Method for detecting defects in continuously cast and rolled billets based on optical profilometry

Abstract

The present application relates to continuous casting and rolling blank detection technical field, especially in kind based on optical profile measurement's continuous casting and rolling blank defect detection method, including: with multi-angle polarization state coding light compensation way irradiation to be measured metal blank surface, and in the same field of view obtains corresponding polarization state image;Based on the polarization state image of different polarization direction obtained, the surface polarization degree image is calculated, and the specular reflection component and the diffuse reflection component are separated according to the polarization decoupling reflection model, and the mirrorless surface diffuse reflection enhancement image is obtained;The multiscale geometric features of the enhanced image are extracted to determine the candidate defect area, and the local polarization degree attenuation rate of the candidate area is extracted in the surface polarization degree image to check the authenticity, and the false defects are filtered out.Through the present application, the micron level micro defect detection rate of high light reflection oxygen-free copper flat blank surface can be solved, and the defect detection accuracy is improved to meet the quality control requirements of high-precision enameled wire production.

Description

Technical Field

[0001] This invention relates to the technical field of continuous casting and rolling billet inspection, and in particular to a method for detecting defects in continuous casting and rolling billets based on optical profile measurement. Background Technology

[0002] Oxygen-free copper flat billets produced by continuous casting and rolling processes are the basic material for manufacturing high-precision enameled wires. Their surface quality directly affects the yield of subsequent wire drawing and coating processes, as well as the electrical performance and withstand voltage rating of the final enameled wire. During the continuous casting and rolling process, due to factors such as billet solidification, rolling deformation, and surface oxidation, micro-defects such as oxide scale indentation, hot cracks, bubble exposure holes, and inclusion precipitation marks may occur on the surface of the oxygen-free copper flat billet.

[0003] Currently, the main methods for detecting surface defects in metal billets include eddy current testing, magnetic flux leakage testing, and automated optical inspection based on two-dimensional vision. Among these, two-dimensional vision inspection is widely used in the field of metal surface defect detection due to its advantages such as non-contact operation, high efficiency, and the ability to record images. A typical two-dimensional vision inspection scheme usually uses a ring light source or a strip light source to uniformly illuminate the surface under test, uses an industrial camera to acquire grayscale images of the surface, and then uses image processing algorithms such as grayscale thresholding, edge extraction, and template matching to identify defect areas.

[0004] However, after rolling, the surface of oxygen-free copper billets exhibits highly reflective, mirror-like properties, with specular reflection being the dominant surface reflection behavior. On such highly reflective surfaces, the imaging characteristics of micro-defect areas no longer present stable contrasts or fixed shapes, but rather undergo significant optical distortion with minute changes in the incident angle of the light source, the camera's shooting angle, and the local surface normal. Specifically, the same micro-defect may appear as a bright spot, a dark spot, or even completely blend into the background under different illumination angles. This makes it difficult for existing two-dimensional vision algorithms based on grayscale thresholds or simple shape matching to reliably capture and describe the essential characteristics of defects, resulting in an extremely low detection rate for micron-level micro-defects, which cannot meet the surface quality control requirements of high-precision enameled wire production for oxygen-free copper billets. Summary of the Invention

[0005] This invention provides a method for detecting defects in continuously cast and rolled billets based on optical profile measurement, which can effectively solve the problems in the background art.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows: A method for detecting defects in continuously cast and rolled billets based on optical profile measurement includes: The surface of the metal billet under test is illuminated by multi-angle polarization state encoding supplementary light, which includes at least three light sources with different polarization directions. Acquire polarization state images of the surface of the metal billet under test in at least three different polarization directions under the same field of view; Based on the obtained polarization state images of at least three different polarization directions, the surface polarization degree image is calculated, and the specular reflection component and diffuse reflection component are separated according to the polarization decoupled reflection model to obtain a de-specularized surface diffuse reflection enhancement image. Multi-scale geometric features are extracted from the surface diffuse reflection enhanced image, and candidate defect regions are determined based on the multi-scale geometric features; For candidate defect regions, the local polarization attenuation rate is extracted from the surface polarization image; The candidate defect regions are verified based on the local polarization attenuation rate. False defect regions with polarization attenuation rates lower than the preset attenuation rate threshold are filtered out, and the candidate defect regions that pass the verification are identified as micro-defects.

[0007] Furthermore, the polarization directions of the three light sources with different polarization directions are set to 0°, 45° and 90°, respectively.

[0008] Furthermore, the surface of the metal billet under test is illuminated by multi-angle polarization state encoding supplementary light, which is achieved through a coaxial ring light source; The coaxial ring light source consists of a first ring polarized light source, a second ring polarized light source, and a third ring polarized light source arranged sequentially from the inside out, corresponding to polarization directions of 0°, 45°, and 90°, respectively.

[0009] Furthermore, the incident angles of the first, second, and third annular polarized light sources are respectively set to correspond to the time-varying Brewster angle range of the metal billet under test within the continuous casting temperature range, so as to differentially suppress the specular reflection light component in each polarization channel.

[0010] Furthermore, based on the acquired polarization state images of at least three different polarization directions, a surface polarization degree image is calculated, including: Based on polarization state images of at least three different polarization directions, a cosine function of polarization intensity as a function of polarization angle is fitted to solve for the polarization degree parameter and generate a surface polarization degree image.

[0011] Furthermore, the multi-scale geometric features include micro-defect curvature direction field features and / or texture anomaly response features.

[0012] Furthermore, multi-scale geometric features are extracted from the surface diffuse reflection enhancement image, including: A Hessian matrix is ​​constructed from the surface diffuse reflection enhancement image, and the eigenvector direction corresponding to the largest eigenvalue of the Hessian matrix is ​​calculated as the field feature of the curvature direction of the micro-defect. And / or, A Gabor filter bank with multiple directional parameters and multiple center frequency parameters is used to perform convolution filtering on the surface diffuse reflection enhancement image, and the texture response amplitude of each filter channel is extracted as a texture anomaly response feature.

[0013] Furthermore, the local polarization attenuation rate is the rate at which the polarization value at the edge of the candidate defect region decreases relative to the polarization value in the background region.

[0014] Furthermore, the preset attenuation rate threshold is determined based on the statistical distribution of the polarization degree attenuation rate of typical micro-defects on the surface of the metal billet to be tested.

[0015] Furthermore, typical micro-defects include at least one of the following: oxide scale indentation, hot cracks, bubble exposure pores, and inclusion precipitation marks.

[0016] The technical solution of this invention can achieve the following technical effects: By using multi-angle polarization state encoding supplementary lighting, the surface of the billet under test is illuminated by at least three light sources with different polarization directions. Combined with the polarization decoupled reflection model, the specular reflection component and diffuse reflection component are separated to obtain a de-spectrified surface diffuse reflection enhanced image. This can eliminate the interference of high reflectivity on micro-defect imaging, making the imaging characteristics of micron-level micro-defects clearer and more stable. By extracting multi-scale geometric features from the surface diffuse reflection enhanced image to determine candidate defect regions, and combining the local polarization attenuation rate in the surface polarization degree image for authenticity verification, the interference of false defects is filtered out. This can improve the detection stability and accuracy of typical micro-defects such as oxide scale indentation, hot cracks, bubble exposure holes, and inclusion precipitation marks, meeting the surface quality control requirements of oxygen-free copper flat billets in high-precision enameled wire production. At the same time, it maintains the advantages of non-contact, high efficiency, and imaging recording, and is suitable for the continuous inspection needs of continuous casting and rolling production lines.

[0017] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the following are specific embodiments of this application. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a schematic flowchart of the continuous casting and rolling billet defect detection method based on optical profile measurement according to the present invention. Detailed Implementation

[0020] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0021] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0022] like Figure 1 As shown, the method for detecting defects in continuously cast and rolled billets based on optical profile measurement of the present invention specifically includes the following steps: Step S100: Illuminate the surface of the metal billet to be tested with multi-angle polarization state encoding supplementary light. The multi-angle polarization state encoding supplementary light includes at least three light sources with different polarization directions. Step S200: Obtain polarization state images of the surface of the metal billet under test in at least three different polarization directions under the same field of view; Step S300: Based on the obtained polarization state images of at least three different polarization directions, calculate the surface polarization degree image, and separate the specular reflection component and diffuse reflection component according to the polarization decoupled reflection model to obtain a de-specimenized surface diffuse reflection enhancement image. Step S400: Extract multi-scale geometric features from the surface diffuse reflection enhancement image, and determine candidate defect regions based on the multi-scale geometric features; Step S500: For the candidate defect region, extract the local polarization attenuation rate from the surface polarization image; Step S600: Verify the authenticity of candidate defect regions based on local polarization attenuation rate, filter out false defect regions with polarization attenuation rate lower than preset attenuation rate threshold, and take the candidate defect regions that pass the verification as the identified micro-defects.

[0023] In this embodiment, incident light sources with different polarization directions cause the anisotropic specular reflection component of the surface to exhibit decoupled differential responses in different polarization images, while the diffuse reflection component remains relatively stable. Based on this, the surface polarization degree image is calculated and the specular reflection and diffuse reflection components are separated to obtain a de-spectrified surface diffuse reflection enhanced image. This eliminates the interference of specular highlights on defect imaging, making micro-defects that originally appeared as bright spots, dark spots, or blended into the background due to changes in illumination angle appear as stable and repeatable geometric features in the diffuse reflection enhanced image. Furthermore, candidate defect regions are determined by extracting multi-scale geometric features, and the local polarization degree attenuation rate is extracted from the polarization degree image for authenticity verification. Ultimately, a high detection rate and low false alarm rate for micron-level micro-defects are achieved, and the detection process is insensitive to changes in illumination angle and local surface normal, meeting the requirements for surface quality control of oxygen-free copper blanks in high-precision enameled wire production.

[0024] In a specific implementation, as one example, given that specular reflection carries information about the minute undulations of the surface normal, while diffuse reflection carries information about the intrinsic absorption of the material, separating the specular reflection component from the total reflected light can suppress the interference of highly reflective background on defect imaging. The polarization states respond differently to specular and diffuse reflection; specular reflection retains the polarization state of the incident light to a higher degree, while diffuse reflection produces depolarization. Therefore, by illuminating with multi-polarization directions and analyzing the polarization characteristics of the reflected light, a polarization-encoded illumination-imaging link is constructed, causing the specular reflection light to be differentially attenuated in different polarization channels, thus providing a physical basis for separating the diffuse reflection component. This embodiment uses a coaxial ring multi-polarization-encoded illumination device to achieve multi-angle polarization state encoded supplementary lighting, as detailed below: Step S110: Configure the physical structure of the coaxial ring illumination device. The device consists of a first ring-polarized light source, a second ring-polarized light source, and a third ring-polarized light source arranged sequentially from the inside out. The three ring-polarized light sources share the same optical axis, which is perpendicular to the average plane of the surface of the metal blank to be tested. Each ring-polarized light source can be independently controlled for on / off states and intensity, enabling time-division lighting and independent adjustment. A linear polarizing film is covered on the emitting surface of the first ring-polarized light source, with its transmission axis direction set to the 0° reference direction. A linear polarizing film is covered on the emitting surface of the second ring-polarized light source, with its transmission axis direction set to 45°. A linear polarizing film is covered on the emitting surface of the third ring-polarized light source, with its transmission axis direction set to 90°. A linear polarizer is fixedly installed in front of the camera lens as an analyzer, with its transmission axis direction aligned with the 0° reference direction. Step S120: Set the incident angle of each annular polarization light source so that the incident angles of the first, second, and third annular polarization light sources correspond to the time-varying Brewster angle range of the metal billet under test within the continuous casting temperature range. According to the physical characteristics of the Brewster angle, when the incident angle is equal to the Brewster angle, the p-polarization component in the reflected light is completely transmitted, and the reflected light only contains the s-polarization component. For metal conductors, the Brewster angle does not completely polarize the reflected light, but the ratio of the s-component to the p-component in the specular reflected light reaches an extreme value near this angle, that is, the specular reflected light is polarized to the maximum extent. Accordingly, the incident angles of the first, second, and third annular light sources are set near the lower limit, median, and upper limit of this time-varying Brewster angle range, respectively, to achieve differentiated suppression of the specular reflected light component. The method for obtaining the time-varying Brewster angle intervals corresponding to the three light sources is as follows: First, the nominal alloy composition of the billet to be tested is read from the process control system of the continuous casting and rolling production line; second, the complex refractive index of the alloy at multiple discrete temperature points within the continuous casting temperature range is measured using a spectroscopic ellipsometry, and then the Brewster angle at each temperature point is calculated; Brewster angle satisfy =n, where n is the real part of the refractive index of the metal at this temperature for the wavelength of incident light; in practice, the imaginary part of the refractive index of the metal will cause a slight shift in the Brewster angle position, so the incident angle corresponding to the peak value of the measured reflected light s / p ratio is used as the actual standard for the Brewster angle; then, the Brewster angle at each temperature point is fitted as a function of temperature to obtain a time-varying Brewster angle curve; finally, the minimum and maximum values ​​of this curve in the continuous casting temperature range are used as the incident angle reference values ​​of the first and third annular light sources, respectively, and the median value is used as the incident angle reference value of the second annular light source; Step S130: Set the working mode of the lighting device to achieve time-sharing lighting and synchronous imaging; the three ring polarized light sources adopt a sequential lighting mode, for example, switching on and off at a time-sharing frequency of 50Hz, and the camera synchronously acquires polarization state images under each light source to ensure that the acquisition of polarization state images in each polarization direction is not interfered with; since the polarization directions and incident angles of the three light sources are different, the same micro-defect will present different brightness responses in the three images.

[0025] In this embodiment, by coupling the illumination angle, polarization direction and Brewster angle, the specular reflection component of the flat background area exhibits a predictable stepped attenuation mode in the three channels; the defective area disrupts the specular reflection condition due to the local normal abrupt change, and its three-channel grayscale vector deviates from the background mode; this deviation provides input for the polarization decoupled reflection model, so that the diffuse reflection component can be separated and extracted, thereby obtaining a de-specularized surface diffuse reflection enhancement image.

[0026] In some embodiments of the present invention, based on the multi-angle polarization state encoding supplementary lighting method, three annular polarized light sources have different incident angles and polarization directions of 0°, 45°, and 90°. Polarization state images corresponding to the illumination of each light source need to be acquired separately under the same field of view, and the three images should be spatially aligned to avoid registration errors introduced by billet movement or vibration. This embodiment achieves this through a single camera combined with sequential alternating illumination, specifically by performing the following operations: Step S210: Configure the synchronization timing of the image acquisition system; set up a timing controller, which is simultaneously connected to the driving circuits of the first, second, and third annular polarized light sources and the camera's external trigger interface; the timing controller internally generates a clock signal with a fixed period, which outputs three high-level pulses of the same width in each period, with a fixed-duration low-level interval between adjacent pulses; the first high-level pulse is used to drive the first annular polarized light source to light up, and simultaneously sends the first frame trigger signal to the camera; the second high-level pulse is used to drive the second annular polarized light source to light up, and simultaneously sends the second frame trigger signal to the camera; the third high-level pulse is used to drive the third annular polarized light source to light up, and simultaneously sends the third frame trigger signal to the camera; after receiving each trigger signal, the camera starts exposure after a fixed delay, which is equal to the time required for the light source to reach a stable luminous intensity after being lit up, and the exposure time is set to not exceed the width of the high-level pulse to ensure that the exposure is completed within the window when the light source is still lit up; Step S220: Determine the parameter configuration of the camera and lens; use a global exposure type area array industrial camera with a monochrome CMOS sensor that does not include a Bayer filter to avoid spatial resolution loss introduced by color interpolation; install a linear polarizer in front of the camera lens as an analyzer, with the transmission axis of the analyzer fixed at 0°; use a telecentric lens for the camera, with its object-side telecentricity controlled within a preset accuracy range to ensure that the incident principal rays at different positions in the field of view are approximately parallel to the optical axis, thereby ensuring consistent polarization efficiency at the edge and center of the field of view; set the working distance between the camera and the surface of the metal billet to be measured to the nominal working distance value designed by the lens, and set the field of view size to cover the distribution range of typical micro-defects on the surface of the continuously cast and rolled billet, while ensuring that the physical size corresponding to a single pixel is no larger than the micrometer level to meet the spatial sampling requirements of micrometer-level micro-defects; Step S230: Continuously acquire three polarization state images under the same field of view; place the metal billet to be tested at the inspection station, ensuring its surface average plane is perpendicular to the camera's optical axis; start the timing controller, continuously run the aforementioned fixed-period clock signal, and the camera continuously outputs three frames of images according to this fixed period, denoted as the first polarization state image P0, the second polarization state image P45, and the third polarization state image P90, respectively; where P0 corresponds to the surface reflected light intensity distribution under the illumination of the first annular polarized light source, and P45 corresponds to the surface reflected light intensity under the illumination of the second annular polarized light source. The distribution of surface reflected light intensity under the illumination of the third annular polarized light source is shown in P90. Since there is a certain time span between the acquisition of the three frames, if the billet is in motion within this time span, the pixel coordinates of the same physical point on the billet surface in P0, P45, and P90 will be offset. To eliminate this offset, the encoder pulse count value at the start of each frame acquisition is recorded in the timing controller. The encoder is installed on the conveyor belt drive roller. Each pulse corresponds to a fixed small distance movement of the billet, and the count value is stored together with the corresponding image frame. Step S240: Spatial registration of the three polarization images; read three frames, P0, P45, and P90, and their corresponding encoder pulse count values. Let the encoder count value at the time of acquisition P0 be N0, the count value at the time of acquisition P45 be N45, and the count value at the time of acquisition P90 be N90; calculate the pixel offset of P45 relative to P0, which is calculated using the encoder pulse difference, the blank movement distance corresponding to a single encoder pulse, and the physical size corresponding to a single pixel of the camera; similarly calculate the pixel offset of P90 relative to P0; translate the P45 image along the opposite direction of the blank movement by the corresponding pixel offset, and translate the P90 image along the opposite direction of the blank movement by the corresponding pixel offset, so that the same physical point in the three frames is aligned to the same pixel coordinates; the translation operation uses bilinear interpolation, and the interpolated pixel values ​​maintain the original grayscale resolution; the three registered images are denoted as... , , Together, they form a set of polarization state images with three different polarization directions under the same field of view.

[0027] In this embodiment, by combining a single global exposure camera with time-sequential alternating illumination, and coordinating motion compensation registration with encoder feedback, three polarization images under the same field of view are acquired under online continuous inspection conditions. This enables the output of three-channel polarization images at the frame rate required for production line inspection, adapting to the inspection needs of continuously cast and rolled billets during online movement. In the three registered images, the same flat background area is... , , The stable grayscale ratio provides accurate input data for fitting a cosine function of polarization intensity as a function of polarization angle.

[0028] In practical implementation, as one example, existing polarization processing schemes often employ single polarization channel analysis or simple grayscale fusion, failing to fully utilize the essential difference in response between specular and diffuse reflection. This results in the inability to accurately separate specular and diffuse reflection components, leading to residual specular reflection in the processed image, insufficient differentiation between defect areas and the background, and a lack of reliable fitting basis for polarization degree calculation, failing to accurately reflect the surface polarization state distribution characteristics. Therefore, precise polarization degree calculation and a reasonable polarization decoupling reflection model are needed to effectively separate specular and diffuse reflection components while preserving the diffuse reflection characteristics of micro-defects. This embodiment obtains a de-specularized surface diffuse reflection enhanced image through the synergy of polarization intensity fitting, polarization degree parameter solving, polarization decoupling reflection model construction, and component separation. The specific implementation steps are as follows: Step S310: Preprocess the registered polarization state image. The preprocessing operations include grayscale normalization and noise smoothing. Grayscale normalization is used to eliminate the grayscale reference deviation caused by the slight difference in light source intensity in different polarization directions and the uneven camera response. It maps the grayscale values ​​of the three images to the same preset grayscale range to ensure the consistency of polarization intensity contrast. Noise smoothing adopts the median filtering algorithm to filter out residual random noise and noise generated by industrial environment interference in the image, while retaining the edge and polarization state features of micro-defect areas in the image, avoiding blurring of defect features during the filtering process. The size of the filtering window is determined according to the previous experiment, with the goal of achieving a balance between noise suppression effect and defect feature preservation effect. Step S320: Fit a cosine function of polarization intensity as a function of polarization angle to solve for the polarization degree parameter of each pixel; for each pixel in the three preprocessed images, extract the gray value of the pixel, and use the three gray values ​​as the observed polarization intensity values ​​of the pixel at the corresponding polarization angle; based on the least squares method, with the polarization angle as the independent variable and the observed polarization intensity value as the dependent variable, fit a cosine function of polarization intensity as a function of polarization angle of the pixel; solve for the polarization degree parameter of the pixel using the fitted cosine function; the polarization degree parameter is used to characterize the strength of the polarization state of the surface region corresponding to the pixel, and its value is positively correlated with the degree of polarization of the surface reflected light; Step S330: The polarization degree parameter of each pixel obtained by the solution is matched one-to-one with the pixel coordinates of the original image to generate a surface polarization degree image; the gray value of the surface polarization degree image has a linear correspondence with the polarization degree parameter value. The higher the gray value, the higher the degree of polarization of the corresponding surface area, that is, the higher the proportion of specular reflection; the lower the gray value, the lower the degree of polarization of the corresponding surface area, that is, the higher the proportion of diffuse reflection. Step S340: Input the polarization state image and the surface polarization degree image into the polarization decoupled reflection model. Combine the model parameters to calculate the specular reflection intensity and diffuse reflection intensity corresponding to each pixel. The construction of the polarization decoupled reflection model is based on the physical mechanism of metal surface reflection. Combined with the multi-angle polarization state encoding supplementary light characteristics, the total surface reflected light intensity is decomposed into two parts: specular reflection intensity and diffuse reflection intensity. The model parameters include the specular reflection polarization coefficient and the diffuse reflection polarization coefficient. The specular reflection polarization coefficient is determined according to the incident angle of each annular polarized light source, and the diffuse reflection polarization coefficient is determined according to the material characteristics of the oxygen-free copper blank to be tested. The specular reflection light has the highest polarization degree under Brewster angle, and its polarization coefficient has a fixed correlation with the incident angle. It can be obtained through previous high-temperature polarization reflection experiments. The diffuse reflection of oxygen-free copper has a fixed depolarization characteristic. Its polarization coefficient can be determined by measuring the diffuse reflection polarization state of the standard oxygen-free copper sample to ensure that the model parameters match the actual material characteristics. Step S350: Generate an initial diffuse reflection image by mapping the diffuse reflection light intensity of each pixel to the pixel coordinates of the original image; perform enhancement processing on the initial diffuse reflection image using a gray-scale stretching algorithm to map the gray-scale range of diffuse reflection light intensity to a preset high-contrast gray-scale range, thereby improving the gray-scale difference between the micro-defect area and the background area while preserving the detailed features of the micro-defects; the enhanced image is the de-spectrified surface diffuse reflection enhancement image, in which the specular reflection component is effectively separated and removed, and only the diffuse reflection component is retained.

[0029] In this embodiment, the generated surface polarization image can accurately reflect the surface polarization state distribution. The polarization parameter is uniform in the flat background area, and the polarization parameter in the defect area is significantly lower than that in the background area. The generated diffuse reflection enhanced image can remove specular reflection interference, and micro-defect features with a depth and width of micrometers can be clearly identified. Using this diffuse reflection enhanced image for multi-scale geometric feature extraction can accurately extract the curvature direction field features and texture anomaly response features of the defects.

[0030] In a specific implementation, as one example, given the morphological and dimensional differences of various types of micro-defects on the surface of oxygen-free copper blanks, single-scale, single-type feature analysis cannot comprehensively capture the geometric and textural features of various micro-defects, resulting in incomplete feature extraction, insufficient accuracy in determining candidate defect regions, and a tendency for missed or false judgments. Therefore, multi-scale adaptation design is needed to extract the curvature direction field features and / or texture anomaly response features of micro-defects, and combine the characteristics of the selected features to determine candidate defect regions, ensuring that all types of micro-defects can be accurately captured. This embodiment achieves the determination of candidate defect regions through multi-scale parameter setting, Hessian matrix construction, Gabor filter bank design, and feature screening. The specific implementation steps are as follows: Step S410: Set multi-scale parameters to adapt to the extraction of micro-defect features of different types and sizes. The multi-scale parameters include a scale factor sequence and a scale adaptation range. The scale factor sequence is determined through previous experiments and covers the size range of typical micro-defects on the surface of oxygen-free copper blanks. Each scale factor corresponds to a feature extraction scale to ensure that micro-defects of different sizes can be captured by the corresponding scale. The scale adaptation range is determined based on the actual size statistics of typical micro-defects. Combined with the pixel-level accuracy of diffuse reflection enhanced images, it ensures that the physical size corresponding to each scale factor matches the size of the micro-defect, avoiding the fusion of small defect features due to excessively large scales and the segmentation of large defect features due to excessively small scales. Step S420: Construct a multi-scale Hessian matrix for the surface diffuse reflection enhancement image. At each scale, a Hessian matrix is ​​constructed for each pixel in the image. The Hessian matrix can accurately describe the second-order gray-level changes of image pixels and capture the gray-level gradient differences between micro-defect areas and background areas. When constructing the Hessian matrix, a neighborhood window matching the current scale factor is selected with each pixel as the center. The second-order partial derivatives of the pixels within the window are calculated to form the Hessian matrix. The size of the neighborhood window is positively correlated with the scale factor. Step S430: For the Hessian matrix of each pixel at each scale, calculate its maximum eigenvalue and the corresponding eigenvector direction. Use the eigenvector direction as the curvature direction field feature of the micro-defect corresponding to that pixel. The maximum eigenvalue of the Hessian matrix can reflect the intensity of grayscale changes in the region where the pixel is located. The maximum eigenvalue of the defect region is significantly higher than that of the background region, which can distinguish the defect from the background. The eigenvector direction corresponding to the maximum eigenvalue can reflect the extension direction of the defect and is directly related to the geometric shape of the micro-defect. During the calculation, a numerically stable eigenvalue solution method is adopted to avoid deviation of the direction field feature caused by calculation error, ensuring that the curvature direction field feature can truly reflect the geometric shape of the micro-defect. At the same time, the curvature direction field feature at each scale is normalized to eliminate the feature deviation caused by scale differences and ensure the comparability of multi-scale features. Step S440: Design a multi-parameter Gabor filter bank to adapt to the extraction of texture anomaly features of micro-defects; the texture distribution in the micro-defect region is irregular and the grayscale changes frequently, while the texture distribution in the background region is uniform and the grayscale changes are gradual. The Gabor filter can capture this texture anomaly by setting specific parameters; the parameters of the filter bank include multiple directional parameters and multiple center frequency parameters. The number and angle of the directional parameters are determined statistically based on the extension direction of typical micro-defects, covering all possible extension directions of micro-defects to ensure that micro-defects with different orientations can be captured; the number and range of the center frequency parameters are determined based on the grayscale change frequency of the micro-defect texture, covering the frequency difference between the defect texture and the background texture to ensure that the defect texture and the background texture can be effectively distinguished; the specific settings of the directional parameters and the center frequency parameters are determined through preliminary experiments, combined with the texture characteristics of standard micro-defect samples, to ensure that the response sensitivity of the filter bank to the micro-defect texture is higher than that to the background texture. Step S450: Perform convolution filtering on the surface diffuse reflection enhancement image. Each filter corresponds to a filtering channel, and outputs a filtered response image. During the filtering process, maintain pixel-level alignment between the filter bank and the image to ensure that the filtered response of each pixel can accurately reflect its texture features. For the response image of each filtering channel, extract the texture response amplitude of each pixel as the texture anomaly response feature corresponding to that pixel. At the same time, fuse the texture response amplitudes of multiple channels to obtain a comprehensive texture anomaly response feature with multiple scales and directions. Step S460: Fuse the multi-scale curvature direction field features with the comprehensive texture anomaly response features, or directly use a single type of feature; the fusion method adopts feature concatenation, combining the curvature direction field features and texture anomaly response features of each pixel into the comprehensive feature vector of that pixel, and determining the pixels with feature values ​​as defective pixels. Step S470: Perform connected component analysis on the obtained defect pixels, grouping adjacent defect pixels into a connected component, with each connected component corresponding to a candidate defect region; as a complete geometric entity, the micro-defect has connected defect pixels, while the background pixels and defect pixels are not connected. Through connected component analysis, the scattered defect pixels can be integrated into a complete candidate defect region.

[0031] In this embodiment, the curvature direction field feature mainly reflects the geometric shape of the micro-defect, while the texture anomaly response feature mainly reflects the texture characteristics of the micro-defect. Both the geometric shape and texture characteristics of the micro-defect can serve as core features to distinguish the defect from the background. Different types of micro-defects respond differently to the two types of features. For example, thermal cracks are more easily captured by the curvature direction field feature, while inclusion precipitation marks are more easily captured by the texture anomaly feature. The feature type can be flexibly selected according to the actual detection needs. Furthermore, by determining the candidate defect region through connected component analysis, it can be ensured that all types of micro-defects can be captured.

[0032] In practical implementation, as an example, existing polarization degree-related extraction schemes mostly adopt global polarization degree statistics or single-pixel polarization degree extraction. They do not specifically distinguish between the edge of the candidate defect region and the background region, nor do they extract the relative change relationship of polarization degree between the two, thus failing to form a quantifiable attenuation parameter. Furthermore, they do not consider the essential difference in polarization state between typical micro-defects on the surface of oxygen-free copper blanks and the background, resulting in extracted parameters that cannot effectively support authenticity verification. This easily leads to false defects being misjudged as valid defects, and valid defects being misjudged as false defects. It is necessary to accurately extract the local polarization degree attenuation rate by locating the edge of the candidate defect region, selecting the background region, and scientifically calculating the polarization degree attenuation rate. This ensures that the parameter can truly reflect the difference in polarization state between valid micro-defects and false defects. The specific implementation steps are as follows: Step S510: Locate the edge of the candidate defect region and determine the edge sampling range; Align the candidate defect region data with the surface polarization degree image at the pixel level; Based on the aligned candidate defect region and surface polarization degree image, use an edge detection algorithm to locate the edge of each candidate defect region; The located edge is a continuous set of pixels; For each candidate defect region, set the edge sampling range, which is the edge pixel and its adjacent preset number of pixels. The number of adjacent pixels is determined according to the edge width of a typical micro-defect to ensure that the sampling range can cover the complete polarization degree distribution of the defect edge; Step S520: Select the background area corresponding to the candidate defect area and determine the background sampling range. The selection of the background area follows the principles of adjacency, uniformity, and no interference. A flat area adjacent to the candidate defect area and without other defects or noise interference is selected as the background area. Each candidate defect area corresponds to an independent background area. The area of ​​the background area is determined according to the area of ​​the candidate defect area to ensure that the polarization degree distribution of the background area is representative and to avoid polarization degree statistical deviation caused by the background area being too small. At the same time, the background sampling range is set, and a uniform sampling method is adopted. A preset number of sampling points are selected in the background area. The number of sampling points is determined according to the previous experiment. The polarization degree statistical value of the sampling points is used as the standard to represent the overall polarization degree level of the background area. Uniform sampling can avoid statistical deviation caused by the concentration of sampling points and ensure the accuracy of the background polarization degree value. Step S530: For the determined edge sampling range, extract the polarization degree value of each sampled pixel, and calculate the average polarization degree value of the edge sampling area using the arithmetic mean method; for the determined background sampling range, calculate the average polarization degree value of the background sampling area using the same arithmetic mean method to ensure that the calculation methods for the average polarization degree values ​​of the edge and the background are consistent, and avoid comparison deviations caused by differences in calculation methods; wherein, the average polarization degree value of the edge reflects the polarization state level of the edge of the candidate defect area, and the average polarization degree value of the background reflects the polarization state level of the normal area around the defect. The diffuse reflection ratio of the effective micro-defect area is high, and the polarization degree value is lower than that of the normal surface. The pseudo-defect area does not have such reflection characteristic differences, and the polarization degree value is basically consistent with that of the background area; Step S540: Based on the average polarization degree value of the edge and the average polarization degree value of the background, the local polarization degree attenuation rate is obtained by the ratio of the difference between the two to the average polarization degree value of the background. Effective micro-defects will cause changes in surface reflection characteristics. The polarization degree value of the defect edge will decrease significantly compared with the background area. The decrease rate can quantify the degree of this polarization state change. The decrease rate of effective micro-defects has a fixed statistical range, while the decrease rate of false defects is significantly lower than this range, which can be used as a quantitative basis for verification of authenticity.

[0033] In this embodiment, effective micro-defects on the surface of oxygen-free copper blanks alter surface reflection characteristics. The polarization degree value at the defect edge exhibits a fixed decreasing trend compared to the background area, which can be quantified by the attenuation ratio. False defects, however, do not exhibit this trend. By accurately locating the edge of the candidate defect area and reasonably selecting adjacent background areas, targeted polarization degree sampling and averaging calculations are performed to obtain a quantifiable local polarization degree attenuation rate. The attenuation rate parameter of effective micro-defects falls within a preset reasonable range, while the attenuation rate parameter of false defects is significantly lower than this range. Using the extracted local polarization degree attenuation rate for authenticity verification effectively distinguishes between effective micro-defects and false defects, ensuring the accuracy of defect identification.

[0034] In a specific implementation, as one example, to achieve accurate identification of micron-level defects on the surface of oxygen-free copper blanks, a threshold needs to be set based on the statistical law of polarization attenuation rate of typical micro-defects on the surface of oxygen-free copper blanks, taking into account the differences in attenuation rate and polarization response characteristics of different types of micro-defects. This embodiment achieves accurate differentiation between genuine and fake candidate defect regions through typical micro-defect sample testing and threshold setting. The specific implementation steps are as follows: Step S610: Prepare typical micro-defect samples and obtain local polarization attenuation rate data of the samples; select a billet with the same material and production process as the oxygen-free copper flat billet to be tested as the sample substrate, and artificially prepare four types of typical micro-defects on the surface of the sample substrate: oxide scale indentation, hot cracks, bubble exposure holes, and inclusion precipitation marks. Prepare multiple samples of different sizes and severity for each type of micro-defect. The number of samples is determined according to the statistical reliability requirements to ensure that the statistical results are representative; for each typical micro-defect sample, according to the process of steps S100 to S500, obtain the polarization state image and surface polarization image of the sample surface, determine the candidate defect area and extract the corresponding local polarization attenuation rate, and form a sample dataset of local polarization attenuation rate of typical micro-defects. The dataset contains the attenuation rate value, defect type, defect size and severity of each micro-defect, and other related information; at the same time, prepare blank samples without defects, and extract their local polarization attenuation rate according to the same process as reference data for pseudo-defects. The number of blank samples prepared is consistent with that of typical micro-defect samples to ensure the reliability of the reference data. Step S620: Perform statistical analysis on the obtained sample dataset of local polarization attenuation rate of typical micro-defects. Use methods such as statistical histograms and distribution curve fitting to obtain the attenuation rate distribution range, peak value and dispersion of typical micro-defects, and clarify the distribution characteristics of attenuation rate of each micro-defect. Step S630: Based on the statistical results and combined with the attenuation rate reference data of the blank sample, set a preset attenuation rate threshold. The threshold setting follows the principle of covering effective micro-defects and excluding false defects. That is, the threshold must be lower than the minimum value of the attenuation rate distribution of each typical micro-defect and higher than the maximum value of the attenuation rate of the blank sample, so as to ensure that the attenuation rate of all effective micro-defects is higher than the threshold and the attenuation rate of all false defects is lower than the threshold. Step S640: For each candidate defect region, compare its corresponding local polarization attenuation rate with a preset attenuation rate threshold; if the attenuation rate is higher than the preset threshold, the candidate defect region is determined to be a real micro-defect and retained; if the attenuation rate is lower than the preset threshold, the candidate defect region is determined to be a pseudo-defect and filtered out; the candidate defect regions that pass the verification are taken as the finally identified micro-defects, and the association information of each micro-defect is output.

[0035] In this embodiment, the preset attenuation rate threshold is determined based on the statistical distribution of polarization attenuation rate of typical micro-defects from samples of the same process and material. This allows for adaptation to surface characteristic fluctuations in different production batches of oxygen-free copper flat blanks, ensuring the applicability and stability of the threshold and avoiding incomplete filtering of false defects or mis-filtering of real micro-defects due to a fixed threshold. The verification of authenticity is performed using the local polarization attenuation rate, directly targeting the essential characteristic of real micro-defects that disrupt surface specular reflection and cause significant attenuation of polarization. This effectively filters out false defects caused by minor differences in surface roughness, light intensity fluctuations, and image noise, ensuring that all identified micro-defects are substantial defects. This provides a reliable quality basis for subsequent wire drawing and painting processes. The verification process retains the local polarization attenuation rate data for each real micro-defect. Since the polarization attenuation rate distribution of typical micro-defects of the same type differs, this data can be used to further distinguish micro-defect types, providing data support for defect cause analysis and assisting in the optimization of the production process.

[0036] Although this application has been described in conjunction with specific features and embodiments, it is obvious that various modifications and combinations can be made thereto without departing from the spirit and scope of this application. Accordingly, this specification and drawings are merely exemplary illustrations of the application as defined herein, and are to be considered as covering any and all modifications, variations, combinations, or equivalents within the scope of this application. Clearly, those skilled in the art can make various alterations and modifications to this application without departing from its scope. Thus, if such modifications and modifications fall within the scope of this application and its equivalents, this application intends to include such modifications and modifications.

Claims

1. A method for detecting defects in continuously cast and rolled billets based on optical profile measurement, characterized in that, include: The surface of the metal billet to be tested is illuminated by multi-angle polarization state encoding supplementary light, wherein the multi-angle polarization state encoding supplementary light includes at least three light sources with different polarization directions. Acquire polarization state images of the surface of the metal billet under test in at least three different polarization directions under the same field of view; Based on the obtained polarization state images of at least three different polarization directions, the surface polarization degree image is calculated, and the specular reflection component and diffuse reflection component are separated according to the polarization decoupled reflection model to obtain a de-specularized surface diffuse reflection enhancement image. Multi-scale geometric features are extracted from the surface diffuse reflection enhancement image, and candidate defect regions are determined based on the multi-scale geometric features; For the candidate defect region, the local polarization attenuation rate is extracted from the surface polarization image; The candidate defect regions are verified based on the local polarization attenuation rate. False defect regions with polarization attenuation rates lower than a preset attenuation rate threshold are filtered out, and the candidate defect regions that pass the verification are identified as micro-defects.

2. The method for detecting defects in continuously cast and rolled billets based on optical profile measurement according to claim 1, characterized in that, The polarization directions of the three light sources, which have different polarization directions, are set to 0°, 45°, and 90°, respectively.

3. The method for detecting defects in continuously cast and rolled billets based on optical profile measurement according to claim 2, characterized in that, The surface of the metal billet under test is illuminated by multi-angle polarization state encoding supplementary light, which is achieved through a coaxial ring light source; The coaxial ring light source is provided with a first ring polarization light source, a second ring polarization light source and a third ring polarization light source arranged sequentially from the inside to the outside, corresponding to the polarization directions of 0°, 45° and 90° respectively.

4. The method for detecting defects in continuously cast and rolled billets based on optical profile measurement according to claim 3, characterized in that, The incident angles of the first, second, and third annular polarized light sources are respectively set to correspond to the time-varying Brewster angle range of the metal billet under test within the continuous casting temperature range, so as to differentially suppress the specular reflection light component in each polarization channel.

5. The method for detecting defects in continuously cast and rolled billets based on optical profile measurement according to claim 1, characterized in that, The calculation of the surface polarization degree image based on the acquired polarization state images of at least three different polarization directions includes: Based on the polarization state images of at least three different polarization directions, a cosine function of polarization intensity as a function of polarization angle is fitted, the polarization degree parameter is solved, and the surface polarization degree image is generated.

6. The method for detecting defects in continuously cast and rolled billets based on optical profile measurement according to claim 1, characterized in that, The multi-scale geometric features include micro-defect curvature direction field features and / or texture anomaly response features.

7. The method for detecting defects in continuously cast and rolled billets based on optical profile measurement according to claim 6, characterized in that, The extraction of multi-scale geometric features from the surface diffuse reflection enhancement image includes: A Hessian matrix is ​​constructed from the surface diffuse reflection enhancement image, and the eigenvector direction corresponding to the largest eigenvalue of the Hessian matrix is ​​calculated as the field feature of the curvature direction of the micro-defect. And / or, The surface diffuse reflection enhancement image is convolutionally filtered using a Gabor filter bank with multiple directional parameters and multiple center frequency parameters, and the texture response amplitude of each filter channel is extracted as a texture anomaly response feature.

8. The method for detecting defects in continuously cast and rolled billets based on optical profile measurement according to claim 1, characterized in that, The local polarization attenuation rate is the rate at which the polarization value at the edge of the candidate defect region decreases relative to the polarization value in the background region.

9. The method for detecting defects in continuously cast and rolled billets based on optical profile measurement according to claim 1, characterized in that, The preset attenuation rate threshold is determined based on the statistical distribution of the polarization degree attenuation rate of typical micro-defects on the surface of the metal billet to be tested.

10. The method for detecting defects in continuously cast and rolled billets based on optical profile measurement according to claim 9, characterized in that, The typical micro-defects include at least one of the following: oxide scale indentation, hot cracks, bubble exposure pores, and inclusion precipitation marks.

Images