient and direction:

[0052]

[0053]

[0054] Fusion of temperature field and visible light image:

[0055] The preprocessed temperature field data is fused with the visible light image. The two data sources are aligned through image registration technology to ensure the consistency of spatial position. During the fusion process, the weighted average method is used to assign weights according to the confidence of the temperature field data and the visible light image. The effective flange position of the rolled piece is identified by combining the temperature gradient and color information. The specific operations include:

[0056] Extract high temperature areas in the temperature gradient.

[0057] Identify red areas in the HSV color space.

[0058] Superimpose the two results, remove irrelevant areas, and retain the valid flange position.

[0059] Effective flange position determination:

[0060] The analysis results of temperature field data and visible light image data are integrated to eliminate invalid flange image data, restore useful real information, and determine the effective flange position of the rolled piece. In the positioning process, edge detection algorithms are used to extract edge information in the image and compare it with the temperature change rate analysis results to further accurately locate. Multi-scale analysis technology is used to analyze images of different scales to improve the accuracy and stability of positioning.

[0061] Preferably, the detection of the rolled piece entering the monitoring area and the adjustment of the infrared spectrum detection device

[0062] Rolling inspection:

[0063] The continuous temperature field change rate and color information in the HSV color space are used to detect whether the rolled piece has entered the monitoring area. The system determines the entry of the rolled piece through a preset threshold, and automatically starts the exposure and frame rate adaptive adjustment function of the infrared spectrum detection device when the rolled piece enters the monitoring area.

[0064] Device adjustment:

[0065] The parameters of the infrared spectrum detection device are automatically adjusted according to the real-time detected changes in the position and temperature field of the rolled piece. The exposure time is adjusted according to the surface reflectivity and lighting conditions of the rolled piece to ensure clear images. The frame rate is adjusted according to the moving speed of the rolled piece to ensure that sufficient image data can be obtained even at high speeds.

[0066] Preferably, the effective flange and side temperature fields are generated and the panoramic image of the rolled piece is generated;

[0067] The effective flange position of the rolled piece is determined by combining the analysis results of the temperature field data and visible light image data.

[0068] The optimal translation amount is calculated through the similarity measurement algorithm, and the continuously shot images of the rolled piece are spliced ​​to generate a complete panoramic image of the rolled piece. After the panoramic image is generated, the image clarity and contrast are improved through image enhancement technology to facilitate subsequent analysis.

[0069] Preferably, the effective flange is used to improve the recognition accuracy of uneven cooling

[0070] Uneven cooling detection:

[0071] The temperature data clustering algorithm is used to identify areas of uneven cooling. The temperature change rate is determined by calculating the temperature difference between adjacent temperature points. The clustering algorithm classifies the temperature data into multiple categories to identify areas of uneven cooling. The color and shape information in the visible light image is combined to compare and verify the uneven cooling phenomenon.

[0072] Improved accuracy:

[0073] The temperature field data and visible light image data are combined by weight fusion to improve the accuracy of uneven cooling recognition. The specific method includes assigning different weights to different data sources, comprehensively evaluating the detection results, and eliminating noise and misjudgment. The fusion results are trained and optimized through machine learning algorithms to improve the accuracy and robustness of the detection model.

[0074] Preferably, build a machine neural network ES structure, design a learning algorithm, continuously optimize the temperature change rate analysis and image fusion algorithm, and improve the system's adaptability and detection accuracy. According to historical data and actual detection conditions, adjust the algorithm parameters to ensure that the system can operate stably under different working conditions. Specific methods include:

[0075] Use supervised learning algorithms to train historical inspection data and construct a mapping relationship between temperature change rate and effective flange position.

[0076] Use reinforcement learning algorithms to adjust system parameters based on real-time detection results and improve the system's adaptability.

[0077] (III) Beneficial effects

[0078] The present invention provides a method for positioning the effective flange of a rolled piece side based on infrared imaging. It has the following beneficial effects:

[0079] This method for locating the effective flange of the rolled piece side based on infrared imaging locates the position of the effective flange of the rolled piece by analyzing the temperature-position change rate of the continuous temperature field in the horizontal and vertical directions. At the same time, the edge information in the image is extracted using the edge detection algorithm using the image color information, and compared with the temperature change rate analysis result, thereby improving the accuracy and reliability of the detection. In the process of visual measurement and positioning on the side of the rolled piece, the camera can locate the effective flange of the rolled piece of different sizes at a fixed viewing angle, avoiding interference caused by another measured edge. It aims to meet the accuracy and efficiency of surface quality monitoring of high-temperature rolled pieces and improve the production quality of rolled pieces.

[0080] The effective flange positioning of rolled products of different sizes is achieved through a camera with a fixed viewing angle, ensuring the accuracy and efficiency of surface quality monitoring of rolled products. This method not only improves the level of automation in the production process of rolled products, but also significantly improves production quality. Through precise positioning and real-time monitoring, human errors and quality defects are effectively reduced, the consistency and reliability of products are improved, and the overall production efficiency is improved, and production costs are reduced. BRIEF DESCRIPTION OF THE DRAWINGS

[0081] Figure 1 This is a schematic diagram of the installation of the infrared temperature measuring device of the present invention;

[0082] Figure 2This is a schematic diagram of the temperature field gradient of an image of the present invention;

[0083] Figure 3 A schematic diagram of the gradient intensity of the local image color parameter of the present invention;

[0084] Figure 4 This is a site diagram of the heavy QST rolled piece of the present invention;

[0085] Figure 5 The infrared imaging of the present invention shows two different edge images;

[0086] Figure 6 It is the basic structure of the neural network ES of the present invention. DETAILED DESCRIPTION

[0087] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0088] See also Figure 1-6 The present invention provides a technical solution: a method for positioning the effective flange of a rolled piece side based on infrared imaging, and the specific operation is as follows:

[0089] S1. Install an infrared temperature measuring device within a certain range on the side of the high-temperature rolled piece, and use an infrared spectrum detection device to obtain temperature field data on the side of the object in real time;

[0090] S2. Using the temperature field information, the high temperature area of ​​the rolled piece is identified according to the threshold value, and the moving speed of the object is calculated. According to the relationship between the moving speed of the object and the temperature change, the temperature judgment threshold is adaptively adjusted, abnormal data is eliminated, and a temperature field diagram of the moving object and its surroundings is generated;

[0091] S3, comprehensive analysis of temperature field data and visible light images. The acquired temperature field data and visible light images are first preprocessed. For visible light images, in order to simplify the color feature extraction of the image, color space conversion is performed. The spectral color data is statistically analyzed, and the threshold is set according to the statistical spectral range and color distance to perform image color segmentation.

[0092] S4. According to the analysis of the temperature-position change rate in the vertical direction of the continuous temperature field, the position where the temperature point data change rate suddenly changes is the edge point, that is, the position where the temperature point data changes the most. The marginal extraction operator is used to calculate the first-order partial derivatives in the horizontal and vertical directions of the temperature field data. First, a three-dimensional coordinate system is established to generate a three-dimensional surface function image z=f(x,y), where x represents the horizontal coordinate, y represents the vertical coordinate, and z represents the temperature value of the point. Partial derivatives are calculated in the horizontal and vertical directions, and extreme values ​​are calculated based on the partial derivative results. The extreme value points in the interval are the discrete points where the temperature point data change rate suddenly changes. In the image, we use gradient to represent the rate and direction of change of the temperature field. After obtaining the temperature field gradient, the gradient of the entire image is scanned, and each temperature point is checked. According to the two principles of whether the gradient of the point is the largest among the points with the same gradient direction and whether the gradient direction is perpendicular to the edge, the edge point is found; the coordinates of the discrete mutation points are gathered into a line through the trajectory filtering algorithm, which is the flange. The effective flange or invalid flange is determined according to the ordinate value of the discrete points in the flange. The two flanges with the lowest ordinate are effective flanges. Figure 2 As shown, the arrow direction represents the gradient direction. Taking the third temperature point in the third row as an example, the intensity of this point [8] is compared with the intensity of the point above it [2]. Since this point has the largest change, it is the edge point. This is used to preliminarily locate the flange position of the rolled piece;

[0093] S5. Use the edge detection algorithm to extract edge information from the image at the same time, and compare it with the temperature change rate analysis results to further accurately locate the image. Combined with the color information in the object image, use the edge detection algorithm to calculate the gradient size and direction of the color parameters, and find the gradient by calculating the derivatives of the color parameters in the horizontal and vertical directions. The degree and direction of change of the color parameters are determined according to the gradient. Use the edge extraction operator to calculate the first-order derivatives in the horizontal and vertical directions to obtain the edge gradient and direction. The gradient direction is perpendicular to the edge. Secondly, based on the gradient size and direction, scan the gradient of the entire image, remove those non-edge numerical points, and retain possible edge points. Finally, compare the gradient value of the color point with the set threshold to determine the position and direction of the edge; such as Figure 3 As shown, the direction of the arrow represents the gradient direction. Taking the fifth point in the third row as an example, since the gradient direction is upward, the intensity of this point [9] is compared with the intensities of the two points above and below it [6] and [3]. Since this point has the largest intensity, it is retained;

[0094] S6. During the positioning process, edge detection algorithms are used to extract edge information from the image, and compared with the temperature change rate analysis results to further accurately locate the object. Multi-scale analysis technology is used to analyze images of different scales to improve the accuracy and stability of positioning. The temperature field and visible light image are fused, and the two data sources are aligned through image registration technology to ensure the consistency of spatial position. The weighted average method is used to assign weights according to the confidence of the temperature field data and the visible light image. Combined with the gradient information and color information of the object's continuous temperature field, the effective flange position of the rolled piece is identified, the limited flange and effective side data are determined, and the invalid flange and interference area image data are eliminated;

[0095] S7. Use the continuous temperature field change rate to detect whether the rolled piece has entered the monitoring area, and automatically adjust the parameters of the infrared spectrum detection device according to the real-time detected rolled piece position and temperature field changes. Combine the temperature field data and visible light image data analysis results to determine the effective flange position of the rolled piece and generate a panoramic image of the rolled piece;

[0096] S8. After removing the invalid flange, the recognition accuracy of uneven cooling is improved. The temperature data clustering algorithm is used to identify the area of ​​uneven cooling. The temperature change rate is determined by calculating the temperature difference between adjacent temperature points. The clustering algorithm divides the temperature data into multiple categories to identify the area of ​​uneven cooling. Combined with the color and shape information in the visible light image, the uneven cooling phenomenon is compared and verified;

[0097] S9. During the temperature field data analysis, a machine neural network ES structure is built, such as Figure 6 As shown in the figure, designing a learning algorithm and continuously optimizing the temperature change rate analysis and image fusion algorithm can help improve the system's adaptability and detection accuracy. According to historical data and actual detection conditions, the algorithm parameters are adjusted to ensure that the system can operate stably under different working conditions.

[0098] Among them, an infrared temperature measuring device is installed on the side of the rolled piece to obtain temperature field data within a certain range, such as Figure 1 , Figure 4 When installing the device, attention should be paid to the adjustment of angle and height to avoid blind areas. The device is fixed on a special bracket, which can be adjusted according to the height and width of the rolled piece to accommodate rolled pieces of different specifications;

[0099] The infrared spectrum detection device is used to obtain the temperature field data of the side of the rolled piece in real time. The data collected by the device is transmitted to the central control system through a high-speed data transmission interface. The sampling frequency is usually set to more than 10 times per second, and the specific frequency is adjusted according to the moving speed of the rolled piece and the cooling process requirements. The data transmission process uses shielded cables and anti-interference filters to ensure the stability and accuracy of data transmission, and to ensure the integrity and anti-interference ability of the signal.

[0100] The acquired temperature field data is first preprocessed, including denoising, smoothing and normalization. The denoising process uses a median filter to eliminate random noise. The smoothing process uses a Gaussian filter to ensure the continuity and smoothness of the temperature field data. The normalization process converts the temperature data to a uniform scale for subsequent analysis;

[0101] A visible light camera is installed next to the infrared temperature measuring device to capture visible light images of the side of the rolled piece in real time. The visible light image undergoes image preprocessing steps, including denoising, graying, and edge detection. The visible light image is converted to color space, and the RGB space is converted to HSV space to simplify the color feature extraction of the image, so as to express the brightness and tone of the color and facilitate the contrast between colors. The contour information of the rolled piece is extracted using color filtering and edge detection algorithms. Color filtering filters out the red area on the surface of the rolled piece by setting the range of H values ​​(such as 0-10 and 160-180). The principle of converting RGB space to HSV space is:

[0102] Assume the input RGB values ​​are R, G, B, and the value range is 0, 255. First, they need to be normalized:

[0103]

[0104] Calculate the maximum and minimum values: C max =max(r,g,b),C min =min(r,g,b)

[0105] Calculate Hue: Let Δ = C max -C min ,

[0106]

[0107] Calculate Saturation:

[0108]

[0109] Calculate the value: V = C max .

[0110] Among them, the temperature change rate analysis:

[0111] Due to the different sizes of rolled products, two flange images may appear under a fixed viewing angle, which may interfere with infrared imaging detection. Figure 5Represents. Therefore, by analyzing the temperature-position change rate in the vertical direction of the continuous temperature field, the position where the temperature point data change rate suddenly changes is the edge point, and the flange position of the rolled piece is preliminarily located. First, a three-dimensional coordinate system is established to generate a three-dimensional surface function image z=f(x,y), where x represents the horizontal coordinate, y represents the vertical coordinate, and z represents the temperature value of the point. Partial derivatives are calculated in the horizontal and vertical directions, and extreme values ​​are calculated based on the partial derivative results. The extreme point in the interval is the position where the temperature point data change rate suddenly changes. The marginal extraction operator is used to calculate the first-order partial derivatives in the horizontal and vertical directions of the temperature field data to obtain the temperature field gradient and direction. In the image, we use the gradient to represent the degree and direction of the temperature field change. After obtaining the temperature field gradient, the gradient of the entire image is scanned, and each temperature point is checked. According to the two principles of whether the gradient of the point is the largest among the points with the same gradient direction and whether the gradient direction is perpendicular to the edge, the edge point is found. The coordinates of the discrete mutation points are gathered into a line through the trajectory filtering algorithm, which is the flange. The flange is judged as a valid flange or an invalid flange based on the ordinate value of the discrete point in the flange. The two flanges with the lowest ordinate values ​​are valid flanges.

[0112] 1) For three-dimensional space:

[0113] F(x,y,z)=0

[0114] 2) First-order partial derivative:

[0115]

[0116] 3) Differential operator of temperature field in horizontal direction:

[0117]

[0118] 4) Differential operator in the vertical direction of the temperature field:

[0119]

[0120] 5) Calculate the gradient and direction:

[0121]

[0122]

[0123] By calculating the rate of change of the unit temperature field, the area with significant temperature change is identified and the position of the flange is preliminarily determined. Figure 2 As shown in the figure, the arrow direction represents the gradient direction. Taking the third temperature point in the third row as an example, the intensity of this point [8] is compared with the intensity of the point above it [2]. Since this point has the largest change, it is an edge point. This is used to preliminarily locate the flange position of the rolled piece.

[0124] Visible light image analysis:

[0125] For visible light color information, an edge detection algorithm is used to extract edge information in the image, which is compared with the temperature change rate analysis results for further precise positioning. Combined with the color information in the object image, the edge detection algorithm is used to calculate the gradient size and direction of the color parameters. The gradient is found by calculating the derivatives of the color parameters in the horizontal and vertical directions. The degree and direction of change of the color parameters are determined based on the gradient. The edge extraction operator is used to calculate the first-order derivatives in the horizontal and vertical directions to obtain the edge gradient and direction. The gradient direction is perpendicular to the edge. Secondly, based on the gradient size and direction, the gradient of the entire image is scanned to remove those non-edge numerical points and retain possible edge points. Finally, the double threshold method is used to determine real and potential edges. The position and direction of the edge are determined by comparing the gradient value of the color point with the set threshold (high threshold and low threshold). For example Figure 3 As shown in the figure, the direction of the arrow represents the gradient direction. Taking the fifth point in the third row as an example, since the gradient direction is upward, the intensity of this point [9] is compared with the intensities of the two points above and below it [6] and [3]. Since this point has the largest intensity, it is retained.

[0126] The color change rate calculation formula is: where G is the color gradient, X is the vertical coordinate, and Y is the vertical coordinate.

[0127] 1) Differential operator in the horizontal direction of the parameter:

[0128]

[0129] 2) Differential operator in the vertical direction of the parameter:

[0130]

[0131] 3) Calculate the gradient and direction:

[0132]

[0133]

[0134] Fusion of temperature field and visible light image:

[0135] The preprocessed temperature field data is fused with the visible light image. The two data sources are aligned through image registration technology to ensure the consistency of spatial position. During the fusion process, the weighted average method is used to assign weights according to the confidence of the temperature field data and the visible light image. The effective flange position of the rolled piece is identified by combining the temperature gradient and color information. The specific operations include:

[0136] Extract high temperature areas in the temperature gradient.

[0137] Identify red areas in the HSV color space.

[0138] Superimpose the two results, remove irrelevant areas, and retain the valid flange position.

[0139] Effective flange position determination:

[0140] The analysis results of temperature field data and visible light image data are integrated to eliminate invalid flange image data, restore useful real information, and determine the effective flange position of the rolled piece. In the positioning process, edge detection algorithms are used to extract edge information in the image and compare it with the temperature change rate analysis results to further accurately locate. Multi-scale analysis technology is used to analyze images of different scales to improve the accuracy and stability of positioning.

[0141] Among them, the detection of the rolled piece entering the monitoring area and the adjustment of the infrared spectrum detection device

[0142] Rolling inspection:

[0143] The continuous temperature field change rate and color information in the HSV color space are used to detect whether the rolled piece has entered the monitoring area. The system determines the entry of the rolled piece through a preset threshold, and automatically starts the exposure and frame rate adaptive adjustment function of the infrared spectrum detection device when the rolled piece enters the monitoring area.

[0144] Device adjustment:

[0145] The parameters of the infrared spectrum detection device are automatically adjusted according to the real-time detected changes in the position and temperature field of the rolled piece. The exposure time is adjusted according to the surface reflectivity and lighting conditions of the rolled piece to ensure clear images. The frame rate is adjusted according to the moving speed of the rolled piece to ensure that sufficient image data can be obtained even at high speeds.

[0146] Among them, the effective flange and side temperature field generation and the rolled piece panoramic image generation

[0147] Temperature field data processing:

[0148] The effective flange position of the rolled piece is determined by combining the analysis results of the temperature field data and the visible light image data. The optimal translation is calculated by the similarity measurement algorithm, and the continuously shot rolled piece images are spliced ​​to generate a complete rolled piece panoramic image. After the panoramic image is generated, the image clarity and contrast are improved by image enhancement technology to facilitate subsequent analysis.

[0149] Among them, the effective flange is used to improve the recognition accuracy of uneven cooling.

[0150] Uneven cooling detection:

[0151] The temperature data clustering algorithm is used to identify areas of uneven cooling. The temperature change rate is determined by calculating the temperature difference between adjacent temperature points. The clustering algorithm classifies the temperature data into multiple categories to identify areas of uneven cooling. The color and shape information in the visible light image is combined to compare and verify the uneven cooling phenomenon.

[0152] Improved accuracy:

[0153] The temperature field data and visible light image data are combined by weight fusion to improve the accuracy of uneven cooling recognition. The specific method includes assigning different weights to different data sources, comprehensively evaluating the detection results, and eliminating noise and misjudgment. The fusion results are trained and optimized through machine learning algorithms to improve the accuracy and robustness of the detection model.

[0154] Among them, build the machine neural network ES structure, such as Figure 6 As shown, a learning algorithm is designed to continuously optimize the temperature change rate analysis and image fusion algorithm to improve the system's adaptability and detection accuracy. According to historical data and actual detection conditions, the algorithm parameters are adjusted to ensure that the system can operate stably under different working conditions. Specific methods include:

[0155] Use supervised learning algorithms to train historical inspection data and construct a mapping relationship between temperature change rate and effective flange position.

[0156] Use reinforcement learning algorithms to adjust system parameters based on real-time detection results and improve the system's adaptability.

[0157] The present invention realizes effective flange positioning of rolled pieces of different sizes by using a camera at a fixed viewing angle during visual measurement and positioning of the side of the rolled piece. The method improves the accuracy of the side positioning of the rolled piece, avoids interference caused by the edge of the other side of the rolled piece, and ensures the stability and reliability of the positioning process. The system is designed to meet the high accuracy and efficiency requirements for the surface quality monitoring of the rolled piece, thereby improving the production quality of the rolled piece.

[0158] This method effectively handles the positioning error caused by the size difference of the rolled piece and realizes accurate and real-time flange positioning in the high-speed moving rolled piece. This strategy can still accurately determine the effective flange position of the rolled piece under complex background and undesirable lighting conditions. This method plays an important role in improving the accuracy and efficiency of infrared image processing of rolled pieces.

[0159] A new positioning algorithm was invented, which collects infrared spectral signals as the basis for calculation. This method obtains the temperature field data of the side of the rolled piece in real time through an infrared spectral detection device, utilizes the gradient information of the continuous temperature field on the side of the rolled piece, and analyzes the temperature change rate in the horizontal and vertical directions of the continuous temperature field to extract invalid flange image data and eliminate them.

[0160] Based on the infrared spectrum signal, the color information of visible light is combined for auxiliary positioning. The edge information in the image is extracted using the edge detection algorithm using the image color information to extract the invalid wing image data, and to eliminate and restore the useful real information, thereby improving the accuracy of positioning.

[0161] Based on the neural network ES structure, a learning algorithm is designed to optimize the temperature change rate analysis algorithm and image fusion algorithm to continuously improve the positioning accuracy.

[0162] In the process of visual measurement and positioning of the side of the rolled piece, the present invention realizes effective flange positioning of rolled pieces of different sizes through a camera with a fixed viewing angle, ensuring the accuracy and efficiency of surface quality monitoring of the rolled piece. This method not only improves the level of automation in the production process of rolled pieces, but also significantly improves production quality. Through precise positioning and real-time monitoring, human errors and quality defects are effectively reduced, the consistency and reliability of products are improved, and the overall production efficiency is improved, and the production cost is reduced.

[0163] In summary, the method for locating the effective flange of the rolled piece side based on infrared imaging locates the position of the effective flange of the rolled piece by analyzing the temperature-position change rate of the continuous temperature field in the horizontal and vertical directions. At the same time, the edge information in the image is extracted using the edge detection algorithm using the image color information, and compared with the temperature change rate analysis result, thereby improving the accuracy and reliability of the detection. In the process of visual measurement and positioning on the side of the rolled piece, the camera can locate the effective flange of the rolled piece of different sizes at a fixed viewing angle, avoiding interference caused by another measured edge. It aims to meet the accuracy and efficiency of surface quality monitoring of high-temperature rolled pieces and improve the production quality of rolled pieces.

[0164] The effective flange positioning of rolled products of different sizes is achieved through a camera with a fixed viewing angle, ensuring the accuracy and efficiency of surface quality monitoring of rolled products. This method not only improves the level of automation in the production process of rolled products, but also significantly improves production quality. Through precise positioning and real-time monitoring, human errors and quality defects are effectively reduced, the consistency and reliability of products are improved, and the overall production efficiency is improved, and production costs are reduced.

[0165] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0166] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for locating the effective flange of a rolled piece side based on infrared imaging, characterized in that: The specific operations are as follows: S1. Install an infrared temperature measuring device within a certain range on the side of the high-temperature rolled piece, and use an infrared spectrum detection device to obtain temperature field data on the side of the object in real time; S2. Using the temperature field information, identify the high temperature area of ​​the rolled piece, and calculate the moving speed of the object. According to the relationship between the moving speed of the object and the temperature change, the temperature judgment threshold is adaptively adjusted to eliminate abnormal data and generate a temperature field map of the moving object and its surroundings. S3, comprehensive analysis of temperature field data and visible light images. The acquired temperature field data and visible light images are first preprocessed. For visible light images, in order to simplify the color feature extraction of the image, color space conversion is performed. The spectral color data is statistically analyzed, and the threshold is set according to the statistical spectral range and color distance to perform image color segmentation. S4. According to the analysis of the temperature-position change rate in the vertical direction of the continuous temperature field, the position where the temperature point data change rate suddenly changes is the edge point, that is, the position where the temperature point data changes the most. The marginal extraction operator is used to calculate the first-order partial derivatives in the horizontal and vertical directions of the temperature field data. First, a three-dimensional coordinate system is established to generate a three-dimensional surface function image z=f(x,y), where x represents the horizontal coordinate, y represents the vertical coordinate, and z represents the temperature value of the point. Partial derivatives are calculated in the horizontal and vertical directions, and extreme values ​​are calculated based on the partial derivative results. The extreme value points in the interval are the discrete points where the temperature point data change rate suddenly changes. In the image, we use gradient to represent the rate and direction of change of the temperature field. After obtaining the temperature field gradient, the gradient of the entire image is scanned, and each temperature point is checked. According to the two principles of whether the gradient of the point is the largest among the points with the same gradient direction and whether the gradient direction is perpendicular to the edge, the edge point is found; the coordinates of the discrete mutation points are gathered into a line through the trajectory filtering algorithm, which is the flange. The effective flange or invalid flange is determined according to the ordinate value of the discrete points in the flange. The two flanges with the lowest ordinate are effective flanges. S5. Use the edge detection algorithm to extract edge information from the image at the same time, and compare it with the temperature change rate analysis results to further accurately locate the image. Combined with the color information in the object image, use the edge detection algorithm to calculate the gradient size and direction of the color parameters, find the gradient by calculating the derivatives of the color parameters in the horizontal and vertical directions, determine the degree and direction of change of the color parameters based on the gradient, and use the edge extraction operator to calculate the first-order derivatives in the horizontal and vertical directions to obtain the edge gradient and direction. The gradient direction is perpendicular to the edge. Secondly, based on the gradient size and direction, scan the gradient of the entire image, remove those non-edge numerical points, and retain possible edge points. Finally, compare the gradient value of the color point with the set threshold to determine the position and direction of the edge; S6. During the positioning process, edge detection algorithms are used to extract edge information from the image, and compared with the temperature change rate analysis results to further accurately locate the object. Multi-scale analysis technology is used to analyze images of different scales to improve the accuracy and stability of positioning. The temperature field and visible light image are fused, and the two data sources are aligned through image registration technology to ensure the consistency of spatial position. The weighted average method is used to assign weights according to the confidence of the temperature field data and the visible light image. Combined with the gradient information and color information of the object's continuous temperature field, the effective flange position of the rolled piece is identified, the limited flange and effective side data are determined, and the invalid flange and interference area image data are eliminated; S7. Use the continuous temperature field change rate to detect whether the rolled piece has entered the monitoring area, and automatically adjust the parameters of the infrared spectrum detection device according to the real-time detected rolled piece position and temperature field changes. Combine the temperature field data and visible light image data analysis results to determine the effective flange position of the rolled piece and generate a panoramic image of the rolled piece; S8. After removing the invalid flange, the recognition accuracy of uneven cooling is improved. The temperature data clustering algorithm is used to identify the area of ​​uneven cooling. The temperature change rate is determined by calculating the temperature difference between adjacent temperature points. The clustering algorithm divides the temperature data into multiple categories to identify the area of ​​uneven cooling. Combined with the color and shape information in the visible light image, the uneven cooling phenomenon is compared and verified; S9. During the temperature field data analysis, we build the machine neural network ES structure, design the learning algorithm, and continuously optimize the temperature change rate analysis and image fusion algorithm, which helps to improve the system's adaptability and detection accuracy. According to historical data and actual detection conditions, we adjust the algorithm parameters to ensure that the system can operate stably under different working conditions.

2. The method for locating the effective flange of a rolled piece side based on infrared imaging according to claim 1, characterized in that: An infrared temperature measuring device is installed on the side of the rolled piece to obtain temperature field data within a certain range. When installing the device, attention should be paid to the adjustment of angle and height to avoid blind areas. The device is fixed on a special bracket, which can be adjusted according to the height and width of the rolled piece to adapt to rolled pieces of different specifications; The infrared spectrum detection device is used to obtain the temperature field data of the side of the rolled piece in real time. The data collected by the device is transmitted to the central control system through a high-speed data transmission interface. The sampling frequency is usually set to more than 10 times per second, and the specific frequency is adjusted according to the moving speed of the rolled piece and the cooling process requirements. The data transmission process uses shielded cables and anti-interference filters to ensure the stability and accuracy of data transmission, and to ensure the integrity and anti-interference ability of the signal.

3. The method for locating the effective flange of a rolled piece side based on infrared imaging according to claim 1, characterized in that: The acquired temperature field data is first preprocessed, including denoising, smoothing and normalization. Denoising uses a median filter to eliminate random noise. Smoothing uses a Gaussian filter to ensure the continuity and smoothness of the temperature field data. Normalization converts the temperature data to a uniform scale for subsequent analysis; A visible light camera is installed next to the infrared temperature measuring device to capture visible light images of the side of the rolled piece in real time. The visible light image undergoes image preprocessing steps, including denoising, graying, and edge detection. The visible light image is converted to color space, and the RGB space is converted to HSV space to simplify the color feature extraction of the image, so as to express the brightness and tone of the color and facilitate the contrast between colors. The contour information of the rolled piece is extracted using color filtering and edge detection algorithms. Color filtering filters out the red area on the surface of the rolled piece by setting the range of H values ​​(such as 0-10 and 160-180). The principle of converting RGB space to HSV space is: Assume the input RGB values ​​are R, G, B, and the value range is 0, 255. First, they need to be normalized: Calculate the maximum and minimum values: C max =max(r,g,b),C min =min(r,g,b) Calculate Hue: Let Δ = C max -C min , Calculate Saturation: Calculate the value: V = C max .

4. The method for locating the effective flange of a rolled piece side based on infrared imaging according to claim 1, characterized in that: Temperature change rate analysis; By analyzing the temperature-position change rate in the vertical direction of the continuous temperature field, the position where the temperature point data change rate suddenly changes is the edge point, and the flange position of the rolled piece is preliminarily located. First, a three-dimensional coordinate system is established to generate a three-dimensional surface function image z=f(x,y), where x represents the horizontal coordinate, y represents the vertical coordinate, and z represents the temperature value of the point. Partial derivatives are calculated in the horizontal and vertical directions, and the extreme values ​​are calculated based on the partial derivative results. The extreme value point in the interval is the position where the temperature point data change rate suddenly changes. The marginal extraction operator is used to calculate the first-order partial derivatives in the horizontal and vertical directions of the temperature field data to obtain the temperature field gradient and direction. In the image, we use the gradient to represent the degree and direction of the temperature field change. After obtaining the temperature field gradient, the gradient of the entire image is scanned, and each temperature point is checked. According to the two principles of whether the gradient of the point is the largest among the points with the same gradient direction and whether the gradient direction is perpendicular to the edge, the edge point is found. The coordinates of the discrete mutation points are gathered into a line through the trajectory filtering algorithm, which is the flange. The flange is judged as a valid flange or an invalid flange based on the ordinate value of the discrete point in the flange. The two flanges with the lowest ordinate values ​​are valid flanges. 1) For three-dimensional space: F(x,y,z)=0 2) First-order partial derivative: 3) Differential operator of temperature field in horizontal direction: 4) Differential operator in the vertical direction of the temperature field: 5) Calculate the gradient and direction: By calculating the rate of change of the unit temperature field, the area with significant temperature change is identified and the position of the flange is preliminarily determined. Visible light image analysis: For visible light color information, edge detection algorithms are used to extract edge information in the image, and compared with the results of temperature change rate analysis to further accurately locate. Combined with the color information in the object image, the edge detection algorithm is used to calculate the gradient size and direction of the color parameters. The gradient is found by calculating the derivatives of the color parameters in the horizontal and vertical directions. The degree and direction of change of the color parameters are determined according to the gradient. The edge extraction operator is used to calculate the first-order derivatives in the horizontal and vertical directions to obtain the edge gradient and direction. The gradient direction is perpendicular to the edge. Secondly, based on the gradient size and direction, the gradient of the entire image is scanned to remove those non-edge numerical points and retain possible edge points. Finally, the double threshold method is used to determine the real and potential edges. The position and direction of the edge are determined by comparing the gradient value of the color point with the set threshold (high threshold and low threshold). The color change rate calculation formula is: where G is the color gradient, X is the vertical coordinate, and Y is the vertical coordinate. 1) Differential operator in the horizontal direction of the parameter: 2) Differential operator in the vertical direction of the parameter: 3) Calculate the gradient and direction: Fusion of temperature field and visible light image: The preprocessed temperature field data is fused with the visible light image. The two data sources are aligned through image registration technology to ensure the consistency of spatial position. During the fusion process, the weighted average method is used to assign weights according to the confidence of the temperature field data and the visible light image. The effective flange position of the rolled piece is identified by combining the temperature gradient and color information. The specific operations include: Extract high temperature areas in the temperature gradient. Identify red areas in the HSV color space. Superimpose the two results, remove irrelevant areas, and retain the valid flange position. Effective flange position determination: The analysis results of temperature field data and visible light image data are integrated to eliminate invalid flange image data, restore useful real information, and determine the effective flange position of the rolled piece. In the positioning process, edge detection algorithms are used to extract edge information in the image and compare it with the temperature change rate analysis results to further accurately locate. Multi-scale analysis technology is used to analyze images of different scales to improve the accuracy and stability of positioning.

5. The method for locating the effective flange of a rolled piece side based on infrared imaging according to claim 1, characterized in that: Detection of rolled pieces entering the monitoring area and adjustment of infrared spectrum detection device Rolling inspection: The continuous temperature field change rate and color information in the HSV color space are used to detect whether the rolled piece has entered the monitoring area. The system determines the entry of the rolled piece through a preset threshold, and automatically starts the exposure and frame rate adaptive adjustment function of the infrared spectrum detection device when the rolled piece enters the monitoring area. Device adjustment: The parameters of the infrared spectrum detection device are automatically adjusted according to the real-time detected changes in the position and temperature field of the rolled piece. The exposure time is adjusted according to the surface reflectivity and lighting conditions of the rolled piece to ensure that the acquired image is clear. The frame rate is adjusted according to the moving speed of the rolled piece to ensure that sufficient image data can be obtained even at high-speed movement.

6. The method for locating the effective flange of a rolled piece side based on infrared imaging according to claim 1, characterized in that: Generation of effective flange and side temperature fields and panoramic images of rolled products; The effective flange position of the rolled piece is determined by combining the analysis results of the temperature field data and the visible light image data. The optimal translation is calculated by the similarity measurement algorithm, and the continuously shot rolled piece images are spliced ​​to generate a complete rolled piece panoramic image. After the panoramic image is generated, the image clarity and contrast are improved by image enhancement technology to facilitate subsequent analysis.

7. The method for locating the effective flange of a rolled piece side based on infrared imaging according to claim 1, characterized in that: Using effective flange to improve the accuracy of identifying uneven cooling Uneven cooling detection: The temperature data clustering algorithm is used to identify areas of uneven cooling. The temperature change rate is determined by calculating the temperature difference between adjacent temperature points. The clustering algorithm classifies the temperature data into multiple categories to identify areas of uneven cooling. The color and shape information in the visible light image is combined to compare and verify the uneven cooling phenomenon. Improved accuracy: The temperature field data and visible light image data are combined by weight fusion to improve the accuracy of uneven cooling recognition. The specific method includes assigning different weights to different data sources, comprehensively evaluating the detection results, and eliminating noise and misjudgment. The fusion results are trained and optimized through machine learning algorithms to improve the accuracy and robustness of the detection model.

8. The method for positioning the effective flange of a rolled piece side based on infrared imaging according to claim 1, characterized in that: Build the machine neural network ES structure, design the learning algorithm, continuously optimize the temperature change rate analysis and image fusion algorithm, and improve the system's adaptability and detection accuracy. According to historical data and actual detection conditions, adjust the algorithm parameters to ensure that the system can operate stably under different working conditions. Specific methods include: Use supervised learning algorithms to train historical inspection data and construct a mapping relationship between temperature change rate and effective flange position. Use reinforcement learning algorithms to adjust system parameters based on real-time detection results and improve the system's adaptive capabilities.

Images